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_CURRENT || 237 type == SPDK_NVMF_SUBTYPE_DISCOVERY) { 238 if (num_ns != 0) { 239 SPDK_ERRLOG("Discovery subsystem cannot have namespaces.\n"); 240 return NULL; 241 } 242 } else if (num_ns == 0) { 243 num_ns = NVMF_SUBSYSTEM_DEFAULT_NAMESPACES; 244 } 245 246 /* Find a free subsystem id (sid) */ 247 sid = spdk_bit_array_find_first_clear(tgt->subsystem_ids, 0); 248 if (sid == UINT32_MAX) { 249 return NULL; 250 } 251 subsystem = calloc(1, sizeof(struct spdk_nvmf_subsystem)); 252 if (subsystem == NULL) { 253 return NULL; 254 } 255 256 subsystem->thread = spdk_get_thread(); 257 subsystem->state = SPDK_NVMF_SUBSYSTEM_INACTIVE; 258 subsystem->tgt = tgt; 259 subsystem->id = sid; 260 subsystem->subtype = type; 261 subsystem->max_nsid = num_ns; 262 subsystem->next_cntlid = 0; 263 subsystem->min_cntlid = NVMF_MIN_CNTLID; 264 subsystem->max_cntlid = NVMF_MAX_CNTLID; 265 snprintf(subsystem->subnqn, sizeof(subsystem->subnqn), "%s", nqn); 266 pthread_mutex_init(&subsystem->mutex, NULL); 267 TAILQ_INIT(&subsystem->listeners); 268 TAILQ_INIT(&subsystem->hosts); 269 TAILQ_INIT(&subsystem->ctrlrs); 270 subsystem->used_listener_ids = spdk_bit_array_create(NVMF_MAX_LISTENERS_PER_SUBSYSTEM); 271 if (subsystem->used_listener_ids == NULL) { 272 pthread_mutex_destroy(&subsystem->mutex); 273 free(subsystem); 274 return NULL; 275 } 276 277 if (num_ns != 0) { 278 subsystem->ns = calloc(num_ns, sizeof(struct spdk_nvmf_ns *)); 279 if (subsystem->ns == NULL) { 280 SPDK_ERRLOG("Namespace memory allocation failed\n"); 281 pthread_mutex_destroy(&subsystem->mutex); 282 spdk_bit_array_free(&subsystem->used_listener_ids); 283 free(subsystem); 284 return NULL; 285 } 286 subsystem->ana_group = calloc(num_ns, sizeof(uint32_t)); 287 if (subsystem->ana_group == NULL) { 288 SPDK_ERRLOG("ANA group memory allocation failed\n"); 289 pthread_mutex_destroy(&subsystem->mutex); 290 free(subsystem->ns); 291 spdk_bit_array_free(&subsystem->used_listener_ids); 292 free(subsystem); 293 return NULL; 294 } 295 } 296 297 memset(subsystem->sn, '0', sizeof(subsystem->sn) - 1); 298 subsystem->sn[sizeof(subsystem->sn) - 1] = '\0'; 299 300 snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", 301 MODEL_NUMBER_DEFAULT); 302 303 spdk_bit_array_set(tgt->subsystem_ids, sid); 304 RB_INSERT(subsystem_tree, &tgt->subsystems, subsystem); 305 306 SPDK_DTRACE_PROBE1(nvmf_subsystem_create, subsystem->subnqn); 307 308 return subsystem; 309 } 310 311 /* Must hold subsystem->mutex while calling this function */ 312 static void 313 nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_host *host) 314 { 315 TAILQ_REMOVE(&subsystem->hosts, host, link); 316 free(host); 317 } 318 319 static void 320 _nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem, 321 struct spdk_nvmf_subsystem_listener *listener, 322 bool stop) 323 { 324 struct spdk_nvmf_transport *transport; 325 struct spdk_nvmf_ctrlr *ctrlr; 326 327 if (stop) { 328 transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, listener->trid->trstring); 329 if (transport != NULL) { 330 spdk_nvmf_transport_stop_listen(transport, listener->trid); 331 } 332 } 333 334 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 335 if (ctrlr->listener == listener) { 336 ctrlr->listener = NULL; 337 } 338 } 339 340 TAILQ_REMOVE(&subsystem->listeners, listener, link); 341 nvmf_update_discovery_log(listener->subsystem->tgt, NULL); 342 free(listener->ana_state); 343 spdk_bit_array_clear(subsystem->used_listener_ids, listener->id); 344 free(listener); 345 } 346 347 static void 348 _nvmf_subsystem_destroy_msg(void *cb_arg) 349 { 350 struct spdk_nvmf_subsystem *subsystem = cb_arg; 351 352 _nvmf_subsystem_destroy(subsystem); 353 } 354 355 static int 356 _nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem) 357 { 358 struct spdk_nvmf_ns *ns; 359 nvmf_subsystem_destroy_cb async_destroy_cb = NULL; 360 void *async_destroy_cb_arg = NULL; 361 int rc; 362 363 if (!TAILQ_EMPTY(&subsystem->ctrlrs)) { 364 SPDK_DEBUGLOG(nvmf, "subsystem %p %s has active controllers\n", subsystem, subsystem->subnqn); 365 subsystem->async_destroy = true; 366 rc = spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_destroy_msg, subsystem); 367 if (rc) { 368 SPDK_ERRLOG("Failed to send thread msg, rc %d\n", rc); 369 assert(0); 370 return rc; 371 } 372 return -EINPROGRESS; 373 } 374 375 ns = spdk_nvmf_subsystem_get_first_ns(subsystem); 376 while (ns != NULL) { 377 struct spdk_nvmf_ns *next_ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns); 378 379 spdk_nvmf_subsystem_remove_ns(subsystem, ns->opts.nsid); 380 ns = next_ns; 381 } 382 383 free(subsystem->ns); 384 free(subsystem->ana_group); 385 386 RB_REMOVE(subsystem_tree, &subsystem->tgt->subsystems, subsystem); 387 assert(spdk_bit_array_get(subsystem->tgt->subsystem_ids, subsystem->id) == true); 388 spdk_bit_array_clear(subsystem->tgt->subsystem_ids, subsystem->id); 389 390 pthread_mutex_destroy(&subsystem->mutex); 391 392 spdk_bit_array_free(&subsystem->used_listener_ids); 393 394 if (subsystem->async_destroy) { 395 async_destroy_cb = subsystem->async_destroy_cb; 396 async_destroy_cb_arg = subsystem->async_destroy_cb_arg; 397 } 398 399 free(subsystem); 400 401 if (async_destroy_cb) { 402 async_destroy_cb(async_destroy_cb_arg); 403 } 404 405 return 0; 406 } 407 408 static struct spdk_nvmf_ns * 409 _nvmf_subsystem_get_first_zoned_ns(struct spdk_nvmf_subsystem *subsystem) 410 { 411 struct spdk_nvmf_ns *ns = spdk_nvmf_subsystem_get_first_ns(subsystem); 412 while (ns != NULL) { 413 if (ns->csi == SPDK_NVME_CSI_ZNS) { 414 return ns; 415 } 416 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns); 417 } 418 return NULL; 419 } 420 421 int 422 spdk_nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem, nvmf_subsystem_destroy_cb cpl_cb, 423 void *cpl_cb_arg) 424 { 425 struct spdk_nvmf_host *host, *host_tmp; 426 struct spdk_nvmf_transport *transport; 427 428 if (!subsystem) { 429 return -EINVAL; 430 } 431 432 SPDK_DTRACE_PROBE1(nvmf_subsystem_destroy, subsystem->subnqn); 433 434 assert(spdk_get_thread() == subsystem->thread); 435 436 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 437 SPDK_ERRLOG("Subsystem can only be destroyed in inactive state, %s state %d\n", 438 subsystem->subnqn, subsystem->state); 439 return -EAGAIN; 440 } 441 if (subsystem->destroying) { 442 SPDK_ERRLOG("Subsystem destruction is already started\n"); 443 assert(0); 444 return -EALREADY; 445 } 446 447 subsystem->destroying = true; 448 449 SPDK_DEBUGLOG(nvmf, "subsystem is %p %s\n", subsystem, subsystem->subnqn); 450 451 nvmf_subsystem_remove_all_listeners(subsystem, false); 452 453 pthread_mutex_lock(&subsystem->mutex); 454 455 TAILQ_FOREACH_SAFE(host, &subsystem->hosts, link, host_tmp) { 456 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 457 transport = spdk_nvmf_transport_get_next(transport)) { 458 if (transport->ops->subsystem_remove_host) { 459 transport->ops->subsystem_remove_host(transport, subsystem, host->nqn); 460 } 461 } 462 nvmf_subsystem_remove_host(subsystem, host); 463 } 464 465 pthread_mutex_unlock(&subsystem->mutex); 466 467 subsystem->async_destroy_cb = cpl_cb; 468 subsystem->async_destroy_cb_arg = cpl_cb_arg; 469 470 return _nvmf_subsystem_destroy(subsystem); 471 } 472 473 /* we have to use the typedef in the function declaration to appease astyle. */ 474 typedef enum spdk_nvmf_subsystem_state spdk_nvmf_subsystem_state_t; 475 476 static spdk_nvmf_subsystem_state_t 477 nvmf_subsystem_get_intermediate_state(enum spdk_nvmf_subsystem_state current_state, 478 enum spdk_nvmf_subsystem_state requested_state) 479 { 480 switch (requested_state) { 481 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 482 return SPDK_NVMF_SUBSYSTEM_DEACTIVATING; 483 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 484 if (current_state == SPDK_NVMF_SUBSYSTEM_PAUSED) { 485 return SPDK_NVMF_SUBSYSTEM_RESUMING; 486 } else { 487 return SPDK_NVMF_SUBSYSTEM_ACTIVATING; 488 } 489 case SPDK_NVMF_SUBSYSTEM_PAUSED: 490 return SPDK_NVMF_SUBSYSTEM_PAUSING; 491 default: 492 assert(false); 493 return SPDK_NVMF_SUBSYSTEM_NUM_STATES; 494 } 495 } 496 497 static int 498 nvmf_subsystem_set_state(struct spdk_nvmf_subsystem *subsystem, 499 enum spdk_nvmf_subsystem_state state) 500 { 501 enum spdk_nvmf_subsystem_state actual_old_state, expected_old_state; 502 bool exchanged; 503 504 switch (state) { 505 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 506 expected_old_state = SPDK_NVMF_SUBSYSTEM_DEACTIVATING; 507 break; 508 case SPDK_NVMF_SUBSYSTEM_ACTIVATING: 509 expected_old_state = SPDK_NVMF_SUBSYSTEM_INACTIVE; 510 break; 511 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 512 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING; 513 break; 514 case SPDK_NVMF_SUBSYSTEM_PAUSING: 515 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE; 516 break; 517 case SPDK_NVMF_SUBSYSTEM_PAUSED: 518 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSING; 519 break; 520 case SPDK_NVMF_SUBSYSTEM_RESUMING: 521 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED; 522 break; 523 case SPDK_NVMF_SUBSYSTEM_DEACTIVATING: 524 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE; 525 break; 526 default: 527 assert(false); 528 return -1; 529 } 530 531 actual_old_state = expected_old_state; 532 exchanged = __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false, 533 __ATOMIC_RELAXED, __ATOMIC_RELAXED); 534 if (spdk_unlikely(exchanged == false)) { 535 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING && 536 state == SPDK_NVMF_SUBSYSTEM_ACTIVE) { 537 expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING; 538 } 539 /* This is for the case when activating the subsystem fails. */ 540 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_ACTIVATING && 541 state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) { 542 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING; 543 } 544 /* This is for the case when resuming the subsystem fails. */ 545 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING && 546 state == SPDK_NVMF_SUBSYSTEM_PAUSING) { 547 expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING; 548 } 549 /* This is for the case when stopping paused subsystem */ 550 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_PAUSED && 551 state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) { 552 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED; 553 } 554 actual_old_state = expected_old_state; 555 __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false, 556 __ATOMIC_RELAXED, __ATOMIC_RELAXED); 557 } 558 assert(actual_old_state == expected_old_state); 559 return actual_old_state - expected_old_state; 560 } 561 562 struct subsystem_state_change_ctx { 563 struct spdk_nvmf_subsystem *subsystem; 564 uint16_t nsid; 565 566 enum spdk_nvmf_subsystem_state original_state; 567 enum spdk_nvmf_subsystem_state requested_state; 568 569 spdk_nvmf_subsystem_state_change_done cb_fn; 570 void *cb_arg; 571 }; 572 573 static void 574 subsystem_state_change_revert_done(struct spdk_io_channel_iter *i, int status) 575 { 576 struct subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 577 578 /* Nothing to be done here if the state setting fails, we are just screwed. */ 579 if (nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state)) { 580 SPDK_ERRLOG("Unable to revert the subsystem state after operation failure.\n"); 581 } 582 583 ctx->subsystem->changing_state = false; 584 if (ctx->cb_fn) { 585 /* return a failure here. This function only exists in an error path. */ 586 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, -1); 587 } 588 free(ctx); 589 } 590 591 static void 592 subsystem_state_change_done(struct spdk_io_channel_iter *i, int status) 593 { 594 struct subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 595 enum spdk_nvmf_subsystem_state intermediate_state; 596 597 SPDK_DTRACE_PROBE4(nvmf_subsystem_change_state_done, ctx->subsystem->subnqn, 598 ctx->requested_state, ctx->original_state, status); 599 600 if (status == 0) { 601 status = nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state); 602 if (status) { 603 status = -1; 604 } 605 } 606 607 if (status) { 608 intermediate_state = nvmf_subsystem_get_intermediate_state(ctx->requested_state, 609 ctx->original_state); 610 assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES); 611 612 if (nvmf_subsystem_set_state(ctx->subsystem, intermediate_state)) { 613 goto out; 614 } 615 ctx->requested_state = ctx->original_state; 616 spdk_for_each_channel(ctx->subsystem->tgt, 617 subsystem_state_change_on_pg, 618 ctx, 619 subsystem_state_change_revert_done); 620 return; 621 } 622 623 out: 624 ctx->subsystem->changing_state = false; 625 if (ctx->cb_fn) { 626 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status); 627 } 628 free(ctx); 629 } 630 631 static void 632 subsystem_state_change_continue(void *ctx, int status) 633 { 634 struct spdk_io_channel_iter *i = ctx; 635 struct subsystem_state_change_ctx *_ctx __attribute__((unused)); 636 637 _ctx = spdk_io_channel_iter_get_ctx(i); 638 SPDK_DTRACE_PROBE3(nvmf_pg_change_state_done, _ctx->subsystem->subnqn, 639 _ctx->requested_state, spdk_thread_get_id(spdk_get_thread())); 640 641 spdk_for_each_channel_continue(i, status); 642 } 643 644 static void 645 subsystem_state_change_on_pg(struct spdk_io_channel_iter *i) 646 { 647 struct subsystem_state_change_ctx *ctx; 648 struct spdk_io_channel *ch; 649 struct spdk_nvmf_poll_group *group; 650 651 ctx = spdk_io_channel_iter_get_ctx(i); 652 ch = spdk_io_channel_iter_get_channel(i); 653 group = spdk_io_channel_get_ctx(ch); 654 655 SPDK_DTRACE_PROBE3(nvmf_pg_change_state, ctx->subsystem->subnqn, 656 ctx->requested_state, spdk_thread_get_id(spdk_get_thread())); 657 switch (ctx->requested_state) { 658 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 659 nvmf_poll_group_remove_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 660 break; 661 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 662 if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_ACTIVATING) { 663 nvmf_poll_group_add_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 664 } else if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_RESUMING) { 665 nvmf_poll_group_resume_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 666 } 667 break; 668 case SPDK_NVMF_SUBSYSTEM_PAUSED: 669 nvmf_poll_group_pause_subsystem(group, ctx->subsystem, ctx->nsid, subsystem_state_change_continue, 670 i); 671 break; 672 default: 673 assert(false); 674 break; 675 } 676 } 677 678 static int 679 nvmf_subsystem_state_change(struct spdk_nvmf_subsystem *subsystem, 680 uint32_t nsid, 681 enum spdk_nvmf_subsystem_state requested_state, 682 spdk_nvmf_subsystem_state_change_done cb_fn, 683 void *cb_arg) 684 { 685 struct subsystem_state_change_ctx *ctx; 686 enum spdk_nvmf_subsystem_state intermediate_state; 687 int rc; 688 689 if (__sync_val_compare_and_swap(&subsystem->changing_state, false, true)) { 690 return -EBUSY; 691 } 692 693 SPDK_DTRACE_PROBE3(nvmf_subsystem_change_state, subsystem->subnqn, 694 requested_state, subsystem->state); 695 /* If we are already in the requested state, just call the callback immediately. */ 696 if (subsystem->state == requested_state) { 697 subsystem->changing_state = false; 698 if (cb_fn) { 699 cb_fn(subsystem, cb_arg, 0); 700 } 701 return 0; 702 } 703 704 intermediate_state = nvmf_subsystem_get_intermediate_state(subsystem->state, requested_state); 705 assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES); 706 707 ctx = calloc(1, sizeof(*ctx)); 708 if (!ctx) { 709 subsystem->changing_state = false; 710 return -ENOMEM; 711 } 712 713 ctx->original_state = subsystem->state; 714 rc = nvmf_subsystem_set_state(subsystem, intermediate_state); 715 if (rc) { 716 free(ctx); 717 subsystem->changing_state = false; 718 return rc; 719 } 720 721 ctx->subsystem = subsystem; 722 ctx->nsid = nsid; 723 ctx->requested_state = requested_state; 724 ctx->cb_fn = cb_fn; 725 ctx->cb_arg = cb_arg; 726 727 spdk_for_each_channel(subsystem->tgt, 728 subsystem_state_change_on_pg, 729 ctx, 730 subsystem_state_change_done); 731 732 return 0; 733 } 734 735 int 736 spdk_nvmf_subsystem_start(struct spdk_nvmf_subsystem *subsystem, 737 spdk_nvmf_subsystem_state_change_done cb_fn, 738 void *cb_arg) 739 { 740 return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg); 741 } 742 743 int 744 spdk_nvmf_subsystem_stop(struct spdk_nvmf_subsystem *subsystem, 745 spdk_nvmf_subsystem_state_change_done cb_fn, 746 void *cb_arg) 747 { 748 return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_INACTIVE, cb_fn, cb_arg); 749 } 750 751 int 752 spdk_nvmf_subsystem_pause(struct spdk_nvmf_subsystem *subsystem, 753 uint32_t nsid, 754 spdk_nvmf_subsystem_state_change_done cb_fn, 755 void *cb_arg) 756 { 757 return nvmf_subsystem_state_change(subsystem, nsid, SPDK_NVMF_SUBSYSTEM_PAUSED, cb_fn, cb_arg); 758 } 759 760 int 761 spdk_nvmf_subsystem_resume(struct spdk_nvmf_subsystem *subsystem, 762 spdk_nvmf_subsystem_state_change_done cb_fn, 763 void *cb_arg) 764 { 765 return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg); 766 } 767 768 struct spdk_nvmf_subsystem * 769 spdk_nvmf_subsystem_get_first(struct spdk_nvmf_tgt *tgt) 770 { 771 return RB_MIN(subsystem_tree, &tgt->subsystems); 772 } 773 774 struct spdk_nvmf_subsystem * 775 spdk_nvmf_subsystem_get_next(struct spdk_nvmf_subsystem *subsystem) 776 { 777 if (!subsystem) { 778 return NULL; 779 } 780 781 return RB_NEXT(subsystem_tree, &tgt->subsystems, subsystem); 782 } 783 784 /* Must hold subsystem->mutex while calling this function */ 785 static struct spdk_nvmf_host * 786 nvmf_subsystem_find_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 787 { 788 struct spdk_nvmf_host *host = NULL; 789 790 TAILQ_FOREACH(host, &subsystem->hosts, link) { 791 if (strcmp(hostnqn, host->nqn) == 0) { 792 return host; 793 } 794 } 795 796 return NULL; 797 } 798 799 int 800 spdk_nvmf_subsystem_add_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn, 801 const struct spdk_json_val *params) 802 { 803 struct spdk_nvmf_host *host; 804 struct spdk_nvmf_transport *transport; 805 int rc; 806 807 if (!nvmf_valid_nqn(hostnqn)) { 808 return -EINVAL; 809 } 810 811 pthread_mutex_lock(&subsystem->mutex); 812 813 if (nvmf_subsystem_find_host(subsystem, hostnqn)) { 814 /* This subsystem already allows the specified host. */ 815 pthread_mutex_unlock(&subsystem->mutex); 816 return 0; 817 } 818 819 host = calloc(1, sizeof(*host)); 820 if (!host) { 821 pthread_mutex_unlock(&subsystem->mutex); 822 return -ENOMEM; 823 } 824 825 snprintf(host->nqn, sizeof(host->nqn), "%s", hostnqn); 826 827 SPDK_DTRACE_PROBE2(nvmf_subsystem_add_host, subsystem->subnqn, host->nqn); 828 829 TAILQ_INSERT_HEAD(&subsystem->hosts, host, link); 830 831 if (!TAILQ_EMPTY(&subsystem->listeners)) { 832 nvmf_update_discovery_log(subsystem->tgt, hostnqn); 833 } 834 835 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 836 transport = spdk_nvmf_transport_get_next(transport)) { 837 if (transport->ops->subsystem_add_host) { 838 rc = transport->ops->subsystem_add_host(transport, subsystem, hostnqn, params); 839 if (rc) { 840 SPDK_ERRLOG("Unable to add host to %s transport\n", transport->ops->name); 841 /* Remove this host from all transports we've managed to add it to. */ 842 pthread_mutex_unlock(&subsystem->mutex); 843 spdk_nvmf_subsystem_remove_host(subsystem, hostnqn); 844 return rc; 845 } 846 } 847 } 848 849 pthread_mutex_unlock(&subsystem->mutex); 850 851 return 0; 852 } 853 854 int 855 spdk_nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 856 { 857 struct spdk_nvmf_host *host; 858 struct spdk_nvmf_transport *transport; 859 860 pthread_mutex_lock(&subsystem->mutex); 861 862 host = nvmf_subsystem_find_host(subsystem, hostnqn); 863 if (host == NULL) { 864 pthread_mutex_unlock(&subsystem->mutex); 865 return -ENOENT; 866 } 867 868 SPDK_DTRACE_PROBE2(nvmf_subsystem_remove_host, subsystem->subnqn, host->nqn); 869 870 nvmf_subsystem_remove_host(subsystem, host); 871 872 if (!TAILQ_EMPTY(&subsystem->listeners)) { 873 nvmf_update_discovery_log(subsystem->tgt, hostnqn); 874 } 875 876 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 877 transport = spdk_nvmf_transport_get_next(transport)) { 878 if (transport->ops->subsystem_remove_host) { 879 transport->ops->subsystem_remove_host(transport, subsystem, hostnqn); 880 } 881 } 882 883 pthread_mutex_unlock(&subsystem->mutex); 884 885 return 0; 886 } 887 888 struct nvmf_subsystem_disconnect_host_ctx { 889 struct spdk_nvmf_subsystem *subsystem; 890 char *hostnqn; 891 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 892 void *cb_arg; 893 }; 894 895 static void 896 nvmf_subsystem_disconnect_host_fini(struct spdk_io_channel_iter *i, int status) 897 { 898 struct nvmf_subsystem_disconnect_host_ctx *ctx; 899 900 ctx = spdk_io_channel_iter_get_ctx(i); 901 902 if (ctx->cb_fn) { 903 ctx->cb_fn(ctx->cb_arg, status); 904 } 905 free(ctx->hostnqn); 906 free(ctx); 907 } 908 909 static void 910 nvmf_subsystem_disconnect_qpairs_by_host(struct spdk_io_channel_iter *i) 911 { 912 struct nvmf_subsystem_disconnect_host_ctx *ctx; 913 struct spdk_nvmf_poll_group *group; 914 struct spdk_io_channel *ch; 915 struct spdk_nvmf_qpair *qpair, *tmp_qpair; 916 struct spdk_nvmf_ctrlr *ctrlr; 917 918 ctx = spdk_io_channel_iter_get_ctx(i); 919 ch = spdk_io_channel_iter_get_channel(i); 920 group = spdk_io_channel_get_ctx(ch); 921 922 TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, tmp_qpair) { 923 ctrlr = qpair->ctrlr; 924 925 if (ctrlr == NULL || ctrlr->subsys != ctx->subsystem) { 926 continue; 927 } 928 929 if (strncmp(ctrlr->hostnqn, ctx->hostnqn, sizeof(ctrlr->hostnqn)) == 0) { 930 /* Right now this does not wait for the queue pairs to actually disconnect. */ 931 spdk_nvmf_qpair_disconnect(qpair, NULL, NULL); 932 } 933 } 934 spdk_for_each_channel_continue(i, 0); 935 } 936 937 int 938 spdk_nvmf_subsystem_disconnect_host(struct spdk_nvmf_subsystem *subsystem, 939 const char *hostnqn, 940 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 941 void *cb_arg) 942 { 943 struct nvmf_subsystem_disconnect_host_ctx *ctx; 944 945 ctx = calloc(1, sizeof(struct nvmf_subsystem_disconnect_host_ctx)); 946 if (ctx == NULL) { 947 return -ENOMEM; 948 } 949 950 ctx->hostnqn = strdup(hostnqn); 951 if (ctx->hostnqn == NULL) { 952 free(ctx); 953 return -ENOMEM; 954 } 955 956 ctx->subsystem = subsystem; 957 ctx->cb_fn = cb_fn; 958 ctx->cb_arg = cb_arg; 959 960 spdk_for_each_channel(subsystem->tgt, nvmf_subsystem_disconnect_qpairs_by_host, ctx, 961 nvmf_subsystem_disconnect_host_fini); 962 963 return 0; 964 } 965 966 int 967 spdk_nvmf_subsystem_set_allow_any_host(struct spdk_nvmf_subsystem *subsystem, bool allow_any_host) 968 { 969 pthread_mutex_lock(&subsystem->mutex); 970 subsystem->flags.allow_any_host = allow_any_host; 971 if (!TAILQ_EMPTY(&subsystem->listeners)) { 972 nvmf_update_discovery_log(subsystem->tgt, NULL); 973 } 974 pthread_mutex_unlock(&subsystem->mutex); 975 976 return 0; 977 } 978 979 bool 980 spdk_nvmf_subsystem_get_allow_any_host(const struct spdk_nvmf_subsystem *subsystem) 981 { 982 bool allow_any_host; 983 struct spdk_nvmf_subsystem *sub; 984 985 /* Technically, taking the mutex modifies data in the subsystem. But the const 986 * is still important to convey that this doesn't mutate any other data. Cast 987 * it away to work around this. */ 988 sub = (struct spdk_nvmf_subsystem *)subsystem; 989 990 pthread_mutex_lock(&sub->mutex); 991 allow_any_host = sub->flags.allow_any_host; 992 pthread_mutex_unlock(&sub->mutex); 993 994 return allow_any_host; 995 } 996 997 bool 998 spdk_nvmf_subsystem_host_allowed(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 999 { 1000 bool allowed; 1001 1002 if (!hostnqn) { 1003 return false; 1004 } 1005 1006 pthread_mutex_lock(&subsystem->mutex); 1007 1008 if (subsystem->flags.allow_any_host) { 1009 pthread_mutex_unlock(&subsystem->mutex); 1010 return true; 1011 } 1012 1013 allowed = nvmf_subsystem_find_host(subsystem, hostnqn) != NULL; 1014 pthread_mutex_unlock(&subsystem->mutex); 1015 1016 return allowed; 1017 } 1018 1019 struct spdk_nvmf_host * 1020 spdk_nvmf_subsystem_get_first_host(struct spdk_nvmf_subsystem *subsystem) 1021 { 1022 return TAILQ_FIRST(&subsystem->hosts); 1023 } 1024 1025 1026 struct spdk_nvmf_host * 1027 spdk_nvmf_subsystem_get_next_host(struct spdk_nvmf_subsystem *subsystem, 1028 struct spdk_nvmf_host *prev_host) 1029 { 1030 return TAILQ_NEXT(prev_host, link); 1031 } 1032 1033 const char * 1034 spdk_nvmf_host_get_nqn(const struct spdk_nvmf_host *host) 1035 { 1036 return host->nqn; 1037 } 1038 1039 struct spdk_nvmf_subsystem_listener * 1040 nvmf_subsystem_find_listener(struct spdk_nvmf_subsystem *subsystem, 1041 const struct spdk_nvme_transport_id *trid) 1042 { 1043 struct spdk_nvmf_subsystem_listener *listener; 1044 1045 TAILQ_FOREACH(listener, &subsystem->listeners, link) { 1046 if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) { 1047 return listener; 1048 } 1049 } 1050 1051 return NULL; 1052 } 1053 1054 /** 1055 * Function to be called once the target is listening. 1056 * 1057 * \param ctx Context argument passed to this function. 1058 * \param status 0 if it completed successfully, or negative errno if it failed. 1059 */ 1060 static void 1061 _nvmf_subsystem_add_listener_done(void *ctx, int status) 1062 { 1063 struct spdk_nvmf_subsystem_listener *listener = ctx; 1064 1065 if (status) { 1066 listener->cb_fn(listener->cb_arg, status); 1067 free(listener); 1068 return; 1069 } 1070 1071 TAILQ_INSERT_HEAD(&listener->subsystem->listeners, listener, link); 1072 nvmf_update_discovery_log(listener->subsystem->tgt, NULL); 1073 listener->cb_fn(listener->cb_arg, status); 1074 } 1075 1076 void 1077 spdk_nvmf_subsystem_listener_opts_init(struct spdk_nvmf_listener_opts *opts, size_t size) 1078 { 1079 if (opts == NULL) { 1080 SPDK_ERRLOG("opts should not be NULL\n"); 1081 assert(false); 1082 return; 1083 } 1084 if (size == 0) { 1085 SPDK_ERRLOG("size should not be zero\n"); 1086 assert(false); 1087 return; 1088 } 1089 1090 memset(opts, 0, size); 1091 opts->opts_size = size; 1092 1093 #define FIELD_OK(field) \ 1094 offsetof(struct spdk_nvmf_listener_opts, field) + sizeof(opts->field) <= size 1095 1096 #define SET_FIELD(field, value) \ 1097 if (FIELD_OK(field)) { \ 1098 opts->field = value; \ 1099 } \ 1100 1101 SET_FIELD(secure_channel, false); 1102 1103 #undef FIELD_OK 1104 #undef SET_FIELD 1105 } 1106 1107 static int 1108 listener_opts_copy(struct spdk_nvmf_listener_opts *src, struct spdk_nvmf_listener_opts *dst) 1109 { 1110 if (src->opts_size == 0) { 1111 SPDK_ERRLOG("source structure size should not be zero\n"); 1112 assert(false); 1113 return -EINVAL; 1114 } 1115 1116 memset(dst, 0, sizeof(*dst)); 1117 dst->opts_size = src->opts_size; 1118 1119 #define FIELD_OK(field) \ 1120 offsetof(struct spdk_nvmf_listener_opts, field) + sizeof(src->field) <= src->opts_size 1121 1122 #define SET_FIELD(field) \ 1123 if (FIELD_OK(field)) { \ 1124 dst->field = src->field; \ 1125 } \ 1126 1127 SET_FIELD(secure_channel); 1128 1129 /* We should not remove this statement, but need to update the assert statement 1130 * if we add a new field, and also add a corresponding SET_FIELD statement. */ 1131 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listener_opts) == 9, "Incorrect size"); 1132 1133 #undef SET_FIELD 1134 #undef FIELD_OK 1135 1136 return 0; 1137 } 1138 1139 static void 1140 _nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem, 1141 struct spdk_nvme_transport_id *trid, 1142 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 1143 void *cb_arg, struct spdk_nvmf_listener_opts *opts) 1144 { 1145 struct spdk_nvmf_transport *transport; 1146 struct spdk_nvmf_subsystem_listener *listener; 1147 struct spdk_nvmf_listener *tr_listener; 1148 uint32_t i; 1149 uint32_t id; 1150 int rc = 0; 1151 1152 assert(cb_fn != NULL); 1153 1154 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1155 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1156 cb_fn(cb_arg, -EAGAIN); 1157 return; 1158 } 1159 1160 if (nvmf_subsystem_find_listener(subsystem, trid)) { 1161 /* Listener already exists in this subsystem */ 1162 cb_fn(cb_arg, 0); 1163 return; 1164 } 1165 1166 transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, trid->trstring); 1167 if (!transport) { 1168 SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n", 1169 trid->trstring); 1170 cb_fn(cb_arg, -EINVAL); 1171 return; 1172 } 1173 1174 tr_listener = nvmf_transport_find_listener(transport, trid); 1175 if (!tr_listener) { 1176 SPDK_ERRLOG("Cannot find transport listener for %s\n", trid->traddr); 1177 cb_fn(cb_arg, -EINVAL); 1178 return; 1179 } 1180 1181 listener = calloc(1, sizeof(*listener)); 1182 if (!listener) { 1183 cb_fn(cb_arg, -ENOMEM); 1184 return; 1185 } 1186 1187 listener->trid = &tr_listener->trid; 1188 listener->transport = transport; 1189 listener->cb_fn = cb_fn; 1190 listener->cb_arg = cb_arg; 1191 listener->subsystem = subsystem; 1192 listener->ana_state = calloc(subsystem->max_nsid, sizeof(enum spdk_nvme_ana_state)); 1193 if (!listener->ana_state) { 1194 free(listener); 1195 cb_fn(cb_arg, -ENOMEM); 1196 return; 1197 } 1198 1199 spdk_nvmf_subsystem_listener_opts_init(&listener->opts, sizeof(listener->opts)); 1200 if (opts != NULL) { 1201 rc = listener_opts_copy(opts, &listener->opts); 1202 if (rc) { 1203 SPDK_ERRLOG("Unable to copy listener options\n"); 1204 free(listener->ana_state); 1205 free(listener); 1206 cb_fn(cb_arg, -EINVAL); 1207 return; 1208 } 1209 } 1210 1211 id = spdk_bit_array_find_first_clear(subsystem->used_listener_ids, 0); 1212 if (id == UINT32_MAX) { 1213 SPDK_ERRLOG("Cannot add any more listeners\n"); 1214 free(listener->ana_state); 1215 free(listener); 1216 cb_fn(cb_arg, -EINVAL); 1217 return; 1218 } 1219 1220 spdk_bit_array_set(subsystem->used_listener_ids, id); 1221 listener->id = id; 1222 1223 for (i = 0; i < subsystem->max_nsid; i++) { 1224 listener->ana_state[i] = SPDK_NVME_ANA_OPTIMIZED_STATE; 1225 } 1226 1227 if (transport->ops->listen_associate != NULL) { 1228 rc = transport->ops->listen_associate(transport, subsystem, trid); 1229 } 1230 1231 SPDK_DTRACE_PROBE4(nvmf_subsystem_add_listener, subsystem->subnqn, listener->trid->trtype, 1232 listener->trid->traddr, listener->trid->trsvcid); 1233 1234 _nvmf_subsystem_add_listener_done(listener, rc); 1235 } 1236 1237 void 1238 spdk_nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem, 1239 struct spdk_nvme_transport_id *trid, 1240 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 1241 void *cb_arg) 1242 { 1243 _nvmf_subsystem_add_listener(subsystem, trid, cb_fn, cb_arg, NULL); 1244 } 1245 1246 void 1247 spdk_nvmf_subsystem_add_listener_ext(struct spdk_nvmf_subsystem *subsystem, 1248 struct spdk_nvme_transport_id *trid, 1249 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 1250 void *cb_arg, struct spdk_nvmf_listener_opts *opts) 1251 { 1252 _nvmf_subsystem_add_listener(subsystem, trid, cb_fn, cb_arg, opts); 1253 } 1254 1255 int 1256 spdk_nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem, 1257 const struct spdk_nvme_transport_id *trid) 1258 { 1259 struct spdk_nvmf_subsystem_listener *listener; 1260 1261 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1262 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1263 return -EAGAIN; 1264 } 1265 1266 listener = nvmf_subsystem_find_listener(subsystem, trid); 1267 if (listener == NULL) { 1268 return -ENOENT; 1269 } 1270 1271 SPDK_DTRACE_PROBE4(nvmf_subsystem_remove_listener, subsystem->subnqn, listener->trid->trtype, 1272 listener->trid->traddr, listener->trid->trsvcid); 1273 1274 _nvmf_subsystem_remove_listener(subsystem, listener, false); 1275 1276 return 0; 1277 } 1278 1279 void 1280 nvmf_subsystem_remove_all_listeners(struct spdk_nvmf_subsystem *subsystem, 1281 bool stop) 1282 { 1283 struct spdk_nvmf_subsystem_listener *listener, *listener_tmp; 1284 1285 TAILQ_FOREACH_SAFE(listener, &subsystem->listeners, link, listener_tmp) { 1286 _nvmf_subsystem_remove_listener(subsystem, listener, stop); 1287 } 1288 } 1289 1290 bool 1291 spdk_nvmf_subsystem_listener_allowed(struct spdk_nvmf_subsystem *subsystem, 1292 const struct spdk_nvme_transport_id *trid) 1293 { 1294 struct spdk_nvmf_subsystem_listener *listener; 1295 1296 TAILQ_FOREACH(listener, &subsystem->listeners, link) { 1297 if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) { 1298 return true; 1299 } 1300 } 1301 1302 if (!strcmp(subsystem->subnqn, SPDK_NVMF_DISCOVERY_NQN)) { 1303 SPDK_WARNLOG("Allowing connection to discovery subsystem on %s/%s/%s, " 1304 "even though this listener was not added to the discovery " 1305 "subsystem. This behavior is deprecated and will be removed " 1306 "in a future release.\n", 1307 spdk_nvme_transport_id_trtype_str(trid->trtype), trid->traddr, trid->trsvcid); 1308 return true; 1309 } 1310 1311 return false; 1312 } 1313 1314 struct spdk_nvmf_subsystem_listener * 1315 spdk_nvmf_subsystem_get_first_listener(struct spdk_nvmf_subsystem *subsystem) 1316 { 1317 return TAILQ_FIRST(&subsystem->listeners); 1318 } 1319 1320 struct spdk_nvmf_subsystem_listener * 1321 spdk_nvmf_subsystem_get_next_listener(struct spdk_nvmf_subsystem *subsystem, 1322 struct spdk_nvmf_subsystem_listener *prev_listener) 1323 { 1324 return TAILQ_NEXT(prev_listener, link); 1325 } 1326 1327 const struct spdk_nvme_transport_id * 1328 spdk_nvmf_subsystem_listener_get_trid(struct spdk_nvmf_subsystem_listener *listener) 1329 { 1330 return listener->trid; 1331 } 1332 1333 void 1334 spdk_nvmf_subsystem_allow_any_listener(struct spdk_nvmf_subsystem *subsystem, 1335 bool allow_any_listener) 1336 { 1337 subsystem->flags.allow_any_listener = allow_any_listener; 1338 } 1339 1340 SPDK_LOG_DEPRECATION_REGISTER(spdk_nvmf_subsytem_any_listener_allowed, 1341 "spdk_nvmf_subsytem_any_listener_allowed is deprecated", "v24.01", 0); 1342 1343 bool 1344 spdk_nvmf_subsytem_any_listener_allowed(struct spdk_nvmf_subsystem *subsystem) 1345 { 1346 SPDK_LOG_DEPRECATED(spdk_nvmf_subsytem_any_listener_allowed); 1347 return subsystem->flags.allow_any_listener; 1348 } 1349 1350 bool 1351 spdk_nvmf_subsystem_any_listener_allowed(struct spdk_nvmf_subsystem *subsystem) 1352 { 1353 return subsystem->flags.allow_any_listener; 1354 } 1355 1356 struct subsystem_update_ns_ctx { 1357 struct spdk_nvmf_subsystem *subsystem; 1358 1359 spdk_nvmf_subsystem_state_change_done cb_fn; 1360 void *cb_arg; 1361 }; 1362 1363 static void 1364 subsystem_update_ns_done(struct spdk_io_channel_iter *i, int status) 1365 { 1366 struct subsystem_update_ns_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 1367 1368 if (ctx->cb_fn) { 1369 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status); 1370 } 1371 free(ctx); 1372 } 1373 1374 static void 1375 subsystem_update_ns_on_pg(struct spdk_io_channel_iter *i) 1376 { 1377 int rc; 1378 struct subsystem_update_ns_ctx *ctx; 1379 struct spdk_nvmf_poll_group *group; 1380 struct spdk_nvmf_subsystem *subsystem; 1381 1382 ctx = spdk_io_channel_iter_get_ctx(i); 1383 group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i)); 1384 subsystem = ctx->subsystem; 1385 1386 rc = nvmf_poll_group_update_subsystem(group, subsystem); 1387 spdk_for_each_channel_continue(i, rc); 1388 } 1389 1390 static int 1391 nvmf_subsystem_update_ns(struct spdk_nvmf_subsystem *subsystem, spdk_channel_for_each_cpl cpl, 1392 void *ctx) 1393 { 1394 spdk_for_each_channel(subsystem->tgt, 1395 subsystem_update_ns_on_pg, 1396 ctx, 1397 cpl); 1398 1399 return 0; 1400 } 1401 1402 static void 1403 nvmf_subsystem_ns_changed(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1404 { 1405 struct spdk_nvmf_ctrlr *ctrlr; 1406 1407 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 1408 nvmf_ctrlr_ns_changed(ctrlr, nsid); 1409 } 1410 } 1411 1412 static uint32_t nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns); 1413 1414 int 1415 spdk_nvmf_subsystem_remove_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1416 { 1417 struct spdk_nvmf_transport *transport; 1418 struct spdk_nvmf_ns *ns; 1419 1420 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1421 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1422 assert(false); 1423 return -1; 1424 } 1425 1426 if (nsid == 0 || nsid > subsystem->max_nsid) { 1427 return -1; 1428 } 1429 1430 ns = subsystem->ns[nsid - 1]; 1431 if (!ns) { 1432 return -1; 1433 } 1434 1435 subsystem->ns[nsid - 1] = NULL; 1436 1437 assert(ns->anagrpid - 1 < subsystem->max_nsid); 1438 assert(subsystem->ana_group[ns->anagrpid - 1] > 0); 1439 1440 subsystem->ana_group[ns->anagrpid - 1]--; 1441 1442 free(ns->ptpl_file); 1443 nvmf_ns_reservation_clear_all_registrants(ns); 1444 spdk_bdev_module_release_bdev(ns->bdev); 1445 spdk_bdev_close(ns->desc); 1446 free(ns); 1447 1448 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 1449 transport = spdk_nvmf_transport_get_next(transport)) { 1450 if (transport->ops->subsystem_remove_ns) { 1451 transport->ops->subsystem_remove_ns(transport, subsystem, nsid); 1452 } 1453 } 1454 1455 nvmf_subsystem_ns_changed(subsystem, nsid); 1456 1457 return 0; 1458 } 1459 1460 struct subsystem_ns_change_ctx { 1461 struct spdk_nvmf_subsystem *subsystem; 1462 spdk_nvmf_subsystem_state_change_done cb_fn; 1463 uint32_t nsid; 1464 }; 1465 1466 static void 1467 _nvmf_ns_hot_remove(struct spdk_nvmf_subsystem *subsystem, 1468 void *cb_arg, int status) 1469 { 1470 struct subsystem_ns_change_ctx *ctx = cb_arg; 1471 int rc; 1472 1473 rc = spdk_nvmf_subsystem_remove_ns(subsystem, ctx->nsid); 1474 if (rc != 0) { 1475 SPDK_ERRLOG("Failed to make changes to NVME-oF subsystem with id: %u\n", subsystem->id); 1476 } 1477 1478 rc = spdk_nvmf_subsystem_resume(subsystem, NULL, NULL); 1479 if (rc != 0) { 1480 SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id); 1481 } 1482 1483 free(ctx); 1484 } 1485 1486 static void 1487 nvmf_ns_change_msg(void *ns_ctx) 1488 { 1489 struct subsystem_ns_change_ctx *ctx = ns_ctx; 1490 int rc; 1491 1492 SPDK_DTRACE_PROBE2(nvmf_ns_change, ctx->nsid, ctx->subsystem->subnqn); 1493 1494 rc = spdk_nvmf_subsystem_pause(ctx->subsystem, ctx->nsid, ctx->cb_fn, ctx); 1495 if (rc) { 1496 if (rc == -EBUSY) { 1497 /* Try again, this is not a permanent situation. */ 1498 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ctx); 1499 } else { 1500 free(ctx); 1501 SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n"); 1502 } 1503 } 1504 } 1505 1506 static void 1507 nvmf_ns_hot_remove(void *remove_ctx) 1508 { 1509 struct spdk_nvmf_ns *ns = remove_ctx; 1510 struct subsystem_ns_change_ctx *ns_ctx; 1511 int rc; 1512 1513 /* We have to allocate a new context because this op 1514 * is asynchronous and we could lose the ns in the middle. 1515 */ 1516 ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx)); 1517 if (!ns_ctx) { 1518 SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n"); 1519 return; 1520 } 1521 1522 ns_ctx->subsystem = ns->subsystem; 1523 ns_ctx->nsid = ns->opts.nsid; 1524 ns_ctx->cb_fn = _nvmf_ns_hot_remove; 1525 1526 rc = spdk_nvmf_subsystem_pause(ns->subsystem, ns_ctx->nsid, _nvmf_ns_hot_remove, ns_ctx); 1527 if (rc) { 1528 if (rc == -EBUSY) { 1529 /* Try again, this is not a permanent situation. */ 1530 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx); 1531 } else { 1532 SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n"); 1533 free(ns_ctx); 1534 } 1535 } 1536 } 1537 1538 static void 1539 _nvmf_ns_resize(struct spdk_nvmf_subsystem *subsystem, void *cb_arg, int status) 1540 { 1541 struct subsystem_ns_change_ctx *ctx = cb_arg; 1542 1543 nvmf_subsystem_ns_changed(subsystem, ctx->nsid); 1544 if (spdk_nvmf_subsystem_resume(subsystem, NULL, NULL) != 0) { 1545 SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id); 1546 } 1547 1548 free(ctx); 1549 } 1550 1551 static void 1552 nvmf_ns_resize(void *event_ctx) 1553 { 1554 struct spdk_nvmf_ns *ns = event_ctx; 1555 struct subsystem_ns_change_ctx *ns_ctx; 1556 int rc; 1557 1558 /* We have to allocate a new context because this op 1559 * is asynchronous and we could lose the ns in the middle. 1560 */ 1561 ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx)); 1562 if (!ns_ctx) { 1563 SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n"); 1564 return; 1565 } 1566 1567 ns_ctx->subsystem = ns->subsystem; 1568 ns_ctx->nsid = ns->opts.nsid; 1569 ns_ctx->cb_fn = _nvmf_ns_resize; 1570 1571 /* Specify 0 for the nsid here, because we do not need to pause the namespace. 1572 * Namespaces can only be resized bigger, so there is no need to quiesce I/O. 1573 */ 1574 rc = spdk_nvmf_subsystem_pause(ns->subsystem, 0, _nvmf_ns_resize, ns_ctx); 1575 if (rc) { 1576 if (rc == -EBUSY) { 1577 /* Try again, this is not a permanent situation. */ 1578 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx); 1579 } else { 1580 SPDK_ERRLOG("Unable to pause subsystem to process namespace resize!\n"); 1581 free(ns_ctx); 1582 } 1583 } 1584 } 1585 1586 static void 1587 nvmf_ns_event(enum spdk_bdev_event_type type, 1588 struct spdk_bdev *bdev, 1589 void *event_ctx) 1590 { 1591 SPDK_DEBUGLOG(nvmf, "Bdev event: type %d, name %s, subsystem_id %d, ns_id %d\n", 1592 type, 1593 spdk_bdev_get_name(bdev), 1594 ((struct spdk_nvmf_ns *)event_ctx)->subsystem->id, 1595 ((struct spdk_nvmf_ns *)event_ctx)->nsid); 1596 1597 switch (type) { 1598 case SPDK_BDEV_EVENT_REMOVE: 1599 nvmf_ns_hot_remove(event_ctx); 1600 break; 1601 case SPDK_BDEV_EVENT_RESIZE: 1602 nvmf_ns_resize(event_ctx); 1603 break; 1604 default: 1605 SPDK_NOTICELOG("Unsupported bdev event: type %d\n", type); 1606 break; 1607 } 1608 } 1609 1610 void 1611 spdk_nvmf_ns_opts_get_defaults(struct spdk_nvmf_ns_opts *opts, size_t opts_size) 1612 { 1613 if (!opts) { 1614 SPDK_ERRLOG("opts should not be NULL.\n"); 1615 return; 1616 } 1617 1618 if (!opts_size) { 1619 SPDK_ERRLOG("opts_size should not be zero.\n"); 1620 return; 1621 } 1622 1623 memset(opts, 0, opts_size); 1624 opts->opts_size = opts_size; 1625 1626 #define FIELD_OK(field) \ 1627 offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= opts_size 1628 1629 #define SET_FIELD(field, value) \ 1630 if (FIELD_OK(field)) { \ 1631 opts->field = value; \ 1632 } \ 1633 1634 /* All current fields are set to 0 by default. */ 1635 SET_FIELD(nsid, 0); 1636 if (FIELD_OK(nguid)) { 1637 memset(opts->nguid, 0, sizeof(opts->nguid)); 1638 } 1639 if (FIELD_OK(eui64)) { 1640 memset(opts->eui64, 0, sizeof(opts->eui64)); 1641 } 1642 if (FIELD_OK(uuid)) { 1643 spdk_uuid_set_null(&opts->uuid); 1644 } 1645 SET_FIELD(anagrpid, 0); 1646 1647 #undef FIELD_OK 1648 #undef SET_FIELD 1649 } 1650 1651 static void 1652 nvmf_ns_opts_copy(struct spdk_nvmf_ns_opts *opts, 1653 const struct spdk_nvmf_ns_opts *user_opts, 1654 size_t opts_size) 1655 { 1656 #define FIELD_OK(field) \ 1657 offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= user_opts->opts_size 1658 1659 #define SET_FIELD(field) \ 1660 if (FIELD_OK(field)) { \ 1661 opts->field = user_opts->field; \ 1662 } \ 1663 1664 SET_FIELD(nsid); 1665 if (FIELD_OK(nguid)) { 1666 memcpy(opts->nguid, user_opts->nguid, sizeof(opts->nguid)); 1667 } 1668 if (FIELD_OK(eui64)) { 1669 memcpy(opts->eui64, user_opts->eui64, sizeof(opts->eui64)); 1670 } 1671 if (FIELD_OK(uuid)) { 1672 spdk_uuid_copy(&opts->uuid, &user_opts->uuid); 1673 } 1674 SET_FIELD(anagrpid); 1675 1676 opts->opts_size = user_opts->opts_size; 1677 1678 /* We should not remove this statement, but need to update the assert statement 1679 * if we add a new field, and also add a corresponding SET_FIELD statement. 1680 */ 1681 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ns_opts) == 64, "Incorrect size"); 1682 1683 #undef FIELD_OK 1684 #undef SET_FIELD 1685 } 1686 1687 /* Dummy bdev module used to to claim bdevs. */ 1688 static struct spdk_bdev_module ns_bdev_module = { 1689 .name = "NVMe-oF Target", 1690 }; 1691 1692 static int nvmf_ns_load_reservation(const char *file, struct spdk_nvmf_reservation_info *info); 1693 static int nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, 1694 struct spdk_nvmf_reservation_info *info); 1695 1696 uint32_t 1697 spdk_nvmf_subsystem_add_ns_ext(struct spdk_nvmf_subsystem *subsystem, const char *bdev_name, 1698 const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size, 1699 const char *ptpl_file) 1700 { 1701 struct spdk_nvmf_transport *transport; 1702 struct spdk_nvmf_ns_opts opts; 1703 struct spdk_nvmf_ns *ns; 1704 struct spdk_nvmf_reservation_info info = {0}; 1705 int rc; 1706 bool zone_append_supported; 1707 uint64_t max_zone_append_size_kib; 1708 1709 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1710 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1711 return 0; 1712 } 1713 1714 spdk_nvmf_ns_opts_get_defaults(&opts, sizeof(opts)); 1715 if (user_opts) { 1716 nvmf_ns_opts_copy(&opts, user_opts, opts_size); 1717 } 1718 1719 if (opts.nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1720 SPDK_ERRLOG("Invalid NSID %" PRIu32 "\n", opts.nsid); 1721 return 0; 1722 } 1723 1724 if (opts.nsid == 0) { 1725 /* 1726 * NSID not specified - find a free index. 1727 * 1728 * If no free slots are found, opts.nsid will be subsystem->max_nsid + 1, which will 1729 * expand max_nsid if possible. 1730 */ 1731 for (opts.nsid = 1; opts.nsid <= subsystem->max_nsid; opts.nsid++) { 1732 if (_nvmf_subsystem_get_ns(subsystem, opts.nsid) == NULL) { 1733 break; 1734 } 1735 } 1736 } 1737 1738 if (_nvmf_subsystem_get_ns(subsystem, opts.nsid)) { 1739 SPDK_ERRLOG("Requested NSID %" PRIu32 " already in use\n", opts.nsid); 1740 return 0; 1741 } 1742 1743 if (opts.nsid > subsystem->max_nsid) { 1744 SPDK_ERRLOG("NSID greater than maximum not allowed\n"); 1745 return 0; 1746 } 1747 1748 if (opts.anagrpid == 0) { 1749 opts.anagrpid = opts.nsid; 1750 } 1751 1752 if (opts.anagrpid > subsystem->max_nsid) { 1753 SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n"); 1754 return 0; 1755 } 1756 1757 ns = calloc(1, sizeof(*ns)); 1758 if (ns == NULL) { 1759 SPDK_ERRLOG("Namespace allocation failed\n"); 1760 return 0; 1761 } 1762 1763 rc = spdk_bdev_open_ext(bdev_name, true, nvmf_ns_event, ns, &ns->desc); 1764 if (rc != 0) { 1765 SPDK_ERRLOG("Subsystem %s: bdev %s cannot be opened, error=%d\n", 1766 subsystem->subnqn, bdev_name, rc); 1767 free(ns); 1768 return 0; 1769 } 1770 1771 ns->bdev = spdk_bdev_desc_get_bdev(ns->desc); 1772 1773 if (spdk_bdev_get_md_size(ns->bdev) != 0) { 1774 if (!spdk_bdev_is_md_interleaved(ns->bdev)) { 1775 SPDK_ERRLOG("Can't attach bdev with separate metadata.\n"); 1776 spdk_bdev_close(ns->desc); 1777 free(ns); 1778 return 0; 1779 } 1780 1781 if (spdk_bdev_get_md_size(ns->bdev) > SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE) { 1782 SPDK_ERRLOG("Maximum supported interleaved md size %u, current md size %u\n", 1783 SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE, spdk_bdev_get_md_size(ns->bdev)); 1784 spdk_bdev_close(ns->desc); 1785 free(ns); 1786 return 0; 1787 } 1788 } 1789 1790 rc = spdk_bdev_module_claim_bdev(ns->bdev, ns->desc, &ns_bdev_module); 1791 if (rc != 0) { 1792 spdk_bdev_close(ns->desc); 1793 free(ns); 1794 return 0; 1795 } 1796 1797 /* Cache the zcopy capability of the bdev device */ 1798 ns->zcopy = spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 1799 1800 if (spdk_uuid_is_null(&opts.uuid)) { 1801 opts.uuid = *spdk_bdev_get_uuid(ns->bdev); 1802 } 1803 1804 /* if nguid descriptor is supported by bdev module (nvme) then uuid = nguid */ 1805 if (spdk_mem_all_zero(opts.nguid, sizeof(opts.nguid))) { 1806 SPDK_STATIC_ASSERT(sizeof(opts.nguid) == sizeof(opts.uuid), "size mismatch"); 1807 memcpy(opts.nguid, spdk_bdev_get_uuid(ns->bdev), sizeof(opts.nguid)); 1808 } 1809 1810 if (spdk_bdev_is_zoned(ns->bdev)) { 1811 SPDK_DEBUGLOG(nvmf, "The added namespace is backed by a zoned block device.\n"); 1812 ns->csi = SPDK_NVME_CSI_ZNS; 1813 1814 zone_append_supported = spdk_bdev_io_type_supported(ns->bdev, 1815 SPDK_BDEV_IO_TYPE_ZONE_APPEND); 1816 max_zone_append_size_kib = spdk_bdev_get_max_zone_append_size( 1817 ns->bdev) * spdk_bdev_get_block_size(ns->bdev); 1818 1819 if (_nvmf_subsystem_get_first_zoned_ns(subsystem) != NULL && 1820 (subsystem->zone_append_supported != zone_append_supported || 1821 subsystem->max_zone_append_size_kib != max_zone_append_size_kib)) { 1822 SPDK_ERRLOG("Namespaces with different zone append support or different zone append size are not allowed.\n"); 1823 goto err_ns_reservation_restore; 1824 } 1825 1826 subsystem->zone_append_supported = zone_append_supported; 1827 subsystem->max_zone_append_size_kib = max_zone_append_size_kib; 1828 } 1829 1830 ns->opts = opts; 1831 ns->subsystem = subsystem; 1832 subsystem->ns[opts.nsid - 1] = ns; 1833 ns->nsid = opts.nsid; 1834 ns->anagrpid = opts.anagrpid; 1835 subsystem->ana_group[ns->anagrpid - 1]++; 1836 TAILQ_INIT(&ns->registrants); 1837 if (ptpl_file) { 1838 rc = nvmf_ns_load_reservation(ptpl_file, &info); 1839 if (!rc) { 1840 rc = nvmf_ns_reservation_restore(ns, &info); 1841 if (rc) { 1842 SPDK_ERRLOG("Subsystem restore reservation failed\n"); 1843 goto err_ns_reservation_restore; 1844 } 1845 } 1846 ns->ptpl_file = strdup(ptpl_file); 1847 if (!ns->ptpl_file) { 1848 SPDK_ERRLOG("Namespace ns->ptpl_file allocation failed\n"); 1849 goto err_strdup; 1850 } 1851 } 1852 1853 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 1854 transport = spdk_nvmf_transport_get_next(transport)) { 1855 if (transport->ops->subsystem_add_ns) { 1856 rc = transport->ops->subsystem_add_ns(transport, subsystem, ns); 1857 if (rc) { 1858 SPDK_ERRLOG("Namespace attachment is not allowed by %s transport\n", transport->ops->name); 1859 goto err_subsystem_add_ns; 1860 } 1861 } 1862 } 1863 1864 SPDK_DEBUGLOG(nvmf, "Subsystem %s: bdev %s assigned nsid %" PRIu32 "\n", 1865 spdk_nvmf_subsystem_get_nqn(subsystem), 1866 bdev_name, 1867 opts.nsid); 1868 1869 nvmf_subsystem_ns_changed(subsystem, opts.nsid); 1870 1871 SPDK_DTRACE_PROBE2(nvmf_subsystem_add_ns, subsystem->subnqn, ns->nsid); 1872 1873 return opts.nsid; 1874 1875 err_subsystem_add_ns: 1876 free(ns->ptpl_file); 1877 err_strdup: 1878 nvmf_ns_reservation_clear_all_registrants(ns); 1879 err_ns_reservation_restore: 1880 subsystem->ns[opts.nsid - 1] = NULL; 1881 spdk_bdev_module_release_bdev(ns->bdev); 1882 spdk_bdev_close(ns->desc); 1883 free(ns); 1884 1885 return 0; 1886 } 1887 1888 static uint32_t 1889 nvmf_subsystem_get_next_allocated_nsid(struct spdk_nvmf_subsystem *subsystem, 1890 uint32_t prev_nsid) 1891 { 1892 uint32_t nsid; 1893 1894 if (prev_nsid >= subsystem->max_nsid) { 1895 return 0; 1896 } 1897 1898 for (nsid = prev_nsid + 1; nsid <= subsystem->max_nsid; nsid++) { 1899 if (subsystem->ns[nsid - 1]) { 1900 return nsid; 1901 } 1902 } 1903 1904 return 0; 1905 } 1906 1907 struct spdk_nvmf_ns * 1908 spdk_nvmf_subsystem_get_first_ns(struct spdk_nvmf_subsystem *subsystem) 1909 { 1910 uint32_t first_nsid; 1911 1912 first_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, 0); 1913 return _nvmf_subsystem_get_ns(subsystem, first_nsid); 1914 } 1915 1916 struct spdk_nvmf_ns * 1917 spdk_nvmf_subsystem_get_next_ns(struct spdk_nvmf_subsystem *subsystem, 1918 struct spdk_nvmf_ns *prev_ns) 1919 { 1920 uint32_t next_nsid; 1921 1922 next_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, prev_ns->opts.nsid); 1923 return _nvmf_subsystem_get_ns(subsystem, next_nsid); 1924 } 1925 1926 struct spdk_nvmf_ns * 1927 spdk_nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1928 { 1929 return _nvmf_subsystem_get_ns(subsystem, nsid); 1930 } 1931 1932 uint32_t 1933 spdk_nvmf_ns_get_id(const struct spdk_nvmf_ns *ns) 1934 { 1935 return ns->opts.nsid; 1936 } 1937 1938 struct spdk_bdev * 1939 spdk_nvmf_ns_get_bdev(struct spdk_nvmf_ns *ns) 1940 { 1941 return ns->bdev; 1942 } 1943 1944 void 1945 spdk_nvmf_ns_get_opts(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_ns_opts *opts, 1946 size_t opts_size) 1947 { 1948 memset(opts, 0, opts_size); 1949 memcpy(opts, &ns->opts, spdk_min(sizeof(ns->opts), opts_size)); 1950 } 1951 1952 const char * 1953 spdk_nvmf_subsystem_get_sn(const struct spdk_nvmf_subsystem *subsystem) 1954 { 1955 return subsystem->sn; 1956 } 1957 1958 int 1959 spdk_nvmf_subsystem_set_sn(struct spdk_nvmf_subsystem *subsystem, const char *sn) 1960 { 1961 size_t len, max_len; 1962 1963 max_len = sizeof(subsystem->sn) - 1; 1964 len = strlen(sn); 1965 if (len > max_len) { 1966 SPDK_DEBUGLOG(nvmf, "Invalid sn \"%s\": length %zu > max %zu\n", 1967 sn, len, max_len); 1968 return -1; 1969 } 1970 1971 if (!nvmf_valid_ascii_string(sn, len)) { 1972 SPDK_DEBUGLOG(nvmf, "Non-ASCII sn\n"); 1973 SPDK_LOGDUMP(nvmf, "sn", sn, len); 1974 return -1; 1975 } 1976 1977 snprintf(subsystem->sn, sizeof(subsystem->sn), "%s", sn); 1978 1979 return 0; 1980 } 1981 1982 const char * 1983 spdk_nvmf_subsystem_get_mn(const struct spdk_nvmf_subsystem *subsystem) 1984 { 1985 return subsystem->mn; 1986 } 1987 1988 int 1989 spdk_nvmf_subsystem_set_mn(struct spdk_nvmf_subsystem *subsystem, const char *mn) 1990 { 1991 size_t len, max_len; 1992 1993 if (mn == NULL) { 1994 mn = MODEL_NUMBER_DEFAULT; 1995 } 1996 max_len = sizeof(subsystem->mn) - 1; 1997 len = strlen(mn); 1998 if (len > max_len) { 1999 SPDK_DEBUGLOG(nvmf, "Invalid mn \"%s\": length %zu > max %zu\n", 2000 mn, len, max_len); 2001 return -1; 2002 } 2003 2004 if (!nvmf_valid_ascii_string(mn, len)) { 2005 SPDK_DEBUGLOG(nvmf, "Non-ASCII mn\n"); 2006 SPDK_LOGDUMP(nvmf, "mn", mn, len); 2007 return -1; 2008 } 2009 2010 snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", mn); 2011 2012 return 0; 2013 } 2014 2015 const char * 2016 spdk_nvmf_subsystem_get_nqn(const struct spdk_nvmf_subsystem *subsystem) 2017 { 2018 return subsystem->subnqn; 2019 } 2020 2021 /* We have to use the typedef in the function declaration to appease astyle. */ 2022 typedef enum spdk_nvmf_subtype spdk_nvmf_subtype_t; 2023 2024 spdk_nvmf_subtype_t 2025 spdk_nvmf_subsystem_get_type(struct spdk_nvmf_subsystem *subsystem) 2026 { 2027 return subsystem->subtype; 2028 } 2029 2030 uint32_t 2031 spdk_nvmf_subsystem_get_max_nsid(struct spdk_nvmf_subsystem *subsystem) 2032 { 2033 return subsystem->max_nsid; 2034 } 2035 2036 int 2037 nvmf_subsystem_set_cntlid_range(struct spdk_nvmf_subsystem *subsystem, 2038 uint16_t min_cntlid, uint16_t max_cntlid) 2039 { 2040 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 2041 return -EAGAIN; 2042 } 2043 2044 if (min_cntlid > max_cntlid) { 2045 return -EINVAL; 2046 } 2047 /* The spec reserves cntlid values in the range FFF0h to FFFFh. */ 2048 if (min_cntlid < NVMF_MIN_CNTLID || min_cntlid > NVMF_MAX_CNTLID || 2049 max_cntlid < NVMF_MIN_CNTLID || max_cntlid > NVMF_MAX_CNTLID) { 2050 return -EINVAL; 2051 } 2052 subsystem->min_cntlid = min_cntlid; 2053 subsystem->max_cntlid = max_cntlid; 2054 if (subsystem->next_cntlid < min_cntlid || subsystem->next_cntlid > max_cntlid - 1) { 2055 subsystem->next_cntlid = min_cntlid - 1; 2056 } 2057 2058 return 0; 2059 } 2060 2061 static uint16_t 2062 nvmf_subsystem_gen_cntlid(struct spdk_nvmf_subsystem *subsystem) 2063 { 2064 int count; 2065 2066 /* 2067 * In the worst case, we might have to try all CNTLID values between min_cntlid and max_cntlid 2068 * before we find one that is unused (or find that all values are in use). 2069 */ 2070 for (count = 0; count < subsystem->max_cntlid - subsystem->min_cntlid + 1; count++) { 2071 subsystem->next_cntlid++; 2072 if (subsystem->next_cntlid > subsystem->max_cntlid) { 2073 subsystem->next_cntlid = subsystem->min_cntlid; 2074 } 2075 2076 /* Check if a controller with this cntlid currently exists. */ 2077 if (nvmf_subsystem_get_ctrlr(subsystem, subsystem->next_cntlid) == NULL) { 2078 /* Found unused cntlid */ 2079 return subsystem->next_cntlid; 2080 } 2081 } 2082 2083 /* All valid cntlid values are in use. */ 2084 return 0xFFFF; 2085 } 2086 2087 int 2088 nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_ctrlr *ctrlr) 2089 { 2090 2091 if (ctrlr->dynamic_ctrlr) { 2092 ctrlr->cntlid = nvmf_subsystem_gen_cntlid(subsystem); 2093 if (ctrlr->cntlid == 0xFFFF) { 2094 /* Unable to get a cntlid */ 2095 SPDK_ERRLOG("Reached max simultaneous ctrlrs\n"); 2096 return -EBUSY; 2097 } 2098 } else if (nvmf_subsystem_get_ctrlr(subsystem, ctrlr->cntlid) != NULL) { 2099 SPDK_ERRLOG("Ctrlr with cntlid %u already exist\n", ctrlr->cntlid); 2100 return -EEXIST; 2101 } 2102 2103 TAILQ_INSERT_TAIL(&subsystem->ctrlrs, ctrlr, link); 2104 2105 SPDK_DTRACE_PROBE3(nvmf_subsystem_add_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn); 2106 2107 return 0; 2108 } 2109 2110 void 2111 nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem, 2112 struct spdk_nvmf_ctrlr *ctrlr) 2113 { 2114 SPDK_DTRACE_PROBE3(nvmf_subsystem_remove_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn); 2115 2116 assert(spdk_get_thread() == subsystem->thread); 2117 assert(subsystem == ctrlr->subsys); 2118 SPDK_DEBUGLOG(nvmf, "remove ctrlr %p id 0x%x from subsys %p %s\n", ctrlr, ctrlr->cntlid, subsystem, 2119 subsystem->subnqn); 2120 TAILQ_REMOVE(&subsystem->ctrlrs, ctrlr, link); 2121 } 2122 2123 struct spdk_nvmf_ctrlr * 2124 nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, uint16_t cntlid) 2125 { 2126 struct spdk_nvmf_ctrlr *ctrlr; 2127 2128 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 2129 if (ctrlr->cntlid == cntlid) { 2130 return ctrlr; 2131 } 2132 } 2133 2134 return NULL; 2135 } 2136 2137 uint32_t 2138 spdk_nvmf_subsystem_get_max_namespaces(const struct spdk_nvmf_subsystem *subsystem) 2139 { 2140 return subsystem->max_nsid; 2141 } 2142 2143 uint16_t 2144 spdk_nvmf_subsystem_get_min_cntlid(const struct spdk_nvmf_subsystem *subsystem) 2145 { 2146 return subsystem->min_cntlid; 2147 } 2148 2149 uint16_t 2150 spdk_nvmf_subsystem_get_max_cntlid(const struct spdk_nvmf_subsystem *subsystem) 2151 { 2152 return subsystem->max_cntlid; 2153 } 2154 2155 struct _nvmf_ns_registrant { 2156 uint64_t rkey; 2157 char *host_uuid; 2158 }; 2159 2160 struct _nvmf_ns_registrants { 2161 size_t num_regs; 2162 struct _nvmf_ns_registrant reg[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2163 }; 2164 2165 struct _nvmf_ns_reservation { 2166 bool ptpl_activated; 2167 enum spdk_nvme_reservation_type rtype; 2168 uint64_t crkey; 2169 char *bdev_uuid; 2170 char *holder_uuid; 2171 struct _nvmf_ns_registrants regs; 2172 }; 2173 2174 static const struct spdk_json_object_decoder nvmf_ns_pr_reg_decoders[] = { 2175 {"rkey", offsetof(struct _nvmf_ns_registrant, rkey), spdk_json_decode_uint64}, 2176 {"host_uuid", offsetof(struct _nvmf_ns_registrant, host_uuid), spdk_json_decode_string}, 2177 }; 2178 2179 static int 2180 nvmf_decode_ns_pr_reg(const struct spdk_json_val *val, void *out) 2181 { 2182 struct _nvmf_ns_registrant *reg = out; 2183 2184 return spdk_json_decode_object(val, nvmf_ns_pr_reg_decoders, 2185 SPDK_COUNTOF(nvmf_ns_pr_reg_decoders), reg); 2186 } 2187 2188 static int 2189 nvmf_decode_ns_pr_regs(const struct spdk_json_val *val, void *out) 2190 { 2191 struct _nvmf_ns_registrants *regs = out; 2192 2193 return spdk_json_decode_array(val, nvmf_decode_ns_pr_reg, regs->reg, 2194 SPDK_NVMF_MAX_NUM_REGISTRANTS, ®s->num_regs, 2195 sizeof(struct _nvmf_ns_registrant)); 2196 } 2197 2198 static const struct spdk_json_object_decoder nvmf_ns_pr_decoders[] = { 2199 {"ptpl", offsetof(struct _nvmf_ns_reservation, ptpl_activated), spdk_json_decode_bool, true}, 2200 {"rtype", offsetof(struct _nvmf_ns_reservation, rtype), spdk_json_decode_uint32, true}, 2201 {"crkey", offsetof(struct _nvmf_ns_reservation, crkey), spdk_json_decode_uint64, true}, 2202 {"bdev_uuid", offsetof(struct _nvmf_ns_reservation, bdev_uuid), spdk_json_decode_string}, 2203 {"holder_uuid", offsetof(struct _nvmf_ns_reservation, holder_uuid), spdk_json_decode_string, true}, 2204 {"registrants", offsetof(struct _nvmf_ns_reservation, regs), nvmf_decode_ns_pr_regs}, 2205 }; 2206 2207 static int 2208 nvmf_ns_load_reservation(const char *file, struct spdk_nvmf_reservation_info *info) 2209 { 2210 FILE *fd; 2211 size_t json_size; 2212 ssize_t values_cnt, rc; 2213 void *json = NULL, *end; 2214 struct spdk_json_val *values = NULL; 2215 struct _nvmf_ns_reservation res = {}; 2216 uint32_t i; 2217 2218 fd = fopen(file, "r"); 2219 /* It's not an error if the file does not exist */ 2220 if (!fd) { 2221 SPDK_NOTICELOG("File %s does not exist\n", file); 2222 return -ENOENT; 2223 } 2224 2225 /* Load all persist file contents into a local buffer */ 2226 json = spdk_posix_file_load(fd, &json_size); 2227 fclose(fd); 2228 if (!json) { 2229 SPDK_ERRLOG("Load persit file %s failed\n", file); 2230 return -ENOMEM; 2231 } 2232 2233 rc = spdk_json_parse(json, json_size, NULL, 0, &end, 0); 2234 if (rc < 0) { 2235 SPDK_NOTICELOG("Parsing JSON configuration failed (%zd)\n", rc); 2236 goto exit; 2237 } 2238 2239 values_cnt = rc; 2240 values = calloc(values_cnt, sizeof(struct spdk_json_val)); 2241 if (values == NULL) { 2242 goto exit; 2243 } 2244 2245 rc = spdk_json_parse(json, json_size, values, values_cnt, &end, 0); 2246 if (rc != values_cnt) { 2247 SPDK_ERRLOG("Parsing JSON configuration failed (%zd)\n", rc); 2248 goto exit; 2249 } 2250 2251 /* Decode json */ 2252 if (spdk_json_decode_object(values, nvmf_ns_pr_decoders, 2253 SPDK_COUNTOF(nvmf_ns_pr_decoders), 2254 &res)) { 2255 SPDK_ERRLOG("Invalid objects in the persist file %s\n", file); 2256 rc = -EINVAL; 2257 goto exit; 2258 } 2259 2260 if (res.regs.num_regs > SPDK_NVMF_MAX_NUM_REGISTRANTS) { 2261 SPDK_ERRLOG("Can only support up to %u registrants\n", SPDK_NVMF_MAX_NUM_REGISTRANTS); 2262 rc = -ERANGE; 2263 goto exit; 2264 } 2265 2266 rc = 0; 2267 info->ptpl_activated = res.ptpl_activated; 2268 info->rtype = res.rtype; 2269 info->crkey = res.crkey; 2270 snprintf(info->bdev_uuid, sizeof(info->bdev_uuid), "%s", res.bdev_uuid); 2271 snprintf(info->holder_uuid, sizeof(info->holder_uuid), "%s", res.holder_uuid); 2272 info->num_regs = res.regs.num_regs; 2273 for (i = 0; i < res.regs.num_regs; i++) { 2274 info->registrants[i].rkey = res.regs.reg[i].rkey; 2275 snprintf(info->registrants[i].host_uuid, sizeof(info->registrants[i].host_uuid), "%s", 2276 res.regs.reg[i].host_uuid); 2277 } 2278 2279 exit: 2280 free(json); 2281 free(values); 2282 free(res.bdev_uuid); 2283 free(res.holder_uuid); 2284 for (i = 0; i < res.regs.num_regs; i++) { 2285 free(res.regs.reg[i].host_uuid); 2286 } 2287 2288 return rc; 2289 } 2290 2291 static bool nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns); 2292 2293 static int 2294 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info) 2295 { 2296 uint32_t i; 2297 struct spdk_nvmf_registrant *reg, *holder = NULL; 2298 struct spdk_uuid bdev_uuid, holder_uuid; 2299 bool rkey_flag = false; 2300 2301 SPDK_DEBUGLOG(nvmf, "NSID %u, PTPL %u, Number of registrants %u\n", 2302 ns->nsid, info->ptpl_activated, info->num_regs); 2303 2304 /* it's not an error */ 2305 if (!info->ptpl_activated || !info->num_regs) { 2306 return 0; 2307 } 2308 2309 /* Check info->crkey exist or not in info->registrants[i].rkey */ 2310 for (i = 0; i < info->num_regs; i++) { 2311 if (info->crkey == info->registrants[i].rkey) { 2312 rkey_flag = true; 2313 } 2314 } 2315 if (!rkey_flag) { 2316 return -EINVAL; 2317 } 2318 2319 spdk_uuid_parse(&bdev_uuid, info->bdev_uuid); 2320 if (spdk_uuid_compare(&bdev_uuid, spdk_bdev_get_uuid(ns->bdev))) { 2321 SPDK_ERRLOG("Existing bdev UUID is not same with configuration file\n"); 2322 return -EINVAL; 2323 } 2324 2325 ns->crkey = info->crkey; 2326 ns->rtype = info->rtype; 2327 ns->ptpl_activated = info->ptpl_activated; 2328 spdk_uuid_parse(&holder_uuid, info->holder_uuid); 2329 2330 SPDK_DEBUGLOG(nvmf, "Bdev UUID %s\n", info->bdev_uuid); 2331 if (info->rtype) { 2332 SPDK_DEBUGLOG(nvmf, "Holder UUID %s, RTYPE %u, RKEY 0x%"PRIx64"\n", 2333 info->holder_uuid, info->rtype, info->crkey); 2334 } 2335 2336 for (i = 0; i < info->num_regs; i++) { 2337 reg = calloc(1, sizeof(*reg)); 2338 if (!reg) { 2339 return -ENOMEM; 2340 } 2341 spdk_uuid_parse(®->hostid, info->registrants[i].host_uuid); 2342 reg->rkey = info->registrants[i].rkey; 2343 TAILQ_INSERT_TAIL(&ns->registrants, reg, link); 2344 if (!spdk_uuid_compare(&holder_uuid, ®->hostid)) { 2345 holder = reg; 2346 } 2347 SPDK_DEBUGLOG(nvmf, "Registrant RKEY 0x%"PRIx64", Host UUID %s\n", 2348 info->registrants[i].rkey, info->registrants[i].host_uuid); 2349 } 2350 2351 if (nvmf_ns_reservation_all_registrants_type(ns)) { 2352 ns->holder = TAILQ_FIRST(&ns->registrants); 2353 } else { 2354 ns->holder = holder; 2355 } 2356 2357 return 0; 2358 } 2359 2360 static int 2361 nvmf_ns_json_write_cb(void *cb_ctx, const void *data, size_t size) 2362 { 2363 char *file = cb_ctx; 2364 size_t rc; 2365 FILE *fd; 2366 2367 fd = fopen(file, "w"); 2368 if (!fd) { 2369 SPDK_ERRLOG("Can't open file %s for write\n", file); 2370 return -ENOENT; 2371 } 2372 rc = fwrite(data, 1, size, fd); 2373 fclose(fd); 2374 2375 return rc == size ? 0 : -1; 2376 } 2377 2378 static int 2379 nvmf_ns_reservation_update(const char *file, struct spdk_nvmf_reservation_info *info) 2380 { 2381 struct spdk_json_write_ctx *w; 2382 uint32_t i; 2383 int rc = 0; 2384 2385 w = spdk_json_write_begin(nvmf_ns_json_write_cb, (void *)file, 0); 2386 if (w == NULL) { 2387 return -ENOMEM; 2388 } 2389 /* clear the configuration file */ 2390 if (!info->ptpl_activated) { 2391 goto exit; 2392 } 2393 2394 spdk_json_write_object_begin(w); 2395 spdk_json_write_named_bool(w, "ptpl", info->ptpl_activated); 2396 spdk_json_write_named_uint32(w, "rtype", info->rtype); 2397 spdk_json_write_named_uint64(w, "crkey", info->crkey); 2398 spdk_json_write_named_string(w, "bdev_uuid", info->bdev_uuid); 2399 spdk_json_write_named_string(w, "holder_uuid", info->holder_uuid); 2400 2401 spdk_json_write_named_array_begin(w, "registrants"); 2402 for (i = 0; i < info->num_regs; i++) { 2403 spdk_json_write_object_begin(w); 2404 spdk_json_write_named_uint64(w, "rkey", info->registrants[i].rkey); 2405 spdk_json_write_named_string(w, "host_uuid", info->registrants[i].host_uuid); 2406 spdk_json_write_object_end(w); 2407 } 2408 spdk_json_write_array_end(w); 2409 spdk_json_write_object_end(w); 2410 2411 exit: 2412 rc = spdk_json_write_end(w); 2413 return rc; 2414 } 2415 2416 static int 2417 nvmf_ns_update_reservation_info(struct spdk_nvmf_ns *ns) 2418 { 2419 struct spdk_nvmf_reservation_info info; 2420 struct spdk_nvmf_registrant *reg, *tmp; 2421 uint32_t i = 0; 2422 2423 assert(ns != NULL); 2424 2425 if (!ns->bdev || !ns->ptpl_file) { 2426 return 0; 2427 } 2428 2429 memset(&info, 0, sizeof(info)); 2430 spdk_uuid_fmt_lower(info.bdev_uuid, sizeof(info.bdev_uuid), spdk_bdev_get_uuid(ns->bdev)); 2431 2432 if (ns->rtype) { 2433 info.rtype = ns->rtype; 2434 info.crkey = ns->crkey; 2435 if (!nvmf_ns_reservation_all_registrants_type(ns)) { 2436 assert(ns->holder != NULL); 2437 spdk_uuid_fmt_lower(info.holder_uuid, sizeof(info.holder_uuid), &ns->holder->hostid); 2438 } 2439 } 2440 2441 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2442 spdk_uuid_fmt_lower(info.registrants[i].host_uuid, sizeof(info.registrants[i].host_uuid), 2443 ®->hostid); 2444 info.registrants[i++].rkey = reg->rkey; 2445 } 2446 2447 info.num_regs = i; 2448 info.ptpl_activated = ns->ptpl_activated; 2449 2450 return nvmf_ns_reservation_update(ns->ptpl_file, &info); 2451 } 2452 2453 static struct spdk_nvmf_registrant * 2454 nvmf_ns_reservation_get_registrant(struct spdk_nvmf_ns *ns, 2455 struct spdk_uuid *uuid) 2456 { 2457 struct spdk_nvmf_registrant *reg, *tmp; 2458 2459 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2460 if (!spdk_uuid_compare(®->hostid, uuid)) { 2461 return reg; 2462 } 2463 } 2464 2465 return NULL; 2466 } 2467 2468 /* Generate reservation notice log to registered HostID controllers */ 2469 static void 2470 nvmf_subsystem_gen_ctrlr_notification(struct spdk_nvmf_subsystem *subsystem, 2471 struct spdk_nvmf_ns *ns, 2472 struct spdk_uuid *hostid_list, 2473 uint32_t num_hostid, 2474 enum spdk_nvme_reservation_notification_log_page_type type) 2475 { 2476 struct spdk_nvmf_ctrlr *ctrlr; 2477 uint32_t i; 2478 2479 for (i = 0; i < num_hostid; i++) { 2480 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 2481 if (!spdk_uuid_compare(&ctrlr->hostid, &hostid_list[i])) { 2482 nvmf_ctrlr_reservation_notice_log(ctrlr, ns, type); 2483 } 2484 } 2485 } 2486 } 2487 2488 /* Get all registrants' hostid other than the controller who issued the command */ 2489 static uint32_t 2490 nvmf_ns_reservation_get_all_other_hostid(struct spdk_nvmf_ns *ns, 2491 struct spdk_uuid *hostid_list, 2492 uint32_t max_num_hostid, 2493 struct spdk_uuid *current_hostid) 2494 { 2495 struct spdk_nvmf_registrant *reg, *tmp; 2496 uint32_t num_hostid = 0; 2497 2498 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2499 if (spdk_uuid_compare(®->hostid, current_hostid)) { 2500 if (num_hostid == max_num_hostid) { 2501 assert(false); 2502 return max_num_hostid; 2503 } 2504 hostid_list[num_hostid++] = reg->hostid; 2505 } 2506 } 2507 2508 return num_hostid; 2509 } 2510 2511 /* Calculate the unregistered HostID list according to list 2512 * prior to execute preempt command and list after executing 2513 * preempt command. 2514 */ 2515 static uint32_t 2516 nvmf_ns_reservation_get_unregistered_hostid(struct spdk_uuid *old_hostid_list, 2517 uint32_t old_num_hostid, 2518 struct spdk_uuid *remaining_hostid_list, 2519 uint32_t remaining_num_hostid) 2520 { 2521 struct spdk_uuid temp_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2522 uint32_t i, j, num_hostid = 0; 2523 bool found; 2524 2525 if (!remaining_num_hostid) { 2526 return old_num_hostid; 2527 } 2528 2529 for (i = 0; i < old_num_hostid; i++) { 2530 found = false; 2531 for (j = 0; j < remaining_num_hostid; j++) { 2532 if (!spdk_uuid_compare(&old_hostid_list[i], &remaining_hostid_list[j])) { 2533 found = true; 2534 break; 2535 } 2536 } 2537 if (!found) { 2538 spdk_uuid_copy(&temp_hostid_list[num_hostid++], &old_hostid_list[i]); 2539 } 2540 } 2541 2542 if (num_hostid) { 2543 memcpy(old_hostid_list, temp_hostid_list, sizeof(struct spdk_uuid) * num_hostid); 2544 } 2545 2546 return num_hostid; 2547 } 2548 2549 /* current reservation type is all registrants or not */ 2550 static bool 2551 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns) 2552 { 2553 return (ns->rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS || 2554 ns->rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS); 2555 } 2556 2557 /* current registrant is reservation holder or not */ 2558 static bool 2559 nvmf_ns_reservation_registrant_is_holder(struct spdk_nvmf_ns *ns, 2560 struct spdk_nvmf_registrant *reg) 2561 { 2562 if (!reg) { 2563 return false; 2564 } 2565 2566 if (nvmf_ns_reservation_all_registrants_type(ns)) { 2567 return true; 2568 } 2569 2570 return (ns->holder == reg); 2571 } 2572 2573 static int 2574 nvmf_ns_reservation_add_registrant(struct spdk_nvmf_ns *ns, 2575 struct spdk_nvmf_ctrlr *ctrlr, 2576 uint64_t nrkey) 2577 { 2578 struct spdk_nvmf_registrant *reg; 2579 2580 reg = calloc(1, sizeof(*reg)); 2581 if (!reg) { 2582 return -ENOMEM; 2583 } 2584 2585 reg->rkey = nrkey; 2586 /* set hostid for the registrant */ 2587 spdk_uuid_copy(®->hostid, &ctrlr->hostid); 2588 TAILQ_INSERT_TAIL(&ns->registrants, reg, link); 2589 ns->gen++; 2590 2591 return 0; 2592 } 2593 2594 static void 2595 nvmf_ns_reservation_release_reservation(struct spdk_nvmf_ns *ns) 2596 { 2597 ns->rtype = 0; 2598 ns->crkey = 0; 2599 ns->holder = NULL; 2600 } 2601 2602 /* release the reservation if the last registrant was removed */ 2603 static void 2604 nvmf_ns_reservation_check_release_on_remove_registrant(struct spdk_nvmf_ns *ns, 2605 struct spdk_nvmf_registrant *reg) 2606 { 2607 struct spdk_nvmf_registrant *next_reg; 2608 2609 /* no reservation holder */ 2610 if (!ns->holder) { 2611 assert(ns->rtype == 0); 2612 return; 2613 } 2614 2615 next_reg = TAILQ_FIRST(&ns->registrants); 2616 if (next_reg && nvmf_ns_reservation_all_registrants_type(ns)) { 2617 /* the next valid registrant is the new holder now */ 2618 ns->holder = next_reg; 2619 } else if (nvmf_ns_reservation_registrant_is_holder(ns, reg)) { 2620 /* release the reservation */ 2621 nvmf_ns_reservation_release_reservation(ns); 2622 } 2623 } 2624 2625 static void 2626 nvmf_ns_reservation_remove_registrant(struct spdk_nvmf_ns *ns, 2627 struct spdk_nvmf_registrant *reg) 2628 { 2629 TAILQ_REMOVE(&ns->registrants, reg, link); 2630 nvmf_ns_reservation_check_release_on_remove_registrant(ns, reg); 2631 free(reg); 2632 ns->gen++; 2633 return; 2634 } 2635 2636 static uint32_t 2637 nvmf_ns_reservation_remove_registrants_by_key(struct spdk_nvmf_ns *ns, 2638 uint64_t rkey) 2639 { 2640 struct spdk_nvmf_registrant *reg, *tmp; 2641 uint32_t count = 0; 2642 2643 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2644 if (reg->rkey == rkey) { 2645 nvmf_ns_reservation_remove_registrant(ns, reg); 2646 count++; 2647 } 2648 } 2649 return count; 2650 } 2651 2652 static uint32_t 2653 nvmf_ns_reservation_remove_all_other_registrants(struct spdk_nvmf_ns *ns, 2654 struct spdk_nvmf_registrant *reg) 2655 { 2656 struct spdk_nvmf_registrant *reg_tmp, *reg_tmp2; 2657 uint32_t count = 0; 2658 2659 TAILQ_FOREACH_SAFE(reg_tmp, &ns->registrants, link, reg_tmp2) { 2660 if (reg_tmp != reg) { 2661 nvmf_ns_reservation_remove_registrant(ns, reg_tmp); 2662 count++; 2663 } 2664 } 2665 return count; 2666 } 2667 2668 static uint32_t 2669 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns) 2670 { 2671 struct spdk_nvmf_registrant *reg, *reg_tmp; 2672 uint32_t count = 0; 2673 2674 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, reg_tmp) { 2675 nvmf_ns_reservation_remove_registrant(ns, reg); 2676 count++; 2677 } 2678 return count; 2679 } 2680 2681 static void 2682 nvmf_ns_reservation_acquire_reservation(struct spdk_nvmf_ns *ns, uint64_t rkey, 2683 enum spdk_nvme_reservation_type rtype, 2684 struct spdk_nvmf_registrant *holder) 2685 { 2686 ns->rtype = rtype; 2687 ns->crkey = rkey; 2688 assert(ns->holder == NULL); 2689 ns->holder = holder; 2690 } 2691 2692 static bool 2693 nvmf_ns_reservation_register(struct spdk_nvmf_ns *ns, 2694 struct spdk_nvmf_ctrlr *ctrlr, 2695 struct spdk_nvmf_request *req) 2696 { 2697 struct spdk_nvme_reservation_register_data key = { 0 }; 2698 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2699 uint8_t rrega, iekey, cptpl, rtype; 2700 struct spdk_nvmf_registrant *reg; 2701 uint8_t status = SPDK_NVME_SC_SUCCESS; 2702 bool update_sgroup = false; 2703 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2704 uint32_t num_hostid = 0; 2705 int rc; 2706 2707 rrega = cmd->cdw10_bits.resv_register.rrega; 2708 iekey = cmd->cdw10_bits.resv_register.iekey; 2709 cptpl = cmd->cdw10_bits.resv_register.cptpl; 2710 2711 if (req->iovcnt > 0 && req->length >= sizeof(key)) { 2712 struct spdk_iov_xfer ix; 2713 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 2714 spdk_iov_xfer_to_buf(&ix, &key, sizeof(key)); 2715 } else { 2716 SPDK_ERRLOG("No key provided. Failing request.\n"); 2717 status = SPDK_NVME_SC_INVALID_FIELD; 2718 goto exit; 2719 } 2720 2721 SPDK_DEBUGLOG(nvmf, "REGISTER: RREGA %u, IEKEY %u, CPTPL %u, " 2722 "NRKEY 0x%"PRIx64", NRKEY 0x%"PRIx64"\n", 2723 rrega, iekey, cptpl, key.crkey, key.nrkey); 2724 2725 if (cptpl == SPDK_NVME_RESERVE_PTPL_CLEAR_POWER_ON) { 2726 /* Ture to OFF state, and need to be updated in the configuration file */ 2727 if (ns->ptpl_activated) { 2728 ns->ptpl_activated = 0; 2729 update_sgroup = true; 2730 } 2731 } else if (cptpl == SPDK_NVME_RESERVE_PTPL_PERSIST_POWER_LOSS) { 2732 if (ns->ptpl_file == NULL) { 2733 status = SPDK_NVME_SC_INVALID_FIELD; 2734 goto exit; 2735 } else if (ns->ptpl_activated == 0) { 2736 ns->ptpl_activated = 1; 2737 update_sgroup = true; 2738 } 2739 } 2740 2741 /* current Host Identifier has registrant or not */ 2742 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2743 2744 switch (rrega) { 2745 case SPDK_NVME_RESERVE_REGISTER_KEY: 2746 if (!reg) { 2747 /* register new controller */ 2748 if (key.nrkey == 0) { 2749 SPDK_ERRLOG("Can't register zeroed new key\n"); 2750 status = SPDK_NVME_SC_INVALID_FIELD; 2751 goto exit; 2752 } 2753 rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey); 2754 if (rc < 0) { 2755 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2756 goto exit; 2757 } 2758 update_sgroup = true; 2759 } else { 2760 /* register with same key is not an error */ 2761 if (reg->rkey != key.nrkey) { 2762 SPDK_ERRLOG("The same host already register a " 2763 "key with 0x%"PRIx64"\n", 2764 reg->rkey); 2765 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2766 goto exit; 2767 } 2768 } 2769 break; 2770 case SPDK_NVME_RESERVE_UNREGISTER_KEY: 2771 if (!reg || (!iekey && reg->rkey != key.crkey)) { 2772 SPDK_ERRLOG("No registrant or current key doesn't match " 2773 "with existing registrant key\n"); 2774 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2775 goto exit; 2776 } 2777 2778 rtype = ns->rtype; 2779 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2780 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2781 &ctrlr->hostid); 2782 2783 nvmf_ns_reservation_remove_registrant(ns, reg); 2784 2785 if (!ns->rtype && num_hostid && (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_REG_ONLY || 2786 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY)) { 2787 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2788 hostid_list, 2789 num_hostid, 2790 SPDK_NVME_RESERVATION_RELEASED); 2791 } 2792 update_sgroup = true; 2793 break; 2794 case SPDK_NVME_RESERVE_REPLACE_KEY: 2795 if (key.nrkey == 0) { 2796 SPDK_ERRLOG("Can't register zeroed new key\n"); 2797 status = SPDK_NVME_SC_INVALID_FIELD; 2798 goto exit; 2799 } 2800 /* Registrant exists */ 2801 if (reg) { 2802 if (!iekey && reg->rkey != key.crkey) { 2803 SPDK_ERRLOG("Current key doesn't match " 2804 "existing registrant key\n"); 2805 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2806 goto exit; 2807 } 2808 if (reg->rkey == key.nrkey) { 2809 goto exit; 2810 } 2811 reg->rkey = key.nrkey; 2812 } else if (iekey) { /* No registrant but IEKEY is set */ 2813 /* new registrant */ 2814 rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey); 2815 if (rc < 0) { 2816 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2817 goto exit; 2818 } 2819 } else { /* No registrant */ 2820 SPDK_ERRLOG("No registrant\n"); 2821 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2822 goto exit; 2823 2824 } 2825 update_sgroup = true; 2826 break; 2827 default: 2828 status = SPDK_NVME_SC_INVALID_FIELD; 2829 goto exit; 2830 } 2831 2832 exit: 2833 if (update_sgroup) { 2834 rc = nvmf_ns_update_reservation_info(ns); 2835 if (rc != 0) { 2836 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2837 } 2838 } 2839 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2840 req->rsp->nvme_cpl.status.sc = status; 2841 return update_sgroup; 2842 } 2843 2844 static bool 2845 nvmf_ns_reservation_acquire(struct spdk_nvmf_ns *ns, 2846 struct spdk_nvmf_ctrlr *ctrlr, 2847 struct spdk_nvmf_request *req) 2848 { 2849 struct spdk_nvme_reservation_acquire_data key = { 0 }; 2850 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2851 uint8_t racqa, iekey, rtype; 2852 struct spdk_nvmf_registrant *reg; 2853 bool all_regs = false; 2854 uint32_t count = 0; 2855 bool update_sgroup = true; 2856 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2857 uint32_t num_hostid = 0; 2858 struct spdk_uuid new_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2859 uint32_t new_num_hostid = 0; 2860 bool reservation_released = false; 2861 uint8_t status = SPDK_NVME_SC_SUCCESS; 2862 2863 racqa = cmd->cdw10_bits.resv_acquire.racqa; 2864 iekey = cmd->cdw10_bits.resv_acquire.iekey; 2865 rtype = cmd->cdw10_bits.resv_acquire.rtype; 2866 2867 if (req->iovcnt > 0 && req->length >= sizeof(key)) { 2868 struct spdk_iov_xfer ix; 2869 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 2870 spdk_iov_xfer_to_buf(&ix, &key, sizeof(key)); 2871 } else { 2872 SPDK_ERRLOG("No key provided. Failing request.\n"); 2873 status = SPDK_NVME_SC_INVALID_FIELD; 2874 goto exit; 2875 } 2876 2877 SPDK_DEBUGLOG(nvmf, "ACQUIRE: RACQA %u, IEKEY %u, RTYPE %u, " 2878 "NRKEY 0x%"PRIx64", PRKEY 0x%"PRIx64"\n", 2879 racqa, iekey, rtype, key.crkey, key.prkey); 2880 2881 if (iekey || rtype > SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) { 2882 SPDK_ERRLOG("Ignore existing key field set to 1\n"); 2883 status = SPDK_NVME_SC_INVALID_FIELD; 2884 update_sgroup = false; 2885 goto exit; 2886 } 2887 2888 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2889 /* must be registrant and CRKEY must match */ 2890 if (!reg || reg->rkey != key.crkey) { 2891 SPDK_ERRLOG("No registrant or current key doesn't match " 2892 "with existing registrant key\n"); 2893 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2894 update_sgroup = false; 2895 goto exit; 2896 } 2897 2898 all_regs = nvmf_ns_reservation_all_registrants_type(ns); 2899 2900 switch (racqa) { 2901 case SPDK_NVME_RESERVE_ACQUIRE: 2902 /* it's not an error for the holder to acquire same reservation type again */ 2903 if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && ns->rtype == rtype) { 2904 /* do nothing */ 2905 update_sgroup = false; 2906 } else if (ns->holder == NULL) { 2907 /* first time to acquire the reservation */ 2908 nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg); 2909 } else { 2910 SPDK_ERRLOG("Invalid rtype or current registrant is not holder\n"); 2911 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2912 update_sgroup = false; 2913 goto exit; 2914 } 2915 break; 2916 case SPDK_NVME_RESERVE_PREEMPT: 2917 /* no reservation holder */ 2918 if (!ns->holder) { 2919 /* unregister with PRKEY */ 2920 nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2921 break; 2922 } 2923 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2924 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2925 &ctrlr->hostid); 2926 2927 /* only 1 reservation holder and reservation key is valid */ 2928 if (!all_regs) { 2929 /* preempt itself */ 2930 if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && 2931 ns->crkey == key.prkey) { 2932 ns->rtype = rtype; 2933 reservation_released = true; 2934 break; 2935 } 2936 2937 if (ns->crkey == key.prkey) { 2938 nvmf_ns_reservation_remove_registrant(ns, ns->holder); 2939 nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg); 2940 reservation_released = true; 2941 } else if (key.prkey != 0) { 2942 nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2943 } else { 2944 /* PRKEY is zero */ 2945 SPDK_ERRLOG("Current PRKEY is zero\n"); 2946 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2947 update_sgroup = false; 2948 goto exit; 2949 } 2950 } else { 2951 /* release all other registrants except for the current one */ 2952 if (key.prkey == 0) { 2953 nvmf_ns_reservation_remove_all_other_registrants(ns, reg); 2954 assert(ns->holder == reg); 2955 } else { 2956 count = nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2957 if (count == 0) { 2958 SPDK_ERRLOG("PRKEY doesn't match any registrant\n"); 2959 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2960 update_sgroup = false; 2961 goto exit; 2962 } 2963 } 2964 } 2965 break; 2966 default: 2967 status = SPDK_NVME_SC_INVALID_FIELD; 2968 update_sgroup = false; 2969 break; 2970 } 2971 2972 exit: 2973 if (update_sgroup && racqa == SPDK_NVME_RESERVE_PREEMPT) { 2974 new_num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, new_hostid_list, 2975 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2976 &ctrlr->hostid); 2977 /* Preempt notification occurs on the unregistered controllers 2978 * other than the controller who issued the command. 2979 */ 2980 num_hostid = nvmf_ns_reservation_get_unregistered_hostid(hostid_list, 2981 num_hostid, 2982 new_hostid_list, 2983 new_num_hostid); 2984 if (num_hostid) { 2985 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2986 hostid_list, 2987 num_hostid, 2988 SPDK_NVME_REGISTRATION_PREEMPTED); 2989 2990 } 2991 /* Reservation released notification occurs on the 2992 * controllers which are the remaining registrants other than 2993 * the controller who issued the command. 2994 */ 2995 if (reservation_released && new_num_hostid) { 2996 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2997 new_hostid_list, 2998 new_num_hostid, 2999 SPDK_NVME_RESERVATION_RELEASED); 3000 3001 } 3002 } 3003 if (update_sgroup && ns->ptpl_activated) { 3004 if (nvmf_ns_update_reservation_info(ns)) { 3005 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 3006 } 3007 } 3008 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3009 req->rsp->nvme_cpl.status.sc = status; 3010 return update_sgroup; 3011 } 3012 3013 static bool 3014 nvmf_ns_reservation_release(struct spdk_nvmf_ns *ns, 3015 struct spdk_nvmf_ctrlr *ctrlr, 3016 struct spdk_nvmf_request *req) 3017 { 3018 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3019 uint8_t rrela, iekey, rtype; 3020 struct spdk_nvmf_registrant *reg; 3021 uint64_t crkey = 0; 3022 uint8_t status = SPDK_NVME_SC_SUCCESS; 3023 bool update_sgroup = true; 3024 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 3025 uint32_t num_hostid = 0; 3026 3027 rrela = cmd->cdw10_bits.resv_release.rrela; 3028 iekey = cmd->cdw10_bits.resv_release.iekey; 3029 rtype = cmd->cdw10_bits.resv_release.rtype; 3030 3031 if (req->iovcnt > 0 && req->length >= sizeof(crkey)) { 3032 struct spdk_iov_xfer ix; 3033 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 3034 spdk_iov_xfer_to_buf(&ix, &crkey, sizeof(crkey)); 3035 } else { 3036 SPDK_ERRLOG("No key provided. Failing request.\n"); 3037 status = SPDK_NVME_SC_INVALID_FIELD; 3038 goto exit; 3039 } 3040 3041 SPDK_DEBUGLOG(nvmf, "RELEASE: RRELA %u, IEKEY %u, RTYPE %u, " 3042 "CRKEY 0x%"PRIx64"\n", rrela, iekey, rtype, crkey); 3043 3044 if (iekey) { 3045 SPDK_ERRLOG("Ignore existing key field set to 1\n"); 3046 status = SPDK_NVME_SC_INVALID_FIELD; 3047 update_sgroup = false; 3048 goto exit; 3049 } 3050 3051 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 3052 if (!reg || reg->rkey != crkey) { 3053 SPDK_ERRLOG("No registrant or current key doesn't match " 3054 "with existing registrant key\n"); 3055 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 3056 update_sgroup = false; 3057 goto exit; 3058 } 3059 3060 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 3061 SPDK_NVMF_MAX_NUM_REGISTRANTS, 3062 &ctrlr->hostid); 3063 3064 switch (rrela) { 3065 case SPDK_NVME_RESERVE_RELEASE: 3066 if (!ns->holder) { 3067 SPDK_DEBUGLOG(nvmf, "RELEASE: no holder\n"); 3068 update_sgroup = false; 3069 goto exit; 3070 } 3071 if (ns->rtype != rtype) { 3072 SPDK_ERRLOG("Type doesn't match\n"); 3073 status = SPDK_NVME_SC_INVALID_FIELD; 3074 update_sgroup = false; 3075 goto exit; 3076 } 3077 if (!nvmf_ns_reservation_registrant_is_holder(ns, reg)) { 3078 /* not the reservation holder, this isn't an error */ 3079 update_sgroup = false; 3080 goto exit; 3081 } 3082 3083 rtype = ns->rtype; 3084 nvmf_ns_reservation_release_reservation(ns); 3085 3086 if (num_hostid && rtype != SPDK_NVME_RESERVE_WRITE_EXCLUSIVE && 3087 rtype != SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 3088 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 3089 hostid_list, 3090 num_hostid, 3091 SPDK_NVME_RESERVATION_RELEASED); 3092 } 3093 break; 3094 case SPDK_NVME_RESERVE_CLEAR: 3095 nvmf_ns_reservation_clear_all_registrants(ns); 3096 if (num_hostid) { 3097 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 3098 hostid_list, 3099 num_hostid, 3100 SPDK_NVME_RESERVATION_PREEMPTED); 3101 } 3102 break; 3103 default: 3104 status = SPDK_NVME_SC_INVALID_FIELD; 3105 update_sgroup = false; 3106 goto exit; 3107 } 3108 3109 exit: 3110 if (update_sgroup && ns->ptpl_activated) { 3111 if (nvmf_ns_update_reservation_info(ns)) { 3112 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 3113 } 3114 } 3115 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3116 req->rsp->nvme_cpl.status.sc = status; 3117 return update_sgroup; 3118 } 3119 3120 static void 3121 nvmf_ns_reservation_report(struct spdk_nvmf_ns *ns, 3122 struct spdk_nvmf_ctrlr *ctrlr, 3123 struct spdk_nvmf_request *req) 3124 { 3125 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3126 struct spdk_nvmf_registrant *reg, *tmp; 3127 struct spdk_nvme_reservation_status_extended_data status_data = { 0 }; 3128 struct spdk_iov_xfer ix; 3129 uint32_t transfer_len; 3130 uint32_t regctl = 0; 3131 uint8_t status = SPDK_NVME_SC_SUCCESS; 3132 3133 if (req->iovcnt == 0) { 3134 SPDK_ERRLOG("No data transfer specified for request. " 3135 " Unable to transfer back response.\n"); 3136 status = SPDK_NVME_SC_INVALID_FIELD; 3137 goto exit; 3138 } 3139 3140 if (!cmd->cdw11_bits.resv_report.eds) { 3141 SPDK_ERRLOG("NVMeoF uses extended controller data structure, " 3142 "please set EDS bit in cdw11 and try again\n"); 3143 status = SPDK_NVME_SC_HOSTID_INCONSISTENT_FORMAT; 3144 goto exit; 3145 } 3146 3147 /* Number of Dwords of the Reservation Status data structure to transfer */ 3148 transfer_len = (cmd->cdw10 + 1) * sizeof(uint32_t); 3149 3150 if (transfer_len < sizeof(struct spdk_nvme_reservation_status_extended_data)) { 3151 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 3152 goto exit; 3153 } 3154 3155 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 3156 3157 status_data.data.gen = ns->gen; 3158 status_data.data.rtype = ns->rtype; 3159 status_data.data.ptpls = ns->ptpl_activated; 3160 3161 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 3162 regctl++; 3163 } 3164 3165 /* 3166 * We report the number of registrants as per the spec here, even if 3167 * the iov isn't big enough to contain them all. In that case, the 3168 * spdk_iov_xfer_from_buf() won't actually copy any of the remaining 3169 * data; as it keeps track of the iov cursor itself, it's simplest to 3170 * just walk the entire list anyway. 3171 */ 3172 status_data.data.regctl = regctl; 3173 3174 spdk_iov_xfer_from_buf(&ix, &status_data, sizeof(status_data)); 3175 3176 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 3177 struct spdk_nvme_registered_ctrlr_extended_data ctrlr_data = { 0 }; 3178 3179 /* Set to 0xffffh for dynamic controller */ 3180 ctrlr_data.cntlid = 0xffff; 3181 ctrlr_data.rcsts.status = (ns->holder == reg) ? true : false; 3182 ctrlr_data.rkey = reg->rkey; 3183 spdk_uuid_copy((struct spdk_uuid *)ctrlr_data.hostid, ®->hostid); 3184 3185 spdk_iov_xfer_from_buf(&ix, &ctrlr_data, sizeof(ctrlr_data)); 3186 } 3187 3188 exit: 3189 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3190 req->rsp->nvme_cpl.status.sc = status; 3191 return; 3192 } 3193 3194 static void 3195 nvmf_ns_reservation_complete(void *ctx) 3196 { 3197 struct spdk_nvmf_request *req = ctx; 3198 3199 spdk_nvmf_request_complete(req); 3200 } 3201 3202 static void 3203 _nvmf_ns_reservation_update_done(struct spdk_nvmf_subsystem *subsystem, 3204 void *cb_arg, int status) 3205 { 3206 struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)cb_arg; 3207 struct spdk_nvmf_poll_group *group = req->qpair->group; 3208 3209 spdk_thread_send_msg(group->thread, nvmf_ns_reservation_complete, req); 3210 } 3211 3212 void 3213 nvmf_ns_reservation_request(void *ctx) 3214 { 3215 struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)ctx; 3216 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3217 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3218 struct subsystem_update_ns_ctx *update_ctx; 3219 uint32_t nsid; 3220 struct spdk_nvmf_ns *ns; 3221 bool update_sgroup = false; 3222 3223 nsid = cmd->nsid; 3224 ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid); 3225 assert(ns != NULL); 3226 3227 switch (cmd->opc) { 3228 case SPDK_NVME_OPC_RESERVATION_REGISTER: 3229 update_sgroup = nvmf_ns_reservation_register(ns, ctrlr, req); 3230 break; 3231 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 3232 update_sgroup = nvmf_ns_reservation_acquire(ns, ctrlr, req); 3233 break; 3234 case SPDK_NVME_OPC_RESERVATION_RELEASE: 3235 update_sgroup = nvmf_ns_reservation_release(ns, ctrlr, req); 3236 break; 3237 case SPDK_NVME_OPC_RESERVATION_REPORT: 3238 nvmf_ns_reservation_report(ns, ctrlr, req); 3239 break; 3240 default: 3241 break; 3242 } 3243 3244 /* update reservation information to subsystem's poll group */ 3245 if (update_sgroup) { 3246 update_ctx = calloc(1, sizeof(*update_ctx)); 3247 if (update_ctx == NULL) { 3248 SPDK_ERRLOG("Can't alloc subsystem poll group update context\n"); 3249 goto update_done; 3250 } 3251 update_ctx->subsystem = ctrlr->subsys; 3252 update_ctx->cb_fn = _nvmf_ns_reservation_update_done; 3253 update_ctx->cb_arg = req; 3254 3255 nvmf_subsystem_update_ns(ctrlr->subsys, subsystem_update_ns_done, update_ctx); 3256 return; 3257 } 3258 3259 update_done: 3260 _nvmf_ns_reservation_update_done(ctrlr->subsys, (void *)req, 0); 3261 } 3262 3263 int 3264 spdk_nvmf_subsystem_set_ana_reporting(struct spdk_nvmf_subsystem *subsystem, 3265 bool ana_reporting) 3266 { 3267 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 3268 return -EAGAIN; 3269 } 3270 3271 subsystem->flags.ana_reporting = ana_reporting; 3272 3273 return 0; 3274 } 3275 3276 bool 3277 nvmf_subsystem_get_ana_reporting(struct spdk_nvmf_subsystem *subsystem) 3278 { 3279 return subsystem->flags.ana_reporting; 3280 } 3281 3282 struct subsystem_listener_update_ctx { 3283 struct spdk_nvmf_subsystem_listener *listener; 3284 3285 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 3286 void *cb_arg; 3287 }; 3288 3289 static void 3290 subsystem_listener_update_done(struct spdk_io_channel_iter *i, int status) 3291 { 3292 struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 3293 3294 if (ctx->cb_fn) { 3295 ctx->cb_fn(ctx->cb_arg, status); 3296 } 3297 free(ctx); 3298 } 3299 3300 static void 3301 subsystem_listener_update_on_pg(struct spdk_io_channel_iter *i) 3302 { 3303 struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 3304 struct spdk_nvmf_subsystem_listener *listener; 3305 struct spdk_nvmf_poll_group *group; 3306 struct spdk_nvmf_ctrlr *ctrlr; 3307 3308 listener = ctx->listener; 3309 group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i)); 3310 3311 TAILQ_FOREACH(ctrlr, &listener->subsystem->ctrlrs, link) { 3312 if (ctrlr->thread != spdk_get_thread()) { 3313 continue; 3314 } 3315 3316 if (ctrlr->admin_qpair && ctrlr->admin_qpair->group == group && ctrlr->listener == listener) { 3317 nvmf_ctrlr_async_event_ana_change_notice(ctrlr); 3318 } 3319 } 3320 3321 spdk_for_each_channel_continue(i, 0); 3322 } 3323 3324 void 3325 nvmf_subsystem_set_ana_state(struct spdk_nvmf_subsystem *subsystem, 3326 const struct spdk_nvme_transport_id *trid, 3327 enum spdk_nvme_ana_state ana_state, uint32_t anagrpid, 3328 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, void *cb_arg) 3329 { 3330 struct spdk_nvmf_subsystem_listener *listener; 3331 struct subsystem_listener_update_ctx *ctx; 3332 uint32_t i; 3333 3334 assert(cb_fn != NULL); 3335 assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 3336 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED); 3337 3338 if (!subsystem->flags.ana_reporting) { 3339 SPDK_ERRLOG("ANA reporting is disabled\n"); 3340 cb_fn(cb_arg, -EINVAL); 3341 return; 3342 } 3343 3344 /* ANA Change state is not used, ANA Persistent Loss state 3345 * is not supported yet. 3346 */ 3347 if (!(ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE || 3348 ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE || 3349 ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE)) { 3350 SPDK_ERRLOG("ANA state %d is not supported\n", ana_state); 3351 cb_fn(cb_arg, -ENOTSUP); 3352 return; 3353 } 3354 3355 if (anagrpid > subsystem->max_nsid) { 3356 SPDK_ERRLOG("ANA group ID %" PRIu32 " is more than maximum\n", anagrpid); 3357 cb_fn(cb_arg, -EINVAL); 3358 return; 3359 } 3360 3361 listener = nvmf_subsystem_find_listener(subsystem, trid); 3362 if (!listener) { 3363 SPDK_ERRLOG("Unable to find listener.\n"); 3364 cb_fn(cb_arg, -EINVAL); 3365 return; 3366 } 3367 3368 if (anagrpid != 0 && listener->ana_state[anagrpid - 1] == ana_state) { 3369 cb_fn(cb_arg, 0); 3370 return; 3371 } 3372 3373 ctx = calloc(1, sizeof(*ctx)); 3374 if (!ctx) { 3375 SPDK_ERRLOG("Unable to allocate context\n"); 3376 cb_fn(cb_arg, -ENOMEM); 3377 return; 3378 } 3379 3380 for (i = 1; i <= subsystem->max_nsid; i++) { 3381 if (anagrpid == 0 || i == anagrpid) { 3382 listener->ana_state[i - 1] = ana_state; 3383 } 3384 } 3385 listener->ana_state_change_count++; 3386 3387 ctx->listener = listener; 3388 ctx->cb_fn = cb_fn; 3389 ctx->cb_arg = cb_arg; 3390 3391 spdk_for_each_channel(subsystem->tgt, 3392 subsystem_listener_update_on_pg, 3393 ctx, 3394 subsystem_listener_update_done); 3395 } 3396 3397 bool 3398 spdk_nvmf_subsystem_is_discovery(struct spdk_nvmf_subsystem *subsystem) 3399 { 3400 return subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT || 3401 subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY; 3402 } 3403