1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2016 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/stdinc.h" 8 9 #include "nvmf_internal.h" 10 #include "transport.h" 11 12 #include "spdk/assert.h" 13 #include "spdk/likely.h" 14 #include "spdk/string.h" 15 #include "spdk/trace.h" 16 #include "spdk/nvmf_spec.h" 17 #include "spdk/uuid.h" 18 #include "spdk/json.h" 19 #include "spdk/file.h" 20 #include "spdk/bit_array.h" 21 #include "spdk/bdev.h" 22 23 #define __SPDK_BDEV_MODULE_ONLY 24 #include "spdk/bdev_module.h" 25 #include "spdk/log.h" 26 #include "spdk_internal/utf.h" 27 #include "spdk_internal/usdt.h" 28 29 #define MODEL_NUMBER_DEFAULT "SPDK bdev Controller" 30 #define NVMF_SUBSYSTEM_DEFAULT_NAMESPACES 32 31 32 /* 33 * States for parsing valid domains in NQNs according to RFC 1034 34 */ 35 enum spdk_nvmf_nqn_domain_states { 36 /* First character of a domain must be a letter */ 37 SPDK_NVMF_DOMAIN_ACCEPT_LETTER = 0, 38 39 /* Subsequent characters can be any of letter, digit, or hyphen */ 40 SPDK_NVMF_DOMAIN_ACCEPT_LDH = 1, 41 42 /* A domain label must end with either a letter or digit */ 43 SPDK_NVMF_DOMAIN_ACCEPT_ANY = 2 44 }; 45 46 static int _nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem); 47 48 /* Returns true if is a valid ASCII string as defined by the NVMe spec */ 49 static bool 50 nvmf_valid_ascii_string(const void *buf, size_t size) 51 { 52 const uint8_t *str = buf; 53 size_t i; 54 55 for (i = 0; i < size; i++) { 56 if (str[i] < 0x20 || str[i] > 0x7E) { 57 return false; 58 } 59 } 60 61 return true; 62 } 63 64 static bool 65 nvmf_valid_nqn(const char *nqn) 66 { 67 size_t len; 68 struct spdk_uuid uuid_value; 69 uint32_t i; 70 int bytes_consumed; 71 uint32_t domain_label_length; 72 char *reverse_domain_end; 73 uint32_t reverse_domain_end_index; 74 enum spdk_nvmf_nqn_domain_states domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LETTER; 75 76 /* Check for length requirements */ 77 len = strlen(nqn); 78 if (len > SPDK_NVMF_NQN_MAX_LEN) { 79 SPDK_ERRLOG("Invalid NQN \"%s\": length %zu > max %d\n", nqn, len, SPDK_NVMF_NQN_MAX_LEN); 80 return false; 81 } 82 83 /* The nqn must be at least as long as SPDK_NVMF_NQN_MIN_LEN to contain the necessary prefix. */ 84 if (len < SPDK_NVMF_NQN_MIN_LEN) { 85 SPDK_ERRLOG("Invalid NQN \"%s\": length %zu < min %d\n", nqn, len, SPDK_NVMF_NQN_MIN_LEN); 86 return false; 87 } 88 89 /* Check for discovery controller nqn */ 90 if (!strcmp(nqn, SPDK_NVMF_DISCOVERY_NQN)) { 91 return true; 92 } 93 94 /* Check for equality with the generic nqn structure of the form "nqn.2014-08.org.nvmexpress:uuid:11111111-2222-3333-4444-555555555555" */ 95 if (!strncmp(nqn, SPDK_NVMF_NQN_UUID_PRE, SPDK_NVMF_NQN_UUID_PRE_LEN)) { 96 if (len != SPDK_NVMF_NQN_UUID_PRE_LEN + SPDK_NVMF_UUID_STRING_LEN) { 97 SPDK_ERRLOG("Invalid NQN \"%s\": uuid is not the correct length\n", nqn); 98 return false; 99 } 100 101 if (spdk_uuid_parse(&uuid_value, &nqn[SPDK_NVMF_NQN_UUID_PRE_LEN])) { 102 SPDK_ERRLOG("Invalid NQN \"%s\": uuid is not formatted correctly\n", nqn); 103 return false; 104 } 105 return true; 106 } 107 108 /* If the nqn does not match the uuid structure, the next several checks validate the form "nqn.yyyy-mm.reverse.domain:user-string" */ 109 110 if (strncmp(nqn, "nqn.", 4) != 0) { 111 SPDK_ERRLOG("Invalid NQN \"%s\": NQN must begin with \"nqn.\".\n", nqn); 112 return false; 113 } 114 115 /* Check for yyyy-mm. */ 116 if (!(isdigit(nqn[4]) && isdigit(nqn[5]) && isdigit(nqn[6]) && isdigit(nqn[7]) && 117 nqn[8] == '-' && isdigit(nqn[9]) && isdigit(nqn[10]) && nqn[11] == '.')) { 118 SPDK_ERRLOG("Invalid date code in NQN \"%s\"\n", nqn); 119 return false; 120 } 121 122 reverse_domain_end = strchr(nqn, ':'); 123 if (reverse_domain_end != NULL && (reverse_domain_end_index = reverse_domain_end - nqn) < len - 1) { 124 } else { 125 SPDK_ERRLOG("Invalid NQN \"%s\". NQN must contain user specified name with a ':' as a prefix.\n", 126 nqn); 127 return false; 128 } 129 130 /* Check for valid reverse domain */ 131 domain_label_length = 0; 132 for (i = 12; i < reverse_domain_end_index; i++) { 133 if (domain_label_length > SPDK_DOMAIN_LABEL_MAX_LEN) { 134 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". At least one Label is too long.\n", nqn); 135 return false; 136 } 137 138 switch (domain_state) { 139 140 case SPDK_NVMF_DOMAIN_ACCEPT_LETTER: { 141 if (isalpha(nqn[i])) { 142 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY; 143 domain_label_length++; 144 break; 145 } else { 146 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must start with a letter.\n", nqn); 147 return false; 148 } 149 } 150 151 case SPDK_NVMF_DOMAIN_ACCEPT_LDH: { 152 if (isalpha(nqn[i]) || isdigit(nqn[i])) { 153 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY; 154 domain_label_length++; 155 break; 156 } else if (nqn[i] == '-') { 157 if (i == reverse_domain_end_index - 1) { 158 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n", 159 nqn); 160 return false; 161 } 162 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LDH; 163 domain_label_length++; 164 break; 165 } else if (nqn[i] == '.') { 166 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n", 167 nqn); 168 return false; 169 } else { 170 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only [a-z,A-Z,0-9,'-','.'].\n", 171 nqn); 172 return false; 173 } 174 } 175 176 case SPDK_NVMF_DOMAIN_ACCEPT_ANY: { 177 if (isalpha(nqn[i]) || isdigit(nqn[i])) { 178 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY; 179 domain_label_length++; 180 break; 181 } else if (nqn[i] == '-') { 182 if (i == reverse_domain_end_index - 1) { 183 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n", 184 nqn); 185 return false; 186 } 187 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LDH; 188 domain_label_length++; 189 break; 190 } else if (nqn[i] == '.') { 191 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LETTER; 192 domain_label_length = 0; 193 break; 194 } else { 195 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only [a-z,A-Z,0-9,'-','.'].\n", 196 nqn); 197 return false; 198 } 199 } 200 } 201 } 202 203 i = reverse_domain_end_index + 1; 204 while (i < len) { 205 bytes_consumed = utf8_valid(&nqn[i], &nqn[len]); 206 if (bytes_consumed <= 0) { 207 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only valid utf-8.\n", nqn); 208 return false; 209 } 210 211 i += bytes_consumed; 212 } 213 return true; 214 } 215 216 static void subsystem_state_change_on_pg(struct spdk_io_channel_iter *i); 217 218 struct spdk_nvmf_subsystem * 219 spdk_nvmf_subsystem_create(struct spdk_nvmf_tgt *tgt, 220 const char *nqn, 221 enum spdk_nvmf_subtype type, 222 uint32_t num_ns) 223 { 224 struct spdk_nvmf_subsystem *subsystem; 225 uint32_t sid; 226 227 if (spdk_nvmf_tgt_find_subsystem(tgt, nqn)) { 228 SPDK_ERRLOG("Subsystem NQN '%s' already exists\n", nqn); 229 return NULL; 230 } 231 232 if (!nvmf_valid_nqn(nqn)) { 233 return NULL; 234 } 235 236 if (type == SPDK_NVMF_SUBTYPE_DISCOVERY) { 237 if (num_ns != 0) { 238 SPDK_ERRLOG("Discovery subsystem cannot have namespaces.\n"); 239 return NULL; 240 } 241 } else if (num_ns == 0) { 242 num_ns = NVMF_SUBSYSTEM_DEFAULT_NAMESPACES; 243 } 244 245 /* Find a free subsystem id (sid) */ 246 sid = spdk_bit_array_find_first_clear(tgt->subsystem_ids, 0); 247 if (sid == UINT32_MAX) { 248 return NULL; 249 } 250 subsystem = calloc(1, sizeof(struct spdk_nvmf_subsystem)); 251 if (subsystem == NULL) { 252 return NULL; 253 } 254 255 subsystem->thread = spdk_get_thread(); 256 subsystem->state = SPDK_NVMF_SUBSYSTEM_INACTIVE; 257 subsystem->tgt = tgt; 258 subsystem->id = sid; 259 subsystem->subtype = type; 260 subsystem->max_nsid = num_ns; 261 subsystem->next_cntlid = 0; 262 subsystem->min_cntlid = NVMF_MIN_CNTLID; 263 subsystem->max_cntlid = NVMF_MAX_CNTLID; 264 snprintf(subsystem->subnqn, sizeof(subsystem->subnqn), "%s", nqn); 265 pthread_mutex_init(&subsystem->mutex, NULL); 266 TAILQ_INIT(&subsystem->listeners); 267 TAILQ_INIT(&subsystem->hosts); 268 TAILQ_INIT(&subsystem->ctrlrs); 269 subsystem->used_listener_ids = spdk_bit_array_create(NVMF_MAX_LISTENERS_PER_SUBSYSTEM); 270 if (subsystem->used_listener_ids == NULL) { 271 pthread_mutex_destroy(&subsystem->mutex); 272 free(subsystem); 273 return NULL; 274 } 275 276 if (num_ns != 0) { 277 subsystem->ns = calloc(num_ns, sizeof(struct spdk_nvmf_ns *)); 278 if (subsystem->ns == NULL) { 279 SPDK_ERRLOG("Namespace memory allocation failed\n"); 280 pthread_mutex_destroy(&subsystem->mutex); 281 spdk_bit_array_free(&subsystem->used_listener_ids); 282 free(subsystem); 283 return NULL; 284 } 285 subsystem->ana_group = calloc(num_ns, sizeof(uint32_t)); 286 if (subsystem->ana_group == NULL) { 287 SPDK_ERRLOG("ANA group memory allocation failed\n"); 288 pthread_mutex_destroy(&subsystem->mutex); 289 free(subsystem->ns); 290 spdk_bit_array_free(&subsystem->used_listener_ids); 291 free(subsystem); 292 return NULL; 293 } 294 } 295 296 memset(subsystem->sn, '0', sizeof(subsystem->sn) - 1); 297 subsystem->sn[sizeof(subsystem->sn) - 1] = '\0'; 298 299 snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", 300 MODEL_NUMBER_DEFAULT); 301 302 spdk_bit_array_set(tgt->subsystem_ids, sid); 303 RB_INSERT(subsystem_tree, &tgt->subsystems, subsystem); 304 305 SPDK_DTRACE_PROBE1(nvmf_subsystem_create, subsystem->subnqn); 306 307 return subsystem; 308 } 309 310 /* Must hold subsystem->mutex while calling this function */ 311 static void 312 nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_host *host) 313 { 314 TAILQ_REMOVE(&subsystem->hosts, host, link); 315 free(host); 316 } 317 318 static void 319 _nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem, 320 struct spdk_nvmf_subsystem_listener *listener, 321 bool stop) 322 { 323 struct spdk_nvmf_transport *transport; 324 struct spdk_nvmf_ctrlr *ctrlr; 325 326 if (stop) { 327 transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, listener->trid->trstring); 328 if (transport != NULL) { 329 spdk_nvmf_transport_stop_listen(transport, listener->trid); 330 } 331 } 332 333 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 334 if (ctrlr->listener == listener) { 335 ctrlr->listener = NULL; 336 } 337 } 338 339 TAILQ_REMOVE(&subsystem->listeners, listener, link); 340 nvmf_update_discovery_log(listener->subsystem->tgt, NULL); 341 free(listener->ana_state); 342 spdk_bit_array_clear(subsystem->used_listener_ids, listener->id); 343 free(listener); 344 } 345 346 static void 347 _nvmf_subsystem_destroy_msg(void *cb_arg) 348 { 349 struct spdk_nvmf_subsystem *subsystem = cb_arg; 350 351 _nvmf_subsystem_destroy(subsystem); 352 } 353 354 static int 355 _nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem) 356 { 357 struct spdk_nvmf_ns *ns; 358 nvmf_subsystem_destroy_cb async_destroy_cb = NULL; 359 void *async_destroy_cb_arg = NULL; 360 int rc; 361 362 if (!TAILQ_EMPTY(&subsystem->ctrlrs)) { 363 SPDK_DEBUGLOG(nvmf, "subsystem %p %s has active controllers\n", subsystem, subsystem->subnqn); 364 subsystem->async_destroy = true; 365 rc = spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_destroy_msg, subsystem); 366 if (rc) { 367 SPDK_ERRLOG("Failed to send thread msg, rc %d\n", rc); 368 assert(0); 369 return rc; 370 } 371 return -EINPROGRESS; 372 } 373 374 ns = spdk_nvmf_subsystem_get_first_ns(subsystem); 375 while (ns != NULL) { 376 struct spdk_nvmf_ns *next_ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns); 377 378 spdk_nvmf_subsystem_remove_ns(subsystem, ns->opts.nsid); 379 ns = next_ns; 380 } 381 382 free(subsystem->ns); 383 free(subsystem->ana_group); 384 385 RB_REMOVE(subsystem_tree, &subsystem->tgt->subsystems, subsystem); 386 assert(spdk_bit_array_get(subsystem->tgt->subsystem_ids, subsystem->id) == true); 387 spdk_bit_array_clear(subsystem->tgt->subsystem_ids, subsystem->id); 388 389 pthread_mutex_destroy(&subsystem->mutex); 390 391 spdk_bit_array_free(&subsystem->used_listener_ids); 392 393 if (subsystem->async_destroy) { 394 async_destroy_cb = subsystem->async_destroy_cb; 395 async_destroy_cb_arg = subsystem->async_destroy_cb_arg; 396 } 397 398 free(subsystem); 399 400 if (async_destroy_cb) { 401 async_destroy_cb(async_destroy_cb_arg); 402 } 403 404 return 0; 405 } 406 407 static struct spdk_nvmf_ns * 408 _nvmf_subsystem_get_first_zoned_ns(struct spdk_nvmf_subsystem *subsystem) 409 { 410 struct spdk_nvmf_ns *ns = spdk_nvmf_subsystem_get_first_ns(subsystem); 411 while (ns != NULL) { 412 if (ns->csi == SPDK_NVME_CSI_ZNS) { 413 return ns; 414 } 415 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns); 416 } 417 return NULL; 418 } 419 420 int 421 spdk_nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem, nvmf_subsystem_destroy_cb cpl_cb, 422 void *cpl_cb_arg) 423 { 424 struct spdk_nvmf_host *host, *host_tmp; 425 struct spdk_nvmf_transport *transport; 426 427 if (!subsystem) { 428 return -EINVAL; 429 } 430 431 SPDK_DTRACE_PROBE1(nvmf_subsystem_destroy, subsystem->subnqn); 432 433 assert(spdk_get_thread() == subsystem->thread); 434 435 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 436 SPDK_ERRLOG("Subsystem can only be destroyed in inactive state, %s state %d\n", 437 subsystem->subnqn, subsystem->state); 438 return -EAGAIN; 439 } 440 if (subsystem->destroying) { 441 SPDK_ERRLOG("Subsystem destruction is already started\n"); 442 assert(0); 443 return -EALREADY; 444 } 445 446 subsystem->destroying = true; 447 448 SPDK_DEBUGLOG(nvmf, "subsystem is %p %s\n", subsystem, subsystem->subnqn); 449 450 nvmf_subsystem_remove_all_listeners(subsystem, false); 451 452 pthread_mutex_lock(&subsystem->mutex); 453 454 TAILQ_FOREACH_SAFE(host, &subsystem->hosts, link, host_tmp) { 455 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 456 transport = spdk_nvmf_transport_get_next(transport)) { 457 if (transport->ops->subsystem_remove_host) { 458 transport->ops->subsystem_remove_host(transport, subsystem, host->nqn); 459 } 460 } 461 nvmf_subsystem_remove_host(subsystem, host); 462 } 463 464 pthread_mutex_unlock(&subsystem->mutex); 465 466 subsystem->async_destroy_cb = cpl_cb; 467 subsystem->async_destroy_cb_arg = cpl_cb_arg; 468 469 return _nvmf_subsystem_destroy(subsystem); 470 } 471 472 /* we have to use the typedef in the function declaration to appease astyle. */ 473 typedef enum spdk_nvmf_subsystem_state spdk_nvmf_subsystem_state_t; 474 475 static spdk_nvmf_subsystem_state_t 476 nvmf_subsystem_get_intermediate_state(enum spdk_nvmf_subsystem_state current_state, 477 enum spdk_nvmf_subsystem_state requested_state) 478 { 479 switch (requested_state) { 480 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 481 return SPDK_NVMF_SUBSYSTEM_DEACTIVATING; 482 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 483 if (current_state == SPDK_NVMF_SUBSYSTEM_PAUSED) { 484 return SPDK_NVMF_SUBSYSTEM_RESUMING; 485 } else { 486 return SPDK_NVMF_SUBSYSTEM_ACTIVATING; 487 } 488 case SPDK_NVMF_SUBSYSTEM_PAUSED: 489 return SPDK_NVMF_SUBSYSTEM_PAUSING; 490 default: 491 assert(false); 492 return SPDK_NVMF_SUBSYSTEM_NUM_STATES; 493 } 494 } 495 496 static int 497 nvmf_subsystem_set_state(struct spdk_nvmf_subsystem *subsystem, 498 enum spdk_nvmf_subsystem_state state) 499 { 500 enum spdk_nvmf_subsystem_state actual_old_state, expected_old_state; 501 bool exchanged; 502 503 switch (state) { 504 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 505 expected_old_state = SPDK_NVMF_SUBSYSTEM_DEACTIVATING; 506 break; 507 case SPDK_NVMF_SUBSYSTEM_ACTIVATING: 508 expected_old_state = SPDK_NVMF_SUBSYSTEM_INACTIVE; 509 break; 510 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 511 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING; 512 break; 513 case SPDK_NVMF_SUBSYSTEM_PAUSING: 514 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE; 515 break; 516 case SPDK_NVMF_SUBSYSTEM_PAUSED: 517 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSING; 518 break; 519 case SPDK_NVMF_SUBSYSTEM_RESUMING: 520 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED; 521 break; 522 case SPDK_NVMF_SUBSYSTEM_DEACTIVATING: 523 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE; 524 break; 525 default: 526 assert(false); 527 return -1; 528 } 529 530 actual_old_state = expected_old_state; 531 exchanged = __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false, 532 __ATOMIC_RELAXED, __ATOMIC_RELAXED); 533 if (spdk_unlikely(exchanged == false)) { 534 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING && 535 state == SPDK_NVMF_SUBSYSTEM_ACTIVE) { 536 expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING; 537 } 538 /* This is for the case when activating the subsystem fails. */ 539 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_ACTIVATING && 540 state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) { 541 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING; 542 } 543 /* This is for the case when resuming the subsystem fails. */ 544 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING && 545 state == SPDK_NVMF_SUBSYSTEM_PAUSING) { 546 expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING; 547 } 548 /* This is for the case when stopping paused subsystem */ 549 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_PAUSED && 550 state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) { 551 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED; 552 } 553 actual_old_state = expected_old_state; 554 __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false, 555 __ATOMIC_RELAXED, __ATOMIC_RELAXED); 556 } 557 assert(actual_old_state == expected_old_state); 558 return actual_old_state - expected_old_state; 559 } 560 561 struct subsystem_state_change_ctx { 562 struct spdk_nvmf_subsystem *subsystem; 563 uint16_t nsid; 564 565 enum spdk_nvmf_subsystem_state original_state; 566 enum spdk_nvmf_subsystem_state requested_state; 567 568 spdk_nvmf_subsystem_state_change_done cb_fn; 569 void *cb_arg; 570 }; 571 572 static void 573 subsystem_state_change_revert_done(struct spdk_io_channel_iter *i, int status) 574 { 575 struct subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 576 577 /* Nothing to be done here if the state setting fails, we are just screwed. */ 578 if (nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state)) { 579 SPDK_ERRLOG("Unable to revert the subsystem state after operation failure.\n"); 580 } 581 582 ctx->subsystem->changing_state = false; 583 if (ctx->cb_fn) { 584 /* return a failure here. This function only exists in an error path. */ 585 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, -1); 586 } 587 free(ctx); 588 } 589 590 static void 591 subsystem_state_change_done(struct spdk_io_channel_iter *i, int status) 592 { 593 struct subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 594 enum spdk_nvmf_subsystem_state intermediate_state; 595 596 SPDK_DTRACE_PROBE4(nvmf_subsystem_change_state_done, ctx->subsystem->subnqn, 597 ctx->requested_state, ctx->original_state, status); 598 599 if (status == 0) { 600 status = nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state); 601 if (status) { 602 status = -1; 603 } 604 } 605 606 if (status) { 607 intermediate_state = nvmf_subsystem_get_intermediate_state(ctx->requested_state, 608 ctx->original_state); 609 assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES); 610 611 if (nvmf_subsystem_set_state(ctx->subsystem, intermediate_state)) { 612 goto out; 613 } 614 ctx->requested_state = ctx->original_state; 615 spdk_for_each_channel(ctx->subsystem->tgt, 616 subsystem_state_change_on_pg, 617 ctx, 618 subsystem_state_change_revert_done); 619 return; 620 } 621 622 out: 623 ctx->subsystem->changing_state = false; 624 if (ctx->cb_fn) { 625 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status); 626 } 627 free(ctx); 628 } 629 630 static void 631 subsystem_state_change_continue(void *ctx, int status) 632 { 633 struct spdk_io_channel_iter *i = ctx; 634 struct subsystem_state_change_ctx *_ctx __attribute__((unused)); 635 636 _ctx = spdk_io_channel_iter_get_ctx(i); 637 SPDK_DTRACE_PROBE3(nvmf_pg_change_state_done, _ctx->subsystem->subnqn, 638 _ctx->requested_state, spdk_thread_get_id(spdk_get_thread())); 639 640 spdk_for_each_channel_continue(i, status); 641 } 642 643 static void 644 subsystem_state_change_on_pg(struct spdk_io_channel_iter *i) 645 { 646 struct subsystem_state_change_ctx *ctx; 647 struct spdk_io_channel *ch; 648 struct spdk_nvmf_poll_group *group; 649 650 ctx = spdk_io_channel_iter_get_ctx(i); 651 ch = spdk_io_channel_iter_get_channel(i); 652 group = spdk_io_channel_get_ctx(ch); 653 654 SPDK_DTRACE_PROBE3(nvmf_pg_change_state, ctx->subsystem->subnqn, 655 ctx->requested_state, spdk_thread_get_id(spdk_get_thread())); 656 switch (ctx->requested_state) { 657 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 658 nvmf_poll_group_remove_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 659 break; 660 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 661 if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_ACTIVATING) { 662 nvmf_poll_group_add_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 663 } else if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_RESUMING) { 664 nvmf_poll_group_resume_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 665 } 666 break; 667 case SPDK_NVMF_SUBSYSTEM_PAUSED: 668 nvmf_poll_group_pause_subsystem(group, ctx->subsystem, ctx->nsid, subsystem_state_change_continue, 669 i); 670 break; 671 default: 672 assert(false); 673 break; 674 } 675 } 676 677 static int 678 nvmf_subsystem_state_change(struct spdk_nvmf_subsystem *subsystem, 679 uint32_t nsid, 680 enum spdk_nvmf_subsystem_state requested_state, 681 spdk_nvmf_subsystem_state_change_done cb_fn, 682 void *cb_arg) 683 { 684 struct subsystem_state_change_ctx *ctx; 685 enum spdk_nvmf_subsystem_state intermediate_state; 686 int rc; 687 688 if (__sync_val_compare_and_swap(&subsystem->changing_state, false, true)) { 689 return -EBUSY; 690 } 691 692 SPDK_DTRACE_PROBE3(nvmf_subsystem_change_state, subsystem->subnqn, 693 requested_state, subsystem->state); 694 /* If we are already in the requested state, just call the callback immediately. */ 695 if (subsystem->state == requested_state) { 696 subsystem->changing_state = false; 697 if (cb_fn) { 698 cb_fn(subsystem, cb_arg, 0); 699 } 700 return 0; 701 } 702 703 intermediate_state = nvmf_subsystem_get_intermediate_state(subsystem->state, requested_state); 704 assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES); 705 706 ctx = calloc(1, sizeof(*ctx)); 707 if (!ctx) { 708 subsystem->changing_state = false; 709 return -ENOMEM; 710 } 711 712 ctx->original_state = subsystem->state; 713 rc = nvmf_subsystem_set_state(subsystem, intermediate_state); 714 if (rc) { 715 free(ctx); 716 subsystem->changing_state = false; 717 return rc; 718 } 719 720 ctx->subsystem = subsystem; 721 ctx->nsid = nsid; 722 ctx->requested_state = requested_state; 723 ctx->cb_fn = cb_fn; 724 ctx->cb_arg = cb_arg; 725 726 spdk_for_each_channel(subsystem->tgt, 727 subsystem_state_change_on_pg, 728 ctx, 729 subsystem_state_change_done); 730 731 return 0; 732 } 733 734 int 735 spdk_nvmf_subsystem_start(struct spdk_nvmf_subsystem *subsystem, 736 spdk_nvmf_subsystem_state_change_done cb_fn, 737 void *cb_arg) 738 { 739 return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg); 740 } 741 742 int 743 spdk_nvmf_subsystem_stop(struct spdk_nvmf_subsystem *subsystem, 744 spdk_nvmf_subsystem_state_change_done cb_fn, 745 void *cb_arg) 746 { 747 return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_INACTIVE, cb_fn, cb_arg); 748 } 749 750 int 751 spdk_nvmf_subsystem_pause(struct spdk_nvmf_subsystem *subsystem, 752 uint32_t nsid, 753 spdk_nvmf_subsystem_state_change_done cb_fn, 754 void *cb_arg) 755 { 756 return nvmf_subsystem_state_change(subsystem, nsid, SPDK_NVMF_SUBSYSTEM_PAUSED, cb_fn, cb_arg); 757 } 758 759 int 760 spdk_nvmf_subsystem_resume(struct spdk_nvmf_subsystem *subsystem, 761 spdk_nvmf_subsystem_state_change_done cb_fn, 762 void *cb_arg) 763 { 764 return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg); 765 } 766 767 struct spdk_nvmf_subsystem * 768 spdk_nvmf_subsystem_get_first(struct spdk_nvmf_tgt *tgt) 769 { 770 return RB_MIN(subsystem_tree, &tgt->subsystems); 771 } 772 773 struct spdk_nvmf_subsystem * 774 spdk_nvmf_subsystem_get_next(struct spdk_nvmf_subsystem *subsystem) 775 { 776 if (!subsystem) { 777 return NULL; 778 } 779 780 return RB_NEXT(subsystem_tree, &tgt->subsystems, subsystem); 781 } 782 783 /* Must hold subsystem->mutex while calling this function */ 784 static struct spdk_nvmf_host * 785 nvmf_subsystem_find_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 786 { 787 struct spdk_nvmf_host *host = NULL; 788 789 TAILQ_FOREACH(host, &subsystem->hosts, link) { 790 if (strcmp(hostnqn, host->nqn) == 0) { 791 return host; 792 } 793 } 794 795 return NULL; 796 } 797 798 int 799 spdk_nvmf_subsystem_add_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 800 const struct spdk_json_val *params) 801 { 802 struct spdk_nvmf_host *host; 803 struct spdk_nvmf_transport *transport; 804 int rc; 805 806 if (!nvmf_valid_nqn(hostnqn)) { 807 return -EINVAL; 808 } 809 810 pthread_mutex_lock(&subsystem->mutex); 811 812 if (nvmf_subsystem_find_host(subsystem, hostnqn)) { 813 /* This subsystem already allows the specified host. */ 814 pthread_mutex_unlock(&subsystem->mutex); 815 return 0; 816 } 817 818 host = calloc(1, sizeof(*host)); 819 if (!host) { 820 pthread_mutex_unlock(&subsystem->mutex); 821 return -ENOMEM; 822 } 823 824 snprintf(host->nqn, sizeof(host->nqn), "%s", hostnqn); 825 826 SPDK_DTRACE_PROBE2(nvmf_subsystem_add_host, subsystem->subnqn, host->nqn); 827 828 TAILQ_INSERT_HEAD(&subsystem->hosts, host, link); 829 830 if (!TAILQ_EMPTY(&subsystem->listeners)) { 831 nvmf_update_discovery_log(subsystem->tgt, hostnqn); 832 } 833 834 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 835 transport = spdk_nvmf_transport_get_next(transport)) { 836 if (transport->ops->subsystem_add_host) { 837 rc = transport->ops->subsystem_add_host(transport, subsystem, hostnqn, params); 838 if (rc) { 839 SPDK_ERRLOG("Unable to add host to %s transport\n", transport->ops->name); 840 /* Remove this host from all transports we've managed to add it to. */ 841 pthread_mutex_unlock(&subsystem->mutex); 842 spdk_nvmf_subsystem_remove_host(subsystem, hostnqn); 843 return rc; 844 } 845 } 846 } 847 848 pthread_mutex_unlock(&subsystem->mutex); 849 850 return 0; 851 } 852 853 int 854 spdk_nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 855 { 856 struct spdk_nvmf_host *host; 857 struct spdk_nvmf_transport *transport; 858 859 pthread_mutex_lock(&subsystem->mutex); 860 861 host = nvmf_subsystem_find_host(subsystem, hostnqn); 862 if (host == NULL) { 863 pthread_mutex_unlock(&subsystem->mutex); 864 return -ENOENT; 865 } 866 867 SPDK_DTRACE_PROBE2(nvmf_subsystem_remove_host, subsystem->subnqn, host->nqn); 868 869 nvmf_subsystem_remove_host(subsystem, host); 870 871 if (!TAILQ_EMPTY(&subsystem->listeners)) { 872 nvmf_update_discovery_log(subsystem->tgt, hostnqn); 873 } 874 875 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 876 transport = spdk_nvmf_transport_get_next(transport)) { 877 if (transport->ops->subsystem_remove_host) { 878 transport->ops->subsystem_remove_host(transport, subsystem, hostnqn); 879 } 880 } 881 882 pthread_mutex_unlock(&subsystem->mutex); 883 884 return 0; 885 } 886 887 struct nvmf_subsystem_disconnect_host_ctx { 888 struct spdk_nvmf_subsystem *subsystem; 889 char *hostnqn; 890 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 891 void *cb_arg; 892 }; 893 894 static void 895 nvmf_subsystem_disconnect_host_fini(struct spdk_io_channel_iter *i, int status) 896 { 897 struct nvmf_subsystem_disconnect_host_ctx *ctx; 898 899 ctx = spdk_io_channel_iter_get_ctx(i); 900 901 if (ctx->cb_fn) { 902 ctx->cb_fn(ctx->cb_arg, status); 903 } 904 free(ctx->hostnqn); 905 free(ctx); 906 } 907 908 static void 909 nvmf_subsystem_disconnect_qpairs_by_host(struct spdk_io_channel_iter *i) 910 { 911 struct nvmf_subsystem_disconnect_host_ctx *ctx; 912 struct spdk_nvmf_poll_group *group; 913 struct spdk_io_channel *ch; 914 struct spdk_nvmf_qpair *qpair, *tmp_qpair; 915 struct spdk_nvmf_ctrlr *ctrlr; 916 917 ctx = spdk_io_channel_iter_get_ctx(i); 918 ch = spdk_io_channel_iter_get_channel(i); 919 group = spdk_io_channel_get_ctx(ch); 920 921 TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, tmp_qpair) { 922 ctrlr = qpair->ctrlr; 923 924 if (ctrlr == NULL || ctrlr->subsys != ctx->subsystem) { 925 continue; 926 } 927 928 if (strncmp(ctrlr->hostnqn, ctx->hostnqn, sizeof(ctrlr->hostnqn)) == 0) { 929 /* Right now this does not wait for the queue pairs to actually disconnect. */ 930 spdk_nvmf_qpair_disconnect(qpair, NULL, NULL); 931 } 932 } 933 spdk_for_each_channel_continue(i, 0); 934 } 935 936 int 937 spdk_nvmf_subsystem_disconnect_host(struct spdk_nvmf_subsystem *subsystem, 938 const char *hostnqn, 939 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 940 void *cb_arg) 941 { 942 struct nvmf_subsystem_disconnect_host_ctx *ctx; 943 944 ctx = calloc(1, sizeof(struct nvmf_subsystem_disconnect_host_ctx)); 945 if (ctx == NULL) { 946 return -ENOMEM; 947 } 948 949 ctx->hostnqn = strdup(hostnqn); 950 if (ctx->hostnqn == NULL) { 951 free(ctx); 952 return -ENOMEM; 953 } 954 955 ctx->subsystem = subsystem; 956 ctx->cb_fn = cb_fn; 957 ctx->cb_arg = cb_arg; 958 959 spdk_for_each_channel(subsystem->tgt, nvmf_subsystem_disconnect_qpairs_by_host, ctx, 960 nvmf_subsystem_disconnect_host_fini); 961 962 return 0; 963 } 964 965 int 966 spdk_nvmf_subsystem_set_allow_any_host(struct spdk_nvmf_subsystem *subsystem, bool allow_any_host) 967 { 968 pthread_mutex_lock(&subsystem->mutex); 969 subsystem->flags.allow_any_host = allow_any_host; 970 if (!TAILQ_EMPTY(&subsystem->listeners)) { 971 nvmf_update_discovery_log(subsystem->tgt, NULL); 972 } 973 pthread_mutex_unlock(&subsystem->mutex); 974 975 return 0; 976 } 977 978 bool 979 spdk_nvmf_subsystem_get_allow_any_host(const struct spdk_nvmf_subsystem *subsystem) 980 { 981 bool allow_any_host; 982 struct spdk_nvmf_subsystem *sub; 983 984 /* Technically, taking the mutex modifies data in the subsystem. But the const 985 * is still important to convey that this doesn't mutate any other data. Cast 986 * it away to work around this. */ 987 sub = (struct spdk_nvmf_subsystem *)subsystem; 988 989 pthread_mutex_lock(&sub->mutex); 990 allow_any_host = sub->flags.allow_any_host; 991 pthread_mutex_unlock(&sub->mutex); 992 993 return allow_any_host; 994 } 995 996 bool 997 spdk_nvmf_subsystem_host_allowed(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 998 { 999 bool allowed; 1000 1001 if (!hostnqn) { 1002 return false; 1003 } 1004 1005 pthread_mutex_lock(&subsystem->mutex); 1006 1007 if (subsystem->flags.allow_any_host) { 1008 pthread_mutex_unlock(&subsystem->mutex); 1009 return true; 1010 } 1011 1012 allowed = nvmf_subsystem_find_host(subsystem, hostnqn) != NULL; 1013 pthread_mutex_unlock(&subsystem->mutex); 1014 1015 return allowed; 1016 } 1017 1018 struct spdk_nvmf_host * 1019 spdk_nvmf_subsystem_get_first_host(struct spdk_nvmf_subsystem *subsystem) 1020 { 1021 return TAILQ_FIRST(&subsystem->hosts); 1022 } 1023 1024 1025 struct spdk_nvmf_host * 1026 spdk_nvmf_subsystem_get_next_host(struct spdk_nvmf_subsystem *subsystem, 1027 struct spdk_nvmf_host *prev_host) 1028 { 1029 return TAILQ_NEXT(prev_host, link); 1030 } 1031 1032 const char * 1033 spdk_nvmf_host_get_nqn(const struct spdk_nvmf_host *host) 1034 { 1035 return host->nqn; 1036 } 1037 1038 struct spdk_nvmf_subsystem_listener * 1039 nvmf_subsystem_find_listener(struct spdk_nvmf_subsystem *subsystem, 1040 const struct spdk_nvme_transport_id *trid) 1041 { 1042 struct spdk_nvmf_subsystem_listener *listener; 1043 1044 TAILQ_FOREACH(listener, &subsystem->listeners, link) { 1045 if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) { 1046 return listener; 1047 } 1048 } 1049 1050 return NULL; 1051 } 1052 1053 /** 1054 * Function to be called once the target is listening. 1055 * 1056 * \param ctx Context argument passed to this function. 1057 * \param status 0 if it completed successfully, or negative errno if it failed. 1058 */ 1059 static void 1060 _nvmf_subsystem_add_listener_done(void *ctx, int status) 1061 { 1062 struct spdk_nvmf_subsystem_listener *listener = ctx; 1063 1064 if (status) { 1065 listener->cb_fn(listener->cb_arg, status); 1066 free(listener); 1067 return; 1068 } 1069 1070 TAILQ_INSERT_HEAD(&listener->subsystem->listeners, listener, link); 1071 nvmf_update_discovery_log(listener->subsystem->tgt, NULL); 1072 listener->cb_fn(listener->cb_arg, status); 1073 } 1074 1075 void 1076 spdk_nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem, 1077 struct spdk_nvme_transport_id *trid, 1078 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 1079 void *cb_arg) 1080 { 1081 struct spdk_nvmf_transport *transport; 1082 struct spdk_nvmf_subsystem_listener *listener; 1083 struct spdk_nvmf_listener *tr_listener; 1084 uint32_t i; 1085 uint32_t id; 1086 int rc = 0; 1087 1088 assert(cb_fn != NULL); 1089 1090 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1091 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1092 cb_fn(cb_arg, -EAGAIN); 1093 return; 1094 } 1095 1096 if (nvmf_subsystem_find_listener(subsystem, trid)) { 1097 /* Listener already exists in this subsystem */ 1098 cb_fn(cb_arg, 0); 1099 return; 1100 } 1101 1102 transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, trid->trstring); 1103 if (!transport) { 1104 SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n", 1105 trid->trstring); 1106 cb_fn(cb_arg, -EINVAL); 1107 return; 1108 } 1109 1110 tr_listener = nvmf_transport_find_listener(transport, trid); 1111 if (!tr_listener) { 1112 SPDK_ERRLOG("Cannot find transport listener for %s\n", trid->traddr); 1113 cb_fn(cb_arg, -EINVAL); 1114 return; 1115 } 1116 1117 listener = calloc(1, sizeof(*listener)); 1118 if (!listener) { 1119 cb_fn(cb_arg, -ENOMEM); 1120 return; 1121 } 1122 1123 listener->trid = &tr_listener->trid; 1124 listener->transport = transport; 1125 listener->cb_fn = cb_fn; 1126 listener->cb_arg = cb_arg; 1127 listener->subsystem = subsystem; 1128 listener->ana_state = calloc(subsystem->max_nsid, sizeof(enum spdk_nvme_ana_state)); 1129 if (!listener->ana_state) { 1130 free(listener); 1131 cb_fn(cb_arg, -ENOMEM); 1132 return; 1133 } 1134 1135 id = spdk_bit_array_find_first_clear(subsystem->used_listener_ids, 0); 1136 if (id == UINT32_MAX) { 1137 SPDK_ERRLOG("Cannot add any more listeners\n"); 1138 free(listener->ana_state); 1139 free(listener); 1140 cb_fn(cb_arg, -EINVAL); 1141 return; 1142 } 1143 1144 spdk_bit_array_set(subsystem->used_listener_ids, id); 1145 listener->id = id; 1146 1147 for (i = 0; i < subsystem->max_nsid; i++) { 1148 listener->ana_state[i] = SPDK_NVME_ANA_OPTIMIZED_STATE; 1149 } 1150 1151 if (transport->ops->listen_associate != NULL) { 1152 rc = transport->ops->listen_associate(transport, subsystem, trid); 1153 } 1154 1155 SPDK_DTRACE_PROBE4(nvmf_subsystem_add_listener, subsystem->subnqn, listener->trid->trtype, 1156 listener->trid->traddr, listener->trid->trsvcid); 1157 1158 _nvmf_subsystem_add_listener_done(listener, rc); 1159 } 1160 1161 int 1162 spdk_nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem, 1163 const struct spdk_nvme_transport_id *trid) 1164 { 1165 struct spdk_nvmf_subsystem_listener *listener; 1166 1167 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1168 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1169 return -EAGAIN; 1170 } 1171 1172 listener = nvmf_subsystem_find_listener(subsystem, trid); 1173 if (listener == NULL) { 1174 return -ENOENT; 1175 } 1176 1177 SPDK_DTRACE_PROBE4(nvmf_subsystem_remove_listener, subsystem->subnqn, listener->trid->trtype, 1178 listener->trid->traddr, listener->trid->trsvcid); 1179 1180 _nvmf_subsystem_remove_listener(subsystem, listener, false); 1181 1182 return 0; 1183 } 1184 1185 void 1186 nvmf_subsystem_remove_all_listeners(struct spdk_nvmf_subsystem *subsystem, 1187 bool stop) 1188 { 1189 struct spdk_nvmf_subsystem_listener *listener, *listener_tmp; 1190 1191 TAILQ_FOREACH_SAFE(listener, &subsystem->listeners, link, listener_tmp) { 1192 _nvmf_subsystem_remove_listener(subsystem, listener, stop); 1193 } 1194 } 1195 1196 bool 1197 spdk_nvmf_subsystem_listener_allowed(struct spdk_nvmf_subsystem *subsystem, 1198 const struct spdk_nvme_transport_id *trid) 1199 { 1200 struct spdk_nvmf_subsystem_listener *listener; 1201 1202 TAILQ_FOREACH(listener, &subsystem->listeners, link) { 1203 if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) { 1204 return true; 1205 } 1206 } 1207 1208 if (!strcmp(subsystem->subnqn, SPDK_NVMF_DISCOVERY_NQN)) { 1209 SPDK_WARNLOG("Allowing connection to discovery subsystem on %s/%s/%s, " 1210 "even though this listener was not added to the discovery " 1211 "subsystem. This behavior is deprecated and will be removed " 1212 "in a future release.\n", 1213 spdk_nvme_transport_id_trtype_str(trid->trtype), trid->traddr, trid->trsvcid); 1214 return true; 1215 } 1216 1217 return false; 1218 } 1219 1220 struct spdk_nvmf_subsystem_listener * 1221 spdk_nvmf_subsystem_get_first_listener(struct spdk_nvmf_subsystem *subsystem) 1222 { 1223 return TAILQ_FIRST(&subsystem->listeners); 1224 } 1225 1226 struct spdk_nvmf_subsystem_listener * 1227 spdk_nvmf_subsystem_get_next_listener(struct spdk_nvmf_subsystem *subsystem, 1228 struct spdk_nvmf_subsystem_listener *prev_listener) 1229 { 1230 return TAILQ_NEXT(prev_listener, link); 1231 } 1232 1233 const struct spdk_nvme_transport_id * 1234 spdk_nvmf_subsystem_listener_get_trid(struct spdk_nvmf_subsystem_listener *listener) 1235 { 1236 return listener->trid; 1237 } 1238 1239 void 1240 spdk_nvmf_subsystem_allow_any_listener(struct spdk_nvmf_subsystem *subsystem, 1241 bool allow_any_listener) 1242 { 1243 subsystem->flags.allow_any_listener = allow_any_listener; 1244 } 1245 1246 bool 1247 spdk_nvmf_subsytem_any_listener_allowed(struct spdk_nvmf_subsystem *subsystem) 1248 { 1249 return subsystem->flags.allow_any_listener; 1250 } 1251 1252 1253 struct subsystem_update_ns_ctx { 1254 struct spdk_nvmf_subsystem *subsystem; 1255 1256 spdk_nvmf_subsystem_state_change_done cb_fn; 1257 void *cb_arg; 1258 }; 1259 1260 static void 1261 subsystem_update_ns_done(struct spdk_io_channel_iter *i, int status) 1262 { 1263 struct subsystem_update_ns_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 1264 1265 if (ctx->cb_fn) { 1266 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status); 1267 } 1268 free(ctx); 1269 } 1270 1271 static void 1272 subsystem_update_ns_on_pg(struct spdk_io_channel_iter *i) 1273 { 1274 int rc; 1275 struct subsystem_update_ns_ctx *ctx; 1276 struct spdk_nvmf_poll_group *group; 1277 struct spdk_nvmf_subsystem *subsystem; 1278 1279 ctx = spdk_io_channel_iter_get_ctx(i); 1280 group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i)); 1281 subsystem = ctx->subsystem; 1282 1283 rc = nvmf_poll_group_update_subsystem(group, subsystem); 1284 spdk_for_each_channel_continue(i, rc); 1285 } 1286 1287 static int 1288 nvmf_subsystem_update_ns(struct spdk_nvmf_subsystem *subsystem, spdk_channel_for_each_cpl cpl, 1289 void *ctx) 1290 { 1291 spdk_for_each_channel(subsystem->tgt, 1292 subsystem_update_ns_on_pg, 1293 ctx, 1294 cpl); 1295 1296 return 0; 1297 } 1298 1299 static void 1300 nvmf_subsystem_ns_changed(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1301 { 1302 struct spdk_nvmf_ctrlr *ctrlr; 1303 1304 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 1305 nvmf_ctrlr_ns_changed(ctrlr, nsid); 1306 } 1307 } 1308 1309 static uint32_t nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns); 1310 1311 int 1312 spdk_nvmf_subsystem_remove_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1313 { 1314 struct spdk_nvmf_transport *transport; 1315 struct spdk_nvmf_ns *ns; 1316 1317 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1318 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1319 assert(false); 1320 return -1; 1321 } 1322 1323 if (nsid == 0 || nsid > subsystem->max_nsid) { 1324 return -1; 1325 } 1326 1327 ns = subsystem->ns[nsid - 1]; 1328 if (!ns) { 1329 return -1; 1330 } 1331 1332 subsystem->ns[nsid - 1] = NULL; 1333 1334 assert(ns->anagrpid - 1 < subsystem->max_nsid); 1335 assert(subsystem->ana_group[ns->anagrpid - 1] > 0); 1336 1337 subsystem->ana_group[ns->anagrpid - 1]--; 1338 1339 free(ns->ptpl_file); 1340 nvmf_ns_reservation_clear_all_registrants(ns); 1341 spdk_bdev_module_release_bdev(ns->bdev); 1342 spdk_bdev_close(ns->desc); 1343 free(ns); 1344 1345 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 1346 transport = spdk_nvmf_transport_get_next(transport)) { 1347 if (transport->ops->subsystem_remove_ns) { 1348 transport->ops->subsystem_remove_ns(transport, subsystem, nsid); 1349 } 1350 } 1351 1352 nvmf_subsystem_ns_changed(subsystem, nsid); 1353 1354 return 0; 1355 } 1356 1357 struct subsystem_ns_change_ctx { 1358 struct spdk_nvmf_subsystem *subsystem; 1359 spdk_nvmf_subsystem_state_change_done cb_fn; 1360 uint32_t nsid; 1361 }; 1362 1363 static void 1364 _nvmf_ns_hot_remove(struct spdk_nvmf_subsystem *subsystem, 1365 void *cb_arg, int status) 1366 { 1367 struct subsystem_ns_change_ctx *ctx = cb_arg; 1368 int rc; 1369 1370 rc = spdk_nvmf_subsystem_remove_ns(subsystem, ctx->nsid); 1371 if (rc != 0) { 1372 SPDK_ERRLOG("Failed to make changes to NVME-oF subsystem with id: %u\n", subsystem->id); 1373 } 1374 1375 rc = spdk_nvmf_subsystem_resume(subsystem, NULL, NULL); 1376 if (rc != 0) { 1377 SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id); 1378 } 1379 1380 free(ctx); 1381 } 1382 1383 static void 1384 nvmf_ns_change_msg(void *ns_ctx) 1385 { 1386 struct subsystem_ns_change_ctx *ctx = ns_ctx; 1387 int rc; 1388 1389 SPDK_DTRACE_PROBE2(nvmf_ns_change, ctx->nsid, ctx->subsystem->subnqn); 1390 1391 rc = spdk_nvmf_subsystem_pause(ctx->subsystem, ctx->nsid, ctx->cb_fn, ctx); 1392 if (rc) { 1393 if (rc == -EBUSY) { 1394 /* Try again, this is not a permanent situation. */ 1395 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ctx); 1396 } else { 1397 free(ctx); 1398 SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n"); 1399 } 1400 } 1401 } 1402 1403 static void 1404 nvmf_ns_hot_remove(void *remove_ctx) 1405 { 1406 struct spdk_nvmf_ns *ns = remove_ctx; 1407 struct subsystem_ns_change_ctx *ns_ctx; 1408 int rc; 1409 1410 /* We have to allocate a new context because this op 1411 * is asynchronous and we could lose the ns in the middle. 1412 */ 1413 ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx)); 1414 if (!ns_ctx) { 1415 SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n"); 1416 return; 1417 } 1418 1419 ns_ctx->subsystem = ns->subsystem; 1420 ns_ctx->nsid = ns->opts.nsid; 1421 ns_ctx->cb_fn = _nvmf_ns_hot_remove; 1422 1423 rc = spdk_nvmf_subsystem_pause(ns->subsystem, ns_ctx->nsid, _nvmf_ns_hot_remove, ns_ctx); 1424 if (rc) { 1425 if (rc == -EBUSY) { 1426 /* Try again, this is not a permanent situation. */ 1427 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx); 1428 } else { 1429 SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n"); 1430 free(ns_ctx); 1431 } 1432 } 1433 } 1434 1435 static void 1436 _nvmf_ns_resize(struct spdk_nvmf_subsystem *subsystem, void *cb_arg, int status) 1437 { 1438 struct subsystem_ns_change_ctx *ctx = cb_arg; 1439 1440 nvmf_subsystem_ns_changed(subsystem, ctx->nsid); 1441 if (spdk_nvmf_subsystem_resume(subsystem, NULL, NULL) != 0) { 1442 SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id); 1443 } 1444 1445 free(ctx); 1446 } 1447 1448 static void 1449 nvmf_ns_resize(void *event_ctx) 1450 { 1451 struct spdk_nvmf_ns *ns = event_ctx; 1452 struct subsystem_ns_change_ctx *ns_ctx; 1453 int rc; 1454 1455 /* We have to allocate a new context because this op 1456 * is asynchronous and we could lose the ns in the middle. 1457 */ 1458 ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx)); 1459 if (!ns_ctx) { 1460 SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n"); 1461 return; 1462 } 1463 1464 ns_ctx->subsystem = ns->subsystem; 1465 ns_ctx->nsid = ns->opts.nsid; 1466 ns_ctx->cb_fn = _nvmf_ns_resize; 1467 1468 /* Specify 0 for the nsid here, because we do not need to pause the namespace. 1469 * Namespaces can only be resized bigger, so there is no need to quiesce I/O. 1470 */ 1471 rc = spdk_nvmf_subsystem_pause(ns->subsystem, 0, _nvmf_ns_resize, ns_ctx); 1472 if (rc) { 1473 if (rc == -EBUSY) { 1474 /* Try again, this is not a permanent situation. */ 1475 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx); 1476 } else { 1477 SPDK_ERRLOG("Unable to pause subsystem to process namespace resize!\n"); 1478 free(ns_ctx); 1479 } 1480 } 1481 } 1482 1483 static void 1484 nvmf_ns_event(enum spdk_bdev_event_type type, 1485 struct spdk_bdev *bdev, 1486 void *event_ctx) 1487 { 1488 SPDK_DEBUGLOG(nvmf, "Bdev event: type %d, name %s, subsystem_id %d, ns_id %d\n", 1489 type, 1490 spdk_bdev_get_name(bdev), 1491 ((struct spdk_nvmf_ns *)event_ctx)->subsystem->id, 1492 ((struct spdk_nvmf_ns *)event_ctx)->nsid); 1493 1494 switch (type) { 1495 case SPDK_BDEV_EVENT_REMOVE: 1496 nvmf_ns_hot_remove(event_ctx); 1497 break; 1498 case SPDK_BDEV_EVENT_RESIZE: 1499 nvmf_ns_resize(event_ctx); 1500 break; 1501 default: 1502 SPDK_NOTICELOG("Unsupported bdev event: type %d\n", type); 1503 break; 1504 } 1505 } 1506 1507 void 1508 spdk_nvmf_ns_opts_get_defaults(struct spdk_nvmf_ns_opts *opts, size_t opts_size) 1509 { 1510 if (!opts) { 1511 SPDK_ERRLOG("opts should not be NULL.\n"); 1512 return; 1513 } 1514 1515 if (!opts_size) { 1516 SPDK_ERRLOG("opts_size should not be zero.\n"); 1517 return; 1518 } 1519 1520 memset(opts, 0, opts_size); 1521 opts->opts_size = opts_size; 1522 1523 #define FIELD_OK(field) \ 1524 offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= opts_size 1525 1526 #define SET_FIELD(field, value) \ 1527 if (FIELD_OK(field)) { \ 1528 opts->field = value; \ 1529 } \ 1530 1531 /* All current fields are set to 0 by default. */ 1532 SET_FIELD(nsid, 0); 1533 if (FIELD_OK(nguid)) { 1534 memset(opts->nguid, 0, sizeof(opts->nguid)); 1535 } 1536 if (FIELD_OK(eui64)) { 1537 memset(opts->eui64, 0, sizeof(opts->eui64)); 1538 } 1539 if (FIELD_OK(uuid)) { 1540 spdk_uuid_set_null(&opts->uuid); 1541 } 1542 SET_FIELD(anagrpid, 0); 1543 1544 #undef FIELD_OK 1545 #undef SET_FIELD 1546 } 1547 1548 static void 1549 nvmf_ns_opts_copy(struct spdk_nvmf_ns_opts *opts, 1550 const struct spdk_nvmf_ns_opts *user_opts, 1551 size_t opts_size) 1552 { 1553 #define FIELD_OK(field) \ 1554 offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= user_opts->opts_size 1555 1556 #define SET_FIELD(field) \ 1557 if (FIELD_OK(field)) { \ 1558 opts->field = user_opts->field; \ 1559 } \ 1560 1561 SET_FIELD(nsid); 1562 if (FIELD_OK(nguid)) { 1563 memcpy(opts->nguid, user_opts->nguid, sizeof(opts->nguid)); 1564 } 1565 if (FIELD_OK(eui64)) { 1566 memcpy(opts->eui64, user_opts->eui64, sizeof(opts->eui64)); 1567 } 1568 if (FIELD_OK(uuid)) { 1569 spdk_uuid_copy(&opts->uuid, &user_opts->uuid); 1570 } 1571 SET_FIELD(anagrpid); 1572 1573 opts->opts_size = user_opts->opts_size; 1574 1575 /* We should not remove this statement, but need to update the assert statement 1576 * if we add a new field, and also add a corresponding SET_FIELD statement. 1577 */ 1578 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ns_opts) == 64, "Incorrect size"); 1579 1580 #undef FIELD_OK 1581 #undef SET_FIELD 1582 } 1583 1584 /* Dummy bdev module used to to claim bdevs. */ 1585 static struct spdk_bdev_module ns_bdev_module = { 1586 .name = "NVMe-oF Target", 1587 }; 1588 1589 static int nvmf_ns_load_reservation(const char *file, struct spdk_nvmf_reservation_info *info); 1590 static int nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, 1591 struct spdk_nvmf_reservation_info *info); 1592 1593 uint32_t 1594 spdk_nvmf_subsystem_add_ns_ext(struct spdk_nvmf_subsystem *subsystem, const char *bdev_name, 1595 const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size, 1596 const char *ptpl_file) 1597 { 1598 struct spdk_nvmf_transport *transport; 1599 struct spdk_nvmf_ns_opts opts; 1600 struct spdk_nvmf_ns *ns; 1601 struct spdk_nvmf_reservation_info info = {0}; 1602 int rc; 1603 bool zone_append_supported; 1604 uint64_t max_zone_append_size_kib; 1605 1606 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1607 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1608 return 0; 1609 } 1610 1611 spdk_nvmf_ns_opts_get_defaults(&opts, sizeof(opts)); 1612 if (user_opts) { 1613 nvmf_ns_opts_copy(&opts, user_opts, opts_size); 1614 } 1615 1616 if (opts.nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1617 SPDK_ERRLOG("Invalid NSID %" PRIu32 "\n", opts.nsid); 1618 return 0; 1619 } 1620 1621 if (opts.nsid == 0) { 1622 /* 1623 * NSID not specified - find a free index. 1624 * 1625 * If no free slots are found, opts.nsid will be subsystem->max_nsid + 1, which will 1626 * expand max_nsid if possible. 1627 */ 1628 for (opts.nsid = 1; opts.nsid <= subsystem->max_nsid; opts.nsid++) { 1629 if (_nvmf_subsystem_get_ns(subsystem, opts.nsid) == NULL) { 1630 break; 1631 } 1632 } 1633 } 1634 1635 if (_nvmf_subsystem_get_ns(subsystem, opts.nsid)) { 1636 SPDK_ERRLOG("Requested NSID %" PRIu32 " already in use\n", opts.nsid); 1637 return 0; 1638 } 1639 1640 if (opts.nsid > subsystem->max_nsid) { 1641 SPDK_ERRLOG("NSID greater than maximum not allowed\n"); 1642 return 0; 1643 } 1644 1645 if (opts.anagrpid == 0) { 1646 opts.anagrpid = opts.nsid; 1647 } 1648 1649 if (opts.anagrpid > subsystem->max_nsid) { 1650 SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n"); 1651 return 0; 1652 } 1653 1654 ns = calloc(1, sizeof(*ns)); 1655 if (ns == NULL) { 1656 SPDK_ERRLOG("Namespace allocation failed\n"); 1657 return 0; 1658 } 1659 1660 rc = spdk_bdev_open_ext(bdev_name, true, nvmf_ns_event, ns, &ns->desc); 1661 if (rc != 0) { 1662 SPDK_ERRLOG("Subsystem %s: bdev %s cannot be opened, error=%d\n", 1663 subsystem->subnqn, bdev_name, rc); 1664 free(ns); 1665 return 0; 1666 } 1667 1668 ns->bdev = spdk_bdev_desc_get_bdev(ns->desc); 1669 1670 if (spdk_bdev_get_md_size(ns->bdev) != 0) { 1671 if (!spdk_bdev_is_md_interleaved(ns->bdev)) { 1672 SPDK_ERRLOG("Can't attach bdev with separate metadata.\n"); 1673 spdk_bdev_close(ns->desc); 1674 free(ns); 1675 return 0; 1676 } 1677 1678 if (spdk_bdev_get_md_size(ns->bdev) > SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE) { 1679 SPDK_ERRLOG("Maximum supported interleaved md size %u, current md size %u\n", 1680 SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE, spdk_bdev_get_md_size(ns->bdev)); 1681 spdk_bdev_close(ns->desc); 1682 free(ns); 1683 return 0; 1684 } 1685 } 1686 1687 rc = spdk_bdev_module_claim_bdev(ns->bdev, ns->desc, &ns_bdev_module); 1688 if (rc != 0) { 1689 spdk_bdev_close(ns->desc); 1690 free(ns); 1691 return 0; 1692 } 1693 1694 /* Cache the zcopy capability of the bdev device */ 1695 ns->zcopy = spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 1696 1697 if (spdk_uuid_is_null(&opts.uuid)) { 1698 opts.uuid = *spdk_bdev_get_uuid(ns->bdev); 1699 } 1700 1701 /* if nguid descriptor is supported by bdev module (nvme) then uuid = nguid */ 1702 if (spdk_mem_all_zero(opts.nguid, sizeof(opts.nguid))) { 1703 SPDK_STATIC_ASSERT(sizeof(opts.nguid) == sizeof(opts.uuid), "size mismatch"); 1704 memcpy(opts.nguid, spdk_bdev_get_uuid(ns->bdev), sizeof(opts.nguid)); 1705 } 1706 1707 if (spdk_bdev_is_zoned(ns->bdev)) { 1708 SPDK_DEBUGLOG(nvmf, "The added namespace is backed by a zoned block device.\n"); 1709 ns->csi = SPDK_NVME_CSI_ZNS; 1710 1711 zone_append_supported = spdk_bdev_io_type_supported(ns->bdev, 1712 SPDK_BDEV_IO_TYPE_ZONE_APPEND); 1713 max_zone_append_size_kib = spdk_bdev_get_max_zone_append_size( 1714 ns->bdev) * spdk_bdev_get_block_size(ns->bdev); 1715 1716 if (_nvmf_subsystem_get_first_zoned_ns(subsystem) != NULL && 1717 (subsystem->zone_append_supported != zone_append_supported || 1718 subsystem->max_zone_append_size_kib != max_zone_append_size_kib)) { 1719 SPDK_ERRLOG("Namespaces with different zone append support or different zone append size are not allowed.\n"); 1720 goto err_ns_reservation_restore; 1721 } 1722 1723 subsystem->zone_append_supported = zone_append_supported; 1724 subsystem->max_zone_append_size_kib = max_zone_append_size_kib; 1725 } 1726 1727 ns->opts = opts; 1728 ns->subsystem = subsystem; 1729 subsystem->ns[opts.nsid - 1] = ns; 1730 ns->nsid = opts.nsid; 1731 ns->anagrpid = opts.anagrpid; 1732 subsystem->ana_group[ns->anagrpid - 1]++; 1733 TAILQ_INIT(&ns->registrants); 1734 if (ptpl_file) { 1735 rc = nvmf_ns_load_reservation(ptpl_file, &info); 1736 if (!rc) { 1737 rc = nvmf_ns_reservation_restore(ns, &info); 1738 if (rc) { 1739 SPDK_ERRLOG("Subsystem restore reservation failed\n"); 1740 goto err_ns_reservation_restore; 1741 } 1742 } 1743 ns->ptpl_file = strdup(ptpl_file); 1744 if (!ns->ptpl_file) { 1745 SPDK_ERRLOG("Namespace ns->ptpl_file allocation failed\n"); 1746 goto err_strdup; 1747 } 1748 } 1749 1750 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 1751 transport = spdk_nvmf_transport_get_next(transport)) { 1752 if (transport->ops->subsystem_add_ns) { 1753 rc = transport->ops->subsystem_add_ns(transport, subsystem, ns); 1754 if (rc) { 1755 SPDK_ERRLOG("Namespace attachment is not allowed by %s transport\n", transport->ops->name); 1756 goto err_subsystem_add_ns; 1757 } 1758 } 1759 } 1760 1761 SPDK_DEBUGLOG(nvmf, "Subsystem %s: bdev %s assigned nsid %" PRIu32 "\n", 1762 spdk_nvmf_subsystem_get_nqn(subsystem), 1763 bdev_name, 1764 opts.nsid); 1765 1766 nvmf_subsystem_ns_changed(subsystem, opts.nsid); 1767 1768 SPDK_DTRACE_PROBE2(nvmf_subsystem_add_ns, subsystem->subnqn, ns->nsid); 1769 1770 return opts.nsid; 1771 1772 err_subsystem_add_ns: 1773 free(ns->ptpl_file); 1774 err_strdup: 1775 nvmf_ns_reservation_clear_all_registrants(ns); 1776 err_ns_reservation_restore: 1777 subsystem->ns[opts.nsid - 1] = NULL; 1778 spdk_bdev_module_release_bdev(ns->bdev); 1779 spdk_bdev_close(ns->desc); 1780 free(ns); 1781 1782 return 0; 1783 } 1784 1785 static uint32_t 1786 nvmf_subsystem_get_next_allocated_nsid(struct spdk_nvmf_subsystem *subsystem, 1787 uint32_t prev_nsid) 1788 { 1789 uint32_t nsid; 1790 1791 if (prev_nsid >= subsystem->max_nsid) { 1792 return 0; 1793 } 1794 1795 for (nsid = prev_nsid + 1; nsid <= subsystem->max_nsid; nsid++) { 1796 if (subsystem->ns[nsid - 1]) { 1797 return nsid; 1798 } 1799 } 1800 1801 return 0; 1802 } 1803 1804 struct spdk_nvmf_ns * 1805 spdk_nvmf_subsystem_get_first_ns(struct spdk_nvmf_subsystem *subsystem) 1806 { 1807 uint32_t first_nsid; 1808 1809 first_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, 0); 1810 return _nvmf_subsystem_get_ns(subsystem, first_nsid); 1811 } 1812 1813 struct spdk_nvmf_ns * 1814 spdk_nvmf_subsystem_get_next_ns(struct spdk_nvmf_subsystem *subsystem, 1815 struct spdk_nvmf_ns *prev_ns) 1816 { 1817 uint32_t next_nsid; 1818 1819 next_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, prev_ns->opts.nsid); 1820 return _nvmf_subsystem_get_ns(subsystem, next_nsid); 1821 } 1822 1823 struct spdk_nvmf_ns * 1824 spdk_nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1825 { 1826 return _nvmf_subsystem_get_ns(subsystem, nsid); 1827 } 1828 1829 uint32_t 1830 spdk_nvmf_ns_get_id(const struct spdk_nvmf_ns *ns) 1831 { 1832 return ns->opts.nsid; 1833 } 1834 1835 struct spdk_bdev * 1836 spdk_nvmf_ns_get_bdev(struct spdk_nvmf_ns *ns) 1837 { 1838 return ns->bdev; 1839 } 1840 1841 void 1842 spdk_nvmf_ns_get_opts(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_ns_opts *opts, 1843 size_t opts_size) 1844 { 1845 memset(opts, 0, opts_size); 1846 memcpy(opts, &ns->opts, spdk_min(sizeof(ns->opts), opts_size)); 1847 } 1848 1849 const char * 1850 spdk_nvmf_subsystem_get_sn(const struct spdk_nvmf_subsystem *subsystem) 1851 { 1852 return subsystem->sn; 1853 } 1854 1855 int 1856 spdk_nvmf_subsystem_set_sn(struct spdk_nvmf_subsystem *subsystem, const char *sn) 1857 { 1858 size_t len, max_len; 1859 1860 max_len = sizeof(subsystem->sn) - 1; 1861 len = strlen(sn); 1862 if (len > max_len) { 1863 SPDK_DEBUGLOG(nvmf, "Invalid sn \"%s\": length %zu > max %zu\n", 1864 sn, len, max_len); 1865 return -1; 1866 } 1867 1868 if (!nvmf_valid_ascii_string(sn, len)) { 1869 SPDK_DEBUGLOG(nvmf, "Non-ASCII sn\n"); 1870 SPDK_LOGDUMP(nvmf, "sn", sn, len); 1871 return -1; 1872 } 1873 1874 snprintf(subsystem->sn, sizeof(subsystem->sn), "%s", sn); 1875 1876 return 0; 1877 } 1878 1879 const char * 1880 spdk_nvmf_subsystem_get_mn(const struct spdk_nvmf_subsystem *subsystem) 1881 { 1882 return subsystem->mn; 1883 } 1884 1885 int 1886 spdk_nvmf_subsystem_set_mn(struct spdk_nvmf_subsystem *subsystem, const char *mn) 1887 { 1888 size_t len, max_len; 1889 1890 if (mn == NULL) { 1891 mn = MODEL_NUMBER_DEFAULT; 1892 } 1893 max_len = sizeof(subsystem->mn) - 1; 1894 len = strlen(mn); 1895 if (len > max_len) { 1896 SPDK_DEBUGLOG(nvmf, "Invalid mn \"%s\": length %zu > max %zu\n", 1897 mn, len, max_len); 1898 return -1; 1899 } 1900 1901 if (!nvmf_valid_ascii_string(mn, len)) { 1902 SPDK_DEBUGLOG(nvmf, "Non-ASCII mn\n"); 1903 SPDK_LOGDUMP(nvmf, "mn", mn, len); 1904 return -1; 1905 } 1906 1907 snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", mn); 1908 1909 return 0; 1910 } 1911 1912 const char * 1913 spdk_nvmf_subsystem_get_nqn(const struct spdk_nvmf_subsystem *subsystem) 1914 { 1915 return subsystem->subnqn; 1916 } 1917 1918 /* We have to use the typedef in the function declaration to appease astyle. */ 1919 typedef enum spdk_nvmf_subtype spdk_nvmf_subtype_t; 1920 1921 spdk_nvmf_subtype_t 1922 spdk_nvmf_subsystem_get_type(struct spdk_nvmf_subsystem *subsystem) 1923 { 1924 return subsystem->subtype; 1925 } 1926 1927 uint32_t 1928 spdk_nvmf_subsystem_get_max_nsid(struct spdk_nvmf_subsystem *subsystem) 1929 { 1930 return subsystem->max_nsid; 1931 } 1932 1933 int 1934 nvmf_subsystem_set_cntlid_range(struct spdk_nvmf_subsystem *subsystem, 1935 uint16_t min_cntlid, uint16_t max_cntlid) 1936 { 1937 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 1938 return -EAGAIN; 1939 } 1940 1941 if (min_cntlid > max_cntlid) { 1942 return -EINVAL; 1943 } 1944 /* The spec reserves cntlid values in the range FFF0h to FFFFh. */ 1945 if (min_cntlid < NVMF_MIN_CNTLID || min_cntlid > NVMF_MAX_CNTLID || 1946 max_cntlid < NVMF_MIN_CNTLID || max_cntlid > NVMF_MAX_CNTLID) { 1947 return -EINVAL; 1948 } 1949 subsystem->min_cntlid = min_cntlid; 1950 subsystem->max_cntlid = max_cntlid; 1951 if (subsystem->next_cntlid < min_cntlid || subsystem->next_cntlid > max_cntlid - 1) { 1952 subsystem->next_cntlid = min_cntlid - 1; 1953 } 1954 1955 return 0; 1956 } 1957 1958 static uint16_t 1959 nvmf_subsystem_gen_cntlid(struct spdk_nvmf_subsystem *subsystem) 1960 { 1961 int count; 1962 1963 /* 1964 * In the worst case, we might have to try all CNTLID values between min_cntlid and max_cntlid 1965 * before we find one that is unused (or find that all values are in use). 1966 */ 1967 for (count = 0; count < subsystem->max_cntlid - subsystem->min_cntlid + 1; count++) { 1968 subsystem->next_cntlid++; 1969 if (subsystem->next_cntlid > subsystem->max_cntlid) { 1970 subsystem->next_cntlid = subsystem->min_cntlid; 1971 } 1972 1973 /* Check if a controller with this cntlid currently exists. */ 1974 if (nvmf_subsystem_get_ctrlr(subsystem, subsystem->next_cntlid) == NULL) { 1975 /* Found unused cntlid */ 1976 return subsystem->next_cntlid; 1977 } 1978 } 1979 1980 /* All valid cntlid values are in use. */ 1981 return 0xFFFF; 1982 } 1983 1984 int 1985 nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_ctrlr *ctrlr) 1986 { 1987 1988 if (ctrlr->dynamic_ctrlr) { 1989 ctrlr->cntlid = nvmf_subsystem_gen_cntlid(subsystem); 1990 if (ctrlr->cntlid == 0xFFFF) { 1991 /* Unable to get a cntlid */ 1992 SPDK_ERRLOG("Reached max simultaneous ctrlrs\n"); 1993 return -EBUSY; 1994 } 1995 } else if (nvmf_subsystem_get_ctrlr(subsystem, ctrlr->cntlid) != NULL) { 1996 SPDK_ERRLOG("Ctrlr with cntlid %u already exist\n", ctrlr->cntlid); 1997 return -EEXIST; 1998 } 1999 2000 TAILQ_INSERT_TAIL(&subsystem->ctrlrs, ctrlr, link); 2001 2002 SPDK_DTRACE_PROBE3(nvmf_subsystem_add_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn); 2003 2004 return 0; 2005 } 2006 2007 void 2008 nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem, 2009 struct spdk_nvmf_ctrlr *ctrlr) 2010 { 2011 SPDK_DTRACE_PROBE3(nvmf_subsystem_remove_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn); 2012 2013 assert(spdk_get_thread() == subsystem->thread); 2014 assert(subsystem == ctrlr->subsys); 2015 SPDK_DEBUGLOG(nvmf, "remove ctrlr %p id 0x%x from subsys %p %s\n", ctrlr, ctrlr->cntlid, subsystem, 2016 subsystem->subnqn); 2017 TAILQ_REMOVE(&subsystem->ctrlrs, ctrlr, link); 2018 } 2019 2020 struct spdk_nvmf_ctrlr * 2021 nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, uint16_t cntlid) 2022 { 2023 struct spdk_nvmf_ctrlr *ctrlr; 2024 2025 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 2026 if (ctrlr->cntlid == cntlid) { 2027 return ctrlr; 2028 } 2029 } 2030 2031 return NULL; 2032 } 2033 2034 uint32_t 2035 spdk_nvmf_subsystem_get_max_namespaces(const struct spdk_nvmf_subsystem *subsystem) 2036 { 2037 return subsystem->max_nsid; 2038 } 2039 2040 uint16_t 2041 spdk_nvmf_subsystem_get_min_cntlid(const struct spdk_nvmf_subsystem *subsystem) 2042 { 2043 return subsystem->min_cntlid; 2044 } 2045 2046 uint16_t 2047 spdk_nvmf_subsystem_get_max_cntlid(const struct spdk_nvmf_subsystem *subsystem) 2048 { 2049 return subsystem->max_cntlid; 2050 } 2051 2052 struct _nvmf_ns_registrant { 2053 uint64_t rkey; 2054 char *host_uuid; 2055 }; 2056 2057 struct _nvmf_ns_registrants { 2058 size_t num_regs; 2059 struct _nvmf_ns_registrant reg[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2060 }; 2061 2062 struct _nvmf_ns_reservation { 2063 bool ptpl_activated; 2064 enum spdk_nvme_reservation_type rtype; 2065 uint64_t crkey; 2066 char *bdev_uuid; 2067 char *holder_uuid; 2068 struct _nvmf_ns_registrants regs; 2069 }; 2070 2071 static const struct spdk_json_object_decoder nvmf_ns_pr_reg_decoders[] = { 2072 {"rkey", offsetof(struct _nvmf_ns_registrant, rkey), spdk_json_decode_uint64}, 2073 {"host_uuid", offsetof(struct _nvmf_ns_registrant, host_uuid), spdk_json_decode_string}, 2074 }; 2075 2076 static int 2077 nvmf_decode_ns_pr_reg(const struct spdk_json_val *val, void *out) 2078 { 2079 struct _nvmf_ns_registrant *reg = out; 2080 2081 return spdk_json_decode_object(val, nvmf_ns_pr_reg_decoders, 2082 SPDK_COUNTOF(nvmf_ns_pr_reg_decoders), reg); 2083 } 2084 2085 static int 2086 nvmf_decode_ns_pr_regs(const struct spdk_json_val *val, void *out) 2087 { 2088 struct _nvmf_ns_registrants *regs = out; 2089 2090 return spdk_json_decode_array(val, nvmf_decode_ns_pr_reg, regs->reg, 2091 SPDK_NVMF_MAX_NUM_REGISTRANTS, ®s->num_regs, 2092 sizeof(struct _nvmf_ns_registrant)); 2093 } 2094 2095 static const struct spdk_json_object_decoder nvmf_ns_pr_decoders[] = { 2096 {"ptpl", offsetof(struct _nvmf_ns_reservation, ptpl_activated), spdk_json_decode_bool, true}, 2097 {"rtype", offsetof(struct _nvmf_ns_reservation, rtype), spdk_json_decode_uint32, true}, 2098 {"crkey", offsetof(struct _nvmf_ns_reservation, crkey), spdk_json_decode_uint64, true}, 2099 {"bdev_uuid", offsetof(struct _nvmf_ns_reservation, bdev_uuid), spdk_json_decode_string}, 2100 {"holder_uuid", offsetof(struct _nvmf_ns_reservation, holder_uuid), spdk_json_decode_string, true}, 2101 {"registrants", offsetof(struct _nvmf_ns_reservation, regs), nvmf_decode_ns_pr_regs}, 2102 }; 2103 2104 static int 2105 nvmf_ns_load_reservation(const char *file, struct spdk_nvmf_reservation_info *info) 2106 { 2107 FILE *fd; 2108 size_t json_size; 2109 ssize_t values_cnt, rc; 2110 void *json = NULL, *end; 2111 struct spdk_json_val *values = NULL; 2112 struct _nvmf_ns_reservation res = {}; 2113 uint32_t i; 2114 2115 fd = fopen(file, "r"); 2116 /* It's not an error if the file does not exist */ 2117 if (!fd) { 2118 SPDK_NOTICELOG("File %s does not exist\n", file); 2119 return -ENOENT; 2120 } 2121 2122 /* Load all persist file contents into a local buffer */ 2123 json = spdk_posix_file_load(fd, &json_size); 2124 fclose(fd); 2125 if (!json) { 2126 SPDK_ERRLOG("Load persit file %s failed\n", file); 2127 return -ENOMEM; 2128 } 2129 2130 rc = spdk_json_parse(json, json_size, NULL, 0, &end, 0); 2131 if (rc < 0) { 2132 SPDK_NOTICELOG("Parsing JSON configuration failed (%zd)\n", rc); 2133 goto exit; 2134 } 2135 2136 values_cnt = rc; 2137 values = calloc(values_cnt, sizeof(struct spdk_json_val)); 2138 if (values == NULL) { 2139 goto exit; 2140 } 2141 2142 rc = spdk_json_parse(json, json_size, values, values_cnt, &end, 0); 2143 if (rc != values_cnt) { 2144 SPDK_ERRLOG("Parsing JSON configuration failed (%zd)\n", rc); 2145 goto exit; 2146 } 2147 2148 /* Decode json */ 2149 if (spdk_json_decode_object(values, nvmf_ns_pr_decoders, 2150 SPDK_COUNTOF(nvmf_ns_pr_decoders), 2151 &res)) { 2152 SPDK_ERRLOG("Invalid objects in the persist file %s\n", file); 2153 rc = -EINVAL; 2154 goto exit; 2155 } 2156 2157 if (res.regs.num_regs > SPDK_NVMF_MAX_NUM_REGISTRANTS) { 2158 SPDK_ERRLOG("Can only support up to %u registrants\n", SPDK_NVMF_MAX_NUM_REGISTRANTS); 2159 rc = -ERANGE; 2160 goto exit; 2161 } 2162 2163 rc = 0; 2164 info->ptpl_activated = res.ptpl_activated; 2165 info->rtype = res.rtype; 2166 info->crkey = res.crkey; 2167 snprintf(info->bdev_uuid, sizeof(info->bdev_uuid), "%s", res.bdev_uuid); 2168 snprintf(info->holder_uuid, sizeof(info->holder_uuid), "%s", res.holder_uuid); 2169 info->num_regs = res.regs.num_regs; 2170 for (i = 0; i < res.regs.num_regs; i++) { 2171 info->registrants[i].rkey = res.regs.reg[i].rkey; 2172 snprintf(info->registrants[i].host_uuid, sizeof(info->registrants[i].host_uuid), "%s", 2173 res.regs.reg[i].host_uuid); 2174 } 2175 2176 exit: 2177 free(json); 2178 free(values); 2179 free(res.bdev_uuid); 2180 free(res.holder_uuid); 2181 for (i = 0; i < res.regs.num_regs; i++) { 2182 free(res.regs.reg[i].host_uuid); 2183 } 2184 2185 return rc; 2186 } 2187 2188 static bool nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns); 2189 2190 static int 2191 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info) 2192 { 2193 uint32_t i; 2194 struct spdk_nvmf_registrant *reg, *holder = NULL; 2195 struct spdk_uuid bdev_uuid, holder_uuid; 2196 bool rkey_flag = false; 2197 2198 SPDK_DEBUGLOG(nvmf, "NSID %u, PTPL %u, Number of registrants %u\n", 2199 ns->nsid, info->ptpl_activated, info->num_regs); 2200 2201 /* it's not an error */ 2202 if (!info->ptpl_activated || !info->num_regs) { 2203 return 0; 2204 } 2205 2206 /* Check info->crkey exist or not in info->registrants[i].rkey */ 2207 for (i = 0; i < info->num_regs; i++) { 2208 if (info->crkey == info->registrants[i].rkey) { 2209 rkey_flag = true; 2210 } 2211 } 2212 if (!rkey_flag) { 2213 return -EINVAL; 2214 } 2215 2216 spdk_uuid_parse(&bdev_uuid, info->bdev_uuid); 2217 if (spdk_uuid_compare(&bdev_uuid, spdk_bdev_get_uuid(ns->bdev))) { 2218 SPDK_ERRLOG("Existing bdev UUID is not same with configuration file\n"); 2219 return -EINVAL; 2220 } 2221 2222 ns->crkey = info->crkey; 2223 ns->rtype = info->rtype; 2224 ns->ptpl_activated = info->ptpl_activated; 2225 spdk_uuid_parse(&holder_uuid, info->holder_uuid); 2226 2227 SPDK_DEBUGLOG(nvmf, "Bdev UUID %s\n", info->bdev_uuid); 2228 if (info->rtype) { 2229 SPDK_DEBUGLOG(nvmf, "Holder UUID %s, RTYPE %u, RKEY 0x%"PRIx64"\n", 2230 info->holder_uuid, info->rtype, info->crkey); 2231 } 2232 2233 for (i = 0; i < info->num_regs; i++) { 2234 reg = calloc(1, sizeof(*reg)); 2235 if (!reg) { 2236 return -ENOMEM; 2237 } 2238 spdk_uuid_parse(®->hostid, info->registrants[i].host_uuid); 2239 reg->rkey = info->registrants[i].rkey; 2240 TAILQ_INSERT_TAIL(&ns->registrants, reg, link); 2241 if (!spdk_uuid_compare(&holder_uuid, ®->hostid)) { 2242 holder = reg; 2243 } 2244 SPDK_DEBUGLOG(nvmf, "Registrant RKEY 0x%"PRIx64", Host UUID %s\n", 2245 info->registrants[i].rkey, info->registrants[i].host_uuid); 2246 } 2247 2248 if (nvmf_ns_reservation_all_registrants_type(ns)) { 2249 ns->holder = TAILQ_FIRST(&ns->registrants); 2250 } else { 2251 ns->holder = holder; 2252 } 2253 2254 return 0; 2255 } 2256 2257 static int 2258 nvmf_ns_json_write_cb(void *cb_ctx, const void *data, size_t size) 2259 { 2260 char *file = cb_ctx; 2261 size_t rc; 2262 FILE *fd; 2263 2264 fd = fopen(file, "w"); 2265 if (!fd) { 2266 SPDK_ERRLOG("Can't open file %s for write\n", file); 2267 return -ENOENT; 2268 } 2269 rc = fwrite(data, 1, size, fd); 2270 fclose(fd); 2271 2272 return rc == size ? 0 : -1; 2273 } 2274 2275 static int 2276 nvmf_ns_reservation_update(const char *file, struct spdk_nvmf_reservation_info *info) 2277 { 2278 struct spdk_json_write_ctx *w; 2279 uint32_t i; 2280 int rc = 0; 2281 2282 w = spdk_json_write_begin(nvmf_ns_json_write_cb, (void *)file, 0); 2283 if (w == NULL) { 2284 return -ENOMEM; 2285 } 2286 /* clear the configuration file */ 2287 if (!info->ptpl_activated) { 2288 goto exit; 2289 } 2290 2291 spdk_json_write_object_begin(w); 2292 spdk_json_write_named_bool(w, "ptpl", info->ptpl_activated); 2293 spdk_json_write_named_uint32(w, "rtype", info->rtype); 2294 spdk_json_write_named_uint64(w, "crkey", info->crkey); 2295 spdk_json_write_named_string(w, "bdev_uuid", info->bdev_uuid); 2296 spdk_json_write_named_string(w, "holder_uuid", info->holder_uuid); 2297 2298 spdk_json_write_named_array_begin(w, "registrants"); 2299 for (i = 0; i < info->num_regs; i++) { 2300 spdk_json_write_object_begin(w); 2301 spdk_json_write_named_uint64(w, "rkey", info->registrants[i].rkey); 2302 spdk_json_write_named_string(w, "host_uuid", info->registrants[i].host_uuid); 2303 spdk_json_write_object_end(w); 2304 } 2305 spdk_json_write_array_end(w); 2306 spdk_json_write_object_end(w); 2307 2308 exit: 2309 rc = spdk_json_write_end(w); 2310 return rc; 2311 } 2312 2313 static int 2314 nvmf_ns_update_reservation_info(struct spdk_nvmf_ns *ns) 2315 { 2316 struct spdk_nvmf_reservation_info info; 2317 struct spdk_nvmf_registrant *reg, *tmp; 2318 uint32_t i = 0; 2319 2320 assert(ns != NULL); 2321 2322 if (!ns->bdev || !ns->ptpl_file) { 2323 return 0; 2324 } 2325 2326 memset(&info, 0, sizeof(info)); 2327 spdk_uuid_fmt_lower(info.bdev_uuid, sizeof(info.bdev_uuid), spdk_bdev_get_uuid(ns->bdev)); 2328 2329 if (ns->rtype) { 2330 info.rtype = ns->rtype; 2331 info.crkey = ns->crkey; 2332 if (!nvmf_ns_reservation_all_registrants_type(ns)) { 2333 assert(ns->holder != NULL); 2334 spdk_uuid_fmt_lower(info.holder_uuid, sizeof(info.holder_uuid), &ns->holder->hostid); 2335 } 2336 } 2337 2338 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2339 spdk_uuid_fmt_lower(info.registrants[i].host_uuid, sizeof(info.registrants[i].host_uuid), 2340 ®->hostid); 2341 info.registrants[i++].rkey = reg->rkey; 2342 } 2343 2344 info.num_regs = i; 2345 info.ptpl_activated = ns->ptpl_activated; 2346 2347 return nvmf_ns_reservation_update(ns->ptpl_file, &info); 2348 } 2349 2350 static struct spdk_nvmf_registrant * 2351 nvmf_ns_reservation_get_registrant(struct spdk_nvmf_ns *ns, 2352 struct spdk_uuid *uuid) 2353 { 2354 struct spdk_nvmf_registrant *reg, *tmp; 2355 2356 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2357 if (!spdk_uuid_compare(®->hostid, uuid)) { 2358 return reg; 2359 } 2360 } 2361 2362 return NULL; 2363 } 2364 2365 /* Generate reservation notice log to registered HostID controllers */ 2366 static void 2367 nvmf_subsystem_gen_ctrlr_notification(struct spdk_nvmf_subsystem *subsystem, 2368 struct spdk_nvmf_ns *ns, 2369 struct spdk_uuid *hostid_list, 2370 uint32_t num_hostid, 2371 enum spdk_nvme_reservation_notification_log_page_type type) 2372 { 2373 struct spdk_nvmf_ctrlr *ctrlr; 2374 uint32_t i; 2375 2376 for (i = 0; i < num_hostid; i++) { 2377 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 2378 if (!spdk_uuid_compare(&ctrlr->hostid, &hostid_list[i])) { 2379 nvmf_ctrlr_reservation_notice_log(ctrlr, ns, type); 2380 } 2381 } 2382 } 2383 } 2384 2385 /* Get all registrants' hostid other than the controller who issued the command */ 2386 static uint32_t 2387 nvmf_ns_reservation_get_all_other_hostid(struct spdk_nvmf_ns *ns, 2388 struct spdk_uuid *hostid_list, 2389 uint32_t max_num_hostid, 2390 struct spdk_uuid *current_hostid) 2391 { 2392 struct spdk_nvmf_registrant *reg, *tmp; 2393 uint32_t num_hostid = 0; 2394 2395 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2396 if (spdk_uuid_compare(®->hostid, current_hostid)) { 2397 if (num_hostid == max_num_hostid) { 2398 assert(false); 2399 return max_num_hostid; 2400 } 2401 hostid_list[num_hostid++] = reg->hostid; 2402 } 2403 } 2404 2405 return num_hostid; 2406 } 2407 2408 /* Calculate the unregistered HostID list according to list 2409 * prior to execute preempt command and list after executing 2410 * preempt command. 2411 */ 2412 static uint32_t 2413 nvmf_ns_reservation_get_unregistered_hostid(struct spdk_uuid *old_hostid_list, 2414 uint32_t old_num_hostid, 2415 struct spdk_uuid *remaining_hostid_list, 2416 uint32_t remaining_num_hostid) 2417 { 2418 struct spdk_uuid temp_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2419 uint32_t i, j, num_hostid = 0; 2420 bool found; 2421 2422 if (!remaining_num_hostid) { 2423 return old_num_hostid; 2424 } 2425 2426 for (i = 0; i < old_num_hostid; i++) { 2427 found = false; 2428 for (j = 0; j < remaining_num_hostid; j++) { 2429 if (!spdk_uuid_compare(&old_hostid_list[i], &remaining_hostid_list[j])) { 2430 found = true; 2431 break; 2432 } 2433 } 2434 if (!found) { 2435 spdk_uuid_copy(&temp_hostid_list[num_hostid++], &old_hostid_list[i]); 2436 } 2437 } 2438 2439 if (num_hostid) { 2440 memcpy(old_hostid_list, temp_hostid_list, sizeof(struct spdk_uuid) * num_hostid); 2441 } 2442 2443 return num_hostid; 2444 } 2445 2446 /* current reservation type is all registrants or not */ 2447 static bool 2448 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns) 2449 { 2450 return (ns->rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS || 2451 ns->rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS); 2452 } 2453 2454 /* current registrant is reservation holder or not */ 2455 static bool 2456 nvmf_ns_reservation_registrant_is_holder(struct spdk_nvmf_ns *ns, 2457 struct spdk_nvmf_registrant *reg) 2458 { 2459 if (!reg) { 2460 return false; 2461 } 2462 2463 if (nvmf_ns_reservation_all_registrants_type(ns)) { 2464 return true; 2465 } 2466 2467 return (ns->holder == reg); 2468 } 2469 2470 static int 2471 nvmf_ns_reservation_add_registrant(struct spdk_nvmf_ns *ns, 2472 struct spdk_nvmf_ctrlr *ctrlr, 2473 uint64_t nrkey) 2474 { 2475 struct spdk_nvmf_registrant *reg; 2476 2477 reg = calloc(1, sizeof(*reg)); 2478 if (!reg) { 2479 return -ENOMEM; 2480 } 2481 2482 reg->rkey = nrkey; 2483 /* set hostid for the registrant */ 2484 spdk_uuid_copy(®->hostid, &ctrlr->hostid); 2485 TAILQ_INSERT_TAIL(&ns->registrants, reg, link); 2486 ns->gen++; 2487 2488 return 0; 2489 } 2490 2491 static void 2492 nvmf_ns_reservation_release_reservation(struct spdk_nvmf_ns *ns) 2493 { 2494 ns->rtype = 0; 2495 ns->crkey = 0; 2496 ns->holder = NULL; 2497 } 2498 2499 /* release the reservation if the last registrant was removed */ 2500 static void 2501 nvmf_ns_reservation_check_release_on_remove_registrant(struct spdk_nvmf_ns *ns, 2502 struct spdk_nvmf_registrant *reg) 2503 { 2504 struct spdk_nvmf_registrant *next_reg; 2505 2506 /* no reservation holder */ 2507 if (!ns->holder) { 2508 assert(ns->rtype == 0); 2509 return; 2510 } 2511 2512 next_reg = TAILQ_FIRST(&ns->registrants); 2513 if (next_reg && nvmf_ns_reservation_all_registrants_type(ns)) { 2514 /* the next valid registrant is the new holder now */ 2515 ns->holder = next_reg; 2516 } else if (nvmf_ns_reservation_registrant_is_holder(ns, reg)) { 2517 /* release the reservation */ 2518 nvmf_ns_reservation_release_reservation(ns); 2519 } 2520 } 2521 2522 static void 2523 nvmf_ns_reservation_remove_registrant(struct spdk_nvmf_ns *ns, 2524 struct spdk_nvmf_registrant *reg) 2525 { 2526 TAILQ_REMOVE(&ns->registrants, reg, link); 2527 nvmf_ns_reservation_check_release_on_remove_registrant(ns, reg); 2528 free(reg); 2529 ns->gen++; 2530 return; 2531 } 2532 2533 static uint32_t 2534 nvmf_ns_reservation_remove_registrants_by_key(struct spdk_nvmf_ns *ns, 2535 uint64_t rkey) 2536 { 2537 struct spdk_nvmf_registrant *reg, *tmp; 2538 uint32_t count = 0; 2539 2540 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2541 if (reg->rkey == rkey) { 2542 nvmf_ns_reservation_remove_registrant(ns, reg); 2543 count++; 2544 } 2545 } 2546 return count; 2547 } 2548 2549 static uint32_t 2550 nvmf_ns_reservation_remove_all_other_registrants(struct spdk_nvmf_ns *ns, 2551 struct spdk_nvmf_registrant *reg) 2552 { 2553 struct spdk_nvmf_registrant *reg_tmp, *reg_tmp2; 2554 uint32_t count = 0; 2555 2556 TAILQ_FOREACH_SAFE(reg_tmp, &ns->registrants, link, reg_tmp2) { 2557 if (reg_tmp != reg) { 2558 nvmf_ns_reservation_remove_registrant(ns, reg_tmp); 2559 count++; 2560 } 2561 } 2562 return count; 2563 } 2564 2565 static uint32_t 2566 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns) 2567 { 2568 struct spdk_nvmf_registrant *reg, *reg_tmp; 2569 uint32_t count = 0; 2570 2571 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, reg_tmp) { 2572 nvmf_ns_reservation_remove_registrant(ns, reg); 2573 count++; 2574 } 2575 return count; 2576 } 2577 2578 static void 2579 nvmf_ns_reservation_acquire_reservation(struct spdk_nvmf_ns *ns, uint64_t rkey, 2580 enum spdk_nvme_reservation_type rtype, 2581 struct spdk_nvmf_registrant *holder) 2582 { 2583 ns->rtype = rtype; 2584 ns->crkey = rkey; 2585 assert(ns->holder == NULL); 2586 ns->holder = holder; 2587 } 2588 2589 static bool 2590 nvmf_ns_reservation_register(struct spdk_nvmf_ns *ns, 2591 struct spdk_nvmf_ctrlr *ctrlr, 2592 struct spdk_nvmf_request *req) 2593 { 2594 struct spdk_nvme_reservation_register_data key = { 0 }; 2595 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2596 uint8_t rrega, iekey, cptpl, rtype; 2597 struct spdk_nvmf_registrant *reg; 2598 uint8_t status = SPDK_NVME_SC_SUCCESS; 2599 bool update_sgroup = false; 2600 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2601 uint32_t num_hostid = 0; 2602 int rc; 2603 2604 rrega = cmd->cdw10_bits.resv_register.rrega; 2605 iekey = cmd->cdw10_bits.resv_register.iekey; 2606 cptpl = cmd->cdw10_bits.resv_register.cptpl; 2607 2608 if (req->iovcnt > 0 && req->length >= sizeof(key)) { 2609 struct spdk_iov_xfer ix; 2610 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 2611 spdk_iov_xfer_to_buf(&ix, &key, sizeof(key)); 2612 } else { 2613 SPDK_ERRLOG("No key provided. Failing request.\n"); 2614 status = SPDK_NVME_SC_INVALID_FIELD; 2615 goto exit; 2616 } 2617 2618 SPDK_DEBUGLOG(nvmf, "REGISTER: RREGA %u, IEKEY %u, CPTPL %u, " 2619 "NRKEY 0x%"PRIx64", NRKEY 0x%"PRIx64"\n", 2620 rrega, iekey, cptpl, key.crkey, key.nrkey); 2621 2622 if (cptpl == SPDK_NVME_RESERVE_PTPL_CLEAR_POWER_ON) { 2623 /* Ture to OFF state, and need to be updated in the configuration file */ 2624 if (ns->ptpl_activated) { 2625 ns->ptpl_activated = 0; 2626 update_sgroup = true; 2627 } 2628 } else if (cptpl == SPDK_NVME_RESERVE_PTPL_PERSIST_POWER_LOSS) { 2629 if (ns->ptpl_file == NULL) { 2630 status = SPDK_NVME_SC_INVALID_FIELD; 2631 goto exit; 2632 } else if (ns->ptpl_activated == 0) { 2633 ns->ptpl_activated = 1; 2634 update_sgroup = true; 2635 } 2636 } 2637 2638 /* current Host Identifier has registrant or not */ 2639 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2640 2641 switch (rrega) { 2642 case SPDK_NVME_RESERVE_REGISTER_KEY: 2643 if (!reg) { 2644 /* register new controller */ 2645 if (key.nrkey == 0) { 2646 SPDK_ERRLOG("Can't register zeroed new key\n"); 2647 status = SPDK_NVME_SC_INVALID_FIELD; 2648 goto exit; 2649 } 2650 rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey); 2651 if (rc < 0) { 2652 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2653 goto exit; 2654 } 2655 update_sgroup = true; 2656 } else { 2657 /* register with same key is not an error */ 2658 if (reg->rkey != key.nrkey) { 2659 SPDK_ERRLOG("The same host already register a " 2660 "key with 0x%"PRIx64"\n", 2661 reg->rkey); 2662 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2663 goto exit; 2664 } 2665 } 2666 break; 2667 case SPDK_NVME_RESERVE_UNREGISTER_KEY: 2668 if (!reg || (!iekey && reg->rkey != key.crkey)) { 2669 SPDK_ERRLOG("No registrant or current key doesn't match " 2670 "with existing registrant key\n"); 2671 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2672 goto exit; 2673 } 2674 2675 rtype = ns->rtype; 2676 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2677 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2678 &ctrlr->hostid); 2679 2680 nvmf_ns_reservation_remove_registrant(ns, reg); 2681 2682 if (!ns->rtype && num_hostid && (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_REG_ONLY || 2683 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY)) { 2684 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2685 hostid_list, 2686 num_hostid, 2687 SPDK_NVME_RESERVATION_RELEASED); 2688 } 2689 update_sgroup = true; 2690 break; 2691 case SPDK_NVME_RESERVE_REPLACE_KEY: 2692 if (key.nrkey == 0) { 2693 SPDK_ERRLOG("Can't register zeroed new key\n"); 2694 status = SPDK_NVME_SC_INVALID_FIELD; 2695 goto exit; 2696 } 2697 /* Registrant exists */ 2698 if (reg) { 2699 if (!iekey && reg->rkey != key.crkey) { 2700 SPDK_ERRLOG("Current key doesn't match " 2701 "existing registrant key\n"); 2702 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2703 goto exit; 2704 } 2705 if (reg->rkey == key.nrkey) { 2706 goto exit; 2707 } 2708 reg->rkey = key.nrkey; 2709 } else if (iekey) { /* No registrant but IEKEY is set */ 2710 /* new registrant */ 2711 rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey); 2712 if (rc < 0) { 2713 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2714 goto exit; 2715 } 2716 } else { /* No registrant */ 2717 SPDK_ERRLOG("No registrant\n"); 2718 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2719 goto exit; 2720 2721 } 2722 update_sgroup = true; 2723 break; 2724 default: 2725 status = SPDK_NVME_SC_INVALID_FIELD; 2726 goto exit; 2727 } 2728 2729 exit: 2730 if (update_sgroup) { 2731 rc = nvmf_ns_update_reservation_info(ns); 2732 if (rc != 0) { 2733 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2734 } 2735 } 2736 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2737 req->rsp->nvme_cpl.status.sc = status; 2738 return update_sgroup; 2739 } 2740 2741 static bool 2742 nvmf_ns_reservation_acquire(struct spdk_nvmf_ns *ns, 2743 struct spdk_nvmf_ctrlr *ctrlr, 2744 struct spdk_nvmf_request *req) 2745 { 2746 struct spdk_nvme_reservation_acquire_data key = { 0 }; 2747 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2748 uint8_t racqa, iekey, rtype; 2749 struct spdk_nvmf_registrant *reg; 2750 bool all_regs = false; 2751 uint32_t count = 0; 2752 bool update_sgroup = true; 2753 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2754 uint32_t num_hostid = 0; 2755 struct spdk_uuid new_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2756 uint32_t new_num_hostid = 0; 2757 bool reservation_released = false; 2758 uint8_t status = SPDK_NVME_SC_SUCCESS; 2759 2760 racqa = cmd->cdw10_bits.resv_acquire.racqa; 2761 iekey = cmd->cdw10_bits.resv_acquire.iekey; 2762 rtype = cmd->cdw10_bits.resv_acquire.rtype; 2763 2764 if (req->iovcnt > 0 && req->length >= sizeof(key)) { 2765 struct spdk_iov_xfer ix; 2766 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 2767 spdk_iov_xfer_to_buf(&ix, &key, sizeof(key)); 2768 } else { 2769 SPDK_ERRLOG("No key provided. Failing request.\n"); 2770 status = SPDK_NVME_SC_INVALID_FIELD; 2771 goto exit; 2772 } 2773 2774 SPDK_DEBUGLOG(nvmf, "ACQUIRE: RACQA %u, IEKEY %u, RTYPE %u, " 2775 "NRKEY 0x%"PRIx64", PRKEY 0x%"PRIx64"\n", 2776 racqa, iekey, rtype, key.crkey, key.prkey); 2777 2778 if (iekey || rtype > SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) { 2779 SPDK_ERRLOG("Ignore existing key field set to 1\n"); 2780 status = SPDK_NVME_SC_INVALID_FIELD; 2781 update_sgroup = false; 2782 goto exit; 2783 } 2784 2785 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2786 /* must be registrant and CRKEY must match */ 2787 if (!reg || reg->rkey != key.crkey) { 2788 SPDK_ERRLOG("No registrant or current key doesn't match " 2789 "with existing registrant key\n"); 2790 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2791 update_sgroup = false; 2792 goto exit; 2793 } 2794 2795 all_regs = nvmf_ns_reservation_all_registrants_type(ns); 2796 2797 switch (racqa) { 2798 case SPDK_NVME_RESERVE_ACQUIRE: 2799 /* it's not an error for the holder to acquire same reservation type again */ 2800 if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && ns->rtype == rtype) { 2801 /* do nothing */ 2802 update_sgroup = false; 2803 } else if (ns->holder == NULL) { 2804 /* first time to acquire the reservation */ 2805 nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg); 2806 } else { 2807 SPDK_ERRLOG("Invalid rtype or current registrant is not holder\n"); 2808 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2809 update_sgroup = false; 2810 goto exit; 2811 } 2812 break; 2813 case SPDK_NVME_RESERVE_PREEMPT: 2814 /* no reservation holder */ 2815 if (!ns->holder) { 2816 /* unregister with PRKEY */ 2817 nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2818 break; 2819 } 2820 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2821 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2822 &ctrlr->hostid); 2823 2824 /* only 1 reservation holder and reservation key is valid */ 2825 if (!all_regs) { 2826 /* preempt itself */ 2827 if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && 2828 ns->crkey == key.prkey) { 2829 ns->rtype = rtype; 2830 reservation_released = true; 2831 break; 2832 } 2833 2834 if (ns->crkey == key.prkey) { 2835 nvmf_ns_reservation_remove_registrant(ns, ns->holder); 2836 nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg); 2837 reservation_released = true; 2838 } else if (key.prkey != 0) { 2839 nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2840 } else { 2841 /* PRKEY is zero */ 2842 SPDK_ERRLOG("Current PRKEY is zero\n"); 2843 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2844 update_sgroup = false; 2845 goto exit; 2846 } 2847 } else { 2848 /* release all other registrants except for the current one */ 2849 if (key.prkey == 0) { 2850 nvmf_ns_reservation_remove_all_other_registrants(ns, reg); 2851 assert(ns->holder == reg); 2852 } else { 2853 count = nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2854 if (count == 0) { 2855 SPDK_ERRLOG("PRKEY doesn't match any registrant\n"); 2856 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2857 update_sgroup = false; 2858 goto exit; 2859 } 2860 } 2861 } 2862 break; 2863 default: 2864 status = SPDK_NVME_SC_INVALID_FIELD; 2865 update_sgroup = false; 2866 break; 2867 } 2868 2869 exit: 2870 if (update_sgroup && racqa == SPDK_NVME_RESERVE_PREEMPT) { 2871 new_num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, new_hostid_list, 2872 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2873 &ctrlr->hostid); 2874 /* Preempt notification occurs on the unregistered controllers 2875 * other than the controller who issued the command. 2876 */ 2877 num_hostid = nvmf_ns_reservation_get_unregistered_hostid(hostid_list, 2878 num_hostid, 2879 new_hostid_list, 2880 new_num_hostid); 2881 if (num_hostid) { 2882 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2883 hostid_list, 2884 num_hostid, 2885 SPDK_NVME_REGISTRATION_PREEMPTED); 2886 2887 } 2888 /* Reservation released notification occurs on the 2889 * controllers which are the remaining registrants other than 2890 * the controller who issued the command. 2891 */ 2892 if (reservation_released && new_num_hostid) { 2893 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2894 new_hostid_list, 2895 new_num_hostid, 2896 SPDK_NVME_RESERVATION_RELEASED); 2897 2898 } 2899 } 2900 if (update_sgroup && ns->ptpl_activated) { 2901 if (nvmf_ns_update_reservation_info(ns)) { 2902 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2903 } 2904 } 2905 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2906 req->rsp->nvme_cpl.status.sc = status; 2907 return update_sgroup; 2908 } 2909 2910 static bool 2911 nvmf_ns_reservation_release(struct spdk_nvmf_ns *ns, 2912 struct spdk_nvmf_ctrlr *ctrlr, 2913 struct spdk_nvmf_request *req) 2914 { 2915 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2916 uint8_t rrela, iekey, rtype; 2917 struct spdk_nvmf_registrant *reg; 2918 uint64_t crkey = 0; 2919 uint8_t status = SPDK_NVME_SC_SUCCESS; 2920 bool update_sgroup = true; 2921 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2922 uint32_t num_hostid = 0; 2923 2924 rrela = cmd->cdw10_bits.resv_release.rrela; 2925 iekey = cmd->cdw10_bits.resv_release.iekey; 2926 rtype = cmd->cdw10_bits.resv_release.rtype; 2927 2928 if (req->iovcnt > 0 && req->length >= sizeof(crkey)) { 2929 struct spdk_iov_xfer ix; 2930 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 2931 spdk_iov_xfer_to_buf(&ix, &crkey, sizeof(crkey)); 2932 } else { 2933 SPDK_ERRLOG("No key provided. Failing request.\n"); 2934 status = SPDK_NVME_SC_INVALID_FIELD; 2935 goto exit; 2936 } 2937 2938 SPDK_DEBUGLOG(nvmf, "RELEASE: RRELA %u, IEKEY %u, RTYPE %u, " 2939 "CRKEY 0x%"PRIx64"\n", rrela, iekey, rtype, crkey); 2940 2941 if (iekey) { 2942 SPDK_ERRLOG("Ignore existing key field set to 1\n"); 2943 status = SPDK_NVME_SC_INVALID_FIELD; 2944 update_sgroup = false; 2945 goto exit; 2946 } 2947 2948 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2949 if (!reg || reg->rkey != crkey) { 2950 SPDK_ERRLOG("No registrant or current key doesn't match " 2951 "with existing registrant key\n"); 2952 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2953 update_sgroup = false; 2954 goto exit; 2955 } 2956 2957 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2958 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2959 &ctrlr->hostid); 2960 2961 switch (rrela) { 2962 case SPDK_NVME_RESERVE_RELEASE: 2963 if (!ns->holder) { 2964 SPDK_DEBUGLOG(nvmf, "RELEASE: no holder\n"); 2965 update_sgroup = false; 2966 goto exit; 2967 } 2968 if (ns->rtype != rtype) { 2969 SPDK_ERRLOG("Type doesn't match\n"); 2970 status = SPDK_NVME_SC_INVALID_FIELD; 2971 update_sgroup = false; 2972 goto exit; 2973 } 2974 if (!nvmf_ns_reservation_registrant_is_holder(ns, reg)) { 2975 /* not the reservation holder, this isn't an error */ 2976 update_sgroup = false; 2977 goto exit; 2978 } 2979 2980 rtype = ns->rtype; 2981 nvmf_ns_reservation_release_reservation(ns); 2982 2983 if (num_hostid && rtype != SPDK_NVME_RESERVE_WRITE_EXCLUSIVE && 2984 rtype != SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 2985 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2986 hostid_list, 2987 num_hostid, 2988 SPDK_NVME_RESERVATION_RELEASED); 2989 } 2990 break; 2991 case SPDK_NVME_RESERVE_CLEAR: 2992 nvmf_ns_reservation_clear_all_registrants(ns); 2993 if (num_hostid) { 2994 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2995 hostid_list, 2996 num_hostid, 2997 SPDK_NVME_RESERVATION_PREEMPTED); 2998 } 2999 break; 3000 default: 3001 status = SPDK_NVME_SC_INVALID_FIELD; 3002 update_sgroup = false; 3003 goto exit; 3004 } 3005 3006 exit: 3007 if (update_sgroup && ns->ptpl_activated) { 3008 if (nvmf_ns_update_reservation_info(ns)) { 3009 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 3010 } 3011 } 3012 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3013 req->rsp->nvme_cpl.status.sc = status; 3014 return update_sgroup; 3015 } 3016 3017 static void 3018 nvmf_ns_reservation_report(struct spdk_nvmf_ns *ns, 3019 struct spdk_nvmf_ctrlr *ctrlr, 3020 struct spdk_nvmf_request *req) 3021 { 3022 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3023 struct spdk_nvmf_registrant *reg, *tmp; 3024 struct spdk_nvme_reservation_status_extended_data status_data = { 0 }; 3025 struct spdk_iov_xfer ix; 3026 uint32_t transfer_len; 3027 uint32_t regctl = 0; 3028 uint8_t status = SPDK_NVME_SC_SUCCESS; 3029 3030 if (req->iovcnt == 0) { 3031 SPDK_ERRLOG("No data transfer specified for request. " 3032 " Unable to transfer back response.\n"); 3033 status = SPDK_NVME_SC_INVALID_FIELD; 3034 goto exit; 3035 } 3036 3037 if (!cmd->cdw11_bits.resv_report.eds) { 3038 SPDK_ERRLOG("NVMeoF uses extended controller data structure, " 3039 "please set EDS bit in cdw11 and try again\n"); 3040 status = SPDK_NVME_SC_HOSTID_INCONSISTENT_FORMAT; 3041 goto exit; 3042 } 3043 3044 /* Number of Dwords of the Reservation Status data structure to transfer */ 3045 transfer_len = (cmd->cdw10 + 1) * sizeof(uint32_t); 3046 3047 if (transfer_len < sizeof(struct spdk_nvme_reservation_status_extended_data)) { 3048 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 3049 goto exit; 3050 } 3051 3052 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 3053 3054 status_data.data.gen = ns->gen; 3055 status_data.data.rtype = ns->rtype; 3056 status_data.data.ptpls = ns->ptpl_activated; 3057 3058 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 3059 regctl++; 3060 } 3061 3062 /* 3063 * We report the number of registrants as per the spec here, even if 3064 * the iov isn't big enough to contain them all. In that case, the 3065 * spdk_iov_xfer_from_buf() won't actually copy any of the remaining 3066 * data; as it keeps track of the iov cursor itself, it's simplest to 3067 * just walk the entire list anyway. 3068 */ 3069 status_data.data.regctl = regctl; 3070 3071 spdk_iov_xfer_from_buf(&ix, &status_data, sizeof(status_data)); 3072 3073 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 3074 struct spdk_nvme_registered_ctrlr_extended_data ctrlr_data = { 0 }; 3075 3076 /* Set to 0xffffh for dynamic controller */ 3077 ctrlr_data.cntlid = 0xffff; 3078 ctrlr_data.rcsts.status = (ns->holder == reg) ? true : false; 3079 ctrlr_data.rkey = reg->rkey; 3080 spdk_uuid_copy((struct spdk_uuid *)ctrlr_data.hostid, ®->hostid); 3081 3082 spdk_iov_xfer_from_buf(&ix, &ctrlr_data, sizeof(ctrlr_data)); 3083 } 3084 3085 exit: 3086 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3087 req->rsp->nvme_cpl.status.sc = status; 3088 return; 3089 } 3090 3091 static void 3092 nvmf_ns_reservation_complete(void *ctx) 3093 { 3094 struct spdk_nvmf_request *req = ctx; 3095 3096 spdk_nvmf_request_complete(req); 3097 } 3098 3099 static void 3100 _nvmf_ns_reservation_update_done(struct spdk_nvmf_subsystem *subsystem, 3101 void *cb_arg, int status) 3102 { 3103 struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)cb_arg; 3104 struct spdk_nvmf_poll_group *group = req->qpair->group; 3105 3106 spdk_thread_send_msg(group->thread, nvmf_ns_reservation_complete, req); 3107 } 3108 3109 void 3110 nvmf_ns_reservation_request(void *ctx) 3111 { 3112 struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)ctx; 3113 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3114 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3115 struct subsystem_update_ns_ctx *update_ctx; 3116 uint32_t nsid; 3117 struct spdk_nvmf_ns *ns; 3118 bool update_sgroup = false; 3119 3120 nsid = cmd->nsid; 3121 ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid); 3122 assert(ns != NULL); 3123 3124 switch (cmd->opc) { 3125 case SPDK_NVME_OPC_RESERVATION_REGISTER: 3126 update_sgroup = nvmf_ns_reservation_register(ns, ctrlr, req); 3127 break; 3128 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 3129 update_sgroup = nvmf_ns_reservation_acquire(ns, ctrlr, req); 3130 break; 3131 case SPDK_NVME_OPC_RESERVATION_RELEASE: 3132 update_sgroup = nvmf_ns_reservation_release(ns, ctrlr, req); 3133 break; 3134 case SPDK_NVME_OPC_RESERVATION_REPORT: 3135 nvmf_ns_reservation_report(ns, ctrlr, req); 3136 break; 3137 default: 3138 break; 3139 } 3140 3141 /* update reservation information to subsystem's poll group */ 3142 if (update_sgroup) { 3143 update_ctx = calloc(1, sizeof(*update_ctx)); 3144 if (update_ctx == NULL) { 3145 SPDK_ERRLOG("Can't alloc subsystem poll group update context\n"); 3146 goto update_done; 3147 } 3148 update_ctx->subsystem = ctrlr->subsys; 3149 update_ctx->cb_fn = _nvmf_ns_reservation_update_done; 3150 update_ctx->cb_arg = req; 3151 3152 nvmf_subsystem_update_ns(ctrlr->subsys, subsystem_update_ns_done, update_ctx); 3153 return; 3154 } 3155 3156 update_done: 3157 _nvmf_ns_reservation_update_done(ctrlr->subsys, (void *)req, 0); 3158 } 3159 3160 int 3161 spdk_nvmf_subsystem_set_ana_reporting(struct spdk_nvmf_subsystem *subsystem, 3162 bool ana_reporting) 3163 { 3164 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 3165 return -EAGAIN; 3166 } 3167 3168 subsystem->flags.ana_reporting = ana_reporting; 3169 3170 return 0; 3171 } 3172 3173 bool 3174 nvmf_subsystem_get_ana_reporting(struct spdk_nvmf_subsystem *subsystem) 3175 { 3176 return subsystem->flags.ana_reporting; 3177 } 3178 3179 struct subsystem_listener_update_ctx { 3180 struct spdk_nvmf_subsystem_listener *listener; 3181 3182 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 3183 void *cb_arg; 3184 }; 3185 3186 static void 3187 subsystem_listener_update_done(struct spdk_io_channel_iter *i, int status) 3188 { 3189 struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 3190 3191 if (ctx->cb_fn) { 3192 ctx->cb_fn(ctx->cb_arg, status); 3193 } 3194 free(ctx); 3195 } 3196 3197 static void 3198 subsystem_listener_update_on_pg(struct spdk_io_channel_iter *i) 3199 { 3200 struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 3201 struct spdk_nvmf_subsystem_listener *listener; 3202 struct spdk_nvmf_poll_group *group; 3203 struct spdk_nvmf_ctrlr *ctrlr; 3204 3205 listener = ctx->listener; 3206 group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i)); 3207 3208 TAILQ_FOREACH(ctrlr, &listener->subsystem->ctrlrs, link) { 3209 if (ctrlr->thread != spdk_get_thread()) { 3210 continue; 3211 } 3212 3213 if (ctrlr->admin_qpair && ctrlr->admin_qpair->group == group && ctrlr->listener == listener) { 3214 nvmf_ctrlr_async_event_ana_change_notice(ctrlr); 3215 } 3216 } 3217 3218 spdk_for_each_channel_continue(i, 0); 3219 } 3220 3221 void 3222 nvmf_subsystem_set_ana_state(struct spdk_nvmf_subsystem *subsystem, 3223 const struct spdk_nvme_transport_id *trid, 3224 enum spdk_nvme_ana_state ana_state, uint32_t anagrpid, 3225 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, void *cb_arg) 3226 { 3227 struct spdk_nvmf_subsystem_listener *listener; 3228 struct subsystem_listener_update_ctx *ctx; 3229 uint32_t i; 3230 3231 assert(cb_fn != NULL); 3232 assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 3233 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED); 3234 3235 if (!subsystem->flags.ana_reporting) { 3236 SPDK_ERRLOG("ANA reporting is disabled\n"); 3237 cb_fn(cb_arg, -EINVAL); 3238 return; 3239 } 3240 3241 /* ANA Change state is not used, ANA Persistent Loss state 3242 * is not supported yet. 3243 */ 3244 if (!(ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE || 3245 ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE || 3246 ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE)) { 3247 SPDK_ERRLOG("ANA state %d is not supported\n", ana_state); 3248 cb_fn(cb_arg, -ENOTSUP); 3249 return; 3250 } 3251 3252 if (anagrpid > subsystem->max_nsid) { 3253 SPDK_ERRLOG("ANA group ID %" PRIu32 " is more than maximum\n", anagrpid); 3254 cb_fn(cb_arg, -EINVAL); 3255 return; 3256 } 3257 3258 listener = nvmf_subsystem_find_listener(subsystem, trid); 3259 if (!listener) { 3260 SPDK_ERRLOG("Unable to find listener.\n"); 3261 cb_fn(cb_arg, -EINVAL); 3262 return; 3263 } 3264 3265 if (anagrpid != 0 && listener->ana_state[anagrpid - 1] == ana_state) { 3266 cb_fn(cb_arg, 0); 3267 return; 3268 } 3269 3270 ctx = calloc(1, sizeof(*ctx)); 3271 if (!ctx) { 3272 SPDK_ERRLOG("Unable to allocate context\n"); 3273 cb_fn(cb_arg, -ENOMEM); 3274 return; 3275 } 3276 3277 for (i = 1; i <= subsystem->max_nsid; i++) { 3278 if (anagrpid == 0 || i == anagrpid) { 3279 listener->ana_state[i - 1] = ana_state; 3280 } 3281 } 3282 listener->ana_state_change_count++; 3283 3284 ctx->listener = listener; 3285 ctx->cb_fn = cb_fn; 3286 ctx->cb_arg = cb_arg; 3287 3288 spdk_for_each_channel(subsystem->tgt, 3289 subsystem_listener_update_on_pg, 3290 ctx, 3291 subsystem_listener_update_done); 3292 } 3293