1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 36 #include "nvmf_internal.h" 37 #include "transport.h" 38 39 #include "spdk/likely.h" 40 #include "spdk/string.h" 41 #include "spdk/trace.h" 42 #include "spdk/nvmf_spec.h" 43 #include "spdk/uuid.h" 44 #include "spdk/json.h" 45 #include "spdk/file.h" 46 47 #include "spdk/bdev_module.h" 48 #include "spdk/log.h" 49 #include "spdk_internal/utf.h" 50 51 #define MODEL_NUMBER_DEFAULT "SPDK bdev Controller" 52 #define NVMF_SUBSYSTEM_DEFAULT_NAMESPACES 32 53 54 /* 55 * States for parsing valid domains in NQNs according to RFC 1034 56 */ 57 enum spdk_nvmf_nqn_domain_states { 58 /* First character of a domain must be a letter */ 59 SPDK_NVMF_DOMAIN_ACCEPT_LETTER = 0, 60 61 /* Subsequent characters can be any of letter, digit, or hyphen */ 62 SPDK_NVMF_DOMAIN_ACCEPT_LDH = 1, 63 64 /* A domain label must end with either a letter or digit */ 65 SPDK_NVMF_DOMAIN_ACCEPT_ANY = 2 66 }; 67 68 /* Returns true if is a valid ASCII string as defined by the NVMe spec */ 69 static bool 70 nvmf_valid_ascii_string(const void *buf, size_t size) 71 { 72 const uint8_t *str = buf; 73 size_t i; 74 75 for (i = 0; i < size; i++) { 76 if (str[i] < 0x20 || str[i] > 0x7E) { 77 return false; 78 } 79 } 80 81 return true; 82 } 83 84 static bool 85 nvmf_valid_nqn(const char *nqn) 86 { 87 size_t len; 88 struct spdk_uuid uuid_value; 89 uint32_t i; 90 int bytes_consumed; 91 uint32_t domain_label_length; 92 char *reverse_domain_end; 93 uint32_t reverse_domain_end_index; 94 enum spdk_nvmf_nqn_domain_states domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LETTER; 95 96 /* Check for length requirements */ 97 len = strlen(nqn); 98 if (len > SPDK_NVMF_NQN_MAX_LEN) { 99 SPDK_ERRLOG("Invalid NQN \"%s\": length %zu > max %d\n", nqn, len, SPDK_NVMF_NQN_MAX_LEN); 100 return false; 101 } 102 103 /* The nqn must be at least as long as SPDK_NVMF_NQN_MIN_LEN to contain the necessary prefix. */ 104 if (len < SPDK_NVMF_NQN_MIN_LEN) { 105 SPDK_ERRLOG("Invalid NQN \"%s\": length %zu < min %d\n", nqn, len, SPDK_NVMF_NQN_MIN_LEN); 106 return false; 107 } 108 109 /* Check for discovery controller nqn */ 110 if (!strcmp(nqn, SPDK_NVMF_DISCOVERY_NQN)) { 111 return true; 112 } 113 114 /* Check for equality with the generic nqn structure of the form "nqn.2014-08.org.nvmexpress:uuid:11111111-2222-3333-4444-555555555555" */ 115 if (!strncmp(nqn, SPDK_NVMF_NQN_UUID_PRE, SPDK_NVMF_NQN_UUID_PRE_LEN)) { 116 if (len != SPDK_NVMF_NQN_UUID_PRE_LEN + SPDK_NVMF_UUID_STRING_LEN) { 117 SPDK_ERRLOG("Invalid NQN \"%s\": uuid is not the correct length\n", nqn); 118 return false; 119 } 120 121 if (spdk_uuid_parse(&uuid_value, &nqn[SPDK_NVMF_NQN_UUID_PRE_LEN])) { 122 SPDK_ERRLOG("Invalid NQN \"%s\": uuid is not formatted correctly\n", nqn); 123 return false; 124 } 125 return true; 126 } 127 128 /* If the nqn does not match the uuid structure, the next several checks validate the form "nqn.yyyy-mm.reverse.domain:user-string" */ 129 130 if (strncmp(nqn, "nqn.", 4) != 0) { 131 SPDK_ERRLOG("Invalid NQN \"%s\": NQN must begin with \"nqn.\".\n", nqn); 132 return false; 133 } 134 135 /* Check for yyyy-mm. */ 136 if (!(isdigit(nqn[4]) && isdigit(nqn[5]) && isdigit(nqn[6]) && isdigit(nqn[7]) && 137 nqn[8] == '-' && isdigit(nqn[9]) && isdigit(nqn[10]) && nqn[11] == '.')) { 138 SPDK_ERRLOG("Invalid date code in NQN \"%s\"\n", nqn); 139 return false; 140 } 141 142 reverse_domain_end = strchr(nqn, ':'); 143 if (reverse_domain_end != NULL && (reverse_domain_end_index = reverse_domain_end - nqn) < len - 1) { 144 } else { 145 SPDK_ERRLOG("Invalid NQN \"%s\". NQN must contain user specified name with a ':' as a prefix.\n", 146 nqn); 147 return false; 148 } 149 150 /* Check for valid reverse domain */ 151 domain_label_length = 0; 152 for (i = 12; i < reverse_domain_end_index; i++) { 153 if (domain_label_length > SPDK_DOMAIN_LABEL_MAX_LEN) { 154 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". At least one Label is too long.\n", nqn); 155 return false; 156 } 157 158 switch (domain_state) { 159 160 case SPDK_NVMF_DOMAIN_ACCEPT_LETTER: { 161 if (isalpha(nqn[i])) { 162 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY; 163 domain_label_length++; 164 break; 165 } else { 166 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must start with a letter.\n", nqn); 167 return false; 168 } 169 } 170 171 case SPDK_NVMF_DOMAIN_ACCEPT_LDH: { 172 if (isalpha(nqn[i]) || isdigit(nqn[i])) { 173 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY; 174 domain_label_length++; 175 break; 176 } else if (nqn[i] == '-') { 177 if (i == reverse_domain_end_index - 1) { 178 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n", 179 nqn); 180 return false; 181 } 182 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LDH; 183 domain_label_length++; 184 break; 185 } else if (nqn[i] == '.') { 186 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n", 187 nqn); 188 return false; 189 } else { 190 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only [a-z,A-Z,0-9,'-','.'].\n", 191 nqn); 192 return false; 193 } 194 } 195 196 case SPDK_NVMF_DOMAIN_ACCEPT_ANY: { 197 if (isalpha(nqn[i]) || isdigit(nqn[i])) { 198 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY; 199 domain_label_length++; 200 break; 201 } else if (nqn[i] == '-') { 202 if (i == reverse_domain_end_index - 1) { 203 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n", 204 nqn); 205 return false; 206 } 207 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LDH; 208 domain_label_length++; 209 break; 210 } else if (nqn[i] == '.') { 211 domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LETTER; 212 domain_label_length = 0; 213 break; 214 } else { 215 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only [a-z,A-Z,0-9,'-','.'].\n", 216 nqn); 217 return false; 218 } 219 } 220 } 221 } 222 223 i = reverse_domain_end_index + 1; 224 while (i < len) { 225 bytes_consumed = utf8_valid(&nqn[i], &nqn[len]); 226 if (bytes_consumed <= 0) { 227 SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only valid utf-8.\n", nqn); 228 return false; 229 } 230 231 i += bytes_consumed; 232 } 233 return true; 234 } 235 236 static void subsystem_state_change_on_pg(struct spdk_io_channel_iter *i); 237 238 struct spdk_nvmf_subsystem * 239 spdk_nvmf_subsystem_create(struct spdk_nvmf_tgt *tgt, 240 const char *nqn, 241 enum spdk_nvmf_subtype type, 242 uint32_t num_ns) 243 { 244 struct spdk_nvmf_subsystem *subsystem; 245 uint32_t sid; 246 247 if (spdk_nvmf_tgt_find_subsystem(tgt, nqn)) { 248 SPDK_ERRLOG("Subsystem NQN '%s' already exists\n", nqn); 249 return NULL; 250 } 251 252 if (!nvmf_valid_nqn(nqn)) { 253 return NULL; 254 } 255 256 if (type == SPDK_NVMF_SUBTYPE_DISCOVERY) { 257 if (num_ns != 0) { 258 SPDK_ERRLOG("Discovery subsystem cannot have namespaces.\n"); 259 return NULL; 260 } 261 } else if (num_ns == 0) { 262 num_ns = NVMF_SUBSYSTEM_DEFAULT_NAMESPACES; 263 } 264 265 /* Find a free subsystem id (sid) */ 266 for (sid = 0; sid < tgt->max_subsystems; sid++) { 267 if (tgt->subsystems[sid] == NULL) { 268 break; 269 } 270 } 271 if (sid >= tgt->max_subsystems) { 272 return NULL; 273 } 274 275 subsystem = calloc(1, sizeof(struct spdk_nvmf_subsystem)); 276 if (subsystem == NULL) { 277 return NULL; 278 } 279 280 subsystem->thread = spdk_get_thread(); 281 subsystem->state = SPDK_NVMF_SUBSYSTEM_INACTIVE; 282 subsystem->tgt = tgt; 283 subsystem->id = sid; 284 subsystem->subtype = type; 285 subsystem->max_nsid = num_ns; 286 subsystem->next_cntlid = 0; 287 snprintf(subsystem->subnqn, sizeof(subsystem->subnqn), "%s", nqn); 288 pthread_mutex_init(&subsystem->mutex, NULL); 289 TAILQ_INIT(&subsystem->listeners); 290 TAILQ_INIT(&subsystem->hosts); 291 TAILQ_INIT(&subsystem->ctrlrs); 292 293 if (num_ns != 0) { 294 subsystem->ns = calloc(num_ns, sizeof(struct spdk_nvmf_ns *)); 295 if (subsystem->ns == NULL) { 296 SPDK_ERRLOG("Namespace memory allocation failed\n"); 297 pthread_mutex_destroy(&subsystem->mutex); 298 free(subsystem); 299 return NULL; 300 } 301 } 302 303 memset(subsystem->sn, '0', sizeof(subsystem->sn) - 1); 304 subsystem->sn[sizeof(subsystem->sn) - 1] = '\0'; 305 306 snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", 307 MODEL_NUMBER_DEFAULT); 308 309 tgt->subsystems[sid] = subsystem; 310 nvmf_update_discovery_log(tgt, NULL); 311 312 return subsystem; 313 } 314 315 /* Must hold subsystem->mutex while calling this function */ 316 static void 317 nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_host *host) 318 { 319 TAILQ_REMOVE(&subsystem->hosts, host, link); 320 free(host); 321 } 322 323 static void 324 _nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem, 325 struct spdk_nvmf_subsystem_listener *listener, 326 bool stop) 327 { 328 struct spdk_nvmf_transport *transport; 329 330 if (stop) { 331 transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, listener->trid->trstring); 332 if (transport != NULL) { 333 spdk_nvmf_transport_stop_listen(transport, listener->trid); 334 } 335 } 336 337 TAILQ_REMOVE(&subsystem->listeners, listener, link); 338 free(listener); 339 } 340 341 void 342 spdk_nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem) 343 { 344 struct spdk_nvmf_host *host, *host_tmp; 345 struct spdk_nvmf_ctrlr *ctrlr, *ctrlr_tmp; 346 struct spdk_nvmf_ns *ns; 347 348 if (!subsystem) { 349 return; 350 } 351 352 assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE); 353 354 SPDK_DEBUGLOG(nvmf, "subsystem is %p\n", subsystem); 355 356 nvmf_subsystem_remove_all_listeners(subsystem, false); 357 358 pthread_mutex_lock(&subsystem->mutex); 359 360 TAILQ_FOREACH_SAFE(host, &subsystem->hosts, link, host_tmp) { 361 nvmf_subsystem_remove_host(subsystem, host); 362 } 363 364 pthread_mutex_unlock(&subsystem->mutex); 365 366 TAILQ_FOREACH_SAFE(ctrlr, &subsystem->ctrlrs, link, ctrlr_tmp) { 367 nvmf_ctrlr_destruct(ctrlr); 368 } 369 370 ns = spdk_nvmf_subsystem_get_first_ns(subsystem); 371 while (ns != NULL) { 372 struct spdk_nvmf_ns *next_ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns); 373 374 spdk_nvmf_subsystem_remove_ns(subsystem, ns->opts.nsid); 375 ns = next_ns; 376 } 377 378 free(subsystem->ns); 379 380 subsystem->tgt->subsystems[subsystem->id] = NULL; 381 nvmf_update_discovery_log(subsystem->tgt, NULL); 382 383 pthread_mutex_destroy(&subsystem->mutex); 384 385 free(subsystem); 386 } 387 388 389 /* we have to use the typedef in the function declaration to appease astyle. */ 390 typedef enum spdk_nvmf_subsystem_state spdk_nvmf_subsystem_state_t; 391 392 static spdk_nvmf_subsystem_state_t 393 nvmf_subsystem_get_intermediate_state(enum spdk_nvmf_subsystem_state current_state, 394 enum spdk_nvmf_subsystem_state requested_state) 395 { 396 switch (requested_state) { 397 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 398 return SPDK_NVMF_SUBSYSTEM_DEACTIVATING; 399 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 400 if (current_state == SPDK_NVMF_SUBSYSTEM_PAUSED) { 401 return SPDK_NVMF_SUBSYSTEM_RESUMING; 402 } else { 403 return SPDK_NVMF_SUBSYSTEM_ACTIVATING; 404 } 405 case SPDK_NVMF_SUBSYSTEM_PAUSED: 406 return SPDK_NVMF_SUBSYSTEM_PAUSING; 407 default: 408 assert(false); 409 return SPDK_NVMF_SUBSYSTEM_NUM_STATES; 410 } 411 } 412 413 static int 414 nvmf_subsystem_set_state(struct spdk_nvmf_subsystem *subsystem, 415 enum spdk_nvmf_subsystem_state state) 416 { 417 enum spdk_nvmf_subsystem_state actual_old_state, expected_old_state; 418 bool exchanged; 419 420 switch (state) { 421 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 422 expected_old_state = SPDK_NVMF_SUBSYSTEM_DEACTIVATING; 423 break; 424 case SPDK_NVMF_SUBSYSTEM_ACTIVATING: 425 expected_old_state = SPDK_NVMF_SUBSYSTEM_INACTIVE; 426 break; 427 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 428 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING; 429 break; 430 case SPDK_NVMF_SUBSYSTEM_PAUSING: 431 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE; 432 break; 433 case SPDK_NVMF_SUBSYSTEM_PAUSED: 434 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSING; 435 break; 436 case SPDK_NVMF_SUBSYSTEM_RESUMING: 437 expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED; 438 break; 439 case SPDK_NVMF_SUBSYSTEM_DEACTIVATING: 440 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE; 441 break; 442 default: 443 assert(false); 444 return -1; 445 } 446 447 actual_old_state = expected_old_state; 448 exchanged = __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false, 449 __ATOMIC_RELAXED, __ATOMIC_RELAXED); 450 if (spdk_unlikely(exchanged == false)) { 451 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING && 452 state == SPDK_NVMF_SUBSYSTEM_ACTIVE) { 453 expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING; 454 } 455 /* This is for the case when activating the subsystem fails. */ 456 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_ACTIVATING && 457 state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) { 458 expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING; 459 } 460 /* This is for the case when resuming the subsystem fails. */ 461 if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING && 462 state == SPDK_NVMF_SUBSYSTEM_PAUSING) { 463 expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING; 464 } 465 actual_old_state = expected_old_state; 466 __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false, 467 __ATOMIC_RELAXED, __ATOMIC_RELAXED); 468 } 469 assert(actual_old_state == expected_old_state); 470 return actual_old_state - expected_old_state; 471 } 472 473 struct subsystem_state_change_ctx { 474 struct spdk_nvmf_subsystem *subsystem; 475 476 enum spdk_nvmf_subsystem_state original_state; 477 478 enum spdk_nvmf_subsystem_state requested_state; 479 480 spdk_nvmf_subsystem_state_change_done cb_fn; 481 void *cb_arg; 482 }; 483 484 static void 485 subsystem_state_change_revert_done(struct spdk_io_channel_iter *i, int status) 486 { 487 struct subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 488 489 /* Nothing to be done here if the state setting fails, we are just screwed. */ 490 if (nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state)) { 491 SPDK_ERRLOG("Unable to revert the subsystem state after operation failure.\n"); 492 } 493 494 ctx->subsystem->changing_state = false; 495 if (ctx->cb_fn) { 496 /* return a failure here. This function only exists in an error path. */ 497 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, -1); 498 } 499 free(ctx); 500 } 501 502 static void 503 subsystem_state_change_done(struct spdk_io_channel_iter *i, int status) 504 { 505 struct subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 506 enum spdk_nvmf_subsystem_state intermediate_state; 507 508 if (status == 0) { 509 status = nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state); 510 if (status) { 511 status = -1; 512 } 513 } 514 515 if (status) { 516 intermediate_state = nvmf_subsystem_get_intermediate_state(ctx->requested_state, 517 ctx->original_state); 518 assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES); 519 520 if (nvmf_subsystem_set_state(ctx->subsystem, intermediate_state)) { 521 goto out; 522 } 523 ctx->requested_state = ctx->original_state; 524 spdk_for_each_channel(ctx->subsystem->tgt, 525 subsystem_state_change_on_pg, 526 ctx, 527 subsystem_state_change_revert_done); 528 return; 529 } 530 531 out: 532 ctx->subsystem->changing_state = false; 533 if (ctx->cb_fn) { 534 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status); 535 } 536 free(ctx); 537 } 538 539 static void 540 subsystem_state_change_continue(void *ctx, int status) 541 { 542 struct spdk_io_channel_iter *i = ctx; 543 spdk_for_each_channel_continue(i, status); 544 } 545 546 static void 547 subsystem_state_change_on_pg(struct spdk_io_channel_iter *i) 548 { 549 struct subsystem_state_change_ctx *ctx; 550 struct spdk_io_channel *ch; 551 struct spdk_nvmf_poll_group *group; 552 553 ctx = spdk_io_channel_iter_get_ctx(i); 554 ch = spdk_io_channel_iter_get_channel(i); 555 group = spdk_io_channel_get_ctx(ch); 556 557 switch (ctx->requested_state) { 558 case SPDK_NVMF_SUBSYSTEM_INACTIVE: 559 nvmf_poll_group_remove_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 560 break; 561 case SPDK_NVMF_SUBSYSTEM_ACTIVE: 562 if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_ACTIVATING) { 563 nvmf_poll_group_add_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 564 } else if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_RESUMING) { 565 nvmf_poll_group_resume_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 566 } 567 break; 568 case SPDK_NVMF_SUBSYSTEM_PAUSED: 569 nvmf_poll_group_pause_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i); 570 break; 571 default: 572 assert(false); 573 break; 574 } 575 } 576 577 static int 578 nvmf_subsystem_state_change(struct spdk_nvmf_subsystem *subsystem, 579 enum spdk_nvmf_subsystem_state requested_state, 580 spdk_nvmf_subsystem_state_change_done cb_fn, 581 void *cb_arg) 582 { 583 struct subsystem_state_change_ctx *ctx; 584 enum spdk_nvmf_subsystem_state intermediate_state; 585 int rc; 586 587 if (__sync_val_compare_and_swap(&subsystem->changing_state, false, true)) { 588 return -EBUSY; 589 } 590 591 /* If we are already in the requested state, just call the callback immediately. */ 592 if (subsystem->state == requested_state) { 593 subsystem->changing_state = false; 594 if (cb_fn) { 595 cb_fn(subsystem, cb_arg, 0); 596 } 597 return 0; 598 } 599 600 intermediate_state = nvmf_subsystem_get_intermediate_state(subsystem->state, requested_state); 601 assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES); 602 603 ctx = calloc(1, sizeof(*ctx)); 604 if (!ctx) { 605 subsystem->changing_state = false; 606 return -ENOMEM; 607 } 608 609 ctx->original_state = subsystem->state; 610 rc = nvmf_subsystem_set_state(subsystem, intermediate_state); 611 if (rc) { 612 free(ctx); 613 subsystem->changing_state = false; 614 return rc; 615 } 616 617 ctx->subsystem = subsystem; 618 ctx->requested_state = requested_state; 619 ctx->cb_fn = cb_fn; 620 ctx->cb_arg = cb_arg; 621 622 spdk_for_each_channel(subsystem->tgt, 623 subsystem_state_change_on_pg, 624 ctx, 625 subsystem_state_change_done); 626 627 return 0; 628 } 629 630 int 631 spdk_nvmf_subsystem_start(struct spdk_nvmf_subsystem *subsystem, 632 spdk_nvmf_subsystem_state_change_done cb_fn, 633 void *cb_arg) 634 { 635 return nvmf_subsystem_state_change(subsystem, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg); 636 } 637 638 int 639 spdk_nvmf_subsystem_stop(struct spdk_nvmf_subsystem *subsystem, 640 spdk_nvmf_subsystem_state_change_done cb_fn, 641 void *cb_arg) 642 { 643 return nvmf_subsystem_state_change(subsystem, SPDK_NVMF_SUBSYSTEM_INACTIVE, cb_fn, cb_arg); 644 } 645 646 int 647 spdk_nvmf_subsystem_pause(struct spdk_nvmf_subsystem *subsystem, 648 spdk_nvmf_subsystem_state_change_done cb_fn, 649 void *cb_arg) 650 { 651 return nvmf_subsystem_state_change(subsystem, SPDK_NVMF_SUBSYSTEM_PAUSED, cb_fn, cb_arg); 652 } 653 654 int 655 spdk_nvmf_subsystem_resume(struct spdk_nvmf_subsystem *subsystem, 656 spdk_nvmf_subsystem_state_change_done cb_fn, 657 void *cb_arg) 658 { 659 return nvmf_subsystem_state_change(subsystem, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg); 660 } 661 662 struct spdk_nvmf_subsystem * 663 spdk_nvmf_subsystem_get_first(struct spdk_nvmf_tgt *tgt) 664 { 665 struct spdk_nvmf_subsystem *subsystem; 666 uint32_t sid; 667 668 for (sid = 0; sid < tgt->max_subsystems; sid++) { 669 subsystem = tgt->subsystems[sid]; 670 if (subsystem) { 671 return subsystem; 672 } 673 } 674 675 return NULL; 676 } 677 678 struct spdk_nvmf_subsystem * 679 spdk_nvmf_subsystem_get_next(struct spdk_nvmf_subsystem *subsystem) 680 { 681 uint32_t sid; 682 struct spdk_nvmf_tgt *tgt; 683 684 if (!subsystem) { 685 return NULL; 686 } 687 688 tgt = subsystem->tgt; 689 690 for (sid = subsystem->id + 1; sid < tgt->max_subsystems; sid++) { 691 subsystem = tgt->subsystems[sid]; 692 if (subsystem) { 693 return subsystem; 694 } 695 } 696 697 return NULL; 698 } 699 700 /* Must hold subsystem->mutex while calling this function */ 701 static struct spdk_nvmf_host * 702 nvmf_subsystem_find_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 703 { 704 struct spdk_nvmf_host *host = NULL; 705 706 TAILQ_FOREACH(host, &subsystem->hosts, link) { 707 if (strcmp(hostnqn, host->nqn) == 0) { 708 return host; 709 } 710 } 711 712 return NULL; 713 } 714 715 int 716 spdk_nvmf_subsystem_add_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 717 { 718 struct spdk_nvmf_host *host; 719 720 if (!nvmf_valid_nqn(hostnqn)) { 721 return -EINVAL; 722 } 723 724 pthread_mutex_lock(&subsystem->mutex); 725 726 if (nvmf_subsystem_find_host(subsystem, hostnqn)) { 727 /* This subsystem already allows the specified host. */ 728 pthread_mutex_unlock(&subsystem->mutex); 729 return 0; 730 } 731 732 host = calloc(1, sizeof(*host)); 733 if (!host) { 734 pthread_mutex_unlock(&subsystem->mutex); 735 return -ENOMEM; 736 } 737 738 snprintf(host->nqn, sizeof(host->nqn), "%s", hostnqn); 739 740 TAILQ_INSERT_HEAD(&subsystem->hosts, host, link); 741 742 nvmf_update_discovery_log(subsystem->tgt, hostnqn); 743 744 pthread_mutex_unlock(&subsystem->mutex); 745 746 return 0; 747 } 748 749 int 750 spdk_nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 751 { 752 struct spdk_nvmf_host *host; 753 754 pthread_mutex_lock(&subsystem->mutex); 755 756 host = nvmf_subsystem_find_host(subsystem, hostnqn); 757 if (host == NULL) { 758 pthread_mutex_unlock(&subsystem->mutex); 759 return -ENOENT; 760 } 761 762 nvmf_subsystem_remove_host(subsystem, host); 763 pthread_mutex_unlock(&subsystem->mutex); 764 765 return 0; 766 } 767 768 struct nvmf_subsystem_disconnect_host_ctx { 769 struct spdk_nvmf_subsystem *subsystem; 770 char *hostnqn; 771 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 772 void *cb_arg; 773 }; 774 775 static void 776 nvmf_subsystem_disconnect_host_fini(struct spdk_io_channel_iter *i, int status) 777 { 778 struct nvmf_subsystem_disconnect_host_ctx *ctx; 779 780 ctx = spdk_io_channel_iter_get_ctx(i); 781 782 if (ctx->cb_fn) { 783 ctx->cb_fn(ctx->cb_arg, status); 784 } 785 free(ctx->hostnqn); 786 free(ctx); 787 } 788 789 static void 790 nvmf_subsystem_disconnect_qpairs_by_host(struct spdk_io_channel_iter *i) 791 { 792 struct nvmf_subsystem_disconnect_host_ctx *ctx; 793 struct spdk_nvmf_poll_group *group; 794 struct spdk_io_channel *ch; 795 struct spdk_nvmf_qpair *qpair, *tmp_qpair; 796 struct spdk_nvmf_ctrlr *ctrlr; 797 798 ctx = spdk_io_channel_iter_get_ctx(i); 799 ch = spdk_io_channel_iter_get_channel(i); 800 group = spdk_io_channel_get_ctx(ch); 801 802 TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, tmp_qpair) { 803 ctrlr = qpair->ctrlr; 804 805 if (ctrlr == NULL || ctrlr->subsys != ctx->subsystem) { 806 continue; 807 } 808 809 if (strncmp(ctrlr->hostnqn, ctx->hostnqn, sizeof(ctrlr->hostnqn)) == 0) { 810 /* Right now this does not wait for the queue pairs to actually disconnect. */ 811 spdk_nvmf_qpair_disconnect(qpair, NULL, NULL); 812 } 813 } 814 spdk_for_each_channel_continue(i, 0); 815 } 816 817 int 818 spdk_nvmf_subsystem_disconnect_host(struct spdk_nvmf_subsystem *subsystem, 819 const char *hostnqn, 820 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 821 void *cb_arg) 822 { 823 struct nvmf_subsystem_disconnect_host_ctx *ctx; 824 825 ctx = calloc(1, sizeof(struct nvmf_subsystem_disconnect_host_ctx)); 826 if (ctx == NULL) { 827 return -ENOMEM; 828 } 829 830 ctx->subsystem = subsystem; 831 ctx->hostnqn = strdup(hostnqn); 832 ctx->cb_fn = cb_fn; 833 ctx->cb_arg = cb_arg; 834 835 spdk_for_each_channel(subsystem->tgt, nvmf_subsystem_disconnect_qpairs_by_host, ctx, 836 nvmf_subsystem_disconnect_host_fini); 837 838 return 0; 839 } 840 841 int 842 spdk_nvmf_subsystem_set_allow_any_host(struct spdk_nvmf_subsystem *subsystem, bool allow_any_host) 843 { 844 pthread_mutex_lock(&subsystem->mutex); 845 subsystem->flags.allow_any_host = allow_any_host; 846 nvmf_update_discovery_log(subsystem->tgt, NULL); 847 pthread_mutex_unlock(&subsystem->mutex); 848 849 return 0; 850 } 851 852 bool 853 spdk_nvmf_subsystem_get_allow_any_host(const struct spdk_nvmf_subsystem *subsystem) 854 { 855 bool allow_any_host; 856 struct spdk_nvmf_subsystem *sub; 857 858 /* Technically, taking the mutex modifies data in the subsystem. But the const 859 * is still important to convey that this doesn't mutate any other data. Cast 860 * it away to work around this. */ 861 sub = (struct spdk_nvmf_subsystem *)subsystem; 862 863 pthread_mutex_lock(&sub->mutex); 864 allow_any_host = sub->flags.allow_any_host; 865 pthread_mutex_unlock(&sub->mutex); 866 867 return allow_any_host; 868 } 869 870 bool 871 spdk_nvmf_subsystem_host_allowed(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn) 872 { 873 bool allowed; 874 875 if (!hostnqn) { 876 return false; 877 } 878 879 pthread_mutex_lock(&subsystem->mutex); 880 881 if (subsystem->flags.allow_any_host) { 882 pthread_mutex_unlock(&subsystem->mutex); 883 return true; 884 } 885 886 allowed = nvmf_subsystem_find_host(subsystem, hostnqn) != NULL; 887 pthread_mutex_unlock(&subsystem->mutex); 888 889 return allowed; 890 } 891 892 struct spdk_nvmf_host * 893 spdk_nvmf_subsystem_get_first_host(struct spdk_nvmf_subsystem *subsystem) 894 { 895 return TAILQ_FIRST(&subsystem->hosts); 896 } 897 898 899 struct spdk_nvmf_host * 900 spdk_nvmf_subsystem_get_next_host(struct spdk_nvmf_subsystem *subsystem, 901 struct spdk_nvmf_host *prev_host) 902 { 903 return TAILQ_NEXT(prev_host, link); 904 } 905 906 const char * 907 spdk_nvmf_host_get_nqn(const struct spdk_nvmf_host *host) 908 { 909 return host->nqn; 910 } 911 912 struct spdk_nvmf_subsystem_listener * 913 nvmf_subsystem_find_listener(struct spdk_nvmf_subsystem *subsystem, 914 const struct spdk_nvme_transport_id *trid) 915 { 916 struct spdk_nvmf_subsystem_listener *listener; 917 918 TAILQ_FOREACH(listener, &subsystem->listeners, link) { 919 if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) { 920 return listener; 921 } 922 } 923 924 return NULL; 925 } 926 927 /** 928 * Function to be called once the target is listening. 929 * 930 * \param ctx Context argument passed to this function. 931 * \param status 0 if it completed successfully, or negative errno if it failed. 932 */ 933 static void 934 _nvmf_subsystem_add_listener_done(void *ctx, int status) 935 { 936 struct spdk_nvmf_subsystem_listener *listener = ctx; 937 938 if (status) { 939 listener->cb_fn(listener->cb_arg, status); 940 free(listener); 941 return; 942 } 943 944 TAILQ_INSERT_HEAD(&listener->subsystem->listeners, listener, link); 945 nvmf_update_discovery_log(listener->subsystem->tgt, NULL); 946 listener->cb_fn(listener->cb_arg, status); 947 } 948 949 void 950 spdk_nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem, 951 struct spdk_nvme_transport_id *trid, 952 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, 953 void *cb_arg) 954 { 955 struct spdk_nvmf_transport *transport; 956 struct spdk_nvmf_subsystem_listener *listener; 957 struct spdk_nvmf_listener *tr_listener; 958 int rc = 0; 959 960 assert(cb_fn != NULL); 961 962 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 963 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 964 cb_fn(cb_arg, -EAGAIN); 965 return; 966 } 967 968 if (nvmf_subsystem_find_listener(subsystem, trid)) { 969 /* Listener already exists in this subsystem */ 970 cb_fn(cb_arg, 0); 971 return; 972 } 973 974 transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, trid->trstring); 975 if (!transport) { 976 SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n", 977 trid->trstring); 978 cb_fn(cb_arg, -EINVAL); 979 return; 980 } 981 982 tr_listener = nvmf_transport_find_listener(transport, trid); 983 if (!tr_listener) { 984 SPDK_ERRLOG("Cannot find transport listener for %s\n", trid->traddr); 985 cb_fn(cb_arg, -EINVAL); 986 return; 987 } 988 989 listener = calloc(1, sizeof(*listener)); 990 if (!listener) { 991 cb_fn(cb_arg, -ENOMEM); 992 return; 993 } 994 995 listener->trid = &tr_listener->trid; 996 listener->transport = transport; 997 listener->cb_fn = cb_fn; 998 listener->cb_arg = cb_arg; 999 listener->subsystem = subsystem; 1000 listener->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE; 1001 1002 if (transport->ops->listen_associate != NULL) { 1003 rc = transport->ops->listen_associate(transport, subsystem, trid); 1004 } 1005 1006 _nvmf_subsystem_add_listener_done(listener, rc); 1007 } 1008 1009 int 1010 spdk_nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem, 1011 const struct spdk_nvme_transport_id *trid) 1012 { 1013 struct spdk_nvmf_subsystem_listener *listener; 1014 1015 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1016 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1017 return -EAGAIN; 1018 } 1019 1020 listener = nvmf_subsystem_find_listener(subsystem, trid); 1021 if (listener == NULL) { 1022 return -ENOENT; 1023 } 1024 1025 _nvmf_subsystem_remove_listener(subsystem, listener, false); 1026 1027 return 0; 1028 } 1029 1030 void 1031 nvmf_subsystem_remove_all_listeners(struct spdk_nvmf_subsystem *subsystem, 1032 bool stop) 1033 { 1034 struct spdk_nvmf_subsystem_listener *listener, *listener_tmp; 1035 1036 TAILQ_FOREACH_SAFE(listener, &subsystem->listeners, link, listener_tmp) { 1037 _nvmf_subsystem_remove_listener(subsystem, listener, stop); 1038 } 1039 } 1040 1041 bool 1042 spdk_nvmf_subsystem_listener_allowed(struct spdk_nvmf_subsystem *subsystem, 1043 const struct spdk_nvme_transport_id *trid) 1044 { 1045 struct spdk_nvmf_subsystem_listener *listener; 1046 1047 if (!strcmp(subsystem->subnqn, SPDK_NVMF_DISCOVERY_NQN)) { 1048 return true; 1049 } 1050 1051 TAILQ_FOREACH(listener, &subsystem->listeners, link) { 1052 if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) { 1053 return true; 1054 } 1055 } 1056 1057 return false; 1058 } 1059 1060 struct spdk_nvmf_subsystem_listener * 1061 spdk_nvmf_subsystem_get_first_listener(struct spdk_nvmf_subsystem *subsystem) 1062 { 1063 return TAILQ_FIRST(&subsystem->listeners); 1064 } 1065 1066 struct spdk_nvmf_subsystem_listener * 1067 spdk_nvmf_subsystem_get_next_listener(struct spdk_nvmf_subsystem *subsystem, 1068 struct spdk_nvmf_subsystem_listener *prev_listener) 1069 { 1070 return TAILQ_NEXT(prev_listener, link); 1071 } 1072 1073 const struct spdk_nvme_transport_id * 1074 spdk_nvmf_subsystem_listener_get_trid(struct spdk_nvmf_subsystem_listener *listener) 1075 { 1076 return listener->trid; 1077 } 1078 1079 void 1080 spdk_nvmf_subsystem_allow_any_listener(struct spdk_nvmf_subsystem *subsystem, 1081 bool allow_any_listener) 1082 { 1083 subsystem->flags.allow_any_listener = allow_any_listener; 1084 } 1085 1086 bool 1087 spdk_nvmf_subsytem_any_listener_allowed(struct spdk_nvmf_subsystem *subsystem) 1088 { 1089 return subsystem->flags.allow_any_listener; 1090 } 1091 1092 1093 struct subsystem_update_ns_ctx { 1094 struct spdk_nvmf_subsystem *subsystem; 1095 1096 spdk_nvmf_subsystem_state_change_done cb_fn; 1097 void *cb_arg; 1098 }; 1099 1100 static void 1101 subsystem_update_ns_done(struct spdk_io_channel_iter *i, int status) 1102 { 1103 struct subsystem_update_ns_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 1104 1105 if (ctx->cb_fn) { 1106 ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status); 1107 } 1108 free(ctx); 1109 } 1110 1111 static void 1112 subsystem_update_ns_on_pg(struct spdk_io_channel_iter *i) 1113 { 1114 int rc; 1115 struct subsystem_update_ns_ctx *ctx; 1116 struct spdk_nvmf_poll_group *group; 1117 struct spdk_nvmf_subsystem *subsystem; 1118 1119 ctx = spdk_io_channel_iter_get_ctx(i); 1120 group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i)); 1121 subsystem = ctx->subsystem; 1122 1123 rc = nvmf_poll_group_update_subsystem(group, subsystem); 1124 spdk_for_each_channel_continue(i, rc); 1125 } 1126 1127 static int 1128 nvmf_subsystem_update_ns(struct spdk_nvmf_subsystem *subsystem, spdk_channel_for_each_cpl cpl, 1129 void *ctx) 1130 { 1131 spdk_for_each_channel(subsystem->tgt, 1132 subsystem_update_ns_on_pg, 1133 ctx, 1134 cpl); 1135 1136 return 0; 1137 } 1138 1139 static void 1140 nvmf_subsystem_ns_changed(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1141 { 1142 struct spdk_nvmf_ctrlr *ctrlr; 1143 1144 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 1145 nvmf_ctrlr_ns_changed(ctrlr, nsid); 1146 } 1147 } 1148 1149 static uint32_t 1150 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns); 1151 1152 int 1153 spdk_nvmf_subsystem_remove_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1154 { 1155 struct spdk_nvmf_transport *transport; 1156 struct spdk_nvmf_ns *ns; 1157 1158 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1159 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1160 assert(false); 1161 return -1; 1162 } 1163 1164 if (nsid == 0 || nsid > subsystem->max_nsid) { 1165 return -1; 1166 } 1167 1168 ns = subsystem->ns[nsid - 1]; 1169 if (!ns) { 1170 return -1; 1171 } 1172 1173 subsystem->ns[nsid - 1] = NULL; 1174 1175 free(ns->ptpl_file); 1176 nvmf_ns_reservation_clear_all_registrants(ns); 1177 spdk_bdev_module_release_bdev(ns->bdev); 1178 spdk_bdev_close(ns->desc); 1179 free(ns); 1180 1181 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 1182 transport = spdk_nvmf_transport_get_next(transport)) { 1183 if (transport->ops->subsystem_remove_ns) { 1184 transport->ops->subsystem_remove_ns(transport, subsystem, nsid); 1185 } 1186 } 1187 1188 nvmf_subsystem_ns_changed(subsystem, nsid); 1189 1190 return 0; 1191 } 1192 1193 struct subsystem_ns_change_ctx { 1194 struct spdk_nvmf_subsystem *subsystem; 1195 spdk_nvmf_subsystem_state_change_done cb_fn; 1196 uint32_t nsid; 1197 }; 1198 1199 static void 1200 _nvmf_ns_hot_remove(struct spdk_nvmf_subsystem *subsystem, 1201 void *cb_arg, int status) 1202 { 1203 struct subsystem_ns_change_ctx *ctx = cb_arg; 1204 int rc; 1205 1206 rc = spdk_nvmf_subsystem_remove_ns(subsystem, ctx->nsid); 1207 if (rc != 0) { 1208 SPDK_ERRLOG("Failed to make changes to NVME-oF subsystem with id: %u\n", subsystem->id); 1209 } 1210 1211 spdk_nvmf_subsystem_resume(subsystem, NULL, NULL); 1212 1213 free(ctx); 1214 } 1215 1216 static void 1217 nvmf_ns_change_msg(void *ns_ctx) 1218 { 1219 struct subsystem_ns_change_ctx *ctx = ns_ctx; 1220 int rc; 1221 1222 rc = spdk_nvmf_subsystem_pause(ctx->subsystem, ctx->cb_fn, ctx); 1223 if (rc) { 1224 if (rc == -EBUSY) { 1225 /* Try again, this is not a permanent situation. */ 1226 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ctx); 1227 } else { 1228 free(ctx); 1229 SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n"); 1230 } 1231 } 1232 } 1233 1234 static void 1235 nvmf_ns_hot_remove(void *remove_ctx) 1236 { 1237 struct spdk_nvmf_ns *ns = remove_ctx; 1238 struct subsystem_ns_change_ctx *ns_ctx; 1239 int rc; 1240 1241 /* We have to allocate a new context because this op 1242 * is asynchronous and we could lose the ns in the middle. 1243 */ 1244 ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx)); 1245 if (!ns_ctx) { 1246 SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n"); 1247 return; 1248 } 1249 1250 ns_ctx->subsystem = ns->subsystem; 1251 ns_ctx->nsid = ns->opts.nsid; 1252 ns_ctx->cb_fn = _nvmf_ns_hot_remove; 1253 1254 rc = spdk_nvmf_subsystem_pause(ns->subsystem, _nvmf_ns_hot_remove, ns_ctx); 1255 if (rc) { 1256 if (rc == -EBUSY) { 1257 /* Try again, this is not a permanent situation. */ 1258 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx); 1259 } else { 1260 SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n"); 1261 free(ns_ctx); 1262 } 1263 } 1264 } 1265 1266 static void 1267 _nvmf_ns_resize(struct spdk_nvmf_subsystem *subsystem, void *cb_arg, int status) 1268 { 1269 struct subsystem_ns_change_ctx *ctx = cb_arg; 1270 1271 nvmf_subsystem_ns_changed(subsystem, ctx->nsid); 1272 spdk_nvmf_subsystem_resume(subsystem, NULL, NULL); 1273 1274 free(ctx); 1275 } 1276 1277 static void 1278 nvmf_ns_resize(void *event_ctx) 1279 { 1280 struct spdk_nvmf_ns *ns = event_ctx; 1281 struct subsystem_ns_change_ctx *ns_ctx; 1282 int rc; 1283 1284 /* We have to allocate a new context because this op 1285 * is asynchronous and we could lose the ns in the middle. 1286 */ 1287 ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx)); 1288 if (!ns_ctx) { 1289 SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n"); 1290 return; 1291 } 1292 1293 ns_ctx->subsystem = ns->subsystem; 1294 ns_ctx->nsid = ns->opts.nsid; 1295 ns_ctx->cb_fn = _nvmf_ns_resize; 1296 1297 rc = spdk_nvmf_subsystem_pause(ns->subsystem, _nvmf_ns_resize, ns_ctx); 1298 if (rc) { 1299 if (rc == -EBUSY) { 1300 /* Try again, this is not a permanent situation. */ 1301 spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx); 1302 } 1303 SPDK_ERRLOG("Unable to pause subsystem to process namespace resize!\n"); 1304 free(ns_ctx); 1305 } 1306 } 1307 1308 static void 1309 nvmf_ns_event(enum spdk_bdev_event_type type, 1310 struct spdk_bdev *bdev, 1311 void *event_ctx) 1312 { 1313 SPDK_DEBUGLOG(nvmf, "Bdev event: type %d, name %s, subsystem_id %d, ns_id %d\n", 1314 type, 1315 bdev->name, 1316 ((struct spdk_nvmf_ns *)event_ctx)->subsystem->id, 1317 ((struct spdk_nvmf_ns *)event_ctx)->nsid); 1318 1319 switch (type) { 1320 case SPDK_BDEV_EVENT_REMOVE: 1321 nvmf_ns_hot_remove(event_ctx); 1322 break; 1323 case SPDK_BDEV_EVENT_RESIZE: 1324 nvmf_ns_resize(event_ctx); 1325 break; 1326 default: 1327 SPDK_NOTICELOG("Unsupported bdev event: type %d\n", type); 1328 break; 1329 } 1330 } 1331 1332 void 1333 spdk_nvmf_ns_opts_get_defaults(struct spdk_nvmf_ns_opts *opts, size_t opts_size) 1334 { 1335 /* All current fields are set to 0 by default. */ 1336 memset(opts, 0, opts_size); 1337 } 1338 1339 /* Dummy bdev module used to to claim bdevs. */ 1340 static struct spdk_bdev_module ns_bdev_module = { 1341 .name = "NVMe-oF Target", 1342 }; 1343 1344 static int 1345 nvmf_ns_load_reservation(const char *file, struct spdk_nvmf_reservation_info *info); 1346 static int 1347 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info); 1348 1349 uint32_t 1350 spdk_nvmf_subsystem_add_ns_ext(struct spdk_nvmf_subsystem *subsystem, const char *bdev_name, 1351 const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size, 1352 const char *ptpl_file) 1353 { 1354 struct spdk_nvmf_transport *transport; 1355 struct spdk_nvmf_ns_opts opts; 1356 struct spdk_nvmf_ns *ns; 1357 struct spdk_nvmf_reservation_info info = {0}; 1358 int rc; 1359 1360 if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 1361 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) { 1362 return 0; 1363 } 1364 1365 spdk_nvmf_ns_opts_get_defaults(&opts, sizeof(opts)); 1366 if (user_opts) { 1367 memcpy(&opts, user_opts, spdk_min(sizeof(opts), opts_size)); 1368 } 1369 1370 if (opts.nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1371 SPDK_ERRLOG("Invalid NSID %" PRIu32 "\n", opts.nsid); 1372 return 0; 1373 } 1374 1375 if (opts.nsid == 0) { 1376 /* 1377 * NSID not specified - find a free index. 1378 * 1379 * If no free slots are found, opts.nsid will be subsystem->max_nsid + 1, which will 1380 * expand max_nsid if possible. 1381 */ 1382 for (opts.nsid = 1; opts.nsid <= subsystem->max_nsid; opts.nsid++) { 1383 if (_nvmf_subsystem_get_ns(subsystem, opts.nsid) == NULL) { 1384 break; 1385 } 1386 } 1387 } 1388 1389 if (_nvmf_subsystem_get_ns(subsystem, opts.nsid)) { 1390 SPDK_ERRLOG("Requested NSID %" PRIu32 " already in use\n", opts.nsid); 1391 return 0; 1392 } 1393 1394 if (opts.nsid > subsystem->max_nsid) { 1395 SPDK_ERRLOG("NSID greater than maximum not allowed\n"); 1396 return 0; 1397 } 1398 1399 ns = calloc(1, sizeof(*ns)); 1400 if (ns == NULL) { 1401 SPDK_ERRLOG("Namespace allocation failed\n"); 1402 return 0; 1403 } 1404 1405 rc = spdk_bdev_open_ext(bdev_name, true, nvmf_ns_event, ns, &ns->desc); 1406 if (rc != 0) { 1407 SPDK_ERRLOG("Subsystem %s: bdev %s cannot be opened, error=%d\n", 1408 subsystem->subnqn, bdev_name, rc); 1409 free(ns); 1410 return 0; 1411 } 1412 1413 ns->bdev = spdk_bdev_desc_get_bdev(ns->desc); 1414 1415 if (spdk_bdev_get_md_size(ns->bdev) != 0 && !spdk_bdev_is_md_interleaved(ns->bdev)) { 1416 SPDK_ERRLOG("Can't attach bdev with separate metadata.\n"); 1417 spdk_bdev_close(ns->desc); 1418 free(ns); 1419 return 0; 1420 } 1421 1422 rc = spdk_bdev_module_claim_bdev(ns->bdev, ns->desc, &ns_bdev_module); 1423 if (rc != 0) { 1424 spdk_bdev_close(ns->desc); 1425 free(ns); 1426 return 0; 1427 } 1428 1429 if (spdk_mem_all_zero(&opts.uuid, sizeof(opts.uuid))) { 1430 opts.uuid = *spdk_bdev_get_uuid(ns->bdev); 1431 } 1432 1433 ns->opts = opts; 1434 ns->subsystem = subsystem; 1435 subsystem->ns[opts.nsid - 1] = ns; 1436 ns->nsid = opts.nsid; 1437 TAILQ_INIT(&ns->registrants); 1438 1439 if (ptpl_file) { 1440 rc = nvmf_ns_load_reservation(ptpl_file, &info); 1441 if (!rc) { 1442 rc = nvmf_ns_reservation_restore(ns, &info); 1443 if (rc) { 1444 SPDK_ERRLOG("Subsystem restore reservation failed\n"); 1445 subsystem->ns[opts.nsid - 1] = NULL; 1446 spdk_bdev_module_release_bdev(ns->bdev); 1447 spdk_bdev_close(ns->desc); 1448 free(ns); 1449 return 0; 1450 } 1451 } 1452 ns->ptpl_file = strdup(ptpl_file); 1453 } 1454 1455 for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport; 1456 transport = spdk_nvmf_transport_get_next(transport)) { 1457 if (transport->ops->subsystem_add_ns) { 1458 rc = transport->ops->subsystem_add_ns(transport, subsystem, ns); 1459 if (rc) { 1460 SPDK_ERRLOG("Namespace attachment is not allowed by %s transport\n", transport->ops->name); 1461 free(ns->ptpl_file); 1462 nvmf_ns_reservation_clear_all_registrants(ns); 1463 subsystem->ns[opts.nsid - 1] = NULL; 1464 spdk_bdev_module_release_bdev(ns->bdev); 1465 spdk_bdev_close(ns->desc); 1466 free(ns); 1467 return 0; 1468 } 1469 } 1470 } 1471 1472 SPDK_DEBUGLOG(nvmf, "Subsystem %s: bdev %s assigned nsid %" PRIu32 "\n", 1473 spdk_nvmf_subsystem_get_nqn(subsystem), 1474 bdev_name, 1475 opts.nsid); 1476 1477 nvmf_subsystem_ns_changed(subsystem, opts.nsid); 1478 1479 return opts.nsid; 1480 } 1481 1482 uint32_t 1483 spdk_nvmf_subsystem_add_ns(struct spdk_nvmf_subsystem *subsystem, struct spdk_bdev *bdev, 1484 const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size, 1485 const char *ptpl_file) 1486 { 1487 return spdk_nvmf_subsystem_add_ns_ext(subsystem, spdk_bdev_get_name(bdev), 1488 user_opts, opts_size, ptpl_file); 1489 } 1490 1491 static uint32_t 1492 nvmf_subsystem_get_next_allocated_nsid(struct spdk_nvmf_subsystem *subsystem, 1493 uint32_t prev_nsid) 1494 { 1495 uint32_t nsid; 1496 1497 if (prev_nsid >= subsystem->max_nsid) { 1498 return 0; 1499 } 1500 1501 for (nsid = prev_nsid + 1; nsid <= subsystem->max_nsid; nsid++) { 1502 if (subsystem->ns[nsid - 1]) { 1503 return nsid; 1504 } 1505 } 1506 1507 return 0; 1508 } 1509 1510 struct spdk_nvmf_ns * 1511 spdk_nvmf_subsystem_get_first_ns(struct spdk_nvmf_subsystem *subsystem) 1512 { 1513 uint32_t first_nsid; 1514 1515 first_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, 0); 1516 return _nvmf_subsystem_get_ns(subsystem, first_nsid); 1517 } 1518 1519 struct spdk_nvmf_ns * 1520 spdk_nvmf_subsystem_get_next_ns(struct spdk_nvmf_subsystem *subsystem, 1521 struct spdk_nvmf_ns *prev_ns) 1522 { 1523 uint32_t next_nsid; 1524 1525 next_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, prev_ns->opts.nsid); 1526 return _nvmf_subsystem_get_ns(subsystem, next_nsid); 1527 } 1528 1529 struct spdk_nvmf_ns * 1530 spdk_nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 1531 { 1532 return _nvmf_subsystem_get_ns(subsystem, nsid); 1533 } 1534 1535 uint32_t 1536 spdk_nvmf_ns_get_id(const struct spdk_nvmf_ns *ns) 1537 { 1538 return ns->opts.nsid; 1539 } 1540 1541 struct spdk_bdev * 1542 spdk_nvmf_ns_get_bdev(struct spdk_nvmf_ns *ns) 1543 { 1544 return ns->bdev; 1545 } 1546 1547 void 1548 spdk_nvmf_ns_get_opts(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_ns_opts *opts, 1549 size_t opts_size) 1550 { 1551 memset(opts, 0, opts_size); 1552 memcpy(opts, &ns->opts, spdk_min(sizeof(ns->opts), opts_size)); 1553 } 1554 1555 const char * 1556 spdk_nvmf_subsystem_get_sn(const struct spdk_nvmf_subsystem *subsystem) 1557 { 1558 return subsystem->sn; 1559 } 1560 1561 int 1562 spdk_nvmf_subsystem_set_sn(struct spdk_nvmf_subsystem *subsystem, const char *sn) 1563 { 1564 size_t len, max_len; 1565 1566 max_len = sizeof(subsystem->sn) - 1; 1567 len = strlen(sn); 1568 if (len > max_len) { 1569 SPDK_DEBUGLOG(nvmf, "Invalid sn \"%s\": length %zu > max %zu\n", 1570 sn, len, max_len); 1571 return -1; 1572 } 1573 1574 if (!nvmf_valid_ascii_string(sn, len)) { 1575 SPDK_DEBUGLOG(nvmf, "Non-ASCII sn\n"); 1576 SPDK_LOGDUMP(nvmf, "sn", sn, len); 1577 return -1; 1578 } 1579 1580 snprintf(subsystem->sn, sizeof(subsystem->sn), "%s", sn); 1581 1582 return 0; 1583 } 1584 1585 const char * 1586 spdk_nvmf_subsystem_get_mn(const struct spdk_nvmf_subsystem *subsystem) 1587 { 1588 return subsystem->mn; 1589 } 1590 1591 int 1592 spdk_nvmf_subsystem_set_mn(struct spdk_nvmf_subsystem *subsystem, const char *mn) 1593 { 1594 size_t len, max_len; 1595 1596 if (mn == NULL) { 1597 mn = MODEL_NUMBER_DEFAULT; 1598 } 1599 max_len = sizeof(subsystem->mn) - 1; 1600 len = strlen(mn); 1601 if (len > max_len) { 1602 SPDK_DEBUGLOG(nvmf, "Invalid mn \"%s\": length %zu > max %zu\n", 1603 mn, len, max_len); 1604 return -1; 1605 } 1606 1607 if (!nvmf_valid_ascii_string(mn, len)) { 1608 SPDK_DEBUGLOG(nvmf, "Non-ASCII mn\n"); 1609 SPDK_LOGDUMP(nvmf, "mn", mn, len); 1610 return -1; 1611 } 1612 1613 snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", mn); 1614 1615 return 0; 1616 } 1617 1618 const char * 1619 spdk_nvmf_subsystem_get_nqn(const struct spdk_nvmf_subsystem *subsystem) 1620 { 1621 return subsystem->subnqn; 1622 } 1623 1624 enum spdk_nvmf_subtype spdk_nvmf_subsystem_get_type(struct spdk_nvmf_subsystem *subsystem) 1625 { 1626 return subsystem->subtype; 1627 } 1628 1629 uint32_t 1630 spdk_nvmf_subsystem_get_max_nsid(struct spdk_nvmf_subsystem *subsystem) 1631 { 1632 return subsystem->max_nsid; 1633 } 1634 1635 static uint16_t 1636 nvmf_subsystem_gen_cntlid(struct spdk_nvmf_subsystem *subsystem) 1637 { 1638 int count; 1639 1640 /* 1641 * In the worst case, we might have to try all CNTLID values between 1 and 0xFFF0 - 1 1642 * before we find one that is unused (or find that all values are in use). 1643 */ 1644 for (count = 0; count < 0xFFF0 - 1; count++) { 1645 subsystem->next_cntlid++; 1646 if (subsystem->next_cntlid >= 0xFFF0) { 1647 /* The spec reserves cntlid values in the range FFF0h to FFFFh. */ 1648 subsystem->next_cntlid = 1; 1649 } 1650 1651 /* Check if a controller with this cntlid currently exists. */ 1652 if (nvmf_subsystem_get_ctrlr(subsystem, subsystem->next_cntlid) == NULL) { 1653 /* Found unused cntlid */ 1654 return subsystem->next_cntlid; 1655 } 1656 } 1657 1658 /* All valid cntlid values are in use. */ 1659 return 0xFFFF; 1660 } 1661 1662 int 1663 nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_ctrlr *ctrlr) 1664 { 1665 ctrlr->cntlid = nvmf_subsystem_gen_cntlid(subsystem); 1666 if (ctrlr->cntlid == 0xFFFF) { 1667 /* Unable to get a cntlid */ 1668 SPDK_ERRLOG("Reached max simultaneous ctrlrs\n"); 1669 return -EBUSY; 1670 } 1671 1672 TAILQ_INSERT_TAIL(&subsystem->ctrlrs, ctrlr, link); 1673 1674 return 0; 1675 } 1676 1677 void 1678 nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem, 1679 struct spdk_nvmf_ctrlr *ctrlr) 1680 { 1681 assert(subsystem == ctrlr->subsys); 1682 TAILQ_REMOVE(&subsystem->ctrlrs, ctrlr, link); 1683 } 1684 1685 struct spdk_nvmf_ctrlr * 1686 nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, uint16_t cntlid) 1687 { 1688 struct spdk_nvmf_ctrlr *ctrlr; 1689 1690 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 1691 if (ctrlr->cntlid == cntlid) { 1692 return ctrlr; 1693 } 1694 } 1695 1696 return NULL; 1697 } 1698 1699 uint32_t 1700 spdk_nvmf_subsystem_get_max_namespaces(const struct spdk_nvmf_subsystem *subsystem) 1701 { 1702 return subsystem->max_nsid; 1703 } 1704 1705 struct _nvmf_ns_registrant { 1706 uint64_t rkey; 1707 char *host_uuid; 1708 }; 1709 1710 struct _nvmf_ns_registrants { 1711 size_t num_regs; 1712 struct _nvmf_ns_registrant reg[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 1713 }; 1714 1715 struct _nvmf_ns_reservation { 1716 bool ptpl_activated; 1717 enum spdk_nvme_reservation_type rtype; 1718 uint64_t crkey; 1719 char *bdev_uuid; 1720 char *holder_uuid; 1721 struct _nvmf_ns_registrants regs; 1722 }; 1723 1724 static const struct spdk_json_object_decoder nvmf_ns_pr_reg_decoders[] = { 1725 {"rkey", offsetof(struct _nvmf_ns_registrant, rkey), spdk_json_decode_uint64}, 1726 {"host_uuid", offsetof(struct _nvmf_ns_registrant, host_uuid), spdk_json_decode_string}, 1727 }; 1728 1729 static int 1730 nvmf_decode_ns_pr_reg(const struct spdk_json_val *val, void *out) 1731 { 1732 struct _nvmf_ns_registrant *reg = out; 1733 1734 return spdk_json_decode_object(val, nvmf_ns_pr_reg_decoders, 1735 SPDK_COUNTOF(nvmf_ns_pr_reg_decoders), reg); 1736 } 1737 1738 static int 1739 nvmf_decode_ns_pr_regs(const struct spdk_json_val *val, void *out) 1740 { 1741 struct _nvmf_ns_registrants *regs = out; 1742 1743 return spdk_json_decode_array(val, nvmf_decode_ns_pr_reg, regs->reg, 1744 SPDK_NVMF_MAX_NUM_REGISTRANTS, ®s->num_regs, 1745 sizeof(struct _nvmf_ns_registrant)); 1746 } 1747 1748 static const struct spdk_json_object_decoder nvmf_ns_pr_decoders[] = { 1749 {"ptpl", offsetof(struct _nvmf_ns_reservation, ptpl_activated), spdk_json_decode_bool, true}, 1750 {"rtype", offsetof(struct _nvmf_ns_reservation, rtype), spdk_json_decode_uint32, true}, 1751 {"crkey", offsetof(struct _nvmf_ns_reservation, crkey), spdk_json_decode_uint64, true}, 1752 {"bdev_uuid", offsetof(struct _nvmf_ns_reservation, bdev_uuid), spdk_json_decode_string}, 1753 {"holder_uuid", offsetof(struct _nvmf_ns_reservation, holder_uuid), spdk_json_decode_string, true}, 1754 {"registrants", offsetof(struct _nvmf_ns_reservation, regs), nvmf_decode_ns_pr_regs}, 1755 }; 1756 1757 static int 1758 nvmf_ns_load_reservation(const char *file, struct spdk_nvmf_reservation_info *info) 1759 { 1760 FILE *fd; 1761 size_t json_size; 1762 ssize_t values_cnt, rc; 1763 void *json = NULL, *end; 1764 struct spdk_json_val *values = NULL; 1765 struct _nvmf_ns_reservation res = {}; 1766 uint32_t i; 1767 1768 fd = fopen(file, "r"); 1769 /* It's not an error if the file does not exist */ 1770 if (!fd) { 1771 SPDK_NOTICELOG("File %s does not exist\n", file); 1772 return -ENOENT; 1773 } 1774 1775 /* Load all persist file contents into a local buffer */ 1776 json = spdk_posix_file_load(fd, &json_size); 1777 fclose(fd); 1778 if (!json) { 1779 SPDK_ERRLOG("Load persit file %s failed\n", file); 1780 return -ENOMEM; 1781 } 1782 1783 rc = spdk_json_parse(json, json_size, NULL, 0, &end, 0); 1784 if (rc < 0) { 1785 SPDK_NOTICELOG("Parsing JSON configuration failed (%zd)\n", rc); 1786 goto exit; 1787 } 1788 1789 values_cnt = rc; 1790 values = calloc(values_cnt, sizeof(struct spdk_json_val)); 1791 if (values == NULL) { 1792 goto exit; 1793 } 1794 1795 rc = spdk_json_parse(json, json_size, values, values_cnt, &end, 0); 1796 if (rc != values_cnt) { 1797 SPDK_ERRLOG("Parsing JSON configuration failed (%zd)\n", rc); 1798 goto exit; 1799 } 1800 1801 /* Decode json */ 1802 if (spdk_json_decode_object(values, nvmf_ns_pr_decoders, 1803 SPDK_COUNTOF(nvmf_ns_pr_decoders), 1804 &res)) { 1805 SPDK_ERRLOG("Invalid objects in the persist file %s\n", file); 1806 rc = -EINVAL; 1807 goto exit; 1808 } 1809 1810 if (res.regs.num_regs > SPDK_NVMF_MAX_NUM_REGISTRANTS) { 1811 SPDK_ERRLOG("Can only support up to %u registrants\n", SPDK_NVMF_MAX_NUM_REGISTRANTS); 1812 rc = -ERANGE; 1813 goto exit; 1814 } 1815 1816 rc = 0; 1817 info->ptpl_activated = res.ptpl_activated; 1818 info->rtype = res.rtype; 1819 info->crkey = res.crkey; 1820 snprintf(info->bdev_uuid, sizeof(info->bdev_uuid), "%s", res.bdev_uuid); 1821 snprintf(info->holder_uuid, sizeof(info->holder_uuid), "%s", res.holder_uuid); 1822 info->num_regs = res.regs.num_regs; 1823 for (i = 0; i < res.regs.num_regs; i++) { 1824 info->registrants[i].rkey = res.regs.reg[i].rkey; 1825 snprintf(info->registrants[i].host_uuid, sizeof(info->registrants[i].host_uuid), "%s", 1826 res.regs.reg[i].host_uuid); 1827 } 1828 1829 exit: 1830 free(json); 1831 free(values); 1832 free(res.bdev_uuid); 1833 free(res.holder_uuid); 1834 for (i = 0; i < res.regs.num_regs; i++) { 1835 free(res.regs.reg[i].host_uuid); 1836 } 1837 1838 return rc; 1839 } 1840 1841 static bool 1842 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns); 1843 1844 static int 1845 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info) 1846 { 1847 uint32_t i; 1848 struct spdk_nvmf_registrant *reg, *holder = NULL; 1849 struct spdk_uuid bdev_uuid, holder_uuid; 1850 1851 SPDK_DEBUGLOG(nvmf, "NSID %u, PTPL %u, Number of registrants %u\n", 1852 ns->nsid, info->ptpl_activated, info->num_regs); 1853 1854 /* it's not an error */ 1855 if (!info->ptpl_activated || !info->num_regs) { 1856 return 0; 1857 } 1858 1859 spdk_uuid_parse(&bdev_uuid, info->bdev_uuid); 1860 if (spdk_uuid_compare(&bdev_uuid, spdk_bdev_get_uuid(ns->bdev))) { 1861 SPDK_ERRLOG("Existing bdev UUID is not same with configuration file\n"); 1862 return -EINVAL; 1863 } 1864 1865 ns->crkey = info->crkey; 1866 ns->rtype = info->rtype; 1867 ns->ptpl_activated = info->ptpl_activated; 1868 spdk_uuid_parse(&holder_uuid, info->holder_uuid); 1869 1870 SPDK_DEBUGLOG(nvmf, "Bdev UUID %s\n", info->bdev_uuid); 1871 if (info->rtype) { 1872 SPDK_DEBUGLOG(nvmf, "Holder UUID %s, RTYPE %u, RKEY 0x%"PRIx64"\n", 1873 info->holder_uuid, info->rtype, info->crkey); 1874 } 1875 1876 for (i = 0; i < info->num_regs; i++) { 1877 reg = calloc(1, sizeof(*reg)); 1878 if (!reg) { 1879 return -ENOMEM; 1880 } 1881 spdk_uuid_parse(®->hostid, info->registrants[i].host_uuid); 1882 reg->rkey = info->registrants[i].rkey; 1883 TAILQ_INSERT_TAIL(&ns->registrants, reg, link); 1884 if (!spdk_uuid_compare(&holder_uuid, ®->hostid)) { 1885 holder = reg; 1886 } 1887 SPDK_DEBUGLOG(nvmf, "Registrant RKEY 0x%"PRIx64", Host UUID %s\n", 1888 info->registrants[i].rkey, info->registrants[i].host_uuid); 1889 } 1890 1891 if (nvmf_ns_reservation_all_registrants_type(ns)) { 1892 ns->holder = TAILQ_FIRST(&ns->registrants); 1893 } else { 1894 ns->holder = holder; 1895 } 1896 1897 return 0; 1898 } 1899 1900 static int 1901 nvmf_ns_json_write_cb(void *cb_ctx, const void *data, size_t size) 1902 { 1903 char *file = cb_ctx; 1904 size_t rc; 1905 FILE *fd; 1906 1907 fd = fopen(file, "w"); 1908 if (!fd) { 1909 SPDK_ERRLOG("Can't open file %s for write\n", file); 1910 return -ENOENT; 1911 } 1912 rc = fwrite(data, 1, size, fd); 1913 fclose(fd); 1914 1915 return rc == size ? 0 : -1; 1916 } 1917 1918 static int 1919 nvmf_ns_reservation_update(const char *file, struct spdk_nvmf_reservation_info *info) 1920 { 1921 struct spdk_json_write_ctx *w; 1922 uint32_t i; 1923 int rc = 0; 1924 1925 w = spdk_json_write_begin(nvmf_ns_json_write_cb, (void *)file, 0); 1926 if (w == NULL) { 1927 return -ENOMEM; 1928 } 1929 /* clear the configuration file */ 1930 if (!info->ptpl_activated) { 1931 goto exit; 1932 } 1933 1934 spdk_json_write_object_begin(w); 1935 spdk_json_write_named_bool(w, "ptpl", info->ptpl_activated); 1936 spdk_json_write_named_uint32(w, "rtype", info->rtype); 1937 spdk_json_write_named_uint64(w, "crkey", info->crkey); 1938 spdk_json_write_named_string(w, "bdev_uuid", info->bdev_uuid); 1939 spdk_json_write_named_string(w, "holder_uuid", info->holder_uuid); 1940 1941 spdk_json_write_named_array_begin(w, "registrants"); 1942 for (i = 0; i < info->num_regs; i++) { 1943 spdk_json_write_object_begin(w); 1944 spdk_json_write_named_uint64(w, "rkey", info->registrants[i].rkey); 1945 spdk_json_write_named_string(w, "host_uuid", info->registrants[i].host_uuid); 1946 spdk_json_write_object_end(w); 1947 } 1948 spdk_json_write_array_end(w); 1949 spdk_json_write_object_end(w); 1950 1951 exit: 1952 rc = spdk_json_write_end(w); 1953 return rc; 1954 } 1955 1956 static int 1957 nvmf_ns_update_reservation_info(struct spdk_nvmf_ns *ns) 1958 { 1959 struct spdk_nvmf_reservation_info info; 1960 struct spdk_nvmf_registrant *reg, *tmp; 1961 uint32_t i = 0; 1962 1963 assert(ns != NULL); 1964 1965 if (!ns->bdev || !ns->ptpl_file) { 1966 return 0; 1967 } 1968 1969 memset(&info, 0, sizeof(info)); 1970 spdk_uuid_fmt_lower(info.bdev_uuid, sizeof(info.bdev_uuid), spdk_bdev_get_uuid(ns->bdev)); 1971 1972 if (ns->rtype) { 1973 info.rtype = ns->rtype; 1974 info.crkey = ns->crkey; 1975 if (!nvmf_ns_reservation_all_registrants_type(ns)) { 1976 assert(ns->holder != NULL); 1977 spdk_uuid_fmt_lower(info.holder_uuid, sizeof(info.holder_uuid), &ns->holder->hostid); 1978 } 1979 } 1980 1981 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 1982 spdk_uuid_fmt_lower(info.registrants[i].host_uuid, sizeof(info.registrants[i].host_uuid), 1983 ®->hostid); 1984 info.registrants[i++].rkey = reg->rkey; 1985 } 1986 1987 info.num_regs = i; 1988 info.ptpl_activated = ns->ptpl_activated; 1989 1990 return nvmf_ns_reservation_update(ns->ptpl_file, &info); 1991 } 1992 1993 static struct spdk_nvmf_registrant * 1994 nvmf_ns_reservation_get_registrant(struct spdk_nvmf_ns *ns, 1995 struct spdk_uuid *uuid) 1996 { 1997 struct spdk_nvmf_registrant *reg, *tmp; 1998 1999 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2000 if (!spdk_uuid_compare(®->hostid, uuid)) { 2001 return reg; 2002 } 2003 } 2004 2005 return NULL; 2006 } 2007 2008 /* Generate reservation notice log to registered HostID controllers */ 2009 static void 2010 nvmf_subsystem_gen_ctrlr_notification(struct spdk_nvmf_subsystem *subsystem, 2011 struct spdk_nvmf_ns *ns, 2012 struct spdk_uuid *hostid_list, 2013 uint32_t num_hostid, 2014 enum spdk_nvme_reservation_notification_log_page_type type) 2015 { 2016 struct spdk_nvmf_ctrlr *ctrlr; 2017 uint32_t i; 2018 2019 for (i = 0; i < num_hostid; i++) { 2020 TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) { 2021 if (!spdk_uuid_compare(&ctrlr->hostid, &hostid_list[i])) { 2022 nvmf_ctrlr_reservation_notice_log(ctrlr, ns, type); 2023 } 2024 } 2025 } 2026 } 2027 2028 /* Get all registrants' hostid other than the controller who issued the command */ 2029 static uint32_t 2030 nvmf_ns_reservation_get_all_other_hostid(struct spdk_nvmf_ns *ns, 2031 struct spdk_uuid *hostid_list, 2032 uint32_t max_num_hostid, 2033 struct spdk_uuid *current_hostid) 2034 { 2035 struct spdk_nvmf_registrant *reg, *tmp; 2036 uint32_t num_hostid = 0; 2037 2038 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2039 if (spdk_uuid_compare(®->hostid, current_hostid)) { 2040 if (num_hostid == max_num_hostid) { 2041 assert(false); 2042 return max_num_hostid; 2043 } 2044 hostid_list[num_hostid++] = reg->hostid; 2045 } 2046 } 2047 2048 return num_hostid; 2049 } 2050 2051 /* Calculate the unregistered HostID list according to list 2052 * prior to execute preempt command and list after executing 2053 * preempt command. 2054 */ 2055 static uint32_t 2056 nvmf_ns_reservation_get_unregistered_hostid(struct spdk_uuid *old_hostid_list, 2057 uint32_t old_num_hostid, 2058 struct spdk_uuid *remaining_hostid_list, 2059 uint32_t remaining_num_hostid) 2060 { 2061 struct spdk_uuid temp_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2062 uint32_t i, j, num_hostid = 0; 2063 bool found; 2064 2065 if (!remaining_num_hostid) { 2066 return old_num_hostid; 2067 } 2068 2069 for (i = 0; i < old_num_hostid; i++) { 2070 found = false; 2071 for (j = 0; j < remaining_num_hostid; j++) { 2072 if (!spdk_uuid_compare(&old_hostid_list[i], &remaining_hostid_list[j])) { 2073 found = true; 2074 break; 2075 } 2076 } 2077 if (!found) { 2078 spdk_uuid_copy(&temp_hostid_list[num_hostid++], &old_hostid_list[i]); 2079 } 2080 } 2081 2082 if (num_hostid) { 2083 memcpy(old_hostid_list, temp_hostid_list, sizeof(struct spdk_uuid) * num_hostid); 2084 } 2085 2086 return num_hostid; 2087 } 2088 2089 /* current reservation type is all registrants or not */ 2090 static bool 2091 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns) 2092 { 2093 return (ns->rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS || 2094 ns->rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS); 2095 } 2096 2097 /* current registrant is reservation holder or not */ 2098 static bool 2099 nvmf_ns_reservation_registrant_is_holder(struct spdk_nvmf_ns *ns, 2100 struct spdk_nvmf_registrant *reg) 2101 { 2102 if (!reg) { 2103 return false; 2104 } 2105 2106 if (nvmf_ns_reservation_all_registrants_type(ns)) { 2107 return true; 2108 } 2109 2110 return (ns->holder == reg); 2111 } 2112 2113 static int 2114 nvmf_ns_reservation_add_registrant(struct spdk_nvmf_ns *ns, 2115 struct spdk_nvmf_ctrlr *ctrlr, 2116 uint64_t nrkey) 2117 { 2118 struct spdk_nvmf_registrant *reg; 2119 2120 reg = calloc(1, sizeof(*reg)); 2121 if (!reg) { 2122 return -ENOMEM; 2123 } 2124 2125 reg->rkey = nrkey; 2126 /* set hostid for the registrant */ 2127 spdk_uuid_copy(®->hostid, &ctrlr->hostid); 2128 TAILQ_INSERT_TAIL(&ns->registrants, reg, link); 2129 ns->gen++; 2130 2131 return 0; 2132 } 2133 2134 static void 2135 nvmf_ns_reservation_release_reservation(struct spdk_nvmf_ns *ns) 2136 { 2137 ns->rtype = 0; 2138 ns->crkey = 0; 2139 ns->holder = NULL; 2140 } 2141 2142 /* release the reservation if the last registrant was removed */ 2143 static void 2144 nvmf_ns_reservation_check_release_on_remove_registrant(struct spdk_nvmf_ns *ns, 2145 struct spdk_nvmf_registrant *reg) 2146 { 2147 struct spdk_nvmf_registrant *next_reg; 2148 2149 /* no reservation holder */ 2150 if (!ns->holder) { 2151 assert(ns->rtype == 0); 2152 return; 2153 } 2154 2155 next_reg = TAILQ_FIRST(&ns->registrants); 2156 if (next_reg && nvmf_ns_reservation_all_registrants_type(ns)) { 2157 /* the next valid registrant is the new holder now */ 2158 ns->holder = next_reg; 2159 } else if (nvmf_ns_reservation_registrant_is_holder(ns, reg)) { 2160 /* release the reservation */ 2161 nvmf_ns_reservation_release_reservation(ns); 2162 } 2163 } 2164 2165 static void 2166 nvmf_ns_reservation_remove_registrant(struct spdk_nvmf_ns *ns, 2167 struct spdk_nvmf_registrant *reg) 2168 { 2169 TAILQ_REMOVE(&ns->registrants, reg, link); 2170 nvmf_ns_reservation_check_release_on_remove_registrant(ns, reg); 2171 free(reg); 2172 ns->gen++; 2173 return; 2174 } 2175 2176 static uint32_t 2177 nvmf_ns_reservation_remove_registrants_by_key(struct spdk_nvmf_ns *ns, 2178 uint64_t rkey) 2179 { 2180 struct spdk_nvmf_registrant *reg, *tmp; 2181 uint32_t count = 0; 2182 2183 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2184 if (reg->rkey == rkey) { 2185 nvmf_ns_reservation_remove_registrant(ns, reg); 2186 count++; 2187 } 2188 } 2189 return count; 2190 } 2191 2192 static uint32_t 2193 nvmf_ns_reservation_remove_all_other_registrants(struct spdk_nvmf_ns *ns, 2194 struct spdk_nvmf_registrant *reg) 2195 { 2196 struct spdk_nvmf_registrant *reg_tmp, *reg_tmp2; 2197 uint32_t count = 0; 2198 2199 TAILQ_FOREACH_SAFE(reg_tmp, &ns->registrants, link, reg_tmp2) { 2200 if (reg_tmp != reg) { 2201 nvmf_ns_reservation_remove_registrant(ns, reg_tmp); 2202 count++; 2203 } 2204 } 2205 return count; 2206 } 2207 2208 static uint32_t 2209 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns) 2210 { 2211 struct spdk_nvmf_registrant *reg, *reg_tmp; 2212 uint32_t count = 0; 2213 2214 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, reg_tmp) { 2215 nvmf_ns_reservation_remove_registrant(ns, reg); 2216 count++; 2217 } 2218 return count; 2219 } 2220 2221 static void 2222 nvmf_ns_reservation_acquire_reservation(struct spdk_nvmf_ns *ns, uint64_t rkey, 2223 enum spdk_nvme_reservation_type rtype, 2224 struct spdk_nvmf_registrant *holder) 2225 { 2226 ns->rtype = rtype; 2227 ns->crkey = rkey; 2228 assert(ns->holder == NULL); 2229 ns->holder = holder; 2230 } 2231 2232 static bool 2233 nvmf_ns_reservation_register(struct spdk_nvmf_ns *ns, 2234 struct spdk_nvmf_ctrlr *ctrlr, 2235 struct spdk_nvmf_request *req) 2236 { 2237 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2238 uint8_t rrega, iekey, cptpl, rtype; 2239 struct spdk_nvme_reservation_register_data key; 2240 struct spdk_nvmf_registrant *reg; 2241 uint8_t status = SPDK_NVME_SC_SUCCESS; 2242 bool update_sgroup = false; 2243 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2244 uint32_t num_hostid = 0; 2245 int rc; 2246 2247 rrega = cmd->cdw10_bits.resv_register.rrega; 2248 iekey = cmd->cdw10_bits.resv_register.iekey; 2249 cptpl = cmd->cdw10_bits.resv_register.cptpl; 2250 2251 if (req->data && req->length >= sizeof(key)) { 2252 memcpy(&key, req->data, sizeof(key)); 2253 } else { 2254 SPDK_ERRLOG("No key provided. Failing request.\n"); 2255 status = SPDK_NVME_SC_INVALID_FIELD; 2256 goto exit; 2257 } 2258 2259 SPDK_DEBUGLOG(nvmf, "REGISTER: RREGA %u, IEKEY %u, CPTPL %u, " 2260 "NRKEY 0x%"PRIx64", NRKEY 0x%"PRIx64"\n", 2261 rrega, iekey, cptpl, key.crkey, key.nrkey); 2262 2263 if (cptpl == SPDK_NVME_RESERVE_PTPL_CLEAR_POWER_ON) { 2264 /* Ture to OFF state, and need to be updated in the configuration file */ 2265 if (ns->ptpl_activated) { 2266 ns->ptpl_activated = 0; 2267 update_sgroup = true; 2268 } 2269 } else if (cptpl == SPDK_NVME_RESERVE_PTPL_PERSIST_POWER_LOSS) { 2270 if (ns->ptpl_file == NULL) { 2271 status = SPDK_NVME_SC_INVALID_FIELD; 2272 goto exit; 2273 } else if (ns->ptpl_activated == 0) { 2274 ns->ptpl_activated = 1; 2275 update_sgroup = true; 2276 } 2277 } 2278 2279 /* current Host Identifier has registrant or not */ 2280 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2281 2282 switch (rrega) { 2283 case SPDK_NVME_RESERVE_REGISTER_KEY: 2284 if (!reg) { 2285 /* register new controller */ 2286 if (key.nrkey == 0) { 2287 SPDK_ERRLOG("Can't register zeroed new key\n"); 2288 status = SPDK_NVME_SC_INVALID_FIELD; 2289 goto exit; 2290 } 2291 rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey); 2292 if (rc < 0) { 2293 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2294 goto exit; 2295 } 2296 update_sgroup = true; 2297 } else { 2298 /* register with same key is not an error */ 2299 if (reg->rkey != key.nrkey) { 2300 SPDK_ERRLOG("The same host already register a " 2301 "key with 0x%"PRIx64"\n", 2302 reg->rkey); 2303 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2304 goto exit; 2305 } 2306 } 2307 break; 2308 case SPDK_NVME_RESERVE_UNREGISTER_KEY: 2309 if (!reg || (!iekey && reg->rkey != key.crkey)) { 2310 SPDK_ERRLOG("No registrant or current key doesn't match " 2311 "with existing registrant key\n"); 2312 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2313 goto exit; 2314 } 2315 2316 rtype = ns->rtype; 2317 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2318 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2319 &ctrlr->hostid); 2320 2321 nvmf_ns_reservation_remove_registrant(ns, reg); 2322 2323 if (!ns->rtype && num_hostid && (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_REG_ONLY || 2324 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY)) { 2325 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2326 hostid_list, 2327 num_hostid, 2328 SPDK_NVME_RESERVATION_RELEASED); 2329 } 2330 update_sgroup = true; 2331 break; 2332 case SPDK_NVME_RESERVE_REPLACE_KEY: 2333 if (!reg || (!iekey && reg->rkey != key.crkey)) { 2334 SPDK_ERRLOG("No registrant or current key doesn't match " 2335 "with existing registrant key\n"); 2336 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2337 goto exit; 2338 } 2339 if (key.nrkey == 0) { 2340 SPDK_ERRLOG("Can't register zeroed new key\n"); 2341 status = SPDK_NVME_SC_INVALID_FIELD; 2342 goto exit; 2343 } 2344 reg->rkey = key.nrkey; 2345 update_sgroup = true; 2346 break; 2347 default: 2348 status = SPDK_NVME_SC_INVALID_FIELD; 2349 goto exit; 2350 } 2351 2352 exit: 2353 if (update_sgroup) { 2354 rc = nvmf_ns_update_reservation_info(ns); 2355 if (rc != 0) { 2356 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2357 } 2358 } 2359 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2360 req->rsp->nvme_cpl.status.sc = status; 2361 return update_sgroup; 2362 } 2363 2364 static bool 2365 nvmf_ns_reservation_acquire(struct spdk_nvmf_ns *ns, 2366 struct spdk_nvmf_ctrlr *ctrlr, 2367 struct spdk_nvmf_request *req) 2368 { 2369 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2370 uint8_t racqa, iekey, rtype; 2371 struct spdk_nvme_reservation_acquire_data key; 2372 struct spdk_nvmf_registrant *reg; 2373 bool all_regs = false; 2374 uint32_t count = 0; 2375 bool update_sgroup = true; 2376 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2377 uint32_t num_hostid = 0; 2378 struct spdk_uuid new_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2379 uint32_t new_num_hostid = 0; 2380 bool reservation_released = false; 2381 uint8_t status = SPDK_NVME_SC_SUCCESS; 2382 2383 racqa = cmd->cdw10_bits.resv_acquire.racqa; 2384 iekey = cmd->cdw10_bits.resv_acquire.iekey; 2385 rtype = cmd->cdw10_bits.resv_acquire.rtype; 2386 2387 if (req->data && req->length >= sizeof(key)) { 2388 memcpy(&key, req->data, sizeof(key)); 2389 } else { 2390 SPDK_ERRLOG("No key provided. Failing request.\n"); 2391 status = SPDK_NVME_SC_INVALID_FIELD; 2392 goto exit; 2393 } 2394 2395 SPDK_DEBUGLOG(nvmf, "ACQUIRE: RACQA %u, IEKEY %u, RTYPE %u, " 2396 "NRKEY 0x%"PRIx64", PRKEY 0x%"PRIx64"\n", 2397 racqa, iekey, rtype, key.crkey, key.prkey); 2398 2399 if (iekey || rtype > SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) { 2400 SPDK_ERRLOG("Ignore existing key field set to 1\n"); 2401 status = SPDK_NVME_SC_INVALID_FIELD; 2402 update_sgroup = false; 2403 goto exit; 2404 } 2405 2406 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2407 /* must be registrant and CRKEY must match */ 2408 if (!reg || reg->rkey != key.crkey) { 2409 SPDK_ERRLOG("No registrant or current key doesn't match " 2410 "with existing registrant key\n"); 2411 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2412 update_sgroup = false; 2413 goto exit; 2414 } 2415 2416 all_regs = nvmf_ns_reservation_all_registrants_type(ns); 2417 2418 switch (racqa) { 2419 case SPDK_NVME_RESERVE_ACQUIRE: 2420 /* it's not an error for the holder to acquire same reservation type again */ 2421 if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && ns->rtype == rtype) { 2422 /* do nothing */ 2423 update_sgroup = false; 2424 } else if (ns->holder == NULL) { 2425 /* fisrt time to acquire the reservation */ 2426 nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg); 2427 } else { 2428 SPDK_ERRLOG("Invalid rtype or current registrant is not holder\n"); 2429 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2430 update_sgroup = false; 2431 goto exit; 2432 } 2433 break; 2434 case SPDK_NVME_RESERVE_PREEMPT: 2435 /* no reservation holder */ 2436 if (!ns->holder) { 2437 /* unregister with PRKEY */ 2438 nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2439 break; 2440 } 2441 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2442 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2443 &ctrlr->hostid); 2444 2445 /* only 1 reservation holder and reservation key is valid */ 2446 if (!all_regs) { 2447 /* preempt itself */ 2448 if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && 2449 ns->crkey == key.prkey) { 2450 ns->rtype = rtype; 2451 reservation_released = true; 2452 break; 2453 } 2454 2455 if (ns->crkey == key.prkey) { 2456 nvmf_ns_reservation_remove_registrant(ns, ns->holder); 2457 nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg); 2458 reservation_released = true; 2459 } else if (key.prkey != 0) { 2460 nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2461 } else { 2462 /* PRKEY is zero */ 2463 SPDK_ERRLOG("Current PRKEY is zero\n"); 2464 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2465 update_sgroup = false; 2466 goto exit; 2467 } 2468 } else { 2469 /* release all other registrants except for the current one */ 2470 if (key.prkey == 0) { 2471 nvmf_ns_reservation_remove_all_other_registrants(ns, reg); 2472 assert(ns->holder == reg); 2473 } else { 2474 count = nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey); 2475 if (count == 0) { 2476 SPDK_ERRLOG("PRKEY doesn't match any registrant\n"); 2477 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2478 update_sgroup = false; 2479 goto exit; 2480 } 2481 } 2482 } 2483 break; 2484 default: 2485 status = SPDK_NVME_SC_INVALID_FIELD; 2486 update_sgroup = false; 2487 break; 2488 } 2489 2490 exit: 2491 if (update_sgroup && racqa == SPDK_NVME_RESERVE_PREEMPT) { 2492 new_num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, new_hostid_list, 2493 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2494 &ctrlr->hostid); 2495 /* Preempt notification occurs on the unregistered controllers 2496 * other than the controller who issued the command. 2497 */ 2498 num_hostid = nvmf_ns_reservation_get_unregistered_hostid(hostid_list, 2499 num_hostid, 2500 new_hostid_list, 2501 new_num_hostid); 2502 if (num_hostid) { 2503 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2504 hostid_list, 2505 num_hostid, 2506 SPDK_NVME_REGISTRATION_PREEMPTED); 2507 2508 } 2509 /* Reservation released notification occurs on the 2510 * controllers which are the remaining registrants other than 2511 * the controller who issued the command. 2512 */ 2513 if (reservation_released && new_num_hostid) { 2514 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2515 new_hostid_list, 2516 new_num_hostid, 2517 SPDK_NVME_RESERVATION_RELEASED); 2518 2519 } 2520 } 2521 if (update_sgroup && ns->ptpl_activated) { 2522 if (nvmf_ns_update_reservation_info(ns)) { 2523 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2524 } 2525 } 2526 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2527 req->rsp->nvme_cpl.status.sc = status; 2528 return update_sgroup; 2529 } 2530 2531 static bool 2532 nvmf_ns_reservation_release(struct spdk_nvmf_ns *ns, 2533 struct spdk_nvmf_ctrlr *ctrlr, 2534 struct spdk_nvmf_request *req) 2535 { 2536 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2537 uint8_t rrela, iekey, rtype; 2538 struct spdk_nvmf_registrant *reg; 2539 uint64_t crkey; 2540 uint8_t status = SPDK_NVME_SC_SUCCESS; 2541 bool update_sgroup = true; 2542 struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 2543 uint32_t num_hostid = 0; 2544 2545 rrela = cmd->cdw10_bits.resv_release.rrela; 2546 iekey = cmd->cdw10_bits.resv_release.iekey; 2547 rtype = cmd->cdw10_bits.resv_release.rtype; 2548 2549 if (req->data && req->length >= sizeof(crkey)) { 2550 memcpy(&crkey, req->data, sizeof(crkey)); 2551 } else { 2552 SPDK_ERRLOG("No key provided. Failing request.\n"); 2553 status = SPDK_NVME_SC_INVALID_FIELD; 2554 goto exit; 2555 } 2556 2557 SPDK_DEBUGLOG(nvmf, "RELEASE: RRELA %u, IEKEY %u, RTYPE %u, " 2558 "CRKEY 0x%"PRIx64"\n", rrela, iekey, rtype, crkey); 2559 2560 if (iekey) { 2561 SPDK_ERRLOG("Ignore existing key field set to 1\n"); 2562 status = SPDK_NVME_SC_INVALID_FIELD; 2563 update_sgroup = false; 2564 goto exit; 2565 } 2566 2567 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid); 2568 if (!reg || reg->rkey != crkey) { 2569 SPDK_ERRLOG("No registrant or current key doesn't match " 2570 "with existing registrant key\n"); 2571 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 2572 update_sgroup = false; 2573 goto exit; 2574 } 2575 2576 num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list, 2577 SPDK_NVMF_MAX_NUM_REGISTRANTS, 2578 &ctrlr->hostid); 2579 2580 switch (rrela) { 2581 case SPDK_NVME_RESERVE_RELEASE: 2582 if (!ns->holder) { 2583 SPDK_DEBUGLOG(nvmf, "RELEASE: no holder\n"); 2584 update_sgroup = false; 2585 goto exit; 2586 } 2587 if (ns->rtype != rtype) { 2588 SPDK_ERRLOG("Type doesn't match\n"); 2589 status = SPDK_NVME_SC_INVALID_FIELD; 2590 update_sgroup = false; 2591 goto exit; 2592 } 2593 if (!nvmf_ns_reservation_registrant_is_holder(ns, reg)) { 2594 /* not the reservation holder, this isn't an error */ 2595 update_sgroup = false; 2596 goto exit; 2597 } 2598 2599 rtype = ns->rtype; 2600 nvmf_ns_reservation_release_reservation(ns); 2601 2602 if (num_hostid && rtype != SPDK_NVME_RESERVE_WRITE_EXCLUSIVE && 2603 rtype != SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 2604 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2605 hostid_list, 2606 num_hostid, 2607 SPDK_NVME_RESERVATION_RELEASED); 2608 } 2609 break; 2610 case SPDK_NVME_RESERVE_CLEAR: 2611 nvmf_ns_reservation_clear_all_registrants(ns); 2612 if (num_hostid) { 2613 nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns, 2614 hostid_list, 2615 num_hostid, 2616 SPDK_NVME_RESERVATION_PREEMPTED); 2617 } 2618 break; 2619 default: 2620 status = SPDK_NVME_SC_INVALID_FIELD; 2621 update_sgroup = false; 2622 goto exit; 2623 } 2624 2625 exit: 2626 if (update_sgroup && ns->ptpl_activated) { 2627 if (nvmf_ns_update_reservation_info(ns)) { 2628 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2629 } 2630 } 2631 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2632 req->rsp->nvme_cpl.status.sc = status; 2633 return update_sgroup; 2634 } 2635 2636 static void 2637 nvmf_ns_reservation_report(struct spdk_nvmf_ns *ns, 2638 struct spdk_nvmf_ctrlr *ctrlr, 2639 struct spdk_nvmf_request *req) 2640 { 2641 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2642 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2643 struct spdk_nvmf_ctrlr *ctrlr_tmp; 2644 struct spdk_nvmf_registrant *reg, *tmp; 2645 struct spdk_nvme_reservation_status_extended_data *status_data; 2646 struct spdk_nvme_registered_ctrlr_extended_data *ctrlr_data; 2647 uint8_t *payload; 2648 uint32_t len, count = 0; 2649 uint32_t regctl = 0; 2650 uint8_t status = SPDK_NVME_SC_SUCCESS; 2651 2652 if (req->data == NULL) { 2653 SPDK_ERRLOG("No data transfer specified for request. " 2654 " Unable to transfer back response.\n"); 2655 status = SPDK_NVME_SC_INVALID_FIELD; 2656 goto exit; 2657 } 2658 2659 if (!cmd->cdw11_bits.resv_report.eds) { 2660 SPDK_ERRLOG("NVMeoF uses extended controller data structure, " 2661 "please set EDS bit in cdw11 and try again\n"); 2662 status = SPDK_NVME_SC_HOSTID_INCONSISTENT_FORMAT; 2663 goto exit; 2664 } 2665 2666 /* Get number of registerd controllers, one Host may have more than 2667 * one controller based on different ports. 2668 */ 2669 TAILQ_FOREACH(ctrlr_tmp, &subsystem->ctrlrs, link) { 2670 reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr_tmp->hostid); 2671 if (reg) { 2672 regctl++; 2673 } 2674 } 2675 2676 len = sizeof(*status_data) + sizeof(*ctrlr_data) * regctl; 2677 payload = calloc(1, len); 2678 if (!payload) { 2679 status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2680 goto exit; 2681 } 2682 2683 status_data = (struct spdk_nvme_reservation_status_extended_data *)payload; 2684 status_data->data.gen = ns->gen; 2685 status_data->data.rtype = ns->rtype; 2686 status_data->data.regctl = regctl; 2687 status_data->data.ptpls = ns->ptpl_activated; 2688 2689 TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) { 2690 assert(count <= regctl); 2691 ctrlr_data = (struct spdk_nvme_registered_ctrlr_extended_data *) 2692 (payload + sizeof(*status_data) + sizeof(*ctrlr_data) * count); 2693 /* Set to 0xffffh for dynamic controller */ 2694 ctrlr_data->cntlid = 0xffff; 2695 ctrlr_data->rcsts.status = (ns->holder == reg) ? true : false; 2696 ctrlr_data->rkey = reg->rkey; 2697 spdk_uuid_copy((struct spdk_uuid *)ctrlr_data->hostid, ®->hostid); 2698 count++; 2699 } 2700 2701 memcpy(req->data, payload, spdk_min(len, (cmd->cdw10 + 1) * sizeof(uint32_t))); 2702 free(payload); 2703 2704 exit: 2705 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2706 req->rsp->nvme_cpl.status.sc = status; 2707 return; 2708 } 2709 2710 static void 2711 nvmf_ns_reservation_complete(void *ctx) 2712 { 2713 struct spdk_nvmf_request *req = ctx; 2714 2715 spdk_nvmf_request_complete(req); 2716 } 2717 2718 static void 2719 _nvmf_ns_reservation_update_done(struct spdk_nvmf_subsystem *subsystem, 2720 void *cb_arg, int status) 2721 { 2722 struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)cb_arg; 2723 struct spdk_nvmf_poll_group *group = req->qpair->group; 2724 2725 spdk_thread_send_msg(group->thread, nvmf_ns_reservation_complete, req); 2726 } 2727 2728 void 2729 nvmf_ns_reservation_request(void *ctx) 2730 { 2731 struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)ctx; 2732 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2733 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2734 struct subsystem_update_ns_ctx *update_ctx; 2735 uint32_t nsid; 2736 struct spdk_nvmf_ns *ns; 2737 bool update_sgroup = false; 2738 2739 nsid = cmd->nsid; 2740 ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid); 2741 assert(ns != NULL); 2742 2743 switch (cmd->opc) { 2744 case SPDK_NVME_OPC_RESERVATION_REGISTER: 2745 update_sgroup = nvmf_ns_reservation_register(ns, ctrlr, req); 2746 break; 2747 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 2748 update_sgroup = nvmf_ns_reservation_acquire(ns, ctrlr, req); 2749 break; 2750 case SPDK_NVME_OPC_RESERVATION_RELEASE: 2751 update_sgroup = nvmf_ns_reservation_release(ns, ctrlr, req); 2752 break; 2753 case SPDK_NVME_OPC_RESERVATION_REPORT: 2754 nvmf_ns_reservation_report(ns, ctrlr, req); 2755 break; 2756 default: 2757 break; 2758 } 2759 2760 /* update reservation information to subsystem's poll group */ 2761 if (update_sgroup) { 2762 update_ctx = calloc(1, sizeof(*update_ctx)); 2763 if (update_ctx == NULL) { 2764 SPDK_ERRLOG("Can't alloc subsystem poll group update context\n"); 2765 goto update_done; 2766 } 2767 update_ctx->subsystem = ctrlr->subsys; 2768 update_ctx->cb_fn = _nvmf_ns_reservation_update_done; 2769 update_ctx->cb_arg = req; 2770 2771 nvmf_subsystem_update_ns(ctrlr->subsys, subsystem_update_ns_done, update_ctx); 2772 return; 2773 } 2774 2775 update_done: 2776 _nvmf_ns_reservation_update_done(ctrlr->subsys, (void *)req, 0); 2777 } 2778 2779 int 2780 spdk_nvmf_subsystem_set_ana_reporting(struct spdk_nvmf_subsystem *subsystem, 2781 bool ana_reporting) 2782 { 2783 if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) { 2784 return -EAGAIN; 2785 } 2786 2787 subsystem->flags.ana_reporting = ana_reporting; 2788 2789 return 0; 2790 } 2791 2792 struct subsystem_listener_update_ctx { 2793 struct spdk_nvmf_subsystem_listener *listener; 2794 2795 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 2796 void *cb_arg; 2797 }; 2798 2799 static void 2800 subsystem_listener_update_done(struct spdk_io_channel_iter *i, int status) 2801 { 2802 struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 2803 2804 if (ctx->cb_fn) { 2805 ctx->cb_fn(ctx->cb_arg, status); 2806 } 2807 free(ctx); 2808 } 2809 2810 static void 2811 subsystem_listener_update_on_pg(struct spdk_io_channel_iter *i) 2812 { 2813 struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 2814 struct spdk_nvmf_subsystem_listener *listener; 2815 struct spdk_nvmf_poll_group *group; 2816 struct spdk_nvmf_ctrlr *ctrlr; 2817 2818 listener = ctx->listener; 2819 group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i)); 2820 2821 TAILQ_FOREACH(ctrlr, &listener->subsystem->ctrlrs, link) { 2822 if (ctrlr->admin_qpair->group == group && ctrlr->listener == listener) { 2823 nvmf_ctrlr_async_event_ana_change_notice(ctrlr); 2824 } 2825 } 2826 2827 spdk_for_each_channel_continue(i, 0); 2828 } 2829 2830 void 2831 nvmf_subsystem_set_ana_state(struct spdk_nvmf_subsystem *subsystem, 2832 const struct spdk_nvme_transport_id *trid, 2833 enum spdk_nvme_ana_state ana_state, 2834 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, void *cb_arg) 2835 { 2836 struct spdk_nvmf_subsystem_listener *listener; 2837 struct subsystem_listener_update_ctx *ctx; 2838 2839 assert(cb_fn != NULL); 2840 assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE || 2841 subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED); 2842 2843 if (!subsystem->flags.ana_reporting) { 2844 SPDK_ERRLOG("ANA reporting is disabled\n"); 2845 cb_fn(cb_arg, -EINVAL); 2846 return; 2847 } 2848 2849 /* ANA Change state is not used, ANA Persistent Loss state 2850 * is not supported yet. 2851 */ 2852 if (!(ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE || 2853 ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE || 2854 ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE)) { 2855 SPDK_ERRLOG("ANA state %d is not supported\n", ana_state); 2856 cb_fn(cb_arg, -ENOTSUP); 2857 return; 2858 } 2859 2860 listener = nvmf_subsystem_find_listener(subsystem, trid); 2861 if (!listener) { 2862 SPDK_ERRLOG("Unable to find listener.\n"); 2863 cb_fn(cb_arg, -EINVAL); 2864 return; 2865 } 2866 2867 if (listener->ana_state == ana_state) { 2868 cb_fn(cb_arg, 0); 2869 return; 2870 } 2871 2872 ctx = calloc(1, sizeof(*ctx)); 2873 if (!ctx) { 2874 SPDK_ERRLOG("Unable to allocate context\n"); 2875 cb_fn(cb_arg, -ENOMEM); 2876 return; 2877 } 2878 2879 listener->ana_state = ana_state; 2880 listener->ana_state_change_count++; 2881 2882 ctx->listener = listener; 2883 ctx->cb_fn = cb_fn; 2884 ctx->cb_arg = cb_arg; 2885 2886 spdk_for_each_channel(subsystem->tgt, 2887 subsystem_listener_update_on_pg, 2888 ctx, 2889 subsystem_listener_update_done); 2890 } 2891