/*
* Check and create the directory requested
*/
static int orte_create_dir(char *directory)
{
mode_t my_mode = S_IRWXU; /* I'm looking for full rights */
int ret;
/* Sanity check before creating the directory with the proper mode,
* Make sure it doesn't exist already */
if( ORTE_ERR_NOT_FOUND !=
(ret = opal_os_dirpath_access(directory, my_mode)) ) {
/* Failure because opal_os_dirpath_access() indicated that either:
* - The directory exists and we can access it (no need to create it again),
* return OPAL_SUCCESS, or
* - don't have access rights, return OPAL_ERROR
*/
if (ORTE_SUCCESS != ret) {
ORTE_ERROR_LOG(ret);
}
return(ret);
}
/* Get here if the directory doesn't exist, so create it */
if (ORTE_SUCCESS != (ret = opal_os_dirpath_create(directory, my_mode))) {
ORTE_ERROR_LOG(ret);
}
return ret;
}
/*
* Check and create the directory requested
*/
static int orte_create_dir(char *directory)
{
#ifndef __WINDOWS__
mode_t my_mode = S_IRWXU; /* at the least, I need to be able to do anything */
#else
mode_t my_mode = _S_IREAD | _S_IWRITE | _S_IEXEC;
#endif
int ret;
/* Sanity check before creating the directory with the proper mode,
* Make sure it doesn't exist already */
if( OPAL_ERR_NOT_FOUND != (ret = opal_os_dirpath_access(directory, my_mode)) ) {
/* Failure because opal_os_dirpath_access() indicated that either:
* - The directory exists and we can access it (no need to create it again),
* return OPAL_SUCCESS, or
* - don't have access rights, return OPAL_ERROR
*/
return(ret);
}
/* The directory doesn't exist so create it */
else {
return(opal_os_dirpath_create(directory, my_mode));
}
}
/**
* This function attempts to remove a directory along with all the
* files in it. If the recursive variable is non-zero, then it will
* try to recursively remove all directories. If provided, the
* callback function is executed prior to the directory or file being
* removed. If the callback returns non-zero, then no removal is
* done.
*/
int opal_os_dirpath_destroy(const char *path,
bool recursive,
opal_os_dirpath_destroy_callback_fn_t cbfunc)
{
int rc, exit_status = OPAL_SUCCESS;
bool is_dir = false;
DIR *dp;
struct dirent *ep;
char *filenm;
#ifndef HAVE_STRUCT_DIRENT_D_TYPE
struct stat buf;
#endif
if (NULL == path) { /* protect against error */
return OPAL_ERROR;
}
/*
* Make sure we have access to the the base directory
*/
if( OPAL_SUCCESS != (rc = opal_os_dirpath_access(path, 0) ) ) {
exit_status = rc;
goto cleanup;
}
/* Open up the directory */
dp = opendir(path);
if (NULL == dp) {
return OPAL_ERROR;
}
while (NULL != (ep = readdir(dp)) ) {
/* skip:
* - . and ..
*/
if ((0 == strcmp(ep->d_name, ".")) ||
(0 == strcmp(ep->d_name, "..")) ) {
continue;
}
/* Check to see if it is a directory */
is_dir = false;
/* Create a pathname. This is not always needed, but it makes
* for cleaner code just to create it here. Note that we are
* allocating memory here, so we need to free it later on.
*/
filenm = opal_os_path(false, path, ep->d_name, NULL);
#ifdef HAVE_STRUCT_DIRENT_D_TYPE
if (DT_DIR == ep->d_type) {
is_dir = true;
}
#else /* have dirent.d_type */
rc = stat(filenm, &buf);
if (rc < 0 || S_ISDIR(buf.st_mode)) {
is_dir = true;
}
#endif /* have dirent.d_type */
/*
* If not recursively decending, then if we find a directory then fail
* since we were not told to remove it.
*/
if( is_dir && !recursive) {
/* Set the error indicating that we found a directory,
* but continue removing files
*/
exit_status = OPAL_ERROR;
free(filenm);
continue;
}
/* Will the caller allow us to remove this file/directory? */
if(NULL != cbfunc) {
/*
* Caller does not wish to remove this file/directory,
* continue with the rest of the entries
*/
if( ! (cbfunc(path, ep->d_name)) ) {
free(filenm);
continue;
}
}
/* Directories are recursively destroyed */
if(is_dir) {
rc = opal_os_dirpath_destroy(filenm, recursive, cbfunc);
free(filenm);
if (OPAL_SUCCESS != rc) {
exit_status = rc;
closedir(dp);
goto cleanup;
}
}
/* Files are removed right here */
else {
if( 0 != (rc = unlink(filenm) ) ) {
exit_status = OPAL_ERROR;
}
free(filenm);
}
}
/* Done with this directory */
closedir(dp);
cleanup:
/*
* If the directory is empty, them remove it
*/
if(opal_os_dirpath_is_empty(path)) {
rmdir(path);
}
return exit_status;
}
/**
* This function attempts to remove a directory along with all the
* files in it. If the recursive variable is non-zero, then it will
* try to recursively remove all directories. If provided, the
* callback function is executed prior to the directory or file being
* removed. If the callback returns non-zero, then no removal is
* done.
*/
int opal_os_dirpath_destroy(const char *path,
bool recursive,
opal_os_dirpath_destroy_callback_fn_t cbfunc)
{
int rc, exit_status = OPAL_SUCCESS;
bool is_dir = false;
if (NULL == path) { /* protect against error */
return OPAL_ERROR;
}
/*
* Make sure we have access to the the base directory
*/
if( OPAL_SUCCESS != (rc = opal_os_dirpath_access(path, 0) ) ) {
exit_status = rc;
goto cleanup;
}
#ifndef __WINDOWS__
{
DIR *dp;
struct dirent *ep;
char *filenm;
#ifndef HAVE_STRUCT_DIRENT_D_TYPE
struct stat buf;
#endif
/* Open up the directory */
dp = opendir(path);
if (NULL == dp) {
return OPAL_ERROR;
}
while (NULL != (ep = readdir(dp)) ) {
/* skip:
* - . and ..
*/
if ((0 == strcmp(ep->d_name, ".")) ||
(0 == strcmp(ep->d_name, "..")) ) {
continue;
}
/* Check to see if it is a directory */
is_dir = false;
/* Create a pathname. This is not always needed, but it makes
* for cleaner code just to create it here. Note that we are
* allocating memory here, so we need to free it later on.
*/
filenm = opal_os_path(false, path, ep->d_name, NULL);
#ifdef HAVE_STRUCT_DIRENT_D_TYPE
if (DT_DIR == ep->d_type) {
is_dir = true;
}
#else /* have dirent.d_type */
rc = stat(filenm, &buf);
if (rc < 0 || S_ISDIR(buf.st_mode)) {
is_dir = true;
}
#endif /* have dirent.d_type */
/*
* If not recursively decending, then if we find a directory then fail
* since we were not told to remove it.
*/
if( is_dir && !recursive) {
/* Set the error indicating that we found a directory,
* but continue removing files
*/
exit_status = OPAL_ERROR;
free(filenm);
continue;
}
/* Will the caller allow us to remove this file/directory? */
if(NULL != cbfunc) {
/*
* Caller does not wish to remove this file/directory,
* continue with the rest of the entries
*/
if( ! (cbfunc(path, ep->d_name)) ) {
free(filenm);
continue;
}
}
/* Directories are recursively destroyed */
if(is_dir) {
rc = opal_os_dirpath_destroy(filenm, recursive, cbfunc);
free(filenm);
if (OPAL_SUCCESS != rc) {
exit_status = rc;
closedir(dp);
goto cleanup;
}
}
/* Files are removed right here */
else {
if( 0 != (rc = unlink(filenm) ) ) {
exit_status = OPAL_ERROR;
}
free(filenm);
}
}
/* Done with this directory */
closedir(dp);
}
#else
{
char search_path[MAX_PATH];
HANDLE file;
WIN32_FIND_DATA file_data;
TCHAR *file_name;
strncpy(search_path, path, strlen(path)+1);
strncat (search_path, OPAL_PATH_SEP"*", 3);
file = FindFirstFile(search_path, &file_data);
if (INVALID_HANDLE_VALUE == file) {
FindClose(file);
return OPAL_ERROR;
}
do {
/* Skip . and .. */
if ((0 == strcmp(file_data.cFileName, ".")) ||
(0 == strcmp(file_data.cFileName, "..")) ) {
continue;
}
is_dir = false;
if(file_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) {
is_dir = true;
}
/*
* If not recursively decending, then if we find a directory then fail
* since we were not told to remove it.
*/
if( is_dir && !recursive) {
/* Set the error indicating that we found a directory,
* but continue removing files
*/
exit_status = OPAL_ERROR;
continue;
}
/* Will the caller allow us to remove this file/directory */
if( NULL != cbfunc) {
/*
* Caller does not wish to remove this file/directory,
* continue with the rest of the entries
*/
if( !cbfunc(path, file_data.cFileName) ) {
continue;
}
}
file_name = opal_os_path(false, path, file_data.cFileName, NULL);
if( is_dir ) {
if( OPAL_SUCCESS != (rc = opal_os_dirpath_destroy(file_name,
recursive,
cbfunc) ) ) {
exit_status = rc;
free(file_name);
FindClose(file);
goto cleanup;
}
} else {
DeleteFile(file_name);
}
free(file_name);
} while( 0 != FindNextFile(file, &file_data) );
FindClose(file);
}
#endif
cleanup:
/*
* If the directory is empty, them remove it
*/
if(opal_os_dirpath_is_empty(path)) {
rmdir(path);
}
return exit_status;
}
/*
* Construct the session directory and create it if necessary
*/
int orte_session_dir(bool create,
char *prefix, char *hostid,
char *batchid, orte_process_name_t *proc)
{
char *fulldirpath = NULL,
*frontend = NULL,
*sav = NULL;
int rc = ORTE_SUCCESS;
char *local_prefix = NULL;
/* use the specified prefix, if one was given */
if (NULL != prefix) {
local_prefix = strdup(prefix);
}
/*
* Get the session directory full name
*/
if( ORTE_SUCCESS != ( rc = orte_session_dir_get_name(&fulldirpath,
&local_prefix,
&frontend,
hostid,
batchid, proc) ) ) {
if (ORTE_ERR_FATAL == rc) {
/* this indicates we should abort quietly */
rc = ORTE_ERR_SILENT;
goto cleanup;
}
/* otherwise, bark a little first */
ORTE_ERROR_LOG(rc);
goto cleanup;
}
/*
* Now that we have the full path, go ahead and create it if necessary
*/
if( create ) {
if( ORTE_SUCCESS != (rc = orte_create_dir(fulldirpath) ) ) {
ORTE_ERROR_LOG(rc);
goto cleanup;
}
}
/*
* if we are not creating, then just verify that the path is OK
*/
else {
if( ORTE_SUCCESS != (rc = opal_os_dirpath_access(fulldirpath, 0) )) {
/* it is okay for the path not to be found - don't error
* log that case, but do error log others
*/
if (ORTE_ERR_NOT_FOUND != rc) {
ORTE_ERROR_LOG(rc);
}
goto cleanup;
}
}
/*
* If we are creating the directory tree, then force overwrite of the
* global structure fields
*/
if (create) {
if (NULL != orte_process_info.tmpdir_base) {
free(orte_process_info.tmpdir_base);
orte_process_info.tmpdir_base = NULL;
}
if (NULL != orte_process_info.top_session_dir) {
free(orte_process_info.top_session_dir);
orte_process_info.top_session_dir = NULL;
}
}
/*
* Update some of the global structures if they are empty
*/
if (NULL == orte_process_info.tmpdir_base) {
orte_process_info.tmpdir_base = strdup(local_prefix);
}
if (NULL == orte_process_info.top_session_dir &&
NULL != frontend) {
orte_process_info.top_session_dir = strdup(frontend);
}
/*
* Set the process session directory
*/
if (ORTE_VPID_INVALID != proc->vpid) {
if (create) { /* overwrite if creating */
if (NULL != orte_process_info.proc_session_dir) {
free(orte_process_info.proc_session_dir);
orte_process_info.proc_session_dir = NULL;
}
}
if (NULL == orte_process_info.proc_session_dir) {
orte_process_info.proc_session_dir = strdup(fulldirpath);
}
/* Strip off last part of directory structure */
sav = opal_dirname(fulldirpath);
free(fulldirpath);
fulldirpath = sav;
sav = NULL;
}
/*
* Set the job session directory
*/
if (ORTE_JOBID_INVALID != proc->jobid) {
if (create) { /* overwrite if creating */
if (NULL != orte_process_info.job_session_dir) {
free(orte_process_info.job_session_dir);
orte_process_info.job_session_dir = NULL;
}
}
if (NULL == orte_process_info.job_session_dir) {
orte_process_info.job_session_dir = strdup(fulldirpath);
}
}
if (orte_debug_flag) {
opal_output(0, "procdir: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.proc_session_dir));
opal_output(0, "jobdir: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.job_session_dir));
opal_output(0, "top: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.top_session_dir));
opal_output(0, "tmp: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.tmpdir_base));
}
cleanup:
if (NULL != local_prefix) {
free(local_prefix);
}
if(NULL != fulldirpath) {
free(fulldirpath);
}
if(NULL != frontend) {
free(frontend);
}
return rc;
}
int
orte_session_dir_finalize(orte_process_name_t *proc)
{
int rc=ORTE_SUCCESS;
if (!orte_create_session_dirs || orte_process_info.rm_session_dirs ) {
/* we haven't created them or RM will clean them up for us*/
return ORTE_SUCCESS;
}
if (NULL == orte_process_info.job_session_dir ||
NULL == orte_process_info.proc_session_dir) {
/* this should never happen - it means we are calling
* cleanup *before* properly setting up the session
* dir system. This leaves open the possibility of
* accidentally removing directories we shouldn't
* touch
*/
rc = ORTE_ERR_NOT_INITIALIZED;
goto CLEANUP;
}
opal_os_dirpath_destroy(orte_process_info.proc_session_dir,
false, orte_dir_check_file);
opal_os_dirpath_destroy(orte_process_info.job_session_dir,
false, orte_dir_check_file);
if( NULL != orte_process_info.top_session_dir ){
opal_os_dirpath_destroy(orte_process_info.top_session_dir,
false, orte_dir_check_file);
}
if (opal_os_dirpath_is_empty(orte_process_info.proc_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found proc session dir empty - deleting");
}
rmdir(orte_process_info.proc_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.proc_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: proc session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: proc session dir not empty - leaving");
}
}
goto CLEANUP;
}
if (opal_os_dirpath_is_empty(orte_process_info.job_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found job session dir empty - deleting");
}
rmdir(orte_process_info.job_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.job_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: job session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: job session dir not empty - leaving");
}
}
goto CLEANUP;
}
if(NULL != orte_process_info.top_session_dir) {
if (opal_os_dirpath_is_empty(orte_process_info.top_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found top session dir empty - deleting");
}
rmdir(orte_process_info.top_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.top_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: top session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: top session dir not empty - leaving");
}
}
}
}
CLEANUP:
return rc;
}
/*
* A job has aborted - so force cleanup of the session directory
*/
int
orte_session_dir_cleanup(orte_jobid_t jobid)
{
int rc = ORTE_SUCCESS;
if (!orte_create_session_dirs || orte_process_info.rm_session_dirs ) {
/* we haven't created them or RM will clean them up for us*/
return ORTE_SUCCESS;
}
if (NULL == orte_process_info.job_session_dir ||
NULL == orte_process_info.proc_session_dir) {
/* this should never happen - it means we are calling
* cleanup *before* properly setting up the session
* dir system. This leaves open the possibility of
* accidentally removing directories we shouldn't
* touch
*/
rc = ORTE_ERR_NOT_INITIALIZED;
goto CLEANUP;
}
/* recursively blow the whole session away for our job family,
* saving only output files
*/
opal_os_dirpath_destroy(orte_process_info.job_session_dir,
true, orte_dir_check_file);
/* now attempt to eliminate the top level directory itself - this
* will fail if anything is present, but ensures we cleanup if
* we are the last one out
*/
if( NULL != orte_process_info.top_session_dir ){
opal_os_dirpath_destroy(orte_process_info.top_session_dir,
false, orte_dir_check_file);
}
if (opal_os_dirpath_is_empty(orte_process_info.job_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_cleanup: found job session dir empty - deleting");
}
rmdir(orte_process_info.job_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.job_session_dir, 0)) {
opal_output(0, "sess_dir_cleanup: job session dir does not exist");
} else {
opal_output(0, "sess_dir_cleanup: job session dir not empty - leaving");
}
}
goto CLEANUP;
}
if ( NULL != orte_process_info.top_session_dir ){
if( opal_os_dirpath_is_empty(orte_process_info.top_session_dir) ) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_cleanup: found top session dir empty - deleting");
}
rmdir(orte_process_info.top_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.top_session_dir, 0)) {
opal_output(0, "sess_dir_cleanup: top session dir does not exist");
} else {
opal_output(0, "sess_dir_cleanup: top session dir not empty - leaving");
}
}
}
}
CLEANUP:
return rc;
}
int
orte_session_dir_finalize(orte_process_name_t *proc)
{
if (!orte_create_session_dirs || orte_process_info.rm_session_dirs ) {
/* we haven't created them or RM will clean them up for us*/
return ORTE_SUCCESS;
}
if (NULL == orte_process_info.job_session_dir ||
NULL == orte_process_info.proc_session_dir) {
/* this should never happen - it means we are calling
* cleanup *before* properly setting up the session
* dir system. This leaves open the possibility of
* accidentally removing directories we shouldn't
* touch
*/
return ORTE_ERR_NOT_INITIALIZED;
}
opal_os_dirpath_destroy(orte_process_info.proc_session_dir,
false, orte_dir_check_file);
opal_os_dirpath_destroy(orte_process_info.job_session_dir,
false, orte_dir_check_file);
/* only remove the jobfam session dir if we are the
* local daemon and we are finalizing our own session dir */
if ((ORTE_PROC_IS_HNP || ORTE_PROC_IS_DAEMON) &&
(ORTE_PROC_MY_NAME == proc)) {
opal_os_dirpath_destroy(orte_process_info.jobfam_session_dir,
false, orte_dir_check_file);
}
if( NULL != orte_process_info.top_session_dir ){
opal_os_dirpath_destroy(orte_process_info.top_session_dir,
false, orte_dir_check_file);
}
if (opal_os_dirpath_is_empty(orte_process_info.proc_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found proc session dir empty - deleting");
}
rmdir(orte_process_info.proc_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.proc_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: proc session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: proc session dir not empty - leaving");
}
}
}
if (opal_os_dirpath_is_empty(orte_process_info.job_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found job session dir empty - deleting");
}
rmdir(orte_process_info.job_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.job_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: job session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: job session dir not empty - leaving");
}
}
}
if (opal_os_dirpath_is_empty(orte_process_info.jobfam_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found jobfam session dir empty - deleting");
}
rmdir(orte_process_info.jobfam_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.jobfam_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: jobfam session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: jobfam session dir not empty - leaving");
}
}
}
if(NULL != orte_process_info.top_session_dir) {
if (opal_os_dirpath_is_empty(orte_process_info.top_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found top session dir empty - deleting");
}
rmdir(orte_process_info.top_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND ==
opal_os_dirpath_access(orte_process_info.top_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: top session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: top session dir not empty - leaving");
}
}
}
}
return ORTE_SUCCESS;
}
int orte_list_local_hnps(opal_list_t *hnps, bool connect)
{
int ret;
#ifndef __WINDOWS__
DIR *cur_dirp = NULL;
struct dirent * dir_entry;
#else
HANDLE hFind = INVALID_HANDLE_VALUE;
WIN32_FIND_DATA file_data;
#endif /* __WINDOWS__ */
char *contact_filename = NULL;
orte_hnp_contact_t *hnp;
char *headdir;
#if !defined(__WINDOWS__)
/*
* Check to make sure we have access to the top-level directory
*/
headdir = opal_os_path(false, orte_process_info.tmpdir_base, orte_process_info.top_session_dir, NULL);
if( ORTE_SUCCESS != (ret = opal_os_dirpath_access(headdir, 0) )) {
/* it is okay not to find this as there may not be any
* HNP's present, and we don't write our own session dir
*/
if (ORTE_ERR_NOT_FOUND != ret) {
ORTE_ERROR_LOG(ret);
}
goto cleanup;
}
/*
* Open up the base directory so we can get a listing
*/
if( NULL == (cur_dirp = opendir(headdir)) ) {
goto cleanup;
}
/*
* For each directory
*/
while( NULL != (dir_entry = readdir(cur_dirp)) ) {
/*
* Skip the obvious
*/
if( 0 == strncmp(dir_entry->d_name, ".", strlen(".")) ||
0 == strncmp(dir_entry->d_name, "..", strlen("..")) ) {
continue;
}
/*
* See if a contact file exists in this directory and read it
*/
contact_filename = opal_os_path( false, headdir,
dir_entry->d_name, "contact.txt", NULL );
hnp = OBJ_NEW(orte_hnp_contact_t);
if (ORTE_SUCCESS == (ret = orte_read_hnp_contact_file(contact_filename, hnp, connect))) {
opal_list_append(hnps, &(hnp->super));
} else {
OBJ_RELEASE(hnp);
}
}
#else
/*
* Open up the base directory so we can get a listing.
*
* On Windows if we want to parse the content of a directory the filename
* should end with the "*". Otherwise we will only open the directory
* structure (and not the content).
*/
char *subdirs = opal_os_path(false, orte_process_info.tmpdir_base, orte_process_info.top_session_dir, "*", NULL);
headdir = opal_os_path(false, orte_process_info.tmpdir_base, orte_process_info.top_session_dir, NULL);
hFind = FindFirstFile( subdirs, &file_data );
if( INVALID_HANDLE_VALUE == hFind ) {
goto cleanup;
}
/*
* For each directory
*/
do {
/*
* Skip the obvious
*/
if( 0 == strncmp(file_data.cFileName, ".", strlen(".")) ||
0 == strncmp(file_data.cFileName, "..", strlen("..")) ) {
continue;
}
/*
* See if a contact file exists in this directory and read it
*/
contact_filename = opal_os_path( false, headdir,
file_data.cFileName, "contact.txt", NULL );
hnp = OBJ_NEW(orte_hnp_contact_t);
if (ORTE_SUCCESS == (ret = orte_read_hnp_contact_file(contact_filename, hnp, connect))) {
opal_list_append(hnps, &(hnp->super));
} else {
OBJ_RELEASE(hnp);
}
} while( 0 != FindNextFile( hFind, &file_data ) );
#endif /* !defined(__WINDOWS__) */
cleanup:
#ifndef __WINDOWS__
if( NULL != cur_dirp )
closedir(cur_dirp);
#else
FindClose(hFind);
#endif /* __WINDOWS__ */
free(headdir);
if( NULL != contact_filename)
free(contact_filename);
return (opal_list_is_empty(hnps) ? ORTE_ERR_NOT_FOUND : ORTE_SUCCESS);
}
int orte_list_local_hnps(opal_list_t *hnps, bool connect)
{
int ret;
DIR *cur_dirp = NULL;
struct dirent * dir_entry;
char *contact_filename = NULL;
orte_hnp_contact_t *hnp;
char *headdir;
/*
* Check to make sure we have access to the top-level directory
*/
headdir = opal_os_path(false, orte_process_info.tmpdir_base, orte_process_info.top_session_dir, NULL);
if( ORTE_SUCCESS != (ret = opal_os_dirpath_access(headdir, 0) )) {
/* it is okay not to find this as there may not be any
* HNP's present, and we don't write our own session dir
*/
if (ORTE_ERR_NOT_FOUND != ret) {
ORTE_ERROR_LOG(ret);
}
goto cleanup;
}
/*
* Open up the base directory so we can get a listing
*/
if( NULL == (cur_dirp = opendir(headdir)) ) {
goto cleanup;
}
/*
* For each directory
*/
while( NULL != (dir_entry = readdir(cur_dirp)) ) {
/*
* Skip the obvious
*/
if( 0 == strncmp(dir_entry->d_name, ".", strlen(".")) ||
0 == strncmp(dir_entry->d_name, "..", strlen("..")) ) {
continue;
}
/*
* See if a contact file exists in this directory and read it
*/
contact_filename = opal_os_path( false, headdir,
dir_entry->d_name, "contact.txt", NULL );
hnp = OBJ_NEW(orte_hnp_contact_t);
if (ORTE_SUCCESS == (ret = orte_read_hnp_contact_file(contact_filename, hnp, connect))) {
opal_list_append(hnps, &(hnp->super));
} else {
OBJ_RELEASE(hnp);
}
}
cleanup:
if( NULL != cur_dirp )
closedir(cur_dirp);
free(headdir);
if( NULL != contact_filename)
free(contact_filename);
return (opal_list_is_empty(hnps) ? ORTE_ERR_NOT_FOUND : ORTE_SUCCESS);
}
/*
* Construct the session directory and create it if necessary
*/
int orte_session_dir(bool create,
char *prefix, char *usr, char *hostid,
char *batchid, char *univ, char *job, char *proc)
{
char *fulldirpath = NULL,
*frontend = NULL,
*sav = NULL;
int return_code = ORTE_SUCCESS, rtn;
/* This indicates if the prefix was set, and so if it fails then we
* should try with the default prefixes.*/
bool dbl_check_prefix = false;
if( NULL != prefix)
dbl_check_prefix = true;
try_again:
/*
* If the first attempt at the path creation failed, try with a null
* prefix. unless the original prefix was null, then we fail.
*/
if(!dbl_check_prefix && /* an indicator that we are trying a second time */
NULL != prefix) {
free(prefix);
prefix = NULL;
}
/*
* Get the session directory full name
* First try it with the specified prefix.
*/
if( ORTE_SUCCESS != ( rtn = orte_session_dir_get_name(&fulldirpath,
&prefix,
&frontend,
usr, hostid,
batchid, univ, job,
proc) ) ) {
return_code = rtn;
/*
* If the first attempt at the path creation failed, try with a null
* prefix. unless the original prefix was null, then we fail :(
*/
if(dbl_check_prefix) {
dbl_check_prefix = false;
goto try_again;
}
else {
goto cleanup;
}
}
/*
* Now that we have the full path, go ahead and create it if necessary
*/
if( create ) {
if( ORTE_SUCCESS != (rtn = orte_create_dir(fulldirpath) ) ) {
return_code = rtn;
if(dbl_check_prefix) {
dbl_check_prefix = false;
goto try_again;
}
else {
goto cleanup;
}
}
}
/*
* if we are not creating, then just verify that the path is OK
*/
else {
if( ORTE_SUCCESS != (rtn = opal_os_dirpath_access(fulldirpath, 0) )) {
/* It is not valid so we give up and return an error */
return_code = rtn;
if(dbl_check_prefix) {
dbl_check_prefix = false;
goto try_again;
}
else {
goto cleanup;
}
}
}
return_code = ORTE_SUCCESS;
/*
* If we are creating the directory tree, the overwrite the
* global structure fields
*/
if (create) {
if (NULL != orte_process_info.tmpdir_base) {
free(orte_process_info.tmpdir_base);
orte_process_info.tmpdir_base = NULL;
}
if (NULL != orte_process_info.top_session_dir) {
free(orte_process_info.top_session_dir);
orte_process_info.top_session_dir = NULL;
}
}
/*
* Update some of the global structures if they are empty
*/
if (NULL == orte_process_info.tmpdir_base)
orte_process_info.tmpdir_base = strdup(prefix);
if (NULL == orte_process_info.top_session_dir)
orte_process_info.top_session_dir = strdup(frontend);
/*
* Set the process session directory
*/
if (NULL != proc) {
if (create) { /* overwrite if creating */
if (NULL != orte_process_info.proc_session_dir) {
free(orte_process_info.proc_session_dir);
orte_process_info.proc_session_dir = NULL;
}
}
if (NULL == orte_process_info.proc_session_dir) {
orte_process_info.proc_session_dir = strdup(fulldirpath);
}
/* Strip off last part of directory structure */
sav = opal_dirname(fulldirpath);
free(fulldirpath);
fulldirpath = sav;
sav = NULL;
}
/*
* Set the job session directory
*/
if (NULL != job) {
if (create) { /* overwrite if creating */
if (NULL != orte_process_info.job_session_dir) {
free(orte_process_info.job_session_dir);
orte_process_info.job_session_dir = NULL;
}
}
if (NULL == orte_process_info.job_session_dir) {
orte_process_info.job_session_dir = strdup(fulldirpath);
}
/* Strip off last part of directory structure */
sav = opal_dirname(fulldirpath);
free(fulldirpath);
fulldirpath = sav;
sav = NULL;
}
/*
* Set the universe session directory
*/
if (create) { /* overwrite if creating */
if (NULL != orte_process_info.universe_session_dir) {
free(orte_process_info.universe_session_dir);
orte_process_info.universe_session_dir = NULL;
}
}
if (NULL == orte_process_info.universe_session_dir) {
orte_process_info.universe_session_dir = strdup(fulldirpath);
}
if (orte_debug_flag) {
opal_output(0, "procdir: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.proc_session_dir));
opal_output(0, "jobdir: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.job_session_dir));
opal_output(0, "unidir: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.universe_session_dir));
opal_output(0, "top: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.top_session_dir));
opal_output(0, "tmp: %s",
OMPI_PRINTF_FIX_STRING(orte_process_info.tmpdir_base));
}
cleanup:
if(NULL != fulldirpath)
free(fulldirpath);
if(NULL != frontend)
free(frontend);
if(NULL != sav)
free(sav);
return return_code;
}
int
orte_session_dir_finalize(orte_process_name_t *proc)
{
int rc;
char *tmp;
char *job_session_dir, *vpid, *proc_session_dir;
if (!orte_create_session_dirs) {
/* didn't create them */
return ORTE_SUCCESS;
}
if (NULL == orte_process_info.tmpdir_base &&
NULL == orte_process_info.top_session_dir) {
/* this should never happen - it means we are calling
* cleanup *before* properly setting up the session
* dir system. This leaves open the possibility of
* accidentally removing directories we shouldn't
* touch
*/
ORTE_ERROR_LOG(ORTE_ERR_NOT_INITIALIZED);
return ORTE_ERR_NOT_INITIALIZED;
}
/* need to setup the top_session_dir with the prefix */
tmp = opal_os_path(false,
orte_process_info.tmpdir_base,
orte_process_info.top_session_dir, NULL);
/* define the proc and job session directories for this process */
if (ORTE_SUCCESS != (rc = orte_util_convert_vpid_to_string(&vpid, proc->vpid))) {
ORTE_ERROR_LOG(rc);
free(tmp);
return rc;
}
job_session_dir = orte_build_job_session_dir(tmp, proc, proc->jobid);
if( NULL == job_session_dir) {
free(tmp);
free(vpid);
return ORTE_ERR_OUT_OF_RESOURCE;
}
proc_session_dir = opal_os_path( false, job_session_dir, vpid, NULL );
if( NULL == proc_session_dir ) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
free(tmp);
free(vpid);
free(job_session_dir);
return ORTE_ERR_OUT_OF_RESOURCE;
}
opal_os_dirpath_destroy(proc_session_dir,
false, orte_dir_check_file);
opal_os_dirpath_destroy(job_session_dir,
false, orte_dir_check_file);
opal_os_dirpath_destroy(tmp,
false, orte_dir_check_file);
if (opal_os_dirpath_is_empty(proc_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found proc session dir empty - deleting");
}
rmdir(proc_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(proc_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: proc session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: proc session dir not empty - leaving");
}
}
goto CLEANUP;
}
if (opal_os_dirpath_is_empty(job_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found job session dir empty - deleting");
}
rmdir(job_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(job_session_dir, 0)) {
opal_output(0, "sess_dir_finalize: job session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: job session dir not empty - leaving");
}
}
goto CLEANUP;
}
if (opal_os_dirpath_is_empty(tmp)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_finalize: found top session dir empty - deleting");
}
rmdir(tmp);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(tmp, 0)) {
opal_output(0, "sess_dir_finalize: top session dir does not exist");
} else {
opal_output(0, "sess_dir_finalize: top session dir not empty - leaving");
}
}
}
CLEANUP:
free(tmp);
free(vpid);
free(job_session_dir);
free(proc_session_dir);
return ORTE_SUCCESS;
}
/*
* A job has aborted - so force cleanup of the session directory
*/
int
orte_session_dir_cleanup(orte_jobid_t jobid)
{
int rc = ORTE_SUCCESS;
char *tmp = NULL;
char *job_session_dir=NULL;
if (!orte_create_session_dirs) {
/* didn't create them */
return ORTE_SUCCESS;
}
if (NULL == orte_process_info.tmpdir_base &&
NULL == orte_process_info.top_session_dir) {
/* this should never happen - it means we are calling
* cleanup *before* properly setting up the session
* dir system. This leaves open the possibility of
* accidentally removing directories we shouldn't
* touch
*/
rc = ORTE_ERR_NOT_INITIALIZED;
goto CLEANUP;
}
/* need to setup the top_session_dir with the prefix */
tmp = opal_os_path(false,
orte_process_info.tmpdir_base,
orte_process_info.top_session_dir, NULL);
/* we can only blow away session directories for our job family */
job_session_dir = orte_build_job_session_dir(tmp, ORTE_PROC_MY_NAME, jobid);
if (NULL == job_session_dir) {
rc = ORTE_ERR_OUT_OF_RESOURCE;
goto CLEANUP;
}
if (ORTE_JOBID_WILDCARD != jobid) {
opal_os_dirpath_destroy(job_session_dir, true, orte_dir_check_file);
} else {
/* if we want the session_dir removed for ALL jobids, then
* just recursively blow the whole session away for our job family,
* saving only output files
*/
opal_os_dirpath_destroy(job_session_dir, true, orte_dir_check_file_output);
}
/* now attempt to eliminate the top level directory itself - this
* will fail if anything is present, but ensures we cleanup if
* we are the last one out
*/
opal_os_dirpath_destroy(tmp, false, orte_dir_check_file);
if (NULL != job_session_dir && opal_os_dirpath_is_empty(job_session_dir)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_cleanup: found job session dir empty - deleting");
}
rmdir(job_session_dir);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(job_session_dir, 0)) {
opal_output(0, "sess_dir_cleanup: job session dir does not exist");
} else {
opal_output(0, "sess_dir_cleanup: job session dir not empty - leaving");
}
}
goto CLEANUP;
}
if (opal_os_dirpath_is_empty(tmp)) {
if (orte_debug_flag) {
opal_output(0, "sess_dir_cleanup: found top session dir empty - deleting");
}
rmdir(tmp);
} else {
if (orte_debug_flag) {
if (OPAL_ERR_NOT_FOUND == opal_os_dirpath_access(tmp, 0)) {
opal_output(0, "sess_dir_cleanup: top session dir does not exist");
} else {
opal_output(0, "sess_dir_cleanup: top session dir not empty - leaving");
}
}
}
CLEANUP:
if (NULL != tmp) free(tmp);
if (NULL != job_session_dir) free(job_session_dir);
return rc;
}
