static HYD_status exec_args_fn(char *arg, char ***argv)
{
long int i, count;
struct HYD_exec *exec = NULL;
HYD_status status = HYD_SUCCESS;
if (**argv == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "NULL argument to exec args\n");
for (exec = HYD_pmcd_pmip.exec_list; exec->next; exec = exec->next);
count = strtol(**argv, NULL, 10);
if (errno == ERANGE || errno == EINVAL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "Exec arg not convertible to integer\n");
for (i = 0; i < count; i++) {
(*argv)++;
exec->exec[i] = MPL_strdup(**argv);
}
exec->exec[i] = NULL;
(*argv)++;
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_ftb_init(void)
{
int ret;
FTB_client_t ci;
HYD_status status = HYD_SUCCESS;
MPL_strncpy(ci.event_space, "ftb.mpi.hydra", sizeof(ci.event_space));
MPL_strncpy(ci.client_name, "hydra " HYDRA_VERSION, sizeof(ci.client_name));
MPL_strncpy(ci.client_subscription_style, "FTB_SUBSCRIPTION_NONE",
sizeof(ci.client_subscription_style));
ci.client_polling_queue_len = -1;
ret = FTB_Connect(&ci, &ch);
if (ret)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "ftb connect\n");
ret = FTB_Declare_publishable_events(ch, NULL, event_info,
sizeof(event_info) / sizeof(event_info[0]));
if (ret)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "ftb declare publishable\n");
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status match_arg(char ***argv_p, struct HYD_arg_match_table *match_table)
{
struct HYD_arg_match_table *m;
char *arg, *val;
HYD_status status = HYD_SUCCESS;
arg = **argv_p;
while (*arg == '-') /* Remove leading dashes */
arg++;
/* If arg is of the form foo=bar, we separate it out as two
* arguments */
for (val = arg; *val && *val != '='; val++);
if (*val == '=') {
/* Found an '='; use the rest of the argument as a separate
* argument */
**argv_p = val + 1;
}
else {
/* Move to the next argument */
(*argv_p)++;
}
*val = 0; /* close out key */
m = match_table;
while (m->handler_fn) {
if (!strcasecmp(arg, m->arg)) {
if (**argv_p && (!strcmp(**argv_p, "-h") || !strcmp(**argv_p, "-help") ||
!strcmp(**argv_p, "--help"))) {
if (m->help_fn == NULL) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "No help message available\n");
}
else {
m->help_fn();
HYDU_ERR_SETANDJUMP(status, HYD_GRACEFUL_ABORT, "%s", "");
}
}
status = m->handler_fn(arg, argv_p);
HYDU_ERR_POP(status, "match handler returned error\n");
break;
}
m++;
}
if (m->handler_fn == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "unrecognized argument %s\n", arg);
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_parse_hostfile(const char *hostfile, struct HYD_node **node_list,
HYD_status(*process_token) (char *token, int newline,
struct HYD_node ** node_list))
{
char line[HYD_TMP_STRLEN], **tokens;
FILE *fp;
int i;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if ((fp = fopen(hostfile, "r")) == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "unable to open host file: %s\n", hostfile);
if (node_list)
*node_list = NULL;
while (fgets(line, HYD_TMP_STRLEN, fp)) {
char *linep = NULL;
linep = line;
strtok(linep, "#");
while (isspace(*linep))
linep++;
/* Ignore blank lines & comments */
if ((*linep == '#') || (*linep == '\0'))
continue;
tokens = HYDU_str_to_strlist(linep);
if (!tokens)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"Unable to convert host file entry to strlist\n");
for (i = 0; tokens[i]; i++) {
status = process_token(tokens[i], !i, node_list);
HYDU_ERR_POP(status, "unable to process token\n");
}
HYDU_free_strlist(tokens);
MPL_free(tokens);
}
fclose(fp);
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_process_mfile_token(char *token, int newline, struct HYD_node **node_list)
{
int num_procs;
char *hostname, *procs, *binding, *tmp, *user, *saveptr;
struct HYD_node *node;
HYD_status status = HYD_SUCCESS;
if (newline) { /* The first entry gives the hostname and processes */
hostname = strtok_r(token, ":", &saveptr);
procs = strtok_r(NULL, ":", &saveptr);
num_procs = procs ? atoi(procs) : 1;
status = HYDU_add_to_node_list(hostname, num_procs, node_list);
HYDU_ERR_POP(status, "unable to add to node list\n");
}
else { /* Not a new line */
tmp = strtok_r(token, "=", &saveptr);
if (!strcmp(tmp, "binding")) {
binding = strtok_r(NULL, "=", &saveptr);
for (node = *node_list; node->next; node = node->next);
if (node->local_binding)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"duplicate local binding setting\n");
node->local_binding = MPL_strdup(binding);
}
else if (!strcmp(tmp, "user")) {
user = strtok_r(NULL, "=", &saveptr);
for (node = *node_list; node->next; node = node->next);
if (node->user)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "duplicate username setting\n");
node->user = MPL_strdup(user);
}
else {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"token %s not supported at this time\n", token);
}
}
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_sock_get_iface_ip(char *iface, char **ip)
{
HYD_status status = HYD_SUCCESS;
#if defined(HAVE_GETIFADDRS)
struct ifaddrs *ifaddr, *ifa;
char buf[MAX_HOSTNAME_LEN];
struct sockaddr_in *sa;
/* Got the interface name; let's query for the IP address */
if (getifaddrs(&ifaddr) == -1)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "getifaddrs failed\n");
for (ifa = ifaddr; ifa; ifa = ifa->ifa_next)
if (!strcmp(ifa->ifa_name, iface) && (ifa->ifa_addr) &&
(ifa->ifa_addr->sa_family == AF_INET))
break;
if (!ifa)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "unable to find interface %s\n", iface);
sa = (struct sockaddr_in *) ifa->ifa_addr;
#if defined HAVE_INET_NTOP
(*ip) = MPL_strdup((char *)
inet_ntop(AF_INET, (const void *) &(sa->sin_addr), buf, MAX_HOSTNAME_LEN));
#else
(*ip) = NULL;
#endif /* HAVE_INET_NTOP */
if (!*ip)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"unable to find IP for interface %s\n", iface);
freeifaddrs(ifaddr);
#else
/* For now just disable interface selection when getifaddrs isn't
* available, such as on AIX. In the future we can implement in MPL
* something along the lines of MPIDI_GetIPInterface from tcp_getip.c in
* nemesis. */
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"interface selection not supported on this platform\n");
#endif
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDTI_bscd_ll_query_node_count(int *count)
{
char *hostfile;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (MPL_env2str("LOADL_HOSTFILE", (const char **) &hostfile) == 0)
hostfile = NULL;
if (hostfile == NULL) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "No LL nodefile found\n");
}
else {
total_node_count = 0;
status = HYDU_parse_hostfile(hostfile, NULL, process_mfile_count);
HYDU_ERR_POP(status, "error parsing hostfile\n");
*count = total_node_count;
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_topo_bind(int idx)
{
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (idx < 0 || ignore_binding)
goto fn_exit;
#if defined HAVE_HWLOC
if (!strcmp(HYDT_topo_info.topolib, "hwloc")) {
status = HYDT_topo_hwloc_bind(idx);
HYDU_ERR_POP(status, "HWLOC failure binding process to core\n");
goto fn_exit;
}
#endif /* HAVE_HWLOC */
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "no topology library available\n");
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status exec_local_env_fn(char *arg, char ***argv)
{
struct HYD_exec *exec = NULL;
int i, count;
char *str;
HYD_status status = HYD_SUCCESS;
if (**argv == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "NULL argument to exec local env\n");
for (exec = HYD_pmcd_pmip.exec_list; exec->next; exec = exec->next);
count = atoi(**argv);
for (i = 0; i < count; i++) {
(*argv)++;
str = **argv;
/* Environment variables are quoted; remove them */
if (*str == '\'') {
str++;
str[strlen(str) - 1] = 0;
}
HYDU_append_env_str_to_list(str, &exec->user_env);
}
(*argv)++;
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status exec_args_fn(char *arg, char ***argv)
{
int i, count;
struct HYD_exec *exec = NULL;
HYD_status status = HYD_SUCCESS;
if (**argv == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "NULL argument to exec args\n");
for (exec = HYD_pmcd_pmip.exec_list; exec->next; exec = exec->next);
count = atoi(**argv);
for (i = 0; i < count; i++) {
(*argv)++;
exec->exec[i] = MPL_strdup(**argv);
}
exec->exec[i] = NULL;
(*argv)++;
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status process_mfile_token(char *token, int newline, struct HYD_node **node_list)
{
int num_procs;
static int entry_count = 0;
static char *hostname;
HYD_status status = HYD_SUCCESS;
entry_count++;
if (newline) { /* The first entry gives the hostname */
entry_count = 1;
if (hostname)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "unexpected token %s\n", token);
hostname = HYDU_strdup(token);
}
else { /* Not a new line */
if (entry_count != 2)
goto fn_exit;
num_procs = atoi(token);
status = HYDU_add_to_node_list(hostname, num_procs, node_list);
HYDU_ERR_POP(status, "unable to initialize proxy\n");
hostname = NULL;
}
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscu_stdio_cb(int fd, HYD_event_t events, void *userp)
{
int stdfd, closed, i;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
stdfd = (int) (size_t) userp;
status = HYDU_sock_forward_stdio(fd, stdfd, &closed);
HYDU_ERR_POP(status, "stdio forwarding error\n");
if (closed || (events & HYD_POLLHUP)) {
/* connection has closed */
status = HYDT_dmx_deregister_fd(fd);
HYDU_ERR_SETANDJUMP(status, status, "error deregistering fd %d\n", fd);
for (i = 0; i < HYD_bscu_fd_count; i++) {
if (HYD_bscu_fd_list[i] == fd) {
HYD_bscu_fd_list[i] = HYD_FD_CLOSED;
break;
}
}
close(fd);
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_strsplit(char *str, char **str1, char **str2, char sep)
{
int i;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (str == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "");
*str1 = HYDU_strdup(str);
for (i = 0; (*str1)[i] && ((*str1)[i] != sep); i++);
if ((*str1)[i] == 0) /* End of the string */
*str2 = NULL;
else {
*str2 = HYDU_strdup(&((*str1)[i + 1]));
(*str1)[i] = 0;
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_sge_query_node_list(struct HYD_node **node_list)
{
char *hostfile;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (MPL_env2str("PE_HOSTFILE", (const char **) &hostfile) == 0)
hostfile = NULL;
if (hostfile == NULL) {
*node_list = NULL;
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "No SGE nodefile found\n");
}
else {
status = HYDU_parse_hostfile(hostfile, node_list, process_mfile_token);
HYDU_ERR_POP(status, "error parsing hostfile\n");
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_topo_hwloc_init(const char *binding, const char *mapping, const char *membind)
{
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
HYDU_ASSERT(binding, status);
hwloc_topology_init(&topology);
hwloc_topology_load(topology);
HYDT_topo_hwloc_info.total_num_pus = hwloc_get_nbobjs_by_type(topology, HWLOC_OBJ_PU);
hwloc_initialized = 1;
/* bindings that don't require mapping */
if (!strncmp(binding, "user:"******"user:"******"user:"******"error binding to %s\n", binding);
goto fn_exit;
}
else if (!strcmp(binding, "rr")) {
status = handle_rr_binding();
HYDU_ERR_POP(status, "error binding to %s\n", binding);
goto fn_exit;
}
status = handle_bitmap_binding(binding, mapping ? mapping : binding);
HYDU_ERR_POP(status, "error binding with bind \"%s\" and map \"%s\"\n", binding, mapping);
/* Memory binding options */
if (membind == NULL)
HYDT_topo_hwloc_info.membind = HWLOC_MEMBIND_DEFAULT;
else if (!strcmp(membind, "firsttouch"))
HYDT_topo_hwloc_info.membind = HWLOC_MEMBIND_FIRSTTOUCH;
else if (!strcmp(membind, "nexttouch"))
HYDT_topo_hwloc_info.membind = HWLOC_MEMBIND_NEXTTOUCH;
else if (!strncmp(membind, "bind:", strlen("bind:"))) {
HYDT_topo_hwloc_info.membind = HWLOC_MEMBIND_BIND;
}
else if (!strncmp(membind, "interleave:", strlen("interleave:"))) {
HYDT_topo_hwloc_info.membind = HWLOC_MEMBIND_INTERLEAVE;
}
else if (!strncmp(membind, "replicate:", strlen("replicate:"))) {
HYDT_topo_hwloc_info.membind = HWLOC_MEMBIND_REPLICATE;
}
else {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"unrecognized membind policy \"%s\"\n", membind);
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
static HYD_status handle_pmi_cmd(int fd, int pgid, int pid, char *buf, int pmi_version)
{
char **args = NULL, *cmd = NULL;
struct HYD_pmcd_pmi_handle *h;
int i;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (pmi_version == 1)
HYD_pmcd_pmi_handle = HYD_pmcd_pmi_v1;
else
HYD_pmcd_pmi_handle = HYD_pmcd_pmi_v2;
if (HYD_server_info.user_global.debug)
HYDU_dump(stdout, "[pgid: %d] got PMI command: %s\n", pgid, buf);
HYDU_MALLOC(args, char **, MAX_PMI_ARGS * sizeof(char *), status);
for (i = 0; i < MAX_PMI_ARGS; i++)
args[i] = NULL;
status = HYD_pmcd_pmi_parse_pmi_cmd(buf, pmi_version, &cmd, args);
HYDU_ERR_POP(status, "unable to parse PMI command\n");
#if defined ENABLE_PROFILING
if (HYD_server_info.enable_profiling)
HYD_server_info.num_pmi_calls++;
#endif /* ENABLE_PROFILING */
h = HYD_pmcd_pmi_handle;
while (h->handler) {
if (!strcmp(cmd, h->cmd)) {
status = h->handler(fd, pid, pgid, args);
HYDU_ERR_POP(status, "PMI handler returned error\n");
break;
}
h++;
}
if (!h->handler) {
/* We don't understand the command */
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"Unrecognized PMI command: %s | cleaning up processes\n", cmd);
}
fn_exit:
if (cmd)
HYDU_FREE(cmd);
if (args) {
HYDU_free_strlist(args);
HYDU_free(args);
}
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_sock_accept(int listen_fd, int *fd)
{
int one = 1;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
*fd = accept(listen_fd, 0, 0);
if (*fd < 0)
HYDU_ERR_SETANDJUMP(status, HYD_SOCK_ERROR, "accept error (%s)\n", MPL_strerror(errno));
/* Disable nagle */
if (setsockopt(*fd, IPPROTO_TCP, TCP_NODELAY, &one, sizeof(int)) < 0)
HYDU_ERR_SETANDJUMP(status, HYD_SOCK_ERROR, "cannot set TCP_NODELAY\n");
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status
HYDU_sock_create_and_listen_portstr(char *iface, char *hostname, char *port_range,
char **port_str,
HYD_status(*callback) (int fd, HYD_event_t events,
void *userp), void *userp)
{
int listenfd;
char *sport, *real_port_range, *ip = NULL;
uint16_t port;
HYD_status status = HYD_SUCCESS;
/* Listen on a port in the port range */
port = 0;
real_port_range = port_range ? MPL_strdup(port_range) : NULL;
status = HYDU_sock_listen(&listenfd, real_port_range, &port);
HYDU_ERR_POP(status, "unable to listen on port\n");
/* Register the listening socket with the demux engine */
status = HYDT_dmx_register_fd(1, &listenfd, HYD_POLLIN, userp, callback);
HYDU_ERR_POP(status, "unable to register fd\n");
/* Create a port string for MPI processes to use to connect to */
if (iface) {
status = HYDU_sock_get_iface_ip(iface, &ip);
HYDU_ERR_POP(status, "unable to get network interface IP\n");
}
else if (hostname) {
ip = MPL_strdup(hostname);
}
else {
char localhost[MAX_HOSTNAME_LEN] = { 0 };
if (gethostname(localhost, MAX_HOSTNAME_LEN) < 0)
HYDU_ERR_SETANDJUMP(status, HYD_SOCK_ERROR, "unable to get local hostname\n");
ip = MPL_strdup(localhost);
}
sport = HYDU_int_to_str(port);
HYDU_MALLOC_OR_JUMP(*port_str, char *, strlen(ip) + 1 + strlen(sport) + 1, status);
MPL_snprintf(*port_str, strlen(ip) + 1 + strlen(sport) + 1, "%s:%s", ip, sport);
MPL_free(sport);
fn_exit:
if (ip)
MPL_free(ip);
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_sock_read(int fd, void *buf, int maxlen, int *recvd, int *closed,
enum HYDU_sock_comm_flag flag)
{
int tmp;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
HYDU_ASSERT(maxlen, status);
*recvd = 0;
*closed = 0;
while (1) {
do {
tmp = read(fd, (char *) buf + *recvd, maxlen - *recvd);
if (tmp < 0) {
if (errno == ECONNRESET || fd == STDIN_FILENO) {
/* If the remote end closed the socket or if we
* get an EINTR on stdin, set the socket to be
* closed and jump out */
*closed = 1;
status = HYD_SUCCESS;
goto fn_exit;
}
}
} while (tmp < 0 && errno == EINTR);
if (tmp < 0) {
HYDU_ERR_SETANDJUMP(status, HYD_SOCK_ERROR, "read error (%s)\n", MPL_strerror(errno));
}
else if (tmp == 0) {
*closed = 1;
goto fn_exit;
}
else {
*recvd += tmp;
}
if (flag == HYDU_SOCK_COMM_NONE || *recvd == maxlen)
break;
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_lsf_query_node_list(struct HYD_node **node_list)
{
char *hosts, *hostname, *num_procs_str, *thosts = NULL;
int num_procs;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (MPL_env2str("LSB_MCPU_HOSTS", (const char **) &hosts) == 0)
hosts = NULL;
if (hosts == NULL) {
*node_list = NULL;
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "No LSF node list found\n");
}
else {
hosts = HYDU_strdup(hosts);
thosts = hosts;
hostname = strtok(hosts, " ");
while (1) {
if (hostname == NULL)
break;
/* the even fields in the list should be the number of
* cores */
num_procs_str = strtok(NULL, " ");
HYDU_ASSERT(num_procs_str, status);
num_procs = atoi(num_procs_str);
status = HYDU_add_to_node_list(hostname, num_procs, node_list);
HYDU_ERR_POP(status, "unable to add to node list\n");
hostname = strtok(NULL, " ");
}
if (thosts)
HYDU_FREE(thosts);
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_ftb_finalize(void)
{
int ret;
HYD_status status = HYD_SUCCESS;
ret = FTB_Disconnect(ch);
if (ret)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "ftb disconnect\n");
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_set_str(char *arg, char **var, const char *val)
{
HYD_status status = HYD_SUCCESS;
HYDU_ERR_CHKANDJUMP(status, *var, HYD_INTERNAL_ERROR, "duplicate setting: %s\n", arg);
if (val == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "cannot assign NULL object\n");
*var = MPL_strdup(val);
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
static HYD_status version_fn(char *arg, char ***argv)
{
HYD_status status = HYD_SUCCESS;
if (strcmp(**argv, HYDRA_VERSION)) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"UI version string does not match proxy version\n");
}
(*argv)++;
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_ll_query_proxy_id(int *proxy_id)
{
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if (MPL_env2int("MP_CHILD", proxy_id) == 0) {
*proxy_id = -1;
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "cannot find ll proxy ID\n");
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_sock_write(int fd, const void *buf, int maxlen, int *sent, int *closed,
enum HYDU_sock_comm_flag flag)
{
int tmp;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
HYDU_ASSERT(maxlen, status);
*sent = 0;
*closed = 0;
while (1) {
tmp = write(fd, (char *) buf + *sent, maxlen - *sent);
if (tmp <= 0) {
if (errno == EAGAIN) {
if (flag == HYDU_SOCK_COMM_NONE)
goto fn_exit;
else
continue;
}
else if (errno == ECONNRESET) {
*closed = 1;
goto fn_exit;
}
HYDU_ERR_SETANDJUMP(status, HYD_SOCK_ERROR, "write error (%s)\n", MPL_strerror(errno));
}
else {
*sent += tmp;
}
if (flag == HYDU_SOCK_COMM_NONE || *sent == maxlen)
break;
}
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_sock_is_local(char *host, int *is_local)
{
char lhost[MAX_HOSTNAME_LEN];
HYD_status status = HYD_SUCCESS;
*is_local = 0;
if (gethostname(lhost, MAX_HOSTNAME_LEN) < 0) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "gethostname returned an error\n");
}
else if (!strcmp(lhost, host)) {
*is_local = 1;
}
fn_exit:
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_ftb_publish(const char *event_name, const char *event_payload)
{
FTB_event_properties_t event_prop;
FTB_event_handle_t event_handle;
int ret;
HYD_status status = HYD_SUCCESS;
event_prop.event_type = 1;
MPL_strncpy(event_prop.event_payload, event_payload, sizeof(event_prop.event_payload));
ret = FTB_Publish(ch, event_name, &event_prop, &event_handle);
if (ret)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR, "ftb publish\n");
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
HYD_status HYDU_sock_set_nonblock(int fd)
{
int flags;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
if ((flags = fcntl(fd, F_GETFL, 0)) == -1)
flags = 0;
#if defined O_NONBLOCK
if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) < 0)
HYDU_ERR_SETANDJUMP(status, HYD_SOCK_ERROR, "fcntl failed on %d\n", fd);
#endif /* O_NONBLOCK */
fn_exit:
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}
char *HYDU_getcwd(void)
{
char *cwdval, *retval = NULL;
const char *pwdval;
HYD_status status = HYD_SUCCESS;
#if defined HAVE_STAT
struct stat spwd, scwd;
#endif /* HAVE_STAT */
if (MPL_env2str("PWD", &pwdval) == 0)
pwdval = NULL;
HYDU_MALLOC_OR_JUMP(cwdval, char *, HYDRA_MAX_PATH, status);
if (getcwd(cwdval, HYDRA_MAX_PATH) == NULL)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"allocated space is too small for absolute path\n");
#if defined HAVE_STAT
if (pwdval && stat(pwdval, &spwd) != -1 && stat(cwdval, &scwd) != -1 &&
spwd.st_dev == scwd.st_dev && spwd.st_ino == scwd.st_ino) {
/* PWD and getcwd() match; use the PWD value */
retval = MPL_strdup(pwdval);
MPL_free(cwdval);
}
else
#endif /* HAVE_STAT */
{
/* PWD and getcwd() don't match; use the getcwd value and hope
* for the best. */
retval = cwdval;
}
fn_exit:
return retval;
fn_fail:
goto fn_exit;
}
HYD_status HYDT_bscd_ll_launch_procs(char **args, struct HYD_proxy *proxy_list,
int *control_fd)
{
int idx, i, total_procs, node_count;
int *pid, *fd_list, exec_idx;
char *targs[HYD_NUM_TMP_STRINGS], *node_list_str = NULL;
char *path = NULL, *extra_arg_list = NULL, *extra_arg, quoted_exec_string[HYD_TMP_STRLEN];
struct HYD_proxy *proxy;
struct HYDT_topo_cpuset_t cpuset;
HYD_status status = HYD_SUCCESS;
HYDU_FUNC_ENTER();
/* We use the following priority order for the executable path:
* (1) user-specified; (2) search in path; (3) Hard-coded
* location */
if (HYDT_bsci_info.launcher_exec)
path = HYDU_strdup(HYDT_bsci_info.launcher_exec);
if (!path)
path = HYDU_find_full_path("poe");
if (!path)
path = HYDU_strdup("/usr/bin/poe");
idx = 0;
targs[idx++] = HYDU_strdup(path);
if (!strcmp(HYDT_bsci_info.rmk, "ll")) {
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"ll does not support user-defined host lists\n");
}
/* Check how many nodes are being passed for the launch */
status = HYDTI_bscd_ll_query_node_count(&total_procs);
HYDU_ERR_POP(status, "unable to query for the node count\n");
node_count = 0;
for (proxy = proxy_list; proxy; proxy = proxy->next)
node_count++;
if (total_procs != node_count)
HYDU_ERR_SETANDJUMP(status, HYD_INTERNAL_ERROR,
"processes to be launched have to cover all nodes\n");
MPL_env2str("HYDRA_LAUNCHER_EXTRA_ARGS", (const char **) &extra_arg_list);
if (extra_arg_list) {
extra_arg = strtok(extra_arg_list, " ");
while (extra_arg) {
targs[idx++] = HYDU_strdup(extra_arg);
extra_arg = strtok(NULL, " ");
}
}
/* Fill in the remaining arguments */
exec_idx = idx;
for (i = 0; args[i]; i++)
targs[idx++] = HYDU_strdup(args[i]);
/* Create a quoted version of the exec string, which is only used
* when the executable is not launched directly, but through an
* actual launcher */
HYDU_snprintf(quoted_exec_string, HYD_TMP_STRLEN, "\"%s\"", targs[exec_idx]);
HYDU_FREE(targs[exec_idx]);
targs[exec_idx] = quoted_exec_string;
/* Increase pid list to accommodate the new pid */
HYDU_MALLOC(pid, int *, (HYD_bscu_pid_count + 1) * sizeof(int), status);
for (i = 0; i < HYD_bscu_pid_count; i++)
pid[i] = HYD_bscu_pid_list[i];
HYDU_FREE(HYD_bscu_pid_list);
HYD_bscu_pid_list = pid;
/* Increase fd list to accommodate these new fds */
HYDU_MALLOC(fd_list, int *, (HYD_bscu_fd_count + 3) * sizeof(int), status);
for (i = 0; i < HYD_bscu_fd_count; i++)
fd_list[i] = HYD_bscu_fd_list[i];
HYDU_FREE(HYD_bscu_fd_list);
HYD_bscu_fd_list = fd_list;
/* append proxy ID as -1 */
targs[idx++] = HYDU_int_to_str(-1);
targs[idx++] = NULL;
HYDT_topo_cpuset_zero(&cpuset);
status = HYDU_create_process(targs, NULL, NULL, &fd_stdout, &fd_stderr,
&HYD_bscu_pid_list[HYD_bscu_pid_count++], cpuset);
HYDU_ERR_POP(status, "create process returned error\n");
HYD_bscu_fd_list[HYD_bscu_fd_count++] = fd_stdout;
HYD_bscu_fd_list[HYD_bscu_fd_count++] = fd_stderr;
status = HYDT_dmx_register_fd(1, &fd_stdout, HYD_POLLIN,
(void *) (size_t) STDOUT_FILENO, HYDT_bscu_stdio_cb);
HYDU_ERR_POP(status, "demux returned error registering fd\n");
status = HYDT_dmx_register_fd(1, &fd_stderr, HYD_POLLIN,
(void *) (size_t) STDERR_FILENO, HYDT_bscu_stdio_cb);
HYDU_ERR_POP(status, "demux returned error registering fd\n");
fn_exit:
if (node_list_str)
HYDU_FREE(node_list_str);
HYDU_free_strlist(targs);
if (path)
HYDU_FREE(path);
HYDU_FUNC_EXIT();
return status;
fn_fail:
goto fn_exit;
}