/*
FUNCTION: smpd_service_stop
PURPOSE: Stops the service
PARAMETERS:
none
RETURN VALUE:
none
COMMENTS:
If a ServiceStop procedure is going to
take longer than 3 seconds to execute,
it should spawn a thread to execute the
stop code, and return. Otherwise, the
ServiceControlManager will believe that
the service has stopped responding.
*/
void smpd_service_stop()
{
SMPDU_Sock_set_t set;
SMPDU_Sock_t sock;
SMPDU_Sock_event_t event;
char host[SMPD_MAX_HOST_LENGTH];
int iter;
DWORD dwThreadID;
int result;
for (iter=0; iter<10; iter++)
{
smpd_process.hBombThread = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)smpd_bomb_thread, NULL, 0, &dwThreadID);
if (smpd_process.hBombThread != NULL)
break;
Sleep(250);
}
/* stop the main thread */
smpd_process.service_stop = SMPD_TRUE;
smpd_get_hostname(host, SMPD_MAX_HOST_LENGTH);
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_create_set failed,\nsock error: %s\n", get_sock_error_string(result));
SetEvent(smpd_process.hBombDiffuseEvent);
WaitForSingleObject(smpd_process.hBombThread, (DWORD)3000);
CloseHandle(smpd_process.hBombThread);
ExitProcess((UINT)-1);
}
result = SMPDU_Sock_post_connect(set, NULL, host, smpd_process.port, &sock);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to connect to '%s:%d',\nsock error: %s\n",
smpd_process.host_list->host, smpd_process.port, get_sock_error_string(result));
SetEvent(smpd_process.hBombDiffuseEvent);
WaitForSingleObject(smpd_process.hBombThread, (DWORD)3000);
CloseHandle(smpd_process.hBombThread);
ExitProcess((UINT)-1);
}
result = SMPDU_Sock_wait(set, SMPDU_SOCK_INFINITE_TIME, &event);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to connect to '%s:%d',\nsock error: %s\n",
smpd_process.host_list->host, smpd_process.port, get_sock_error_string(result));
SetEvent(smpd_process.hBombDiffuseEvent);
WaitForSingleObject(smpd_process.hBombThread, (DWORD)3000);
CloseHandle(smpd_process.hBombThread);
ExitProcess((UINT)-1);
}
}
int smpd_parse_command_args(int *argcp, char **argvp[])
{
int result = 0;
#ifdef HAVE_WINDOWS_H
char str[20], read_handle_str[20], write_handle_str[20];
int port;
SMPDU_Sock_t listener;
SMPDU_Sock_set_t set;
HANDLE hWrite, hRead;
DWORD num_written, num_read;
#endif
int dbg_flag;
char domain[SMPD_MAX_HOST_LENGTH];
char opt[SMPD_MAX_NAME_LENGTH];
char opt_val[SMPD_MAX_VALUE_LENGTH];
char filename[SMPD_MAX_FILENAME];
int i;
smpd_enter_fn(FCNAME);
/* check for help option */
if (
#ifndef HAVE_WINDOWS_H
*argcp < 2 || /* unix: print the options if no arguments are supplied */
#endif
smpd_get_opt(argcp, argvp, "-help") || smpd_get_opt(argcp, argvp, "-?"))
{
smpd_print_options();
smpd_exit(0);
}
/* check for the printprocs option */
if (smpd_get_opt(argcp, argvp, "-printprocs"))
{
smpd_watch_processes();
smpd_exit(0);
}
if (smpd_get_opt(argcp, argvp, "-hosts"))
{
char first_host[SMPD_MAX_HOST_LENGTH], host[SMPD_MAX_HOST_LENGTH], alt_host[SMPD_MAX_HOST_LENGTH];
smpd_get_default_hosts();
result = smpd_get_next_hostname(first_host, alt_host);
if (result != SMPD_SUCCESS)
smpd_exit(result);
printf("%s\n", first_host);
result = smpd_get_next_hostname(host, alt_host);
if (result != SMPD_SUCCESS)
smpd_exit(result);
while (strcmp(host, first_host) != 0)
{
printf("%s\n", host);
result = smpd_get_next_hostname(host, alt_host);
if (result != SMPD_SUCCESS)
smpd_exit(result);
}
smpd_exit(0);
}
if (smpd_get_opt(argcp, argvp, "-sethosts"))
{
char *buffer, *iter;
int i, length;
length = (*argcp) * SMPD_MAX_HOST_LENGTH;
buffer = MPIU_Malloc(length);
if (buffer == NULL)
{
smpd_err_printf("unable to allocate memory to store the host names.\n");
smpd_exit(-1);
}
iter = buffer;
for (i=1; i<*argcp; i++)
{
result = MPIU_Str_add_string(&iter, &length, (*argvp)[i]);
if (result)
{
printf("unable to add host #%d, %s\n", i, (*argvp)[i]);
MPIU_Free(buffer);
smpd_exit(-1);
}
}
/*printf("hosts: %s\n", buffer);*/
result = smpd_set_smpd_data("hosts", buffer);
if (result == SMPD_SUCCESS)
{
printf("hosts data saved successfully.\n");
}
else
{
printf("Error: unable to save the hosts data.\n");
}
MPIU_Free(buffer);
smpd_exit(0);
}
if (smpd_get_opt_two_strings(argcp, argvp, "-set", opt, SMPD_MAX_NAME_LENGTH, opt_val, SMPD_MAX_VALUE_LENGTH))
{
/* The do loop allows for multiple -set operations to be specified on the command line */
do
{
if (strlen(opt) == 0)
{
printf("invalid option specified.\n");
smpd_exit(-1);
}
if (strlen(opt_val) == 0)
{
result = smpd_delete_smpd_data(opt);
}
else
{
result = smpd_set_smpd_data(opt, opt_val);
}
if (result == SMPD_SUCCESS)
{
printf("%s = %s\n", opt, opt_val);
}
else
{
printf("unable to set %s option.\n", opt);
}
} while (smpd_get_opt_two_strings(argcp, argvp, "-set", opt, SMPD_MAX_NAME_LENGTH, opt_val, SMPD_MAX_VALUE_LENGTH));
smpd_exit(0);
}
if (smpd_get_opt_string(argcp, argvp, "-get", opt, SMPD_MAX_NAME_LENGTH))
{
if (strlen(opt) == 0)
{
printf("invalid option specified.\n");
smpd_exit(-1);
}
result = smpd_get_smpd_data(opt, opt_val, SMPD_MAX_VALUE_LENGTH);
if (result == SMPD_SUCCESS)
{
printf("%s\n", opt_val);
}
else
{
printf("default\n");
}
smpd_exit(0);
}
/* If we've made it here and there still is "-set" or "-get" on the command line then the user
* probably didn't supply the correct number of parameters. So print the usage message
* and exit.
*/
if (smpd_get_opt(argcp, argvp, "-set") || smpd_get_opt(argcp, argvp, "-get"))
{
smpd_print_options();
smpd_exit(-1);
}
if (smpd_get_opt(argcp, argvp, "-enumerate") || smpd_get_opt(argcp, argvp, "-enum"))
{
smpd_data_t *data;
if (smpd_get_all_smpd_data(&data) == SMPD_SUCCESS)
{
smpd_data_t *iter = data;
while (iter != NULL)
{
printf("%s\n%s\n", iter->name, iter->value);
iter = iter->next;
}
while (data != NULL)
{
iter = data;
data = data->next;
MPIU_Free(iter);
}
}
smpd_exit(0);
}
if (smpd_get_opt_string(argcp, argvp, "-query", domain, SMPD_MAX_HOST_LENGTH))
{
printf("querying hosts in the %s domain:\n", domain);
printf("Not implemented.\n");
smpd_exit(0);
}
if (smpd_get_opt(argcp, argvp, "-query"))
{
printf("querying hosts in the default domain:\n");
printf("Not implemented.\n");
smpd_exit(0);
}
/* check for the service/silent option */
#ifdef HAVE_WINDOWS_H
smpd_process.bService = SMPD_TRUE;
#endif
if (smpd_get_opt(argcp, argvp, "-s"))
{
#ifdef HAVE_WINDOWS_H
printf("The -s option is only available under unix.\n");
smpd_print_options();
smpd_exit(0);
#else
smpd_process.bNoTTY = SMPD_TRUE;
#endif
}
if (smpd_get_opt(argcp, argvp, "-r"))
{
#ifdef HAVE_WINDOWS_H
printf("The -r option is only available under unix.\n");
smpd_print_options();
#else
printf("The -r root option is not yet implemented.\n");
#endif
smpd_exit(0);
}
/* check for debug option */
if (smpd_get_opt_int(argcp, argvp, "-d", &dbg_flag))
{
smpd_process.dbg_state = dbg_flag;
smpd_process.bNoTTY = SMPD_FALSE;
smpd_process.bService = SMPD_FALSE;
}
if (smpd_get_opt(argcp, argvp, "-d"))
{
smpd_process.dbg_state = SMPD_DBG_STATE_ERROUT | SMPD_DBG_STATE_STDOUT | SMPD_DBG_STATE_PREPEND_RANK | SMPD_DBG_STATE_TRACE;
smpd_process.bNoTTY = SMPD_FALSE;
smpd_process.bService = SMPD_FALSE;
}
if (smpd_get_opt_int(argcp, argvp, "-debug", &dbg_flag))
{
smpd_process.dbg_state = dbg_flag;
smpd_process.bNoTTY = SMPD_FALSE;
smpd_process.bService = SMPD_FALSE;
}
if (smpd_get_opt(argcp, argvp, "-debug"))
{
smpd_process.dbg_state = SMPD_DBG_STATE_ERROUT | SMPD_DBG_STATE_STDOUT | SMPD_DBG_STATE_PREPEND_RANK | SMPD_DBG_STATE_TRACE;
smpd_process.bNoTTY = SMPD_FALSE;
smpd_process.bService = SMPD_FALSE;
}
/* check for port option */
smpd_get_opt_int(argcp, argvp, "-p", &smpd_process.port);
smpd_get_opt_int(argcp, argvp, "-port", &smpd_process.port);
if (smpd_get_opt(argcp, argvp, "-anyport"))
{
smpd_process.port = 0;
smpd_process.dbg_state = 0; /* turn of debugging or you won't be able to read the port from stdout */
smpd_process.bNoTTY = SMPD_FALSE;
smpd_process.bService = SMPD_FALSE;
}
smpd_process.noprompt = smpd_get_opt(argcp, argvp, "-noprompt");
#ifdef HAVE_WINDOWS_H
/* check for service options */
if (smpd_get_opt(argcp, argvp, "-remove") ||
smpd_get_opt(argcp, argvp, "-unregserver") ||
smpd_get_opt(argcp, argvp, "-uninstall") ||
smpd_get_opt(argcp, argvp, "/Remove") ||
smpd_get_opt(argcp, argvp, "/Uninstall"))
{
/*RegDeleteKey(HKEY_CURRENT_USER, MPICHKEY);*/
smpd_remove_service(SMPD_TRUE);
ExitProcess(0);
}
if (smpd_get_opt(argcp, argvp, "-install") ||
smpd_get_opt(argcp, argvp, "-regserver") ||
smpd_get_opt(argcp, argvp, "/Install") ||
smpd_get_opt(argcp, argvp, "/install") ||
smpd_get_opt(argcp, argvp, "/RegServer"))
{
char phrase[SMPD_PASSPHRASE_MAX_LENGTH]="", port_str[SMPD_MAX_PORT_STR_LENGTH]="";
if (smpd_remove_service(SMPD_FALSE) == SMPD_FALSE)
{
printf("Unable to remove the previous installation, install failed.\n");
ExitProcess(0);
}
if (smpd_get_opt_string(argcp, argvp, "-phrase", phrase, SMPD_PASSPHRASE_MAX_LENGTH) ||
smpd_get_win_opt_string(argcp, argvp, "/phrase", phrase, SMPD_PASSPHRASE_MAX_LENGTH))
{
smpd_set_smpd_data("phrase", phrase);
}
if (smpd_get_opt(argcp, argvp, "-getphrase"))
{
printf("passphrase for smpd: ");fflush(stdout);
smpd_get_password(phrase);
smpd_set_smpd_data("phrase", phrase);
}
if (smpd_process.port != SMPD_LISTENER_PORT)
{
snprintf(port_str, SMPD_MAX_PORT_STR_LENGTH, "%d", smpd_process.port);
smpd_set_smpd_data("port", port_str);
}
smpd_install_service(SMPD_FALSE, SMPD_TRUE, smpd_get_opt(argcp, argvp, "-delegation"));
smpd_set_smpd_data("version", SMPD_VERSION);
ExitProcess(0);
}
if (smpd_get_opt(argcp, argvp, "-start"))
{
smpd_start_service();
ExitProcess(0);
}
if (smpd_get_opt(argcp, argvp, "-stop"))
{
smpd_stop_service();
ExitProcess(0);
}
if (smpd_get_opt(argcp, argvp, "-register_spn"))
{
char filename[SMPD_MAX_FILENAME];
if(smpd_get_opt_string(argcp, argvp, "-f", filename, SMPD_MAX_FILENAME)){
result = smpd_setup_scps_with_file(filename);
if(result != SMPD_SUCCESS){
printf("Failed to register smpd's Service Principal Names (at least one failed) with Domain Controller\n");
ExitProcess((UINT )-1);
}
}
else{
result = smpd_setup_scp(NULL);
if(result != SMPD_SUCCESS){
printf("Failed to register smpd's Service Principal Name with Domain Controller\n");
ExitProcess((UINT )-1);
}
}
printf("Service Principal Name registered with the domain controller.\n");
printf("SMPD is now capable of launching processes using passwordless delegation.\n");
printf("The system administrator must ensure the following:\n");
printf(" 1) This host is trusted for delegation in Active Directory\n");
printf(" 2) All users who will run jobs are trusted for delegation.\n");
printf("Domain administrators can enable these options for hosts and users\nin Active Directory on the domain controller.\n");
ExitProcess(0);
}
if (smpd_get_opt(argcp, argvp, "-remove_spn"))
{
char filename[SMPD_MAX_FILENAME];
smpd_spn_list_hnd_t hnd;
result = smpd_spn_list_init(&hnd);
if(result != SMPD_SUCCESS){
printf("Unable to initialize SPN list\n");
ExitProcess((UINT ) -1);
}
if(smpd_get_opt_string(argcp, argvp, "-f", filename, SMPD_MAX_FILENAME)){
result = smpd_remove_scps_with_file(filename, hnd);
if(result != SMPD_SUCCESS){
printf("Failed to remove smpd's Service Principal Names (at least one failed) with Domain Controller\n");
ExitProcess((UINT )-1);
}
printf("Removed smpd's Service Principal Names successfully\n");
}
else{
result = smpd_remove_scp(NULL, hnd);
if(result != SMPD_SUCCESS){
printf("Failed to remove smpd's Service Principal Name with Domain Controller\n");
ExitProcess((UINT )-1);
}
printf("Removed smpd's Service Principal Name successfully\n");
}
smpd_spn_list_finalize(&hnd);
ExitProcess(0);
}
if (smpd_get_opt(argcp, argvp, "-mgr"))
{
/* Set a ctrl-handler to kill child processes if this smpd is killed */
if (!SetConsoleCtrlHandler(smpd_ctrl_handler, TRUE))
{
result = GetLastError();
smpd_dbg_printf("unable to set the ctrl handler for the smpd manager, error %d.\n", result);
}
#ifdef HAVE_WINDOWS_H
{
BOOL ret;
ret = smpd_init_affinity_table();
if(!ret){
smpd_dbg_printf("Initializing smpd affinity table failed\n");
}
}
#endif
smpd_process.bService = SMPD_FALSE;
if (!smpd_get_opt_string(argcp, argvp, "-read", read_handle_str, 20))
{
smpd_err_printf("manager started without a read pipe handle.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (!smpd_get_opt_string(argcp, argvp, "-write", write_handle_str, 20))
{
smpd_err_printf("manager started without a write pipe handle.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
hRead = smpd_decode_handle(read_handle_str);
hWrite = smpd_decode_handle(write_handle_str);
smpd_dbg_printf("manager creating listener and session sets.\n");
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_create_set(listener) failed,\nsock error: %s\n", get_sock_error_string(result));
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
smpd_process.set = set;
smpd_dbg_printf("created set for manager listener, %d\n", SMPDU_Sock_get_sock_set_id(set));
port = 0;
result = SMPDU_Sock_listen(set, NULL, &port, &listener);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_listen failed,\nsock error: %s\n", get_sock_error_string(result));
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
smpd_dbg_printf("smpd manager listening on port %d\n", port);
result = smpd_create_context(SMPD_CONTEXT_LISTENER, set, listener, -1, &smpd_process.listener_context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create a context for the smpd listener.\n");
smpd_exit_fn(FCNAME);
return result;
}
result = SMPDU_Sock_set_user_ptr(listener, smpd_process.listener_context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_set_user_ptr failed,\nsock error: %s\n", get_sock_error_string(result));
smpd_exit_fn(FCNAME);
return result;
}
smpd_process.listener_context->state = SMPD_MGR_LISTENING;
memset(str, 0, 20);
snprintf(str, 20, "%d", port);
smpd_dbg_printf("manager writing port back to smpd.\n");
if (!WriteFile(hWrite, str, 20, &num_written, NULL))
{
smpd_err_printf("WriteFile failed, error %d\n", GetLastError());
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
CloseHandle(hWrite);
if (num_written != 20)
{
smpd_err_printf("wrote only %d bytes of 20\n", num_written);
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
smpd_dbg_printf("manager reading account and password from smpd.\n");
if (!ReadFile(hRead, smpd_process.UserAccount, SMPD_MAX_ACCOUNT_LENGTH, &num_read, NULL))
{
smpd_err_printf("ReadFile failed, error %d\n", GetLastError());
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (num_read != SMPD_MAX_ACCOUNT_LENGTH)
{
smpd_err_printf("read only %d bytes of %d\n", num_read, SMPD_MAX_ACCOUNT_LENGTH);
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (!ReadFile(hRead, smpd_process.UserPassword, SMPD_MAX_PASSWORD_LENGTH, &num_read, NULL))
{
smpd_err_printf("ReadFile failed, error %d\n", GetLastError());
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (num_read != SMPD_MAX_PASSWORD_LENGTH)
{
smpd_err_printf("read only %d bytes of %d\n", num_read, SMPD_MAX_PASSWORD_LENGTH);
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (!ReadFile(hRead, smpd_process.passphrase, SMPD_PASSPHRASE_MAX_LENGTH, &num_read, NULL))
{
smpd_err_printf("ReadFile failed, error %d\n", GetLastError());
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
if (num_read != SMPD_PASSPHRASE_MAX_LENGTH)
{
smpd_err_printf("read only %d bytes of %d\n", num_read, SMPD_PASSPHRASE_MAX_LENGTH);
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
smpd_process.credentials_prompt = SMPD_FALSE;
result = smpd_enter_at_state(set, SMPD_MGR_LISTENING);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("state machine failed.\n");
}
/*
result = SMPDU_Sock_finalize();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_finalize failed,\nsock error: %s\n", get_sock_error_string(result));
}
*/
smpd_exit(0);
smpd_exit_fn(FCNAME);
ExitProcess(0);
}
#endif
/* check for the status option */
if (smpd_get_opt_string(argcp, argvp, "-status", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
{
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_DO_STATUS;
}
else if (smpd_get_opt(argcp, argvp, "-status"))
{
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_DO_STATUS;
}
/* check for console options */
if (smpd_get_opt_string(argcp, argvp, "-console", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
{
smpd_process.do_console = 1;
}
else if (smpd_get_opt(argcp, argvp, "-console"))
{
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
}
if (smpd_process.do_console)
{
/* This may need to be changed to avoid conflict */
if (smpd_get_opt(argcp, argvp, "-p"))
{
smpd_process.use_process_session = 1;
}
}
if (smpd_get_opt_string(argcp, argvp, "-shutdown", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
{
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_SHUTDOWN;
}
else if (smpd_get_opt(argcp, argvp, "-shutdown"))
{
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_SHUTDOWN;
}
if (smpd_get_opt_string(argcp, argvp, "-restart", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
{
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_RESTART;
}
else if (smpd_get_opt(argcp, argvp, "-restart"))
{
#ifdef HAVE_WINDOWS_H
printf("restarting the smpd service...\n");
smpd_stop_service();
Sleep(1000);
smpd_start_service();
smpd_exit(0);
#else
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_RESTART;
#endif
}
if (smpd_get_opt_string(argcp, argvp, "-version", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
{
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_VERSION;
}
else if (smpd_get_opt(argcp, argvp, "-version"))
{
printf("%s\n", SMPD_VERSION);
fflush(stdout);
smpd_exit(0);
}
/* These commands are handled by mpiexec although doing them here is an alternate solution.
if (smpd_get_opt_two_strings(argcp, argvp, "-add_job", smpd_process.job_key, SMPD_MAX_NAME_LENGTH, smpd_process.job_key_account, SMPD_MAX_ACCOUNT_LENGTH))
{
if (!smpd_get_opt_string(argcp, argvp, "-host", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
if (smpd_get_opt_string(argcp, argvp, "-password", smpd_process.job_key_password, SMPD_MAX_PASSWORD_LENGTH)
smpd_process.builtin_cmd = SMPD_CMD_ADD_JOB_KEY_AND_PASSWORD;
else
smpd_process.builtin_cmd = SMPD_CMD_ADD_JOB_KEY;
smpd_process.do_console = 1;
}
if (smpd_get_opt_string(argcp, argvp, "-remove_job", smpd_process.job_key, SMPD_MAX_NAME_LENGTH))
{
if (!smpd_get_opt_string(argcp, argvp, "-host", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_REMOVE_JOB_KEY;
}
if (smpd_get_opt_string(argcp, argvp, "-associate_job", smpd_process.job_key, SMPD_MAX_NAME_LENGTH))
{
if (!smpd_get_opt_string(argcp, argvp, "-host", smpd_process.console_host, SMPD_MAX_HOST_LENGTH))
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_ASSOCIATE_JOB_KEY;
}
*/
smpd_get_opt_string(argcp, argvp, "-phrase", smpd_process.passphrase, SMPD_PASSPHRASE_MAX_LENGTH);
if (smpd_get_opt(argcp, argvp, "-getphrase"))
{
printf("passphrase for smpd: ");fflush(stdout);
smpd_get_password(smpd_process.passphrase);
}
if (smpd_get_opt_string(argcp, argvp, "-smpdfile", smpd_process.smpd_filename, SMPD_MAX_FILENAME))
{
struct stat s;
if (stat(smpd_process.smpd_filename, &s) == 0)
{
if (s.st_mode & 00077)
{
printf(".smpd file cannot be readable by anyone other than the current user.\n");
smpd_exit_fn(FCNAME);
return SMPD_FAIL;
}
}
}
if (smpd_get_opt_string(argcp, argvp, "-traceon", filename, SMPD_MAX_FILENAME))
{
smpd_process.do_console_returns = SMPD_TRUE;
if (*argcp > 1)
{
for (i=1; i<*argcp; i++)
{
strcpy(smpd_process.console_host, (*argvp)[i]);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_SET;
strcpy(smpd_process.key, "logfile");
strcpy(smpd_process.val, filename);
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to set the logfile name on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_SET;
strcpy(smpd_process.key, "log");
strcpy(smpd_process.val, "yes");
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to set the log option on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_RESTART;
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to restart the smpd on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
}
}
else
{
result = smpd_set_smpd_data("logfile", filename);
result = smpd_set_smpd_data("log", "yes");
#ifdef HAVE_WINDOWS_H
printf("restarting the smpd service...\n");
smpd_stop_service();
Sleep(1000);
smpd_start_service();
#else
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_RESTART;
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to restart the smpd on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
#endif
}
smpd_exit_fn(FCNAME);
smpd_exit(result);
}
if (smpd_get_opt(argcp, argvp, "-traceoff"))
{
smpd_process.do_console_returns = SMPD_TRUE;
if (*argcp > 1)
{
for (i=1; i<*argcp; i++)
{
strcpy(smpd_process.console_host, (*argvp)[i]);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_SET;
strcpy(smpd_process.key, "log");
strcpy(smpd_process.val, "no");
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to set the log option on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_RESTART;
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to restart the smpd on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
}
}
else
{
result = smpd_set_smpd_data("log", "no");
#ifdef HAVE_WINDOWS_H
printf("restarting the smpd service...\n");
smpd_stop_service();
Sleep(1000);
smpd_start_service();
#else
smpd_get_hostname(smpd_process.console_host, SMPD_MAX_HOST_LENGTH);
smpd_process.do_console = 1;
smpd_process.builtin_cmd = SMPD_CMD_RESTART;
result = smpd_do_console();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to restart the smpd on host '%s'\n", smpd_process.console_host);
smpd_exit_fn(FCNAME);
return result;
}
#endif
}
smpd_exit_fn(FCNAME);
smpd_exit(result);
}
if (smpd_process.do_console)
{
result = smpd_do_console();
smpd_exit_fn(FCNAME);
return result;
}
smpd_exit_fn(FCNAME);
return SMPD_SUCCESS;
}
int smpd_do_console()
{
int result = -1;
smpd_context_t *context;
SMPDU_Sock_set_t set;
SMPDU_Sock_t sock;
SMPD_BOOL no_smpd = SMPD_FALSE;
int saved_state = 0;
int exit_code = 0;
smpd_enter_fn(FCNAME);
/* make sure we have a passphrase to authenticate connections to the smpds */
if (smpd_process.passphrase[0] == '\0')
smpd_get_smpd_data("phrase", smpd_process.passphrase, SMPD_PASSPHRASE_MAX_LENGTH);
if (smpd_process.passphrase[0] == '\0')
{
if (smpd_process.noprompt)
{
printf("Error: No smpd passphrase specified through the registry or .smpd file, exiting.\n");
goto quit_job;
}
printf("Please specify an authentication passphrase for smpd: ");
fflush(stdout);
smpd_get_password(smpd_process.passphrase);
}
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_create_set failed,\nsock error: %s\n", get_sock_error_string(result));
goto quit_job;
}
smpd_process.set = set;
/* set the id of the mpiexec node to zero */
smpd_process.id = 0;
/* turn off output if do_status is selected to supress error messages */
if (smpd_process.builtin_cmd == SMPD_CMD_DO_STATUS)
{
saved_state = smpd_process.dbg_state;
smpd_process.dbg_state = 0;
}
/* start connecting the tree by posting a connect to the first host */
result = smpd_create_context(SMPD_CONTEXT_LEFT_CHILD, set, SMPDU_SOCK_INVALID_SOCK/*sock*/, 1, &context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to create a context.\n");
goto quit_job;
}
result = SMPDU_Sock_post_connect(set, context, smpd_process.console_host, smpd_process.port, &sock);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to connect to '%s:%d',\nsock error: %s\n",
smpd_process.console_host, smpd_process.port, get_sock_error_string(result));
no_smpd = SMPD_TRUE;
goto quit_job;
}
context->sock = sock;
/* turn output back on */
if (smpd_process.builtin_cmd == SMPD_CMD_DO_STATUS)
smpd_process.dbg_state = saved_state;
context->state = SMPD_MPIEXEC_CONNECTING_SMPD;
smpd_process.left_context = context;
/* turn off output if do_status is selected to supress error messages */
if (smpd_process.builtin_cmd == SMPD_CMD_DO_STATUS)
smpd_process.dbg_state = 0;
result = smpd_enter_at_state(set, SMPD_MPIEXEC_CONNECTING_SMPD);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("state machine failed.\n");
no_smpd = SMPD_TRUE;
goto quit_job;
}
/* turn output back on */
if (smpd_process.builtin_cmd == SMPD_CMD_DO_STATUS)
smpd_process.dbg_state = saved_state;
quit_job:
if (result != SMPD_SUCCESS)
{
exit_code = result;
}
if (smpd_process.builtin_cmd == SMPD_CMD_DO_STATUS && (no_smpd || smpd_process.state_machine_ret_val != SMPD_SUCCESS))
{
printf("no smpd running on %s\n", smpd_process.console_host);
smpd_process.dbg_state = saved_state;
}
if (smpd_process.do_console_returns == SMPD_TRUE)
{
smpd_exit_fn(FCNAME);
return exit_code;
}
/* finalize */
/*
smpd_dbg_printf("calling SMPDU_Sock_finalize\n");
result = SMPDU_Sock_finalize();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_finalize failed,\nsock error: %s\n", get_sock_error_string(result));
}
*/
#ifdef HAVE_WINDOWS_H
if (smpd_process.hCloseStdinThreadEvent)
SetEvent(smpd_process.hCloseStdinThreadEvent);
if (smpd_process.hStdinThread != NULL)
{
/* close stdin so the input thread will exit */
CloseHandle(GetStdHandle(STD_INPUT_HANDLE));
if (WaitForSingleObject(smpd_process.hStdinThread, 3000) != WAIT_OBJECT_0)
{
TerminateThread(smpd_process.hStdinThread, 321);
}
CloseHandle(smpd_process.hStdinThread);
}
if (smpd_process.hCloseStdinThreadEvent)
{
CloseHandle(smpd_process.hCloseStdinThreadEvent);
smpd_process.hCloseStdinThreadEvent = NULL;
}
#elif defined(USE_PTHREAD_STDIN_REDIRECTION)
smpd_cancel_stdin_thread();
#endif
smpd_exit_fn(FCNAME);
smpd_exit(exit_code);
return SMPD_SUCCESS;
}
int mpiexec_rsh()
{
int i;
smpd_launch_node_t *launch_node_ptr;
smpd_process_t *process, **processes;
int result;
char *iter1, *iter2;
char exe[SMPD_MAX_EXE_LENGTH];
char *p;
char ssh_cmd[100] = "ssh -x";
SMPDU_Sock_set_t set;
SMPD_BOOL escape_escape = SMPD_TRUE;
char *env_str;
int maxlen;
SMPDU_Sock_t abort_sock;
smpd_context_t *abort_context = NULL;
smpd_command_t *cmd_ptr;
PROCESS_HANDLE_TYPE hnd;
smpd_enter_fn("mpiexec_rsh");
#ifdef HAVE_WINDOWS_H
SetConsoleCtrlHandler(mpiexec_rsh_handler, TRUE);
#else
/* setup a signall hander? */
#endif
p = getenv("MPIEXEC_RSH");
if (p != NULL && strlen(p) > 0){
strncpy(ssh_cmd, p, 100);
}
p = getenv("MPIEXEC_RSH_NO_ESCAPE");
if (p != NULL){
if (smpd_is_affirmative(p) || strcmp(p, "1") == 0){
escape_escape = SMPD_FALSE;
}
}
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create a set for the mpiexec_rsh.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
smpd_process.nproc = smpd_process.launch_list->nproc;
if (smpd_process.use_pmi_server){
result = start_pmi_server(smpd_process.launch_list->nproc, root_host, 100, &root_port);
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh is unable to start the local pmi server.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
smpd_dbg_printf("the pmi server is listening on %s:%d\n", root_host, root_port);
}
else{
/* start the root smpd */
result = start_root_smpd(root_host, SMPD_MAX_HOST_LENGTH, &root_port, &hnd);
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh is unable to start the root smpd.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
smpd_dbg_printf("the root smpd is listening on %s:%d\n", root_host, root_port);
/* create a connection to the root smpd used to abort the job */
result = ConnectToHost(root_host, root_port, SMPD_CONNECTING_RPMI, set, &abort_sock, &abort_context);
if (result != SMPD_SUCCESS){
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
}
processes = (smpd_process_t**)MPIU_Malloc(sizeof(smpd_process_t*) * smpd_process.launch_list->nproc);
if (processes == NULL){
smpd_err_printf("unable to allocate process array.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
launch_node_ptr = smpd_process.launch_list;
for (i=0; i<smpd_process.launch_list->nproc; i++){
if (launch_node_ptr == NULL){
smpd_err_printf("Error: not enough launch nodes.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* initialize process structure */
result = smpd_create_process_struct(i, &process);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create a process structure.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* no need for a pmi context */
if (process->pmi){
smpd_free_context(process->pmi);
}
process->pmi = NULL;
/* change stdout and stderr to rsh behavior:
* write stdout/err directly to stdout/err instead of creating
* an smpd stdout/err command
*/
if (process->out != NULL){
process->out->type = SMPD_CONTEXT_STDOUT_RSH;
}
if (process->err != NULL){
process->err->type = SMPD_CONTEXT_STDERR_RSH;
}
MPIU_Strncpy(process->clique, launch_node_ptr->clique, SMPD_MAX_CLIQUE_LENGTH);
MPIU_Strncpy(process->dir, launch_node_ptr->dir, SMPD_MAX_DIR_LENGTH);
MPIU_Strncpy(process->domain_name, smpd_process.kvs_name, SMPD_MAX_DBS_NAME_LEN);
MPIU_Strncpy(process->env, launch_node_ptr->env, SMPD_MAX_ENV_LENGTH);
if (escape_escape == SMPD_TRUE && smpd_process.mpiexec_run_local != SMPD_TRUE){
/* convert \ to \\ to make cygwin ssh happy */
iter1 = launch_node_ptr->exe;
iter2 = exe;
while (*iter1){
if (*iter1 == '\\'){
*iter2 = *iter1;
iter2++;
*iter2 = *iter1;
}
else{
*iter2 = *iter1;
}
iter1++;
iter2++;
}
*iter2 = '\0';
/*printf("[%s] -> [%s]\n", launch_node_ptr->exe, exe);*/
}
else{
MPIU_Strncpy(exe, launch_node_ptr->exe, SMPD_MAX_EXE_LENGTH);
}
/* Two samples for testing on the local machine */
/* static rPMI initialization */
/*sprintf(process->exe, "env PMI_RANK=%d PMI_SIZE=%d PMI_KVS=%s PMI_ROOT_HOST=%s PMI_ROOT_PORT=8888 PMI_ROOT_LOCAL=1 PMI_APPNUM=%d %s",
launch_node_ptr->iproc, launch_node_ptr->nproc, smpd_process.kvs_name, root_host, launch_node_ptr->appnum, exe);*/
/* dynamic rPMI initialization */
/*sprintf(process->exe, "env PMI_RANK=%d PMI_SIZE=%d PMI_KVS=%s PMI_ROOT_HOST=%s PMI_ROOT_PORT=%d PMI_ROOT_LOCAL=0 PMI_APPNUM=%d %s",
launch_node_ptr->iproc, launch_node_ptr->nproc, smpd_process.kvs_name, root_host, root_port, launch_node_ptr->appnum, exe);*/
if (smpd_process.mpiexec_run_local == SMPD_TRUE){
/* -localonly option and dynamic rPMI initialization */
env_str = &process->env[strlen(process->env)];
maxlen = (int)(SMPD_MAX_ENV_LENGTH - strlen(process->env));
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_RANK", launch_node_ptr->iproc);
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_SIZE", launch_node_ptr->nproc);
MPIU_Str_add_string_arg(&env_str, &maxlen, "PMI_KVS", smpd_process.kvs_name);
MPIU_Str_add_string_arg(&env_str, &maxlen, "PMI_ROOT_HOST", root_host);
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_ROOT_PORT", root_port);
MPIU_Str_add_string_arg(&env_str, &maxlen, "PMI_ROOT_LOCAL", "0");
MPIU_Str_add_int_arg(&env_str, &maxlen, "PMI_APPNUM", launch_node_ptr->appnum);
MPIU_Strncpy(process->exe, exe, SMPD_MAX_EXE_LENGTH);
}
else{
/* ssh and dynamic rPMI initialization */
char fmtEnv[SMPD_MAX_ENV_LENGTH];
int fmtEnvLen = SMPD_MAX_ENV_LENGTH;
char *pExe = process->exe;
int curLen = 0;
MPIU_Snprintf(pExe, SMPD_MAX_EXE_LENGTH, "%s %s env", ssh_cmd, launch_node_ptr->hostname);
curLen = strlen(process->exe);
pExe = process->exe + curLen;
if(FmtEnvVarsForSSH(launch_node_ptr->env, fmtEnv, fmtEnvLen)){
/* Add user specified env vars */
MPIU_Snprintf(pExe, SMPD_MAX_EXE_LENGTH - curLen, "%s", fmtEnv);
curLen = strlen(process->exe);
pExe = process->exe + curLen;
}
MPIU_Snprintf(pExe, SMPD_MAX_EXE_LENGTH - curLen, " \"PMI_RANK=%d\" \"PMI_SIZE=%d\" \"PMI_KVS=%s\" \"PMI_ROOT_HOST=%s\" \"PMI_ROOT_PORT=%d\" \"PMI_ROOT_LOCAL=0\" \"PMI_APPNUM=%d\" %s",
launch_node_ptr->iproc, launch_node_ptr->nproc, smpd_process.kvs_name, root_host, root_port, launch_node_ptr->appnum, exe);
}
MPIU_Strncpy(process->kvs_name, smpd_process.kvs_name, SMPD_MAX_DBS_NAME_LEN);
process->nproc = launch_node_ptr->nproc;
MPIU_Strncpy(process->path, launch_node_ptr->path, SMPD_MAX_PATH_LENGTH);
/* call smpd_launch_process */
smpd_dbg_printf("launching: %s\n", process->exe);
result = smpd_launch_process(process, SMPD_DEFAULT_PRIORITY_CLASS, SMPD_DEFAULT_PRIORITY, SMPD_FALSE, set);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to launch process %d <%s>.\n", i, process->exe);
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* save the new process in the list */
process->next = smpd_process.process_list;
smpd_process.process_list = process;
if (i == 0){
/* start the stdin redirection thread to the first process */
setup_stdin_redirection(process, set);
}
smpd_process.nproc_launched++;
processes[i] = process;
launch_node_ptr = launch_node_ptr->next;
} /* for (i=0; i<smpd_process.launch_list->nproc; i++) */
if (launch_node_ptr != NULL){
smpd_err_printf("Error: too many launch nodes.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
/* Start the timeout mechanism if specified */
if (smpd_process.timeout > 0){
smpd_context_t *reader_context;
SMPDU_Sock_t sock_reader;
SMPDU_SOCK_NATIVE_FD reader, writer;
#ifdef HAVE_WINDOWS_H
/*SOCKET reader, writer;*/
smpd_make_socket_loop((SOCKET*)&reader, (SOCKET*)&writer);
#else
/*int reader, writer;*/
int pair[2];
socketpair(AF_UNIX, SOCK_STREAM, 0, pair);
reader = pair[0];
writer = pair[1];
#endif
result = SMPDU_Sock_native_to_sock(set, reader, NULL, &sock_reader);
result = SMPDU_Sock_native_to_sock(set, writer, NULL, &smpd_process.timeout_sock);
result = smpd_create_context(SMPD_CONTEXT_TIMEOUT, set, sock_reader, -1, &reader_context);
reader_context->read_state = SMPD_READING_TIMEOUT;
result = SMPDU_Sock_post_read(sock_reader, &reader_context->read_cmd.cmd, 1, 1, NULL);
#ifdef HAVE_WINDOWS_H
/* create a Windows thread to sleep until the timeout expires */
smpd_process.timeout_thread = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)timeout_thread, NULL, 0, NULL);
if (smpd_process.timeout_thread == NULL){
printf("Error: unable to create a timeout thread, errno %d.\n", GetLastError());
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
#else /* HAVE_WINDOWS_H */
#ifdef SIGALRM
/* create an alarm to signal mpiexec when the timeout expires */
smpd_signal(SIGALRM, timeout_function);
alarm(smpd_process.timeout);
#else /* SIGALARM */
#ifdef HAVE_PTHREAD_H
/* create a pthread to sleep until the timeout expires */
result = pthread_create(&smpd_process.timeout_thread, NULL, timeout_thread, NULL);
if (result != 0){
printf("Error: unable to create a timeout thread, errno %d.\n", result);
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
#else /* HAVE_PTHREAD_H */
/* no timeout mechanism available */
#endif /* HAVE_PTHREAD_H */
#endif /* SIGALARM */
#endif /* HAVE_WINDOWS_H */
} /* if (smpd_process.timeout > 0) */
result = smpd_enter_at_state(set, SMPD_IDLE);
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh state machine failed.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
if (smpd_process.use_pmi_server){
result = stop_pmi_server();
if (result != SMPD_SUCCESS){
smpd_err_printf("mpiexec_rsh unable to stop the pmi server.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
}
else{
/* Send an abort command to the root_smpd thread/process to insure that it exits.
* This only needs to be sent when there is an error or failed process of some sort
* but it is safe to send it in all cases.
*/
result = smpd_create_command("abort", 0, 0, SMPD_FALSE, &cmd_ptr);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create an abort command.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
result = smpd_post_write_command(abort_context, cmd_ptr);
if (result != SMPD_SUCCESS){
/* Only print this as a debug message instead of an error because the root_smpd thread/process may have already exited. */
smpd_dbg_printf("unable to post a write of the abort command to the %s context.\n", smpd_get_context_str(abort_context));
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
result = stop_root_smpd(hnd);
if (result != PMI_SUCCESS){
smpd_err_printf("mpiexec_rsh unable to stop the root smpd.\n");
smpd_exit_fn("mpiexec_rsh");
return SMPD_FAIL;
}
}
smpd_exit_fn("mpiexec_rsh");
return 0;
}
/* FIXME: Why is this func defined here ?
* - shouldn't this be in smpd_util*.lib ?
*/
static int root_smpd(void *p)
{
int result;
SMPDU_Sock_set_t set;
SMPDU_Sock_t listener;
mpiexec_rsh_rsmpd_args_t *pArgs = NULL;
smpd_process_group_t *pg;
int i, rootPort;
#ifndef HAVE_WINDOWS_H
int send_kvs = 0;
int pipe_fd;
#endif
pArgs = (mpiexec_rsh_rsmpd_args_t *)p;
if(!pArgs){
smpd_err_printf("Invalid args - NULL - to root smpd\n");
return SMPD_FAIL;
}
smpd_process.id = 1;
smpd_process.root_smpd = SMPD_FALSE;
smpd_process.map0to1 = SMPD_TRUE;
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS){
smpd_err_printf(result, "SMPDU_Sock_create_set failed.\n");
return SMPD_FAIL;
}
smpd_process.set = set;
smpd_dbg_printf("created a set for the listener: %d\n", SMPDU_Sock_get_sock_set_id(set));
result = SMPDU_Sock_listen(set, NULL, &rootPort, &listener);
if (result != SMPD_SUCCESS){
smpd_err_printf(result, "SMPDU_Sock_listen failed.\n");
return SMPD_FAIL;
}
smpd_dbg_printf("smpd listening on port %d\n", rootPort);
result = smpd_create_context(SMPD_CONTEXT_LISTENER, set, listener, -1, &smpd_process.listener_context);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to create a context for the smpd listener.\n");
return SMPD_FAIL;
}
result = SMPDU_Sock_set_user_ptr(listener, smpd_process.listener_context);
if (result != SMPD_SUCCESS){
smpd_err_printf(result, "SMPDU_Sock_set_user_ptr failed.\n");
return SMPD_FAIL;
}
smpd_process.listener_context->state = SMPD_SMPD_LISTENING;
smpd_dbs_init();
smpd_process.have_dbs = SMPD_TRUE;
if (smpd_process.kvs_name[0] != '\0'){
result = smpd_dbs_create_name_in(smpd_process.kvs_name);
}
else{
result = smpd_dbs_create(smpd_process.kvs_name);
#ifndef HAVE_WINDOWS_H
send_kvs = 1;
#endif
}
if (result != SMPD_DBS_SUCCESS){
smpd_err_printf("unable to create a kvs database: name = <%s>.\n", smpd_process.kvs_name);
return SMPD_FAIL;
}
/* Set up the process group */
/* initialize a new process group structure */
pg = (smpd_process_group_t*)MPIU_Malloc(sizeof(smpd_process_group_t));
if (pg == NULL){
smpd_err_printf("unable to allocate memory for a process group structure.\n");
return SMPD_FAIL;
}
pg->aborted = SMPD_FALSE;
pg->any_init_received = SMPD_FALSE;
pg->any_noinit_process_exited = SMPD_FALSE;
strncpy(pg->kvs, smpd_process.kvs_name, SMPD_MAX_DBS_NAME_LEN);
pg->num_procs = smpd_process.nproc;
pg->processes = (smpd_exit_process_t*)MPIU_Malloc(smpd_process.nproc * sizeof(smpd_exit_process_t));
if (pg->processes == NULL){
smpd_err_printf("unable to allocate an array of %d process exit structures.\n", smpd_process.nproc);
return SMPD_FAIL;
}
for (i=0; i<smpd_process.nproc; i++){
pg->processes[i].ctx_key[0] = '\0';
pg->processes[i].errmsg = NULL;
pg->processes[i].exitcode = -1;
pg->processes[i].exited = SMPD_FALSE;
pg->processes[i].finalize_called = SMPD_FALSE;
pg->processes[i].init_called = SMPD_FALSE;
pg->processes[i].node_id = i+1;
pg->processes[i].host[0] = '\0';
pg->processes[i].suspended = SMPD_FALSE;
pg->processes[i].suspend_cmd = NULL;
}
/* add the process group to the global list */
pg->next = smpd_process.pg_list;
smpd_process.pg_list = pg;
#ifdef HAVE_WINDOWS_H
*(pArgs->pRootPort) = rootPort;
SetEvent(pArgs->hRootSMPDRdyEvent);
#else
if (p != NULL){
/*pipe_fd = *(int*)p;*/
pipe_fd = pArgs->pipe_fd;
/* send the root port back over the pipe */
writebuf(pipe_fd, &rootPort, sizeof(int));
if (send_kvs){
writebuf(pipe_fd, smpd_process.kvs_name, SMPD_MAX_DBS_NAME_LEN);
}
close(pipe_fd);
}
#endif
result = smpd_enter_at_state(set, SMPD_SMPD_LISTENING);
if (result != SMPD_SUCCESS){
smpd_err_printf("root_smpd state machine failed.\n");
return SMPD_FAIL;
}
result = SMPDU_Sock_destroy_set(set);
if (result != SMPD_SUCCESS){
smpd_err_printf("unable to destroy the set (result = %d)\n", result);
}
return SMPD_SUCCESS;
}
int main(int argc, char* argv[])
{
int result = SMPD_SUCCESS;
smpd_host_node_t *host_node_ptr;
smpd_launch_node_t *launch_node_ptr;
smpd_context_t *context;
SMPDU_Sock_set_t set;
SMPDU_Sock_t sock = SMPDU_SOCK_INVALID_SOCK;
smpd_state_t state;
smpd_enter_fn("main");
/* catch an empty command line */
if (argc < 2)
{
mp_print_options();
exit(0);
}
smpd_process.mpiexec_argv0 = argv[0];
/* initialize */
/* FIXME: Get rid of this hack - we already create
* local KVS for all singleton clients by default
*/
putenv("PMI_SMPD_FD=0");
result = PMPI_Init(&argc, &argv);
/* SMPD_CS_ENTER(); */
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPD_Init failed,\nerror: %d\n", result);
smpd_exit_fn("main");
return result;
}
result = SMPDU_Sock_init();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_init failed,\nsock error: %s\n",
get_sock_error_string(result));
smpd_exit_fn("main");
return result;
}
result = smpd_init_process();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("smpd_init_process failed.\n");
goto quit_job;
}
smpd_process.dbg_state = SMPD_DBG_STATE_ERROUT;
/* parse the command line */
smpd_dbg_printf("parsing the command line.\n");
result = mp_parse_command_args(&argc, &argv);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to parse the mpiexec command arguments.\n");
goto quit_job;
}
/* If we are using MS HPC job scheduler we only connect
* to the local SMPD
*/
if(smpd_process.use_ms_hpc){
char host[100];
int id;
/* Free the current host list */
result = smpd_free_host_list();
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to free the global host list\n");
goto quit_job;
}
/* Add local host to the host list */
result = smpd_get_hostname(host, 100);
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to get the local hostname\n");
goto quit_job;
}
result = smpd_get_host_id(host, &id);
if(result != SMPD_SUCCESS){
smpd_err_printf("Unable to get host id for local host\n");
goto quit_job;
}
/* Set the number of PMI procs since they are not launched by mpiexec */
smpd_process.nproc = smpd_process.launch_list->nproc;
smpd_dbg_printf("Adding (%s:%d) == (localhost) to the host list\n", host, id);
}
/* print and see what we've got */
/* debugging output *************/
smpd_dbg_printf("host tree:\n");
host_node_ptr = smpd_process.host_list;
if (!host_node_ptr)
smpd_dbg_printf("<none>\n");
while (host_node_ptr)
{
smpd_dbg_printf(" host: %s, parent: %d, id: %d\n",
host_node_ptr->host,
host_node_ptr->parent, host_node_ptr->id);
host_node_ptr = host_node_ptr->next;
}
smpd_dbg_printf("launch nodes:\n");
launch_node_ptr = smpd_process.launch_list;
if (!launch_node_ptr)
smpd_dbg_printf("<none>\n");
while (launch_node_ptr)
{
smpd_dbg_printf(" iproc: %d, id: %d, exe: %s\n",
launch_node_ptr->iproc, launch_node_ptr->host_id,
launch_node_ptr->exe);
launch_node_ptr = launch_node_ptr->next;
}
/* end debug output *************/
/* set the id of the mpiexec node to zero */
smpd_process.id = 0;
result = SMPDU_Sock_create_set(&set);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_create_set failed,\nsock error: %s\n", get_sock_error_string(result));
goto quit_job;
}
smpd_process.set = set;
/* Check to see if the user wants to use a remote shell mechanism for launching the processes
* instead of using the smpd process managers.
*/
if (smpd_process.rsh_mpiexec == SMPD_TRUE)
{
/* Do rsh or localonly stuff */
result = mpiexec_rsh();
/* skip over the non-rsh code and go to the cleanup section */
goto quit_job;
}
/* Start the timeout mechanism if specified */
/* This code occurs after the rsh_mpiexec option check because the rsh code implementes timeouts differently */
if (smpd_process.timeout > 0)
{
#ifdef HAVE_WINDOWS_H
/* create a Windows thread to sleep until the timeout expires */
if (smpd_process.timeout_thread == NULL)
{
smpd_process.timeout_thread = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)timeout_thread, NULL, 0, NULL);
if (smpd_process.timeout_thread == NULL)
{
printf("Error: unable to create a timeout thread, errno %d.\n", GetLastError());
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
}
#elif defined(SIGALRM)
/* create an alarm to signal mpiexec when the timeout expires */
smpd_signal(SIGALRM, timeout_function);
alarm(smpd_process.timeout);
#elif defined(HAVE_PTHREAD_H)
/* create a pthread to sleep until the timeout expires */
result = pthread_create(&smpd_process.timeout_thread, NULL, timeout_thread, NULL);
if (result != 0)
{
printf("Error: unable to create a timeout thread, errno %d.\n", result);
smpd_exit_fn("mp_parse_command_args");
return SMPD_FAIL;
}
#else
/* no timeout mechanism available */
#endif
}
/* make sure we have a passphrase to authenticate connections to the smpds */
if (smpd_process.passphrase[0] == '\0')
smpd_get_smpd_data("phrase", smpd_process.passphrase, SMPD_PASSPHRASE_MAX_LENGTH);
if (smpd_process.passphrase[0] == '\0')
{
if (smpd_process.noprompt)
{
printf("Error: No smpd passphrase specified through the registry or .smpd file, exiting.\n");
result = SMPD_FAIL;
goto quit_job;
}
printf("Please specify an authentication passphrase for smpd: ");
fflush(stdout);
smpd_get_password(smpd_process.passphrase);
}
/* set the state to create a console session or a job session */
state = smpd_process.do_console ? SMPD_MPIEXEC_CONNECTING_SMPD : SMPD_MPIEXEC_CONNECTING_TREE;
result = smpd_create_context(SMPD_CONTEXT_LEFT_CHILD, set, sock, 1, &context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create a context for the first host in the tree.\n");
goto quit_job;
}
#ifdef HAVE_WINDOWS_H
if (!smpd_process.local_root)
{
#endif
/* start connecting the tree by posting a connect to the first host */
result = SMPDU_Sock_post_connect(set, context, smpd_process.host_list->host, smpd_process.port, &sock);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to connect to '%s:%d',\nsock error: %s\n",
smpd_process.host_list->host, smpd_process.port, get_sock_error_string(result));
goto quit_job;
}
#ifdef HAVE_WINDOWS_H
}
#endif
context->sock = sock;
context->state = state;
context->connect_to = smpd_process.host_list;
#ifdef HAVE_WINDOWS_H
if (smpd_process.local_root)
{
int port;
smpd_context_t *rc_context;
/* The local_root option is implemented by having mpiexec act as the smpd
* and launch the smpd manager. Then mpiexec connects to this manager just
* as if it had been created by a real smpd. This causes all the processes
* destined for the first smpd host to be launched by this child process of
* mpiexec and not the smpd service. This allows for these processes to
* create windows that are visible to the interactive user. It also means
* that the job cannot be run in the context of a user other than the user
* running mpiexec. */
/* get the path to smpd.exe because pszExe is currently mpiexec.exe */
smpd_get_smpd_data("binary", smpd_process.pszExe, SMPD_MAX_EXE_LENGTH);
/* launch the manager process */
result = smpd_start_win_mgr(context, SMPD_FALSE);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to start the local smpd manager.\n");
goto quit_job;
}
/* connect to the manager */
smpd_dbg_printf("connecting a new socket.\n");
port = atol(context->port_str);
if (port < 1)
{
smpd_err_printf("Invalid reconnect port read: %d\n", port);
goto quit_job;
}
result = smpd_create_context(context->type, context->set, SMPDU_SOCK_INVALID_SOCK, context->id, &rc_context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to create a new context for the reconnection.\n");
goto quit_job;
}
rc_context->state = context->state;
rc_context->write_state = SMPD_RECONNECTING;
context->state = SMPD_CLOSING;
rc_context->connect_to = context->connect_to;
rc_context->connect_return_id = context->connect_return_id;
rc_context->connect_return_tag = context->connect_return_tag;
strcpy(rc_context->host, context->host);
smpd_process.left_context = rc_context;
smpd_dbg_printf("posting a re-connect to %s:%d in %s context.\n", rc_context->connect_to->host, port, smpd_get_context_str(rc_context));
result = SMPDU_Sock_post_connect(rc_context->set, rc_context, rc_context->connect_to->host, port, &rc_context->sock);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("Unable to post a connect to '%s:%d',\nsock error: %s\n",
rc_context->connect_to->host, port, get_sock_error_string(result));
if (smpd_post_abort_command("Unable to connect to '%s:%d',\nsock error: %s\n",
rc_context->connect_to->host, port, get_sock_error_string(result)) != SMPD_SUCCESS)
{
goto quit_job;
}
}
}
else
{
#endif
smpd_process.left_context = context;
result = SMPDU_Sock_set_user_ptr(sock, context);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("unable to set the smpd sock user pointer,\nsock error: %s\n",
get_sock_error_string(result));
goto quit_job;
}
#ifdef HAVE_WINDOWS_H
}
#endif
#ifdef HAVE_WINDOWS_H
{
/* Create a break handler and a socket to handle aborting the job when mpiexec receives break signals */
smpd_context_t *reader_context;
SMPDU_Sock_t sock_reader;
SMPDU_SOCK_NATIVE_FD reader, writer;
smpd_make_socket_loop((SOCKET*)&reader, (SOCKET*)&writer);
result = SMPDU_Sock_native_to_sock(set, reader, NULL, &sock_reader);
result = SMPDU_Sock_native_to_sock(set, writer, NULL, &smpd_process.mpiexec_abort_sock);
result = smpd_create_context(SMPD_CONTEXT_MPIEXEC_ABORT, set, sock_reader, -1, &reader_context);
reader_context->read_state = SMPD_READING_MPIEXEC_ABORT;
result = SMPDU_Sock_post_read(sock_reader, &reader_context->read_cmd.cmd, 1, 1, NULL);
if (!SetConsoleCtrlHandler(mpiexec_ctrl_handler, TRUE))
{
/* Don't error out; allow the job to run without a ctrl handler? */
result = GetLastError();
smpd_dbg_printf("unable to set a ctrl handler for mpiexec, error %d\n", result);
}
}
#endif
result = smpd_enter_at_state(set, state);
if (result != SMPD_SUCCESS)
{
smpd_err_printf("state machine failed.\n");
goto quit_job;
}
quit_job:
if ((result != SMPD_SUCCESS) && (smpd_process.mpiexec_exit_code == 0))
{
smpd_process.mpiexec_exit_code = -1;
}
/* finalize */
smpd_dbg_printf("calling SMPDU_Sock_finalize\n");
result = SMPDU_Sock_finalize();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPDU_Sock_finalize failed,\nsock error: %s\n", get_sock_error_string(result));
}
/* SMPD_Finalize called in smpd_exit()
smpd_dbg_printf("calling SMPD_Finalize\n");
result = PMPI_Finalize();
if (result != SMPD_SUCCESS)
{
smpd_err_printf("SMPD_Finalize failed,\nerror: %d\n", result);
}
*/
#ifdef HAVE_WINDOWS_H
if (smpd_process.hCloseStdinThreadEvent)
SetEvent(smpd_process.hCloseStdinThreadEvent);
if (smpd_process.hStdinThread != NULL)
{
/* close stdin so the input thread will exit */
CloseHandle(GetStdHandle(STD_INPUT_HANDLE));
if (WaitForSingleObject(smpd_process.hStdinThread, 3000) != WAIT_OBJECT_0)
{
TerminateThread(smpd_process.hStdinThread, 321);
}
CloseHandle(smpd_process.hStdinThread);
}
if (smpd_process.hCloseStdinThreadEvent)
{
CloseHandle(smpd_process.hCloseStdinThreadEvent);
smpd_process.hCloseStdinThreadEvent = NULL;
}
#elif defined(USE_PTHREAD_STDIN_REDIRECTION)
smpd_cancel_stdin_thread();
#endif
smpd_exit_fn("main");
/* SMPD_CS_EXIT(); */
return smpd_exit(smpd_process.mpiexec_exit_code);
}