static void open_local_file(orte_dfs_request_t *dfs)
{
char *filename;
orte_dfs_tracker_t *trk;
/* extract the filename from the uri */
if (NULL == (filename = opal_filename_from_uri(dfs->uri, NULL))) {
/* something wrong - error was reported, so just get out */
if (NULL != dfs->open_cbfunc) {
dfs->open_cbfunc(-1, dfs->cbdata);
}
OBJ_RELEASE(dfs);
return;
}
opal_output_verbose(1, orte_dfs_base.output,
"%s opening local file %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename);
/* attempt to open the file */
if (0 > (dfs->remote_fd = open(filename, O_RDONLY))) {
ORTE_ERROR_LOG(ORTE_ERR_FILE_OPEN_FAILURE);
if (NULL != dfs->open_cbfunc) {
dfs->open_cbfunc(dfs->remote_fd, dfs->cbdata);
}
return;
}
/* otherwise, create a tracker for this file */
trk = OBJ_NEW(orte_dfs_tracker_t);
trk->requestor.jobid = ORTE_PROC_MY_NAME->jobid;
trk->requestor.vpid = ORTE_PROC_MY_NAME->vpid;
trk->filename = strdup(filename);
/* define the local fd */
trk->local_fd = local_fd++;
/* record the remote file descriptor */
trk->remote_fd = dfs->remote_fd;
/* add it to our list of active files */
opal_list_append(&active_files, &trk->super);
/* the file is locally hosted */
trk->host_daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
trk->host_daemon.vpid = ORTE_PROC_MY_DAEMON->vpid;
opal_output_verbose(1, orte_dfs_base.output,
"%s local file %s mapped localfd %d to remotefd %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename, trk->local_fd, trk->remote_fd);
/* let the caller know */
if (NULL != dfs->open_cbfunc) {
dfs->open_cbfunc(trk->local_fd, dfs->cbdata);
}
/* request will be released by the calling routing */
}
static void process_opens(int fd, short args, void *cbdata)
{
orte_dfs_request_t *dfs = (orte_dfs_request_t*)cbdata;
int rc;
opal_buffer_t *buffer;
char *scheme, *host, *filename, *hostname;
orte_process_name_t daemon;
bool found;
orte_vpid_t v;
/* get the scheme to determine if we can process locally or not */
if (NULL == (scheme = opal_uri_get_scheme(dfs->uri))) {
OBJ_RELEASE(dfs);
return;
}
if (0 == strcmp(scheme, "nfs")) {
open_local_file(dfs);
goto complete;
}
if (0 != strcmp(scheme, "file")) {
/* not yet supported */
orte_show_help("orte_dfs_help.txt", "unsupported-filesystem",
true, dfs->uri);
if (NULL != dfs->open_cbfunc) {
dfs->open_cbfunc(-1, dfs->cbdata);
}
goto complete;
}
/* dissect the uri to extract host and filename/path */
if (NULL == (filename = opal_filename_from_uri(dfs->uri, &host))) {
goto complete;
}
/* if the host is our own, then treat it as a local file */
if (NULL == host ||
0 == strcmp(host, orte_process_info.nodename) ||
0 == strcmp(host, "localhost") ||
opal_ifislocal(host)) {
opal_output_verbose(1, orte_dfs_base.output,
"%s file %s on local host",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename);
open_local_file(dfs);
goto complete;
}
/* ident the daemon on that host */
daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
found = false;
for (v=0; v < orte_process_info.num_daemons; v++) {
daemon.vpid = v;
/* fetch the hostname where this daemon is located */
if (ORTE_SUCCESS != (rc = orte_db.fetch_pointer(&daemon, ORTE_DB_HOSTNAME, (void**)&hostname, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
goto complete;
}
if (0 == strcmp(host, hostname)) {
found = true;
break;
}
}
if (!found) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto complete;
}
opal_output_verbose(1, orte_dfs_base.output,
"%s file %s on host %s daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename, host, ORTE_NAME_PRINT(&daemon));
/* double-check: if it is our local daemon, then we
* treat this as local
*/
if (daemon.vpid == ORTE_PROC_MY_DAEMON->vpid) {
opal_output_verbose(1, orte_dfs_base.output,
"%s local file %s on same daemon",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename);
open_local_file(dfs);
goto complete;
}
/* add this request to our local list so we can
* match it with the returned response when it comes
*/
dfs->id = req_id++;
opal_list_append(&requests, &dfs->super);
/* setup a message for the daemon telling
* them what file we want to access
*/
buffer = OBJ_NEW(opal_buffer_t);
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &dfs->cmd, 1, ORTE_DFS_CMD_T))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
/* pass the request id */
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &dfs->id, 1, OPAL_UINT64))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &filename, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
opal_output_verbose(1, orte_dfs_base.output,
"%s sending open file request to %s file %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&daemon),
filename);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(&daemon, buffer,
ORTE_RML_TAG_DFS_CMD, 0,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buffer);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
/* don't release it */
return;
complete:
OBJ_RELEASE(dfs);
}
static void process_opens(int fd, short args, void *cbdata)
{
orte_dfs_request_t *dfs = (orte_dfs_request_t*)cbdata;
int rc;
opal_buffer_t *buffer;
char *scheme, *host, *filename, *hostname;
orte_process_name_t daemon;
bool found;
orte_vpid_t v;
opal_list_t myvals;
opal_value_t *kv;
opal_output(0, "%s PROCESSING OPEN", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* get the scheme to determine if we can process locally or not */
if (NULL == (scheme = opal_uri_get_scheme(dfs->uri))) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
goto complete;
}
opal_output(0, "%s GOT SCHEME", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
if (0 != strcmp(scheme, "file")) {
/* not yet supported */
orte_show_help("orte_dfs_help.txt", "unsupported-filesystem",
true, dfs->uri);
goto complete;
}
/* dissect the uri to extract host and filename/path */
if (NULL == (filename = opal_filename_from_uri(dfs->uri, &host))) {
goto complete;
}
opal_output(0, "%s GOT FILENAME %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), filename);
if (NULL == host) {
host = strdup(orte_process_info.nodename);
}
/* ident the daemon on that host */
daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
found = false;
for (v=0; v < orte_process_info.num_daemons; v++) {
daemon.vpid = v;
/* fetch the hostname where this daemon is located */
OBJ_CONSTRUCT(&myvals, opal_list_t);
if (ORTE_SUCCESS != (rc = opal_dstore.fetch(opal_dstore_internal,
(opal_identifier_t*)&daemon,
ORTE_DB_HOSTNAME, &myvals))) {
ORTE_ERROR_LOG(rc);
OPAL_LIST_DESTRUCT(&myvals);
goto complete;
}
kv = (opal_value_t*)opal_list_get_first(&myvals);
hostname = strdup(kv->data.string);
OPAL_LIST_DESTRUCT(&myvals);
opal_output(0, "%s GOT HOST %s HOSTNAME %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), host, hostname);
if (0 == strcmp(host, hostname)) {
found = true;
break;
}
}
if (!found) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
goto complete;
}
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s file %s on host %s daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename, host, ORTE_NAME_PRINT(&daemon));
/* add this request to our local list so we can
* match it with the returned response when it comes
*/
dfs->id = req_id++;
opal_list_append(&requests, &dfs->super);
/* setup a message for the daemon telling
* them what file we want to access
*/
buffer = OBJ_NEW(opal_buffer_t);
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &dfs->cmd, 1, ORTE_DFS_CMD_T))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
/* pass the request id */
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &dfs->id, 1, OPAL_UINT64))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &filename, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s sending open file request to %s file %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&daemon),
filename);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(&daemon, buffer,
ORTE_RML_TAG_DFS_CMD,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buffer);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
/* don't release it */
return;
complete:
/* we get here if an error occurred - execute any
* pending callback so the proc doesn't hang
*/
if (NULL != dfs->open_cbfunc) {
dfs->open_cbfunc(-1, dfs->cbdata);
}
OBJ_RELEASE(dfs);
}
static void process_opens(int fd, short args, void *cbdata)
{
orte_dfs_request_t *dfs = (orte_dfs_request_t*)cbdata;
int rc;
opal_buffer_t *buffer;
char *scheme, *host=NULL, *filename=NULL;
orte_process_name_t daemon;
opal_list_t lt;
opal_namelist_t *nm;
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s PROCESSING OPEN", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
/* get the scheme to determine if we can process locally or not */
if (NULL == (scheme = opal_uri_get_scheme(dfs->uri))) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
goto complete;
}
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s GOT SCHEME", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
if (0 != strcmp(scheme, "file")) {
/* not yet supported */
orte_show_help("orte_dfs_help.txt", "unsupported-filesystem",
true, dfs->uri);
free(scheme);
goto complete;
}
free(scheme);
/* dissect the uri to extract host and filename/path */
if (NULL == (filename = opal_filename_from_uri(dfs->uri, &host))) {
goto complete;
}
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s GOT FILENAME %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), filename);
if (NULL == host) {
host = strdup(orte_process_info.nodename);
}
/* ident the daemon on that host */
daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
OBJ_CONSTRUCT(<, opal_list_t);
if (ORTE_SUCCESS != (rc = opal_pmix.resolve_peers(host, daemon.jobid, <))) {
ORTE_ERROR_LOG(rc);
OBJ_DESTRUCT(<);
goto complete;
}
nm = (opal_namelist_t*)opal_list_get_first(<);
daemon.vpid = nm->name.vpid;
OPAL_LIST_DESTRUCT(<);
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s file %s on host %s daemon %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
filename, host, ORTE_NAME_PRINT(&daemon));
/* add this request to our local list so we can
* match it with the returned response when it comes
*/
dfs->id = req_id++;
opal_list_append(&requests, &dfs->super);
/* setup a message for the daemon telling
* them what file we want to access
*/
buffer = OBJ_NEW(opal_buffer_t);
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &dfs->cmd, 1, ORTE_DFS_CMD_T))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
/* pass the request id */
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &dfs->id, 1, OPAL_UINT64))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
if (OPAL_SUCCESS != (rc = opal_dss.pack(buffer, &filename, 1, OPAL_STRING))) {
ORTE_ERROR_LOG(rc);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
opal_output_verbose(1, orte_dfs_base_framework.framework_output,
"%s sending open file request to %s file %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&daemon),
filename);
/* send it */
if (0 > (rc = orte_rml.send_buffer_nb(&daemon, buffer,
ORTE_RML_TAG_DFS_CMD,
orte_rml_send_callback, NULL))) {
ORTE_ERROR_LOG(rc);
OBJ_RELEASE(buffer);
opal_list_remove_item(&requests, &dfs->super);
goto complete;
}
/* don't release it */
free(host);
free(filename);
return;
complete:
/* we get here if an error occurred - execute any
* pending callback so the proc doesn't hang
*/
if (NULL != host) {
free(host);
}
if (NULL != filename) {
free(filename);
}
if (NULL != dfs->open_cbfunc) {
dfs->open_cbfunc(-1, dfs->cbdata);
}
OBJ_RELEASE(dfs);
}
