| 1 | /* -*- Mode: C; c-basic-offset:4 ; -*- */ |
|---|
| 2 | /* |
|---|
| 3 | * (C) 2001 by Argonne National Laboratory. |
|---|
| 4 | * See COPYRIGHT in top-level directory. |
|---|
| 5 | */ |
|---|
| 6 | |
|---|
| 7 | /*********************** PMI implementation ********************************/ |
|---|
| 8 | /* |
|---|
| 9 | * This file implements the client-side of the PMI interface. |
|---|
| 10 | * |
|---|
| 11 | * Note that the PMI client code must not print error messages (except |
|---|
| 12 | * when an abort is required) because MPI error handling is based on |
|---|
| 13 | * reporting error codes to which messages are attached. |
|---|
| 14 | * |
|---|
| 15 | * In v2, we should require a PMI client interface to use MPI error codes |
|---|
| 16 | * to provide better integration with MPICH2. |
|---|
| 17 | */ |
|---|
| 18 | /***************************************************************************/ |
|---|
| 19 | |
|---|
| 20 | #include "pmiconf.h" |
|---|
| 21 | |
|---|
| 22 | #define PMI_VERSION 1 |
|---|
| 23 | #define PMI_SUBVERSION 1 |
|---|
| 24 | |
|---|
| 25 | #include <stdio.h> |
|---|
| 26 | #ifdef HAVE_UNISTD_H |
|---|
| 27 | #include <unistd.h> |
|---|
| 28 | #endif |
|---|
| 29 | #ifdef HAVE_STDLIB_H |
|---|
| 30 | #include <stdlib.h> |
|---|
| 31 | #endif |
|---|
| 32 | #ifdef HAVE_STRING_H |
|---|
| 33 | #include <string.h> |
|---|
| 34 | #endif |
|---|
| 35 | #ifdef HAVE_STRINGS_H |
|---|
| 36 | #include <strings.h> |
|---|
| 37 | #endif |
|---|
| 38 | #ifdef USE_PMI_PORT |
|---|
| 39 | #ifndef MAXHOSTNAME |
|---|
| 40 | #define MAXHOSTNAME 256 |
|---|
| 41 | #endif |
|---|
| 42 | #endif |
|---|
| 43 | /* This should be moved to pmiu for shutdown */ |
|---|
| 44 | #if defined(HAVE_SYS_SOCKET_H) |
|---|
| 45 | #include <sys/socket.h> |
|---|
| 46 | #endif |
|---|
| 47 | |
|---|
| 48 | #include "mpibase.h" /* Get ATTRIBUTE, some base functions */ |
|---|
| 49 | /* mpimem includes the definitions for MPIU_Snprintf, MPIU_Malloc, and |
|---|
| 50 | MPIU_Free */ |
|---|
| 51 | #include "mpimem.h" |
|---|
| 52 | |
|---|
| 53 | /* Temporary debug definitions */ |
|---|
| 54 | #if 0 |
|---|
| 55 | #define DBG_PRINTF(args) printf args ; fflush(stdout) |
|---|
| 56 | #else |
|---|
| 57 | #define DBG_PRINTF(args) |
|---|
| 58 | #endif |
|---|
| 59 | |
|---|
| 60 | #include "pmi.h" |
|---|
| 61 | #include "simple_pmiutil.h" |
|---|
| 62 | #include "mpi.h" /* to get MPI_MAX_PORT_NAME */ |
|---|
| 63 | |
|---|
| 64 | /* |
|---|
| 65 | These are global variable used *ONLY* in this file, and are hence |
|---|
| 66 | declared static. |
|---|
| 67 | */ |
|---|
| 68 | |
|---|
| 69 | |
|---|
| 70 | static int PMI_fd = -1; |
|---|
| 71 | static int PMI_size = 1; |
|---|
| 72 | static int PMI_rank = 0; |
|---|
| 73 | |
|---|
| 74 | /* Set PMI_initialized to 1 for singleton init but no process manager |
|---|
| 75 | to help. Initialized to 2 for normal initialization. Initialized |
|---|
| 76 | to values higher than 2 when singleton_init by a process manager. |
|---|
| 77 | All values higher than 1 invlove a PM in some way. |
|---|
| 78 | */ |
|---|
| 79 | typedef enum { PMI_UNINITIALIZED = 0, |
|---|
| 80 | SINGLETON_INIT_BUT_NO_PM = 1, |
|---|
| 81 | NORMAL_INIT_WITH_PM, |
|---|
| 82 | SINGLETON_INIT_WITH_PM } PMIState; |
|---|
| 83 | static PMIState PMI_initialized = PMI_UNINITIALIZED; |
|---|
| 84 | |
|---|
| 85 | /* ALL GLOBAL VARIABLES MUST BE INITIALIZED TO AVOID POLLUTING THE |
|---|
| 86 | LIBRARY WITH COMMON SYMBOLS */ |
|---|
| 87 | static int PMI_kvsname_max = 0; |
|---|
| 88 | static int PMI_keylen_max = 0; |
|---|
| 89 | static int PMI_vallen_max = 0; |
|---|
| 90 | |
|---|
| 91 | static int PMI_iter_next_idx = 0; |
|---|
| 92 | static int PMI_debug = 0; |
|---|
| 93 | static int PMI_debug_init = 0; /* Set this to true to debug the init |
|---|
| 94 | handshakes */ |
|---|
| 95 | static int PMI_spawned = 0; |
|---|
| 96 | |
|---|
| 97 | /* Function prototypes for internal routines */ |
|---|
| 98 | static int PMII_getmaxes( int *kvsname_max, int *keylen_max, int *vallen_max ); |
|---|
| 99 | static int PMII_iter( const char *kvsname, const int idx, int *nextidx, |
|---|
| 100 | char *key, int key_len, char *val, int val_len ); |
|---|
| 101 | static int PMII_Set_from_port( int, int ); |
|---|
| 102 | static int PMII_Connect_to_pm( char *, int ); |
|---|
| 103 | |
|---|
| 104 | static int GetResponse( const char [], const char [], int ); |
|---|
| 105 | static int getPMIFD( int * ); |
|---|
| 106 | |
|---|
| 107 | #ifdef USE_PMI_PORT |
|---|
| 108 | static int PMII_singinit(void); |
|---|
| 109 | static int PMI_totalview = 0; |
|---|
| 110 | #endif |
|---|
| 111 | static int PMIi_InitIfSingleton(void); |
|---|
| 112 | static int accept_one_connection(int); |
|---|
| 113 | static char cached_singinit_key[PMIU_MAXLINE]; |
|---|
| 114 | static char cached_singinit_val[PMIU_MAXLINE]; |
|---|
| 115 | static char singinit_kvsname[256]; |
|---|
| 116 | |
|---|
| 117 | /******************************** Group functions *************************/ |
|---|
| 118 | |
|---|
| 119 | int PMI_Init( int *spawned ) |
|---|
| 120 | { |
|---|
| 121 | char *p; |
|---|
| 122 | int notset = 1; |
|---|
| 123 | int rc; |
|---|
| 124 | |
|---|
| 125 | /* FIXME: Why is setvbuf commented out? */ |
|---|
| 126 | /* FIXME: What if the output should be fully buffered (directed to file)? |
|---|
| 127 | unbuffered (user explicitly set?) */ |
|---|
| 128 | /* setvbuf(stdout,0,_IONBF,0); */ |
|---|
| 129 | setbuf(stdout,NULL); |
|---|
| 130 | /* PMIU_printf( 1, "PMI_INIT\n" ); */ |
|---|
| 131 | |
|---|
| 132 | /* Get the value of PMI_DEBUG from the environment if possible, since |
|---|
| 133 | we may have set it to help debug the setup process */ |
|---|
| 134 | p = getenv( "PMI_DEBUG" ); |
|---|
| 135 | if (p) PMI_debug = atoi( p ); |
|---|
| 136 | |
|---|
| 137 | /* Get the fd for PMI commands; if none, we're a singleton */ |
|---|
| 138 | rc = getPMIFD(¬set); |
|---|
| 139 | if (rc) { |
|---|
| 140 | return rc; |
|---|
| 141 | } |
|---|
| 142 | |
|---|
| 143 | if ( PMI_fd == -1 ) { |
|---|
| 144 | /* Singleton init: Process not started with mpiexec, |
|---|
| 145 | so set size to 1, rank to 0 */ |
|---|
| 146 | PMI_size = 1; |
|---|
| 147 | PMI_rank = 0; |
|---|
| 148 | *spawned = 0; |
|---|
| 149 | |
|---|
| 150 | PMI_initialized = SINGLETON_INIT_BUT_NO_PM; |
|---|
| 151 | /* 256 is picked as the minimum allowed length by the PMI servers */ |
|---|
| 152 | PMI_kvsname_max = 256; |
|---|
| 153 | PMI_keylen_max = 256; |
|---|
| 154 | PMI_vallen_max = 256; |
|---|
| 155 | |
|---|
| 156 | return( 0 ); |
|---|
| 157 | } |
|---|
| 158 | |
|---|
| 159 | /* If size, rank, and debug are not set from a communication port, |
|---|
| 160 | use the environment */ |
|---|
| 161 | if (notset) { |
|---|
| 162 | if ( ( p = getenv( "PMI_SIZE" ) ) ) |
|---|
| 163 | PMI_size = atoi( p ); |
|---|
| 164 | else |
|---|
| 165 | PMI_size = 1; |
|---|
| 166 | |
|---|
| 167 | if ( ( p = getenv( "PMI_RANK" ) ) ) { |
|---|
| 168 | PMI_rank = atoi( p ); |
|---|
| 169 | /* Let the util routine know the rank of this process for |
|---|
| 170 | any messages (usually debugging or error) */ |
|---|
| 171 | PMIU_Set_rank( PMI_rank ); |
|---|
| 172 | } |
|---|
| 173 | else |
|---|
| 174 | PMI_rank = 0; |
|---|
| 175 | |
|---|
| 176 | if ( ( p = getenv( "PMI_DEBUG" ) ) ) |
|---|
| 177 | PMI_debug = atoi( p ); |
|---|
| 178 | else |
|---|
| 179 | PMI_debug = 0; |
|---|
| 180 | |
|---|
| 181 | /* Leave unchanged otherwise, which indicates that no value |
|---|
| 182 | was set */ |
|---|
| 183 | } |
|---|
| 184 | |
|---|
| 185 | /* FIXME: Why does this depend on their being a port??? */ |
|---|
| 186 | /* FIXME: What is this for? */ |
|---|
| 187 | #ifdef USE_PMI_PORT |
|---|
| 188 | if ( ( p = getenv( "PMI_TOTALVIEW" ) ) ) |
|---|
| 189 | PMI_totalview = atoi( p ); |
|---|
| 190 | if ( PMI_totalview ) { |
|---|
| 191 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 192 | /* FIXME: This should use a cmd/response rather than a expecting the |
|---|
| 193 | server to set a value in this and only this case */ |
|---|
| 194 | /* FIXME: And it most ceratainly should not happen *before* the |
|---|
| 195 | initialization handshake */ |
|---|
| 196 | PMIU_readline( PMI_fd, buf, PMIU_MAXLINE ); |
|---|
| 197 | PMIU_parse_keyvals( buf ); |
|---|
| 198 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 199 | if ( strncmp( cmd, "tv_ready", PMIU_MAXLINE ) != 0 ) { |
|---|
| 200 | PMIU_printf( 1, "expecting cmd=tv_ready, got %s\n", buf ); |
|---|
| 201 | return( PMI_FAIL ); |
|---|
| 202 | } |
|---|
| 203 | } |
|---|
| 204 | #endif |
|---|
| 205 | |
|---|
| 206 | PMII_getmaxes( &PMI_kvsname_max, &PMI_keylen_max, &PMI_vallen_max ); |
|---|
| 207 | |
|---|
| 208 | /* FIXME: This is something that the PM should tell the process, |
|---|
| 209 | rather than deliver it through the environment */ |
|---|
| 210 | if ( ( p = getenv( "PMI_SPAWNED" ) ) ) |
|---|
| 211 | PMI_spawned = atoi( p ); |
|---|
| 212 | else |
|---|
| 213 | PMI_spawned = 0; |
|---|
| 214 | if (PMI_spawned) |
|---|
| 215 | *spawned = 1; |
|---|
| 216 | else |
|---|
| 217 | *spawned = 0; |
|---|
| 218 | |
|---|
| 219 | if ( ! PMI_initialized ) |
|---|
| 220 | PMI_initialized = NORMAL_INIT_WITH_PM; |
|---|
| 221 | |
|---|
| 222 | return( 0 ); |
|---|
| 223 | } |
|---|
| 224 | |
|---|
| 225 | int PMI_Initialized( PMI_BOOL *initialized ) |
|---|
| 226 | { |
|---|
| 227 | /* Turn this into a logical value (1 or 0) . This allows us |
|---|
| 228 | to use PMI_initialized to distinguish between initialized with |
|---|
| 229 | an PMI service (e.g., via mpiexec) and the singleton init, |
|---|
| 230 | which has no PMI service */ |
|---|
| 231 | *initialized = PMI_initialized != 0 ? PMI_TRUE : PMI_FALSE; |
|---|
| 232 | return PMI_SUCCESS; |
|---|
| 233 | } |
|---|
| 234 | |
|---|
| 235 | int PMI_Get_size( int *size ) |
|---|
| 236 | { |
|---|
| 237 | if ( PMI_initialized ) |
|---|
| 238 | *size = PMI_size; |
|---|
| 239 | else |
|---|
| 240 | *size = 1; |
|---|
| 241 | return( 0 ); |
|---|
| 242 | } |
|---|
| 243 | |
|---|
| 244 | int PMI_Get_rank( int *rank ) |
|---|
| 245 | { |
|---|
| 246 | if ( PMI_initialized ) |
|---|
| 247 | *rank = PMI_rank; |
|---|
| 248 | else |
|---|
| 249 | *rank = 0; |
|---|
| 250 | return( 0 ); |
|---|
| 251 | } |
|---|
| 252 | |
|---|
| 253 | /* |
|---|
| 254 | * Get_universe_size is one of the routines that needs to communicate |
|---|
| 255 | * with the process manager. If we started as a singleton init, then |
|---|
| 256 | * we first need to connect to the process manager and acquire the |
|---|
| 257 | * needed information. |
|---|
| 258 | */ |
|---|
| 259 | int PMI_Get_universe_size( int *size) |
|---|
| 260 | { |
|---|
| 261 | int err; |
|---|
| 262 | char size_c[PMIU_MAXLINE]; |
|---|
| 263 | |
|---|
| 264 | /* Connect to the PM if we haven't already */ |
|---|
| 265 | if (PMIi_InitIfSingleton() != 0) return -1; |
|---|
| 266 | |
|---|
| 267 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 268 | err = GetResponse( "cmd=get_universe_size\n", "universe_size", 0 ); |
|---|
| 269 | if (err == PMI_SUCCESS) { |
|---|
| 270 | PMIU_getval( "size", size_c, PMIU_MAXLINE ); |
|---|
| 271 | *size = atoi(size_c); |
|---|
| 272 | return( PMI_SUCCESS ); |
|---|
| 273 | } |
|---|
| 274 | else return err; |
|---|
| 275 | } |
|---|
| 276 | else |
|---|
| 277 | *size = 1; |
|---|
| 278 | return( PMI_SUCCESS ); |
|---|
| 279 | } |
|---|
| 280 | |
|---|
| 281 | int PMI_Get_appnum( int *appnum ) |
|---|
| 282 | { |
|---|
| 283 | int err; |
|---|
| 284 | char appnum_c[PMIU_MAXLINE]; |
|---|
| 285 | |
|---|
| 286 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 287 | err = GetResponse( "cmd=get_appnum\n", "appnum", 0 ); |
|---|
| 288 | if (err == PMI_SUCCESS) { |
|---|
| 289 | PMIU_getval( "appnum", appnum_c, PMIU_MAXLINE ); |
|---|
| 290 | *appnum = atoi(appnum_c); |
|---|
| 291 | return( PMI_SUCCESS ); |
|---|
| 292 | } |
|---|
| 293 | else return err; |
|---|
| 294 | |
|---|
| 295 | } |
|---|
| 296 | else |
|---|
| 297 | *appnum = -1; |
|---|
| 298 | |
|---|
| 299 | return( PMI_SUCCESS ); |
|---|
| 300 | } |
|---|
| 301 | |
|---|
| 302 | int PMI_Barrier( void ) |
|---|
| 303 | { |
|---|
| 304 | int err = PMI_SUCCESS; |
|---|
| 305 | |
|---|
| 306 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 307 | err = GetResponse( "cmd=barrier_in\n", "barrier_out", 0 ); |
|---|
| 308 | } |
|---|
| 309 | |
|---|
| 310 | return err; |
|---|
| 311 | } |
|---|
| 312 | |
|---|
| 313 | /* */ |
|---|
| 314 | static int clique_size=-2, *clique_ranks =0; |
|---|
| 315 | |
|---|
| 316 | /* pmiPrivateLocalRanks_<r> gets the local ranks for this process */ |
|---|
| 317 | int PMI_Get_clique_size( int *size ) |
|---|
| 318 | { |
|---|
| 319 | #if 1 |
|---|
| 320 | char buf[PMIU_MAXLINE]; |
|---|
| 321 | char pmi_kvsname[1024]; |
|---|
| 322 | int i, rc, err; |
|---|
| 323 | |
|---|
| 324 | /* As the server for the information on the */ |
|---|
| 325 | if (clique_size == -2 && PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 326 | PMI_KVS_Get_my_name( pmi_kvsname, sizeof(pmi_kvsname) ); |
|---|
| 327 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, |
|---|
| 328 | "cmd=get kvsname=%s key=pmiPrivateLocalRanks_%d\n", |
|---|
| 329 | pmi_kvsname, PMI_rank ); |
|---|
| 330 | if (rc < 0) return PMI_FAIL; |
|---|
| 331 | |
|---|
| 332 | err = GetResponse( buf, "get_result", 0 ); |
|---|
| 333 | if (err == PMI_SUCCESS) { |
|---|
| 334 | PMIU_getval( "rc", buf, PMIU_MAXLINE ); |
|---|
| 335 | rc = atoi( buf ); |
|---|
| 336 | if ( rc == 0 ) { |
|---|
| 337 | char *p = buf, *p0; |
|---|
| 338 | /* Allocate clique_ranks and fill it in */ |
|---|
| 339 | PMIU_getval( "value", buf, PMIU_MAXLINE ); |
|---|
| 340 | /* Count the number of ranks and allocate the space for them */ |
|---|
| 341 | clique_size = 1; |
|---|
| 342 | while (*p) { |
|---|
| 343 | if (*p++ == ',') clique_size++; |
|---|
| 344 | } |
|---|
| 345 | clique_ranks = (int *)MPIU_Malloc( clique_size * sizeof(int) ); |
|---|
| 346 | DBG_PRINTF( ("Clique_size = %d\n", clique_size) ); |
|---|
| 347 | p0 = p = buf; |
|---|
| 348 | i = 0; |
|---|
| 349 | while (*p) { |
|---|
| 350 | while (*p && *p != ',') p++; |
|---|
| 351 | if (*p == ',') *p++ = 0; |
|---|
| 352 | clique_ranks[i++] = atoi(p0); |
|---|
| 353 | p0 = p; |
|---|
| 354 | } |
|---|
| 355 | } |
|---|
| 356 | else { |
|---|
| 357 | /* Default case (PM did not understand request) */ |
|---|
| 358 | clique_size = 1; |
|---|
| 359 | } |
|---|
| 360 | } |
|---|
| 361 | } |
|---|
| 362 | if (clique_size < 0) *size = 1; |
|---|
| 363 | else *size = clique_size; |
|---|
| 364 | #else |
|---|
| 365 | *size = 1; |
|---|
| 366 | #endif |
|---|
| 367 | return PMI_SUCCESS; |
|---|
| 368 | } |
|---|
| 369 | |
|---|
| 370 | int PMI_Get_clique_ranks( int ranks[], int length ) |
|---|
| 371 | { |
|---|
| 372 | #if 1 |
|---|
| 373 | int i; |
|---|
| 374 | if (length < 1) |
|---|
| 375 | return PMI_ERR_INVALID_ARG; |
|---|
| 376 | |
|---|
| 377 | if (clique_size > 0 && clique_ranks) { |
|---|
| 378 | for (i=0; i<length && i<clique_size; i++) |
|---|
| 379 | ranks[i] = clique_ranks[i]; |
|---|
| 380 | } |
|---|
| 381 | else |
|---|
| 382 | ranks[0] = PMI_rank; |
|---|
| 383 | return PMI_SUCCESS; |
|---|
| 384 | #else |
|---|
| 385 | if ( length < 1 ) |
|---|
| 386 | return PMI_ERR_INVALID_ARG; |
|---|
| 387 | else |
|---|
| 388 | return PMI_Get_rank( &ranks[0] ); |
|---|
| 389 | #endif |
|---|
| 390 | } |
|---|
| 391 | |
|---|
| 392 | /* Inform the process manager that we're in finalize */ |
|---|
| 393 | int PMI_Finalize( void ) |
|---|
| 394 | { |
|---|
| 395 | int err = PMI_SUCCESS; |
|---|
| 396 | |
|---|
| 397 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 398 | err = GetResponse( "cmd=finalize\n", "finalize_ack", 0 ); |
|---|
| 399 | shutdown( PMI_fd, SHUT_RDWR ); |
|---|
| 400 | close( PMI_fd ); |
|---|
| 401 | } |
|---|
| 402 | /* Free any memory that we've allocated */ |
|---|
| 403 | if (clique_ranks) MPIU_Free( clique_ranks ); |
|---|
| 404 | |
|---|
| 405 | return err; |
|---|
| 406 | } |
|---|
| 407 | |
|---|
| 408 | int PMI_Abort(int exit_code, const char error_msg[]) |
|---|
| 409 | { |
|---|
| 410 | PMIU_printf(1, "aborting job:\n%s\n", error_msg); |
|---|
| 411 | MPIU_Exit(exit_code); |
|---|
| 412 | return -1; |
|---|
| 413 | } |
|---|
| 414 | |
|---|
| 415 | /************************************* Keymap functions **********************/ |
|---|
| 416 | |
|---|
| 417 | /*FIXME: need to return an error if the value of the kvs name returned is |
|---|
| 418 | truncated because it is larger than length */ |
|---|
| 419 | /* FIXME: My name should be cached rather than re-acquired, as it is |
|---|
| 420 | unchanging (after singleton init) */ |
|---|
| 421 | int PMI_KVS_Get_my_name( char kvsname[], int length ) |
|---|
| 422 | { |
|---|
| 423 | int err; |
|---|
| 424 | |
|---|
| 425 | if (PMI_initialized == SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 426 | /* Return a dummy name */ |
|---|
| 427 | /* FIXME: We need to support a distinct kvsname for each |
|---|
| 428 | process group */ |
|---|
| 429 | /* FIXME: Should the length be length (from the arg list) |
|---|
| 430 | instead of PMIU_MAXLINE? */ |
|---|
| 431 | MPIU_Snprintf( kvsname, PMIU_MAXLINE, "singinit_kvs_%d_0", |
|---|
| 432 | (int)getpid() ); |
|---|
| 433 | return 0; |
|---|
| 434 | } |
|---|
| 435 | err = GetResponse( "cmd=get_my_kvsname\n", "my_kvsname", 0 ); |
|---|
| 436 | if (err == PMI_SUCCESS) { |
|---|
| 437 | PMIU_getval( "kvsname", kvsname, length ); |
|---|
| 438 | } |
|---|
| 439 | return err; |
|---|
| 440 | } |
|---|
| 441 | |
|---|
| 442 | int PMI_KVS_Get_name_length_max( int *maxlen ) |
|---|
| 443 | { |
|---|
| 444 | if (maxlen == NULL) |
|---|
| 445 | return PMI_ERR_INVALID_ARG; |
|---|
| 446 | *maxlen = PMI_kvsname_max; |
|---|
| 447 | return PMI_SUCCESS; |
|---|
| 448 | } |
|---|
| 449 | |
|---|
| 450 | int PMI_KVS_Get_key_length_max( int *maxlen ) |
|---|
| 451 | { |
|---|
| 452 | if (maxlen == NULL) |
|---|
| 453 | return PMI_ERR_INVALID_ARG; |
|---|
| 454 | *maxlen = PMI_keylen_max; |
|---|
| 455 | return PMI_SUCCESS; |
|---|
| 456 | } |
|---|
| 457 | |
|---|
| 458 | int PMI_KVS_Get_value_length_max( int *maxlen ) |
|---|
| 459 | { |
|---|
| 460 | if (maxlen == NULL) |
|---|
| 461 | return PMI_ERR_INVALID_ARG; |
|---|
| 462 | *maxlen = PMI_vallen_max; |
|---|
| 463 | return PMI_SUCCESS; |
|---|
| 464 | } |
|---|
| 465 | |
|---|
| 466 | /* We will use the default kvsname for both the kvs_domain_id and for the id */ |
|---|
| 467 | /* Hence the implementation of the following three functions */ |
|---|
| 468 | |
|---|
| 469 | int PMI_Get_id_length_max( int *length ) |
|---|
| 470 | { |
|---|
| 471 | if (length == NULL) |
|---|
| 472 | return PMI_ERR_INVALID_ARG; |
|---|
| 473 | *length = PMI_kvsname_max; |
|---|
| 474 | return PMI_SUCCESS; |
|---|
| 475 | } |
|---|
| 476 | |
|---|
| 477 | int PMI_Get_id( char id_str[], int length ) |
|---|
| 478 | { |
|---|
| 479 | int rc = PMI_KVS_Get_my_name( id_str, length ); |
|---|
| 480 | return rc; |
|---|
| 481 | } |
|---|
| 482 | |
|---|
| 483 | /* FIXME: What is this function? How is it defined and used? */ |
|---|
| 484 | int PMI_Get_kvs_domain_id( char id_str[], int length ) |
|---|
| 485 | { |
|---|
| 486 | return PMI_KVS_Get_my_name( id_str, length ); |
|---|
| 487 | } |
|---|
| 488 | |
|---|
| 489 | /*FIXME: need to return an error if the value of the kvs name returned is |
|---|
| 490 | truncated because it is larger than length */ |
|---|
| 491 | int PMI_KVS_Create( char kvsname[], int length ) |
|---|
| 492 | { |
|---|
| 493 | int err = PMI_SUCCESS; |
|---|
| 494 | |
|---|
| 495 | if (PMI_initialized == SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 496 | /* It is ok to pretend to *create* a kvs space */ |
|---|
| 497 | return 0; |
|---|
| 498 | } |
|---|
| 499 | |
|---|
| 500 | err = GetResponse( "cmd=create_kvs\n", "newkvs", 0 ); |
|---|
| 501 | if (err == PMI_SUCCESS) { |
|---|
| 502 | PMIU_getval( "kvsname", kvsname, length ); |
|---|
| 503 | } |
|---|
| 504 | return err; |
|---|
| 505 | } |
|---|
| 506 | |
|---|
| 507 | int PMI_KVS_Destroy( const char kvsname[] ) |
|---|
| 508 | { |
|---|
| 509 | char buf[PMIU_MAXLINE]; |
|---|
| 510 | int err = PMI_SUCCESS; |
|---|
| 511 | |
|---|
| 512 | if (PMI_initialized == SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 513 | return 0; |
|---|
| 514 | } |
|---|
| 515 | |
|---|
| 516 | /* FIXME: Check for tempbuf too short */ |
|---|
| 517 | MPIU_Snprintf( buf, PMIU_MAXLINE, "cmd=destroy_kvs kvsname=%s\n", |
|---|
| 518 | kvsname ); |
|---|
| 519 | err = GetResponse( buf, "kvs_destroyed", 1 ); |
|---|
| 520 | return err; |
|---|
| 521 | |
|---|
| 522 | } |
|---|
| 523 | |
|---|
| 524 | int PMI_KVS_Put( const char kvsname[], const char key[], const char value[] ) |
|---|
| 525 | { |
|---|
| 526 | char buf[PMIU_MAXLINE]; |
|---|
| 527 | int err = PMI_SUCCESS; |
|---|
| 528 | int rc; |
|---|
| 529 | |
|---|
| 530 | /* This is a special hack to support singleton initialization */ |
|---|
| 531 | if (PMI_initialized == SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 532 | rc = MPIU_Strncpy(cached_singinit_key,key,PMI_keylen_max); |
|---|
| 533 | if (rc != 0) return PMI_FAIL; |
|---|
| 534 | rc = MPIU_Strncpy(cached_singinit_val,value,PMI_vallen_max); |
|---|
| 535 | if (rc != 0) return PMI_FAIL; |
|---|
| 536 | return 0; |
|---|
| 537 | } |
|---|
| 538 | |
|---|
| 539 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, |
|---|
| 540 | "cmd=put kvsname=%s key=%s value=%s\n", |
|---|
| 541 | kvsname, key, value); |
|---|
| 542 | if (rc < 0) return PMI_FAIL; |
|---|
| 543 | err = GetResponse( buf, "put_result", 1 ); |
|---|
| 544 | return err; |
|---|
| 545 | } |
|---|
| 546 | |
|---|
| 547 | int PMI_KVS_Commit( const char kvsname[] ATTRIBUTE((unused))) |
|---|
| 548 | { |
|---|
| 549 | /* no-op in this implementation */ |
|---|
| 550 | return( 0 ); |
|---|
| 551 | } |
|---|
| 552 | |
|---|
| 553 | /*FIXME: need to return an error if the value returned is truncated |
|---|
| 554 | because it is larger than length */ |
|---|
| 555 | int PMI_KVS_Get( const char kvsname[], const char key[], char value[], |
|---|
| 556 | int length) |
|---|
| 557 | { |
|---|
| 558 | char buf[PMIU_MAXLINE]; |
|---|
| 559 | int err = PMI_SUCCESS; |
|---|
| 560 | int rc; |
|---|
| 561 | |
|---|
| 562 | /* Connect to the PM if we haven't already. This is needed in case |
|---|
| 563 | we're doing an MPI_Comm_join or MPI_Comm_connect/accept from |
|---|
| 564 | the singleton init case. This test is here because, in the way in |
|---|
| 565 | which MPICH2 uses PMI, this is where the test needs to be. */ |
|---|
| 566 | if (PMIi_InitIfSingleton() != 0) return -1; |
|---|
| 567 | |
|---|
| 568 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, "cmd=get kvsname=%s key=%s\n", |
|---|
| 569 | kvsname, key ); |
|---|
| 570 | if (rc < 0) return PMI_FAIL; |
|---|
| 571 | |
|---|
| 572 | err = GetResponse( buf, "get_result", 0 ); |
|---|
| 573 | if (err == PMI_SUCCESS) { |
|---|
| 574 | PMIU_getval( "rc", buf, PMIU_MAXLINE ); |
|---|
| 575 | rc = atoi( buf ); |
|---|
| 576 | if ( rc == 0 ) { |
|---|
| 577 | PMIU_getval( "value", value, length ); |
|---|
| 578 | return( 0 ); |
|---|
| 579 | } |
|---|
| 580 | else { |
|---|
| 581 | return( -1 ); |
|---|
| 582 | } |
|---|
| 583 | } |
|---|
| 584 | |
|---|
| 585 | return err; |
|---|
| 586 | } |
|---|
| 587 | |
|---|
| 588 | int PMI_KVS_Iter_first(const char kvsname[], char key[], int key_len, |
|---|
| 589 | char val[], int val_len) |
|---|
| 590 | { |
|---|
| 591 | int rc; |
|---|
| 592 | |
|---|
| 593 | rc = PMII_iter( kvsname, 0, &PMI_iter_next_idx, key, key_len, val, val_len ); |
|---|
| 594 | return( rc ); |
|---|
| 595 | } |
|---|
| 596 | |
|---|
| 597 | int PMI_KVS_Iter_next(const char kvsname[], char key[], int key_len, |
|---|
| 598 | char val[], int val_len) |
|---|
| 599 | { |
|---|
| 600 | int rc; |
|---|
| 601 | |
|---|
| 602 | rc = PMII_iter( kvsname, PMI_iter_next_idx, &PMI_iter_next_idx, |
|---|
| 603 | key, key_len, val, val_len ); |
|---|
| 604 | if ( rc == -2 ) |
|---|
| 605 | PMI_iter_next_idx = 0; |
|---|
| 606 | return( rc ); |
|---|
| 607 | } |
|---|
| 608 | |
|---|
| 609 | /*************************** Name Publishing functions **********************/ |
|---|
| 610 | |
|---|
| 611 | int PMI_Publish_name( const char service_name[], const char port[] ) |
|---|
| 612 | { |
|---|
| 613 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 614 | int err; |
|---|
| 615 | |
|---|
| 616 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 617 | MPIU_Snprintf( cmd, PMIU_MAXLINE, |
|---|
| 618 | "cmd=publish_name service=%s port=%s\n", |
|---|
| 619 | service_name, port ); |
|---|
| 620 | err = GetResponse( cmd, "publish_result", 0 ); |
|---|
| 621 | /* FIXME: This should have used rc and msg */ |
|---|
| 622 | if (err == PMI_SUCCESS) { |
|---|
| 623 | PMIU_getval( "info", buf, PMIU_MAXLINE ); |
|---|
| 624 | if ( strcmp(buf,"ok") != 0 ) { |
|---|
| 625 | PMIU_printf( 1, "publish failed; reason = %s\n", buf ); |
|---|
| 626 | return( PMI_FAIL ); |
|---|
| 627 | } |
|---|
| 628 | } |
|---|
| 629 | } |
|---|
| 630 | else |
|---|
| 631 | { |
|---|
| 632 | PMIU_printf( 1, "PMI_Publish_name called before init\n" ); |
|---|
| 633 | return( PMI_FAIL ); |
|---|
| 634 | } |
|---|
| 635 | |
|---|
| 636 | return( PMI_SUCCESS ); |
|---|
| 637 | } |
|---|
| 638 | |
|---|
| 639 | int PMI_Unpublish_name( const char service_name[] ) |
|---|
| 640 | { |
|---|
| 641 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 642 | int err = PMI_SUCCESS; |
|---|
| 643 | |
|---|
| 644 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 645 | MPIU_Snprintf( cmd, PMIU_MAXLINE, "cmd=unpublish_name service=%s\n", |
|---|
| 646 | service_name ); |
|---|
| 647 | err = GetResponse( cmd, "unpublish_result", 0 ); |
|---|
| 648 | if (err == PMI_SUCCESS) { |
|---|
| 649 | PMIU_getval( "info", buf, PMIU_MAXLINE ); |
|---|
| 650 | if ( strcmp(buf,"ok") != 0 ) { |
|---|
| 651 | /* FIXME: Do correct error reporting */ |
|---|
| 652 | /* |
|---|
| 653 | PMIU_printf( 1, "unpublish failed; reason = %s\n", buf ); |
|---|
| 654 | */ |
|---|
| 655 | return( PMI_FAIL ); |
|---|
| 656 | } |
|---|
| 657 | } |
|---|
| 658 | } |
|---|
| 659 | else |
|---|
| 660 | { |
|---|
| 661 | PMIU_printf( 1, "PMI_Unpublish_name called before init\n" ); |
|---|
| 662 | return( PMI_FAIL ); |
|---|
| 663 | } |
|---|
| 664 | |
|---|
| 665 | return( PMI_SUCCESS ); |
|---|
| 666 | } |
|---|
| 667 | |
|---|
| 668 | int PMI_Lookup_name( const char service_name[], char port[] ) |
|---|
| 669 | { |
|---|
| 670 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 671 | int err; |
|---|
| 672 | |
|---|
| 673 | if ( PMI_initialized > SINGLETON_INIT_BUT_NO_PM) { |
|---|
| 674 | MPIU_Snprintf( cmd, PMIU_MAXLINE, "cmd=lookup_name service=%s\n", |
|---|
| 675 | service_name ); |
|---|
| 676 | err = GetResponse( cmd, "lookup_result", 0 ); |
|---|
| 677 | if (err == PMI_SUCCESS) { |
|---|
| 678 | PMIU_getval( "info", buf, PMIU_MAXLINE ); |
|---|
| 679 | if ( strcmp(buf,"ok") != 0 ) { |
|---|
| 680 | /* FIXME: Do correct error reporting */ |
|---|
| 681 | /**** |
|---|
| 682 | PMIU_printf( 1, "lookup failed; reason = %s\n", buf ); |
|---|
| 683 | ****/ |
|---|
| 684 | return( PMI_FAIL ); |
|---|
| 685 | } |
|---|
| 686 | PMIU_getval( "port", port, MPI_MAX_PORT_NAME ); |
|---|
| 687 | } |
|---|
| 688 | } |
|---|
| 689 | else |
|---|
| 690 | { |
|---|
| 691 | PMIU_printf( 1, "PMI_Lookup_name called before init\n" ); |
|---|
| 692 | return( PMI_FAIL ); |
|---|
| 693 | } |
|---|
| 694 | |
|---|
| 695 | return( PMI_SUCCESS ); |
|---|
| 696 | } |
|---|
| 697 | |
|---|
| 698 | |
|---|
| 699 | /************************** Process Creation functions **********************/ |
|---|
| 700 | |
|---|
| 701 | int PMI_Spawn_multiple(int count, |
|---|
| 702 | const char * cmds[], |
|---|
| 703 | const char ** argvs[], |
|---|
| 704 | const int maxprocs[], |
|---|
| 705 | const int info_keyval_sizes[], |
|---|
| 706 | const PMI_keyval_t * info_keyval_vectors[], |
|---|
| 707 | int preput_keyval_size, |
|---|
| 708 | const PMI_keyval_t preput_keyval_vector[], |
|---|
| 709 | int errors[]) |
|---|
| 710 | { |
|---|
| 711 | int i,rc,argcnt,spawncnt,total_num_processes,num_errcodes_found; |
|---|
| 712 | char buf[PMIU_MAXLINE], tempbuf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 713 | char *lead, *lag; |
|---|
| 714 | |
|---|
| 715 | /* Connect to the PM if we haven't already */ |
|---|
| 716 | if (PMIi_InitIfSingleton() != 0) return -1; |
|---|
| 717 | |
|---|
| 718 | total_num_processes = 0; |
|---|
| 719 | |
|---|
| 720 | for (spawncnt=0; spawncnt < count; spawncnt++) |
|---|
| 721 | { |
|---|
| 722 | total_num_processes += maxprocs[spawncnt]; |
|---|
| 723 | |
|---|
| 724 | rc = MPIU_Snprintf(buf, PMIU_MAXLINE, |
|---|
| 725 | "mcmd=spawn\nnprocs=%d\nexecname=%s\n", |
|---|
| 726 | maxprocs[spawncnt], cmds[spawncnt] ); |
|---|
| 727 | if (rc < 0) { |
|---|
| 728 | return PMI_FAIL; |
|---|
| 729 | } |
|---|
| 730 | |
|---|
| 731 | rc = MPIU_Snprintf(tempbuf, PMIU_MAXLINE, |
|---|
| 732 | "totspawns=%d\nspawnssofar=%d\n", |
|---|
| 733 | count, spawncnt+1); |
|---|
| 734 | |
|---|
| 735 | if (rc < 0) { |
|---|
| 736 | return PMI_FAIL; |
|---|
| 737 | } |
|---|
| 738 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 739 | if (rc != 0) { |
|---|
| 740 | return PMI_FAIL; |
|---|
| 741 | } |
|---|
| 742 | |
|---|
| 743 | argcnt = 0; |
|---|
| 744 | if ((argvs != NULL) && (argvs[spawncnt] != NULL)) { |
|---|
| 745 | for (i=0; argvs[spawncnt][i] != NULL; i++) |
|---|
| 746 | { |
|---|
| 747 | /* FIXME (protocol design flaw): command line arguments |
|---|
| 748 | may contain both = and <space> (and even tab!). |
|---|
| 749 | */ |
|---|
| 750 | /* Note that part of this fixme was really a design error - |
|---|
| 751 | because this uses the mcmd form, the data can be |
|---|
| 752 | sent in multiple writelines. This code now takes |
|---|
| 753 | advantage of that. Note also that a correct parser |
|---|
| 754 | of the commands will permit any character other than a |
|---|
| 755 | new line in the argument, since the form is |
|---|
| 756 | argn=<any nonnewline><newline> */ |
|---|
| 757 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"arg%d=%s\n", |
|---|
| 758 | i+1,argvs[spawncnt][i]); |
|---|
| 759 | if (rc < 0) { |
|---|
| 760 | return PMI_FAIL; |
|---|
| 761 | } |
|---|
| 762 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 763 | if (rc != 0) { |
|---|
| 764 | return PMI_FAIL; |
|---|
| 765 | } |
|---|
| 766 | argcnt++; |
|---|
| 767 | rc = PMIU_writeline( PMI_fd, buf ); |
|---|
| 768 | buf[0] = 0; |
|---|
| 769 | |
|---|
| 770 | } |
|---|
| 771 | } |
|---|
| 772 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"argcnt=%d\n",argcnt); |
|---|
| 773 | if (rc < 0) { |
|---|
| 774 | return PMI_FAIL; |
|---|
| 775 | } |
|---|
| 776 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 777 | if (rc != 0) { |
|---|
| 778 | return PMI_FAIL; |
|---|
| 779 | } |
|---|
| 780 | |
|---|
| 781 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"preput_num=%d\n", |
|---|
| 782 | preput_keyval_size); |
|---|
| 783 | if (rc < 0) { |
|---|
| 784 | return PMI_FAIL; |
|---|
| 785 | } |
|---|
| 786 | |
|---|
| 787 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 788 | if (rc != 0) { |
|---|
| 789 | return PMI_FAIL; |
|---|
| 790 | } |
|---|
| 791 | for (i=0; i < preput_keyval_size; i++) { |
|---|
| 792 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"preput_key_%d=%s\n", |
|---|
| 793 | i,preput_keyval_vector[i].key); |
|---|
| 794 | if (rc < 0) { |
|---|
| 795 | return PMI_FAIL; |
|---|
| 796 | } |
|---|
| 797 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 798 | if (rc != 0) { |
|---|
| 799 | return PMI_FAIL; |
|---|
| 800 | } |
|---|
| 801 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"preput_val_%d=%s\n", |
|---|
| 802 | i,preput_keyval_vector[i].val); |
|---|
| 803 | if (rc < 0) { |
|---|
| 804 | return PMI_FAIL; |
|---|
| 805 | } |
|---|
| 806 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 807 | if (rc != 0) { |
|---|
| 808 | return PMI_FAIL; |
|---|
| 809 | } |
|---|
| 810 | } |
|---|
| 811 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"info_num=%d\n", |
|---|
| 812 | info_keyval_sizes[spawncnt]); |
|---|
| 813 | if (rc < 0) { |
|---|
| 814 | return PMI_FAIL; |
|---|
| 815 | } |
|---|
| 816 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 817 | if (rc != 0) { |
|---|
| 818 | return PMI_FAIL; |
|---|
| 819 | } |
|---|
| 820 | for (i=0; i < info_keyval_sizes[spawncnt]; i++) |
|---|
| 821 | { |
|---|
| 822 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"info_key_%d=%s\n", |
|---|
| 823 | i,info_keyval_vectors[spawncnt][i].key); |
|---|
| 824 | if (rc < 0) { |
|---|
| 825 | return PMI_FAIL; |
|---|
| 826 | } |
|---|
| 827 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 828 | if (rc != 0) { |
|---|
| 829 | return PMI_FAIL; |
|---|
| 830 | } |
|---|
| 831 | rc = MPIU_Snprintf(tempbuf,PMIU_MAXLINE,"info_val_%d=%s\n", |
|---|
| 832 | i,info_keyval_vectors[spawncnt][i].val); |
|---|
| 833 | if (rc < 0) { |
|---|
| 834 | return PMI_FAIL; |
|---|
| 835 | } |
|---|
| 836 | rc = MPIU_Strnapp(buf,tempbuf,PMIU_MAXLINE); |
|---|
| 837 | if (rc != 0) { |
|---|
| 838 | return PMI_FAIL; |
|---|
| 839 | } |
|---|
| 840 | } |
|---|
| 841 | |
|---|
| 842 | rc = MPIU_Strnapp(buf, "endcmd\n", PMIU_MAXLINE); |
|---|
| 843 | if (rc != 0) { |
|---|
| 844 | return PMI_FAIL; |
|---|
| 845 | } |
|---|
| 846 | PMIU_writeline( PMI_fd, buf ); |
|---|
| 847 | } |
|---|
| 848 | |
|---|
| 849 | PMIU_readline( PMI_fd, buf, PMIU_MAXLINE ); |
|---|
| 850 | PMIU_parse_keyvals( buf ); |
|---|
| 851 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 852 | if ( strncmp( cmd, "spawn_result", PMIU_MAXLINE ) != 0 ) { |
|---|
| 853 | PMIU_printf( 1, "got unexpected response to spawn :%s:\n", buf ); |
|---|
| 854 | return( -1 ); |
|---|
| 855 | } |
|---|
| 856 | else { |
|---|
| 857 | PMIU_getval( "rc", buf, PMIU_MAXLINE ); |
|---|
| 858 | rc = atoi( buf ); |
|---|
| 859 | if ( rc != 0 ) { |
|---|
| 860 | /**** |
|---|
| 861 | PMIU_getval( "status", tempbuf, PMIU_MAXLINE ); |
|---|
| 862 | PMIU_printf( 1, "pmi_spawn_mult failed; status: %s\n",tempbuf); |
|---|
| 863 | ****/ |
|---|
| 864 | return( -1 ); |
|---|
| 865 | } |
|---|
| 866 | } |
|---|
| 867 | |
|---|
| 868 | PMIU_Assert(errors != NULL); |
|---|
| 869 | if (PMIU_getval( "errcodes", tempbuf, PMIU_MAXLINE )) { |
|---|
| 870 | num_errcodes_found = 0; |
|---|
| 871 | lag = &tempbuf[0]; |
|---|
| 872 | do { |
|---|
| 873 | lead = strchr(lag, ','); |
|---|
| 874 | if (lead) *lead = '\0'; |
|---|
| 875 | errors[num_errcodes_found++] = atoi(lag); |
|---|
| 876 | lag = lead + 1; /* move past the null char */ |
|---|
| 877 | PMIU_Assert(num_errcodes_found <= total_num_processes); |
|---|
| 878 | } while (lead != NULL); |
|---|
| 879 | PMIU_Assert(num_errcodes_found == total_num_processes); |
|---|
| 880 | } |
|---|
| 881 | else { |
|---|
| 882 | /* gforker doesn't return errcodes, so we'll just pretend that means |
|---|
| 883 | that it was going to send all `0's. */ |
|---|
| 884 | for (i = 0; i < total_num_processes; ++i) { |
|---|
| 885 | errors[i] = 0; |
|---|
| 886 | } |
|---|
| 887 | } |
|---|
| 888 | |
|---|
| 889 | return( 0 ); |
|---|
| 890 | } |
|---|
| 891 | |
|---|
| 892 | /***************** Internal routines not part of PMI interface ***************/ |
|---|
| 893 | |
|---|
| 894 | /* get a keyval pair by specific index */ |
|---|
| 895 | |
|---|
| 896 | static int PMII_iter( const char *kvsname, const int idx, int *next_idx, |
|---|
| 897 | char *key, int key_len, char *val, int val_len) |
|---|
| 898 | { |
|---|
| 899 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 900 | int rc, err; |
|---|
| 901 | |
|---|
| 902 | /* FIXME: Check for tempbuf too short */ |
|---|
| 903 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, "cmd=getbyidx kvsname=%s idx=%d\n", |
|---|
| 904 | kvsname, idx ); |
|---|
| 905 | if (rc < 0) { |
|---|
| 906 | return PMI_FAIL; |
|---|
| 907 | } |
|---|
| 908 | err = GetResponse( cmd, "getbyidx_results", 0 ); |
|---|
| 909 | if (err == PMI_SUCCESS) { |
|---|
| 910 | PMIU_getval( "rc", buf, PMIU_MAXLINE ); |
|---|
| 911 | rc = atoi( buf ); |
|---|
| 912 | if ( rc == 0 ) { |
|---|
| 913 | PMIU_getval( "nextidx", buf, PMIU_MAXLINE ); |
|---|
| 914 | *next_idx = atoi( buf ); |
|---|
| 915 | PMIU_getval( "key", key, key_len ); |
|---|
| 916 | PMIU_getval( "val", val, val_len ); |
|---|
| 917 | return( PMI_SUCCESS ); |
|---|
| 918 | } |
|---|
| 919 | else { |
|---|
| 920 | PMIU_getval( "reason", buf, PMIU_MAXLINE ); |
|---|
| 921 | if ( strncmp( buf, "no_more_keyvals", PMIU_MAXLINE ) == 0 ) { |
|---|
| 922 | key[0] = '\0'; |
|---|
| 923 | return( PMI_SUCCESS ); |
|---|
| 924 | } |
|---|
| 925 | else { |
|---|
| 926 | PMIU_printf( 1, "iter failed; reason = %s\n", buf ); |
|---|
| 927 | return( PMI_FAIL ); |
|---|
| 928 | } |
|---|
| 929 | } |
|---|
| 930 | } |
|---|
| 931 | return err; |
|---|
| 932 | } |
|---|
| 933 | |
|---|
| 934 | /* to get all maxes in one message */ |
|---|
| 935 | /* FIXME: This mixes init with get maxes */ |
|---|
| 936 | static int PMII_getmaxes( int *kvsname_max, int *keylen_max, int *vallen_max ) |
|---|
| 937 | { |
|---|
| 938 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE], errmsg[PMIU_MAXLINE]; |
|---|
| 939 | int err, rc; |
|---|
| 940 | |
|---|
| 941 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, |
|---|
| 942 | "cmd=init pmi_version=%d pmi_subversion=%d\n", |
|---|
| 943 | PMI_VERSION, PMI_SUBVERSION ); |
|---|
| 944 | if (rc < 0) { |
|---|
| 945 | return PMI_FAIL; |
|---|
| 946 | } |
|---|
| 947 | |
|---|
| 948 | rc = PMIU_writeline( PMI_fd, buf ); |
|---|
| 949 | if (rc != 0) { |
|---|
| 950 | PMIU_printf( 1, "Unable to write to PMI_fd\n" ); |
|---|
| 951 | return PMI_FAIL; |
|---|
| 952 | } |
|---|
| 953 | buf[0] = 0; /* Ensure buffer is empty if read fails */ |
|---|
| 954 | err = PMIU_readline( PMI_fd, buf, PMIU_MAXLINE ); |
|---|
| 955 | if (err < 0) { |
|---|
| 956 | PMIU_printf( 1, "Error reading initack on %d\n", PMI_fd ); |
|---|
| 957 | perror( "Error on readline:" ); |
|---|
| 958 | PMI_Abort(-1, "Above error when reading after init" ); |
|---|
| 959 | } |
|---|
| 960 | PMIU_parse_keyvals( buf ); |
|---|
| 961 | cmd[0] = 0; |
|---|
| 962 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 963 | if ( strncmp( cmd, "response_to_init", PMIU_MAXLINE ) != 0 ) { |
|---|
| 964 | MPIU_Snprintf(errmsg, PMIU_MAXLINE, |
|---|
| 965 | "got unexpected response to init :%s: (full line = %s)", |
|---|
| 966 | cmd, buf ); |
|---|
| 967 | PMI_Abort( -1, errmsg ); |
|---|
| 968 | } |
|---|
| 969 | else { |
|---|
| 970 | char buf1[PMIU_MAXLINE]; |
|---|
| 971 | PMIU_getval( "rc", buf, PMIU_MAXLINE ); |
|---|
| 972 | if ( strncmp( buf, "0", PMIU_MAXLINE ) != 0 ) { |
|---|
| 973 | PMIU_getval( "pmi_version", buf, PMIU_MAXLINE ); |
|---|
| 974 | PMIU_getval( "pmi_subversion", buf1, PMIU_MAXLINE ); |
|---|
| 975 | MPIU_Snprintf(errmsg, PMIU_MAXLINE, |
|---|
| 976 | "pmi_version mismatch; client=%d.%d mgr=%s.%s", |
|---|
| 977 | PMI_VERSION, PMI_SUBVERSION, buf, buf1 ); |
|---|
| 978 | PMI_Abort( -1, errmsg ); |
|---|
| 979 | } |
|---|
| 980 | } |
|---|
| 981 | err = GetResponse( "cmd=get_maxes\n", "maxes", 0 ); |
|---|
| 982 | if (err == PMI_SUCCESS) { |
|---|
| 983 | PMIU_getval( "kvsname_max", buf, PMIU_MAXLINE ); |
|---|
| 984 | *kvsname_max = atoi( buf ); |
|---|
| 985 | PMIU_getval( "keylen_max", buf, PMIU_MAXLINE ); |
|---|
| 986 | *keylen_max = atoi( buf ); |
|---|
| 987 | PMIU_getval( "vallen_max", buf, PMIU_MAXLINE ); |
|---|
| 988 | *vallen_max = atoi( buf ); |
|---|
| 989 | } |
|---|
| 990 | return err; |
|---|
| 991 | } |
|---|
| 992 | |
|---|
| 993 | /* ----------------------------------------------------------------------- */ |
|---|
| 994 | /* |
|---|
| 995 | * This function is used to request information from the server and check |
|---|
| 996 | * that the response uses the expected command name. On a successful |
|---|
| 997 | * return from this routine, additional PMIU_getval calls may be used |
|---|
| 998 | * to access information about the returned value. |
|---|
| 999 | * |
|---|
| 1000 | * If checkRc is true, this routine also checks that the rc value returned |
|---|
| 1001 | * was 0. If not, it uses the "msg" value to report on the reason for |
|---|
| 1002 | * the failure. |
|---|
| 1003 | */ |
|---|
| 1004 | static int GetResponse( const char request[], const char expectedCmd[], |
|---|
| 1005 | int checkRc ) |
|---|
| 1006 | { |
|---|
| 1007 | int err, n; |
|---|
| 1008 | char *p; |
|---|
| 1009 | char recvbuf[PMIU_MAXLINE]; |
|---|
| 1010 | char cmdName[PMIU_MAXLINE]; |
|---|
| 1011 | |
|---|
| 1012 | /* FIXME: This is an example of an incorrect fix - writeline can change |
|---|
| 1013 | the second argument in some cases, and that will break the const'ness |
|---|
| 1014 | of request. Instead, writeline should take a const item and return |
|---|
| 1015 | an error in the case in which it currently truncates the data. */ |
|---|
| 1016 | err = PMIU_writeline( PMI_fd, (char *)request ); |
|---|
| 1017 | if (err) { |
|---|
| 1018 | return err; |
|---|
| 1019 | } |
|---|
| 1020 | n = PMIU_readline( PMI_fd, recvbuf, sizeof(recvbuf) ); |
|---|
| 1021 | if (n <= 0) { |
|---|
| 1022 | PMIU_printf( 1, "readline failed\n" ); |
|---|
| 1023 | return PMI_FAIL; |
|---|
| 1024 | } |
|---|
| 1025 | err = PMIU_parse_keyvals( recvbuf ); |
|---|
| 1026 | if (err) { |
|---|
| 1027 | PMIU_printf( 1, "parse_kevals failed %d\n", err ); |
|---|
| 1028 | return err; |
|---|
| 1029 | } |
|---|
| 1030 | p = PMIU_getval( "cmd", cmdName, sizeof(cmdName) ); |
|---|
| 1031 | if (!p) { |
|---|
| 1032 | PMIU_printf( 1, "getval cmd failed\n" ); |
|---|
| 1033 | return PMI_FAIL; |
|---|
| 1034 | } |
|---|
| 1035 | if (strcmp( expectedCmd, cmdName ) != 0) { |
|---|
| 1036 | PMIU_printf( 1, "expecting cmd=%s, got %s\n", expectedCmd, cmdName ); |
|---|
| 1037 | return PMI_FAIL; |
|---|
| 1038 | } |
|---|
| 1039 | if (checkRc) { |
|---|
| 1040 | p = PMIU_getval( "rc", cmdName, PMIU_MAXLINE ); |
|---|
| 1041 | if ( p && strcmp(cmdName,"0") != 0 ) { |
|---|
| 1042 | PMIU_getval( "msg", cmdName, PMIU_MAXLINE ); |
|---|
| 1043 | PMIU_printf( 1, "Command %s failed, reason='%s'\n", |
|---|
| 1044 | request, cmdName ); |
|---|
| 1045 | return PMI_FAIL; |
|---|
| 1046 | } |
|---|
| 1047 | } |
|---|
| 1048 | |
|---|
| 1049 | return err; |
|---|
| 1050 | } |
|---|
| 1051 | /* ----------------------------------------------------------------------- */ |
|---|
| 1052 | |
|---|
| 1053 | |
|---|
| 1054 | #ifdef USE_PMI_PORT |
|---|
| 1055 | /* |
|---|
| 1056 | * This code allows a program to contact a host/port for the PMI socket. |
|---|
| 1057 | */ |
|---|
| 1058 | #include <errno.h> |
|---|
| 1059 | #if defined(HAVE_SYS_TYPES_H) |
|---|
| 1060 | #include <sys/types.h> |
|---|
| 1061 | #endif |
|---|
| 1062 | #include <sys/param.h> |
|---|
| 1063 | #include <sys/socket.h> |
|---|
| 1064 | |
|---|
| 1065 | /* sockaddr_in (Internet) */ |
|---|
| 1066 | #include <netinet/in.h> |
|---|
| 1067 | /* TCP_NODELAY */ |
|---|
| 1068 | #include <netinet/tcp.h> |
|---|
| 1069 | |
|---|
| 1070 | /* sockaddr_un (Unix) */ |
|---|
| 1071 | #include <sys/un.h> |
|---|
| 1072 | |
|---|
| 1073 | /* defs of gethostbyname */ |
|---|
| 1074 | #include <netdb.h> |
|---|
| 1075 | |
|---|
| 1076 | /* fcntl, F_GET/SETFL */ |
|---|
| 1077 | #include <fcntl.h> |
|---|
| 1078 | |
|---|
| 1079 | /* This is really IP!? */ |
|---|
| 1080 | #ifndef TCP |
|---|
| 1081 | #define TCP 0 |
|---|
| 1082 | #endif |
|---|
| 1083 | |
|---|
| 1084 | /* stub for connecting to a specified host/port instead of using a |
|---|
| 1085 | specified fd inherited from a parent process */ |
|---|
| 1086 | static int PMII_Connect_to_pm( char *hostname, int portnum ) |
|---|
| 1087 | { |
|---|
| 1088 | struct hostent *hp; |
|---|
| 1089 | struct sockaddr_in sa; |
|---|
| 1090 | int fd; |
|---|
| 1091 | int optval = 1; |
|---|
| 1092 | int q_wait = 1; |
|---|
| 1093 | |
|---|
| 1094 | hp = gethostbyname( hostname ); |
|---|
| 1095 | if (!hp) { |
|---|
| 1096 | PMIU_printf( 1, "Unable to get host entry for %s\n", hostname ); |
|---|
| 1097 | return -1; |
|---|
| 1098 | } |
|---|
| 1099 | |
|---|
| 1100 | memset( (void *)&sa, 0, sizeof(sa) ); |
|---|
| 1101 | /* POSIX might define h_addr_list only and node define h_addr */ |
|---|
| 1102 | #ifdef HAVE_H_ADDR_LIST |
|---|
| 1103 | memcpy( (void *)&sa.sin_addr, (void *)hp->h_addr_list[0], hp->h_length); |
|---|
| 1104 | #else |
|---|
| 1105 | memcpy( (void *)&sa.sin_addr, (void *)hp->h_addr, hp->h_length); |
|---|
| 1106 | #endif |
|---|
| 1107 | sa.sin_family = hp->h_addrtype; |
|---|
| 1108 | sa.sin_port = htons( (unsigned short) portnum ); |
|---|
| 1109 | |
|---|
| 1110 | fd = socket( AF_INET, SOCK_STREAM, TCP ); |
|---|
| 1111 | if (fd < 0) { |
|---|
| 1112 | PMIU_printf( 1, "Unable to get AF_INET socket\n" ); |
|---|
| 1113 | return -1; |
|---|
| 1114 | } |
|---|
| 1115 | |
|---|
| 1116 | if (setsockopt( fd, IPPROTO_TCP, TCP_NODELAY, |
|---|
| 1117 | (char *)&optval, sizeof(optval) )) { |
|---|
| 1118 | perror( "Error calling setsockopt:" ); |
|---|
| 1119 | } |
|---|
| 1120 | |
|---|
| 1121 | /* We wait here for the connection to succeed */ |
|---|
| 1122 | if (connect( fd, (struct sockaddr *)&sa, sizeof(sa) ) < 0) { |
|---|
| 1123 | switch (errno) { |
|---|
| 1124 | case ECONNREFUSED: |
|---|
| 1125 | PMIU_printf( 1, "connect failed with connection refused\n" ); |
|---|
| 1126 | /* (close socket, get new socket, try again) */ |
|---|
| 1127 | if (q_wait) |
|---|
| 1128 | close(fd); |
|---|
| 1129 | return -1; |
|---|
| 1130 | |
|---|
| 1131 | case EINPROGRESS: /* (nonblocking) - select for writing. */ |
|---|
| 1132 | break; |
|---|
| 1133 | |
|---|
| 1134 | case EISCONN: /* (already connected) */ |
|---|
| 1135 | break; |
|---|
| 1136 | |
|---|
| 1137 | case ETIMEDOUT: /* timed out */ |
|---|
| 1138 | PMIU_printf( 1, "connect failed with timeout\n" ); |
|---|
| 1139 | return -1; |
|---|
| 1140 | |
|---|
| 1141 | default: |
|---|
| 1142 | PMIU_printf( 1, "connect failed with errno %d\n", errno ); |
|---|
| 1143 | return -1; |
|---|
| 1144 | } |
|---|
| 1145 | } |
|---|
| 1146 | |
|---|
| 1147 | return fd; |
|---|
| 1148 | } |
|---|
| 1149 | |
|---|
| 1150 | static int PMII_Set_from_port( int fd, int id ) |
|---|
| 1151 | { |
|---|
| 1152 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 1153 | int err, rc; |
|---|
| 1154 | |
|---|
| 1155 | /* We start by sending a startup message to the server */ |
|---|
| 1156 | |
|---|
| 1157 | if (PMI_debug) { |
|---|
| 1158 | PMIU_printf( 1, "Writing initack to destination fd %d\n", fd ); |
|---|
| 1159 | } |
|---|
| 1160 | /* Handshake and initialize from a port */ |
|---|
| 1161 | |
|---|
| 1162 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, "cmd=initack pmiid=%d\n", id ); |
|---|
| 1163 | if (rc < 0) { |
|---|
| 1164 | return PMI_FAIL; |
|---|
| 1165 | } |
|---|
| 1166 | PMIU_printf( PMI_debug, "writing on fd %d line :%s:\n", fd, buf ); |
|---|
| 1167 | err = PMIU_writeline( fd, buf ); |
|---|
| 1168 | if (err) { |
|---|
| 1169 | PMIU_printf( 1, "Error in writeline initack\n" ); |
|---|
| 1170 | return -1; |
|---|
| 1171 | } |
|---|
| 1172 | |
|---|
| 1173 | /* cmd=initack */ |
|---|
| 1174 | buf[0] = 0; |
|---|
| 1175 | PMIU_printf( PMI_debug, "reading initack\n" ); |
|---|
| 1176 | err = PMIU_readline( fd, buf, PMIU_MAXLINE ); |
|---|
| 1177 | if (err < 0) { |
|---|
| 1178 | PMIU_printf( 1, "Error reading initack on %d\n", fd ); |
|---|
| 1179 | perror( "Error on readline:" ); |
|---|
| 1180 | return -1; |
|---|
| 1181 | } |
|---|
| 1182 | PMIU_parse_keyvals( buf ); |
|---|
| 1183 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 1184 | if ( strcmp( cmd, "initack" ) ) { |
|---|
| 1185 | PMIU_printf( 1, "got unexpected input %s\n", buf ); |
|---|
| 1186 | return -1; |
|---|
| 1187 | } |
|---|
| 1188 | |
|---|
| 1189 | /* Read, in order, size, rank, and debug. Eventually, we'll want |
|---|
| 1190 | the handshake to include a version number */ |
|---|
| 1191 | |
|---|
| 1192 | /* size */ |
|---|
| 1193 | PMIU_printf( PMI_debug, "reading size\n" ); |
|---|
| 1194 | err = PMIU_readline( fd, buf, PMIU_MAXLINE ); |
|---|
| 1195 | if (err < 0) { |
|---|
| 1196 | PMIU_printf( 1, "Error reading size on %d\n", fd ); |
|---|
| 1197 | perror( "Error on readline:" ); |
|---|
| 1198 | return -1; |
|---|
| 1199 | } |
|---|
| 1200 | PMIU_parse_keyvals( buf ); |
|---|
| 1201 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 1202 | if ( strcmp(cmd,"set")) { |
|---|
| 1203 | PMIU_printf( 1, "got unexpected command %s in %s\n", cmd, buf ); |
|---|
| 1204 | return -1; |
|---|
| 1205 | } |
|---|
| 1206 | /* cmd=set size=n */ |
|---|
| 1207 | PMIU_getval( "size", cmd, PMIU_MAXLINE ); |
|---|
| 1208 | PMI_size = atoi(cmd); |
|---|
| 1209 | |
|---|
| 1210 | /* rank */ |
|---|
| 1211 | PMIU_printf( PMI_debug, "reading rank\n" ); |
|---|
| 1212 | err = PMIU_readline( fd, buf, PMIU_MAXLINE ); |
|---|
| 1213 | if (err < 0) { |
|---|
| 1214 | PMIU_printf( 1, "Error reading rank on %d\n", fd ); |
|---|
| 1215 | perror( "Error on readline:" ); |
|---|
| 1216 | return -1; |
|---|
| 1217 | } |
|---|
| 1218 | PMIU_parse_keyvals( buf ); |
|---|
| 1219 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 1220 | if ( strcmp(cmd,"set")) { |
|---|
| 1221 | PMIU_printf( 1, "got unexpected command %s in %s\n", cmd, buf ); |
|---|
| 1222 | return -1; |
|---|
| 1223 | } |
|---|
| 1224 | /* cmd=set rank=n */ |
|---|
| 1225 | PMIU_getval( "rank", cmd, PMIU_MAXLINE ); |
|---|
| 1226 | PMI_rank = atoi(cmd); |
|---|
| 1227 | PMIU_Set_rank( PMI_rank ); |
|---|
| 1228 | |
|---|
| 1229 | /* debug flag */ |
|---|
| 1230 | err = PMIU_readline( fd, buf, PMIU_MAXLINE ); |
|---|
| 1231 | if (err < 0) { |
|---|
| 1232 | PMIU_printf( 1, "Error reading debug on %d\n", fd ); |
|---|
| 1233 | return -1; |
|---|
| 1234 | } |
|---|
| 1235 | PMIU_parse_keyvals( buf ); |
|---|
| 1236 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 1237 | if ( strcmp(cmd,"set")) { |
|---|
| 1238 | PMIU_printf( 1, "got unexpected command %s in %s\n", cmd, buf ); |
|---|
| 1239 | return -1; |
|---|
| 1240 | } |
|---|
| 1241 | /* cmd=set debug=n */ |
|---|
| 1242 | PMIU_getval( "debug", cmd, PMIU_MAXLINE ); |
|---|
| 1243 | PMI_debug = atoi(cmd); |
|---|
| 1244 | |
|---|
| 1245 | if (PMI_debug) { |
|---|
| 1246 | DBG_PRINTF( ("end of handshake, rank = %d, size = %d\n", |
|---|
| 1247 | PMI_rank, PMI_size )); |
|---|
| 1248 | DBG_PRINTF( ("Completed init\n" ) ); |
|---|
| 1249 | } |
|---|
| 1250 | |
|---|
| 1251 | return 0; |
|---|
| 1252 | } |
|---|
| 1253 | |
|---|
| 1254 | /* ------------------------------------------------------------------------- */ |
|---|
| 1255 | /* |
|---|
| 1256 | * Singleton Init. |
|---|
| 1257 | * |
|---|
| 1258 | * MPI-2 allows processes to become MPI processes and then make MPI calls, |
|---|
| 1259 | * such as MPI_Comm_spawn, that require a process manager (this is different |
|---|
| 1260 | * than the much simpler case of allowing MPI programs to run with an |
|---|
| 1261 | * MPI_COMM_WORLD of size 1 without an mpiexec or process manager). |
|---|
| 1262 | * |
|---|
| 1263 | * The process starts when either the client or the process manager contacts |
|---|
| 1264 | * the other. If the client starts, it sends a singinit command and |
|---|
| 1265 | * waits for the server to respond with its own singinit command. |
|---|
| 1266 | * If the server start, it send a singinit command and waits for the |
|---|
| 1267 | * client to respond with its own singinit command |
|---|
| 1268 | * |
|---|
| 1269 | * client sends singinit with these required values |
|---|
| 1270 | * pmi_version=<value of PMI_VERSION> |
|---|
| 1271 | * pmi_subversion=<value of PMI_SUBVERSION> |
|---|
| 1272 | * |
|---|
| 1273 | * and these optional values |
|---|
| 1274 | * stdio=[yes|no] |
|---|
| 1275 | * authtype=[none|shared|<other-to-be-defined>] |
|---|
| 1276 | * authstring=<string> |
|---|
| 1277 | * |
|---|
| 1278 | * server sends singinit with the same required and optional values as |
|---|
| 1279 | * above. |
|---|
| 1280 | * |
|---|
| 1281 | * At this point, the protocol is now the same in both cases, and has the |
|---|
| 1282 | * following components: |
|---|
| 1283 | * |
|---|
| 1284 | * server sends singinit_info with these required fields |
|---|
| 1285 | * versionok=[yes|no] |
|---|
| 1286 | * stdio=[yes|no] |
|---|
| 1287 | * kvsname=<string> |
|---|
| 1288 | * |
|---|
| 1289 | * The client then issues the init command (see PMII_getmaxes) |
|---|
| 1290 | * |
|---|
| 1291 | * cmd=init pmi_version=<val> pmi_subversion=<val> |
|---|
| 1292 | * |
|---|
| 1293 | * and expects to receive a |
|---|
| 1294 | * |
|---|
| 1295 | * cmd=response_to_init rc=0 pmi_version=<val> pmi_subversion=<val> |
|---|
| 1296 | * |
|---|
| 1297 | * (This is the usual init sequence). |
|---|
| 1298 | * |
|---|
| 1299 | */ |
|---|
| 1300 | /* ------------------------------------------------------------------------- */ |
|---|
| 1301 | /* This is a special routine used to re-initialize PMI when it is in |
|---|
| 1302 | the singleton init case. That is, the executable was started without |
|---|
| 1303 | mpiexec, and PMI_Init returned as if there was only one process. |
|---|
| 1304 | |
|---|
| 1305 | Note that PMI routines should not call PMII_singinit; they should |
|---|
| 1306 | call PMIi_InitIfSingleton(), which both connects to the process mangager |
|---|
| 1307 | and sets up the initial KVS connection entry. |
|---|
| 1308 | */ |
|---|
| 1309 | |
|---|
| 1310 | static int PMII_singinit(void) |
|---|
| 1311 | { |
|---|
| 1312 | int pid, rc; |
|---|
| 1313 | int singinit_listen_sock, stdin_sock, stdout_sock, stderr_sock; |
|---|
| 1314 | const char *newargv[8]; |
|---|
| 1315 | char charpid[8], port_c[8]; |
|---|
| 1316 | struct sockaddr_in sin; |
|---|
| 1317 | socklen_t len; |
|---|
| 1318 | |
|---|
| 1319 | /* Create a socket on which to allow an mpiexec to connect back to |
|---|
| 1320 | us */ |
|---|
| 1321 | sin.sin_family = AF_INET; |
|---|
| 1322 | sin.sin_addr.s_addr = INADDR_ANY; |
|---|
| 1323 | sin.sin_port = htons(0); /* anonymous port */ |
|---|
| 1324 | singinit_listen_sock = socket(AF_INET, SOCK_STREAM, 0); |
|---|
| 1325 | rc = bind(singinit_listen_sock, (struct sockaddr *)&sin ,sizeof(sin)); |
|---|
| 1326 | len = sizeof(struct sockaddr_in); |
|---|
| 1327 | rc = getsockname( singinit_listen_sock, (struct sockaddr *) &sin, &len ); |
|---|
| 1328 | MPIU_Snprintf(port_c, sizeof(port_c), "%d",ntohs(sin.sin_port)); |
|---|
| 1329 | rc = listen(singinit_listen_sock, 5); |
|---|
| 1330 | |
|---|
| 1331 | PMIU_printf( PMI_debug_init, "Starting mpiexec with %s\n", port_c ); |
|---|
| 1332 | |
|---|
| 1333 | /* Launch the mpiexec process with the name of this port */ |
|---|
| 1334 | pid = fork(); |
|---|
| 1335 | if (pid < 0) { |
|---|
| 1336 | perror("PMII_singinit: fork failed"); |
|---|
| 1337 | exit(-1); |
|---|
| 1338 | } |
|---|
| 1339 | else if (pid == 0) { |
|---|
| 1340 | newargv[0] = "mpiexec"; |
|---|
| 1341 | newargv[1] = "-pmi_args"; |
|---|
| 1342 | newargv[2] = port_c; |
|---|
| 1343 | /* FIXME: Use a valid hostname */ |
|---|
| 1344 | newargv[3] = "default_interface"; /* default interface name, for now */ |
|---|
| 1345 | newargv[4] = "default_key"; /* default authentication key, for now */ |
|---|
| 1346 | MPIU_Snprintf(charpid, sizeof(charpid), "%d",getpid()); |
|---|
| 1347 | newargv[5] = charpid; |
|---|
| 1348 | newargv[6] = NULL; |
|---|
| 1349 | rc = execvp(newargv[0], (char **)newargv); |
|---|
| 1350 | perror("PMII_singinit: execv failed"); |
|---|
| 1351 | PMIU_printf(1, " This singleton init program attempted to access some feature\n"); |
|---|
| 1352 | PMIU_printf(1, " for which process manager support was required, e.g. spawn or universe_size.\n"); |
|---|
| 1353 | PMIU_printf(1, " But the necessary mpiexec is not in your path.\n"); |
|---|
| 1354 | return(-1); |
|---|
| 1355 | } |
|---|
| 1356 | else |
|---|
| 1357 | { |
|---|
| 1358 | char buf[PMIU_MAXLINE], cmd[PMIU_MAXLINE]; |
|---|
| 1359 | char *p; |
|---|
| 1360 | int connectStdio = 0; |
|---|
| 1361 | |
|---|
| 1362 | /* Allow one connection back from the created mpiexec program */ |
|---|
| 1363 | PMI_fd = accept_one_connection(singinit_listen_sock); |
|---|
| 1364 | if (PMI_fd < 0) { |
|---|
| 1365 | PMIU_printf( 1, "Failed to establish singleton init connection\n" ); |
|---|
| 1366 | return PMI_FAIL; |
|---|
| 1367 | } |
|---|
| 1368 | /* Execute the singleton init protocol */ |
|---|
| 1369 | rc = PMIU_readline( PMI_fd, buf, PMIU_MAXLINE ); |
|---|
| 1370 | PMIU_printf( PMI_debug_init, "Singinit: read %s\n", buf ); |
|---|
| 1371 | |
|---|
| 1372 | PMIU_parse_keyvals( buf ); |
|---|
| 1373 | PMIU_getval( "cmd", cmd, PMIU_MAXLINE ); |
|---|
| 1374 | if (strcmp( cmd, "singinit" ) != 0) { |
|---|
| 1375 | PMIU_printf( 1, "unexpected command from PM: %s\n", cmd ); |
|---|
| 1376 | return PMI_FAIL; |
|---|
| 1377 | } |
|---|
| 1378 | p = PMIU_getval( "authtype", cmd, PMIU_MAXLINE ); |
|---|
| 1379 | if (p && strcmp( cmd, "none" ) != 0) { |
|---|
| 1380 | PMIU_printf( 1, "unsupported authentication method %s\n", cmd ); |
|---|
| 1381 | return PMI_FAIL; |
|---|
| 1382 | } |
|---|
| 1383 | /* p = PMIU_getval( "authstring", cmd, PMIU_MAXLINE ); */ |
|---|
| 1384 | |
|---|
| 1385 | /* If we're successful, send back our own singinit */ |
|---|
| 1386 | rc = MPIU_Snprintf( buf, PMIU_MAXLINE, |
|---|
| 1387 | "cmd=singinit pmi_version=%d pmi_subversion=%d stdio=yes authtype=none\n", |
|---|
| 1388 | PMI_VERSION, PMI_SUBVERSION ); |
|---|
| 1389 | if (rc < 0) { |
|---|
| 1390 | return PMI_FAIL; |
|---|
| 1391 | } |
|---|
| 1392 | PMIU_printf( PMI_debug_init, "GetResponse with %s\n", buf ); |
|---|
| 1393 | |
|---|
| 1394 | rc = GetResponse( buf, "singinit_info", 0 ); |
|---|
| 1395 | if (rc != 0) { |
|---|
| 1396 | PMIU_printf( 1, "GetResponse failed\n" ); |
|---|
| 1397 | return PMI_FAIL; |
|---|
| 1398 | } |
|---|
| 1399 | p = PMIU_getval( "versionok", cmd, PMIU_MAXLINE ); |
|---|
| 1400 | if (p && strcmp( cmd, "yes" ) != 0) { |
|---|
| 1401 | PMIU_printf( 1, "Process manager needs a different PMI version\n" ); |
|---|
| 1402 | return PMI_FAIL; |
|---|
| 1403 | } |
|---|
| 1404 | p = PMIU_getval( "stdio", cmd, PMIU_MAXLINE ); |
|---|
| 1405 | if (p && strcmp( cmd, "yes" ) == 0) { |
|---|
| 1406 | PMIU_printf( PMI_debug_init, "PM agreed to connect stdio\n" ); |
|---|
| 1407 | connectStdio = 1; |
|---|
| 1408 | } |
|---|
| 1409 | p = PMIU_getval( "kvsname", singinit_kvsname, sizeof(singinit_kvsname) ); |
|---|
| 1410 | PMIU_printf( PMI_debug_init, "kvsname to use is %s\n", |
|---|
| 1411 | singinit_kvsname ); |
|---|
| 1412 | |
|---|
| 1413 | if (connectStdio) { |
|---|
| 1414 | PMIU_printf( PMI_debug_init, |
|---|
| 1415 | "Accepting three connections for stdin, out, err\n" ); |
|---|
| 1416 | stdin_sock = accept_one_connection(singinit_listen_sock); |
|---|
| 1417 | dup2(stdin_sock, 0); |
|---|
| 1418 | stdout_sock = accept_one_connection(singinit_listen_sock); |
|---|
| 1419 | dup2(stdout_sock,1); |
|---|
| 1420 | stderr_sock = accept_one_connection(singinit_listen_sock); |
|---|
| 1421 | dup2(stderr_sock,2); |
|---|
| 1422 | } |
|---|
| 1423 | PMIU_printf( PMI_debug_init, "Done with singinit handshake\n" ); |
|---|
| 1424 | } |
|---|
| 1425 | return 0; |
|---|
| 1426 | } |
|---|
| 1427 | |
|---|
| 1428 | /* Promote PMI to a fully initialized version if it was started as |
|---|
| 1429 | a singleton init */ |
|---|
| 1430 | static int PMIi_InitIfSingleton(void) |
|---|
| 1431 | { |
|---|
| 1432 | int rc; |
|---|
| 1433 | static int firstcall = 1; |
|---|
| 1434 | |
|---|
| 1435 | if (PMI_initialized != SINGLETON_INIT_BUT_NO_PM || !firstcall) return 0; |
|---|
| 1436 | |
|---|
| 1437 | /* We only try to init as a singleton the first time */ |
|---|
| 1438 | firstcall = 0; |
|---|
| 1439 | |
|---|
| 1440 | /* First, start (if necessary) an mpiexec, connect to it, |
|---|
| 1441 | and start the singleton init handshake */ |
|---|
| 1442 | rc = PMII_singinit(); |
|---|
| 1443 | |
|---|
| 1444 | if (rc < 0) |
|---|
| 1445 | return(-1); |
|---|
| 1446 | PMI_initialized = SINGLETON_INIT_WITH_PM; /* do this right away */ |
|---|
| 1447 | PMI_size = 1; |
|---|
| 1448 | PMI_rank = 0; |
|---|
| 1449 | PMI_debug = 0; |
|---|
| 1450 | PMI_spawned = 0; |
|---|
| 1451 | |
|---|
| 1452 | PMII_getmaxes( &PMI_kvsname_max, &PMI_keylen_max, &PMI_vallen_max ); |
|---|
| 1453 | |
|---|
| 1454 | /* FIXME: We need to support a distinct kvsname for each |
|---|
| 1455 | process group */ |
|---|
| 1456 | PMI_KVS_Put( singinit_kvsname, cached_singinit_key, cached_singinit_val ); |
|---|
| 1457 | |
|---|
| 1458 | return 0; |
|---|
| 1459 | } |
|---|
| 1460 | |
|---|
| 1461 | static int accept_one_connection(int list_sock) |
|---|
| 1462 | { |
|---|
| 1463 | int gotit, new_sock; |
|---|
| 1464 | struct sockaddr_in from; |
|---|
| 1465 | socklen_t len; |
|---|
| 1466 | |
|---|
| 1467 | len = sizeof(from); |
|---|
| 1468 | gotit = 0; |
|---|
| 1469 | while ( ! gotit ) |
|---|
| 1470 | { |
|---|
| 1471 | new_sock = accept(list_sock, (struct sockaddr *)&from, &len); |
|---|
| 1472 | if (new_sock == -1) |
|---|
| 1473 | { |
|---|
| 1474 | if (errno == EINTR) /* interrupted? If so, try again */ |
|---|
| 1475 | continue; |
|---|
| 1476 | else |
|---|
| 1477 | { |
|---|
| 1478 | PMIU_printf(1, "accept failed in accept_one_connection\n"); |
|---|
| 1479 | exit(-1); |
|---|
| 1480 | } |
|---|
| 1481 | } |
|---|
| 1482 | else |
|---|
| 1483 | gotit = 1; |
|---|
| 1484 | } |
|---|
| 1485 | return(new_sock); |
|---|
| 1486 | } |
|---|
| 1487 | |
|---|
| 1488 | #endif |
|---|
| 1489 | /* end USE_PMI_PORT */ |
|---|
| 1490 | |
|---|
| 1491 | /* Get the FD to use for PMI operations. If a port is used, rather than |
|---|
| 1492 | a pre-established FD (i.e., via pipe), this routine will handle the |
|---|
| 1493 | initial handshake. |
|---|
| 1494 | */ |
|---|
| 1495 | static int getPMIFD( int *notset ) |
|---|
| 1496 | { |
|---|
| 1497 | char *p; |
|---|
| 1498 | |
|---|
| 1499 | /* Set the default */ |
|---|
| 1500 | PMI_fd = -1; |
|---|
| 1501 | |
|---|
| 1502 | p = getenv( "PMI_FD" ); |
|---|
| 1503 | |
|---|
| 1504 | if (p) { |
|---|
| 1505 | PMI_fd = atoi( p ); |
|---|
| 1506 | return 0; |
|---|
| 1507 | } |
|---|
| 1508 | |
|---|
| 1509 | #ifdef USE_PMI_PORT |
|---|
| 1510 | p = getenv( "PMI_PORT" ); |
|---|
| 1511 | if (p) { |
|---|
| 1512 | int portnum; |
|---|
| 1513 | char hostname[MAXHOSTNAME+1]; |
|---|
| 1514 | char *pn, *ph; |
|---|
| 1515 | int id = 0; |
|---|
| 1516 | |
|---|
| 1517 | /* Connect to the indicated port (in format hostname:portnumber) |
|---|
| 1518 | and get the fd for the socket */ |
|---|
| 1519 | |
|---|
| 1520 | /* Split p into host and port */ |
|---|
| 1521 | pn = p; |
|---|
| 1522 | ph = hostname; |
|---|
| 1523 | while (*pn && *pn != ':' && (ph - hostname) < MAXHOSTNAME) { |
|---|
| 1524 | *ph++ = *pn++; |
|---|
| 1525 | } |
|---|
| 1526 | *ph = 0; |
|---|
| 1527 | |
|---|
| 1528 | if (PMI_debug) { |
|---|
| 1529 | DBG_PRINTF( ("Connecting to %s\n", p) ); |
|---|
| 1530 | } |
|---|
| 1531 | if (*pn == ':') { |
|---|
| 1532 | portnum = atoi( pn+1 ); |
|---|
| 1533 | /* FIXME: Check for valid integer after : */ |
|---|
| 1534 | /* This routine only gets the fd to use to talk to |
|---|
| 1535 | the process manager. The handshake below is used |
|---|
| 1536 | to setup the initial values */ |
|---|
| 1537 | PMI_fd = PMII_Connect_to_pm( hostname, portnum ); |
|---|
| 1538 | if (PMI_fd < 0) { |
|---|
| 1539 | PMIU_printf( 1, "Unable to connect to %s on %d\n", |
|---|
| 1540 | hostname, portnum ); |
|---|
| 1541 | return -1; |
|---|
| 1542 | } |
|---|
| 1543 | } |
|---|
| 1544 | else { |
|---|
| 1545 | PMIU_printf( 1, "unable to decode hostport from %s\n", p ); |
|---|
| 1546 | return PMI_FAIL; |
|---|
| 1547 | } |
|---|
| 1548 | |
|---|
| 1549 | /* We should first handshake to get size, rank, debug. */ |
|---|
| 1550 | p = getenv( "PMI_ID" ); |
|---|
| 1551 | if (p) { |
|---|
| 1552 | id = atoi( p ); |
|---|
| 1553 | /* PMII_Set_from_port sets up the values that are delivered |
|---|
| 1554 | by enviroment variables when a separate port is not used */ |
|---|
| 1555 | PMII_Set_from_port( PMI_fd, id ); |
|---|
| 1556 | *notset = 0; |
|---|
| 1557 | } |
|---|
| 1558 | return 0; |
|---|
| 1559 | } |
|---|
| 1560 | #endif |
|---|
| 1561 | |
|---|
| 1562 | /* Singleton init case - its ok to return success with no fd set */ |
|---|
| 1563 | return 0; |
|---|
| 1564 | } |
|---|