1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2017 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2021, 2023 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/bdev.h" 13 #include "spdk/bdev_zone.h" 14 #include "spdk/bit_array.h" 15 #include "spdk/endian.h" 16 #include "spdk/thread.h" 17 #include "spdk/nvme_spec.h" 18 #include "spdk/nvmf_cmd.h" 19 #include "spdk/string.h" 20 #include "spdk/util.h" 21 #include "spdk/version.h" 22 #include "spdk/log.h" 23 #include "spdk_internal/usdt.h" 24 25 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000 26 #define NVMF_DISC_KATO_IN_MS 120000 27 #define KAS_TIME_UNIT_IN_MS 100 28 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS) 29 30 #define NVMF_CC_RESET_SHN_TIMEOUT_IN_MS 10000 31 32 #define NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + 5000) 33 34 #define DUPLICATE_QID_RETRY_US 1000 35 36 /* 37 * Report the SPDK version as the firmware revision. 38 * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts. 39 */ 40 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING 41 42 #define ANA_TRANSITION_TIME_IN_SEC 10 43 44 #define NVMF_ABORT_COMMAND_LIMIT 3 45 46 /* 47 * Support for custom admin command handlers 48 */ 49 struct spdk_nvmf_custom_admin_cmd { 50 spdk_nvmf_custom_cmd_hdlr hdlr; 51 uint32_t nsid; /* nsid to forward */ 52 }; 53 54 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1]; 55 56 static void _nvmf_request_complete(void *ctx); 57 int nvmf_passthru_admin_cmd_for_ctrlr(struct spdk_nvmf_request *req, struct spdk_nvmf_ctrlr *ctrlr); 58 59 static inline void 60 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp, 61 uint8_t iattr, uint16_t ipo) 62 { 63 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 64 rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 65 rsp->status_code_specific.invalid.iattr = iattr; 66 rsp->status_code_specific.invalid.ipo = ipo; 67 } 68 69 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field) \ 70 nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field)) 71 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field) \ 72 nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field)) 73 74 75 static void 76 nvmf_ctrlr_stop_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr) 77 { 78 if (!ctrlr) { 79 SPDK_ERRLOG("Controller is NULL\n"); 80 return; 81 } 82 83 if (ctrlr->keep_alive_poller == NULL) { 84 return; 85 } 86 87 SPDK_DEBUGLOG(nvmf, "Stop keep alive poller\n"); 88 spdk_poller_unregister(&ctrlr->keep_alive_poller); 89 } 90 91 static void 92 nvmf_ctrlr_stop_association_timer(struct spdk_nvmf_ctrlr *ctrlr) 93 { 94 if (!ctrlr) { 95 SPDK_ERRLOG("Controller is NULL\n"); 96 assert(false); 97 return; 98 } 99 100 if (ctrlr->association_timer == NULL) { 101 return; 102 } 103 104 SPDK_DEBUGLOG(nvmf, "Stop association timer\n"); 105 spdk_poller_unregister(&ctrlr->association_timer); 106 } 107 108 static void 109 nvmf_ctrlr_disconnect_qpairs_done(struct spdk_io_channel_iter *i, int status) 110 { 111 if (status == 0) { 112 SPDK_DEBUGLOG(nvmf, "ctrlr disconnect qpairs complete successfully\n"); 113 } else { 114 SPDK_ERRLOG("Fail to disconnect ctrlr qpairs\n"); 115 } 116 } 117 118 static int 119 _nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i, bool include_admin) 120 { 121 int rc = 0; 122 struct spdk_nvmf_ctrlr *ctrlr; 123 struct spdk_nvmf_qpair *qpair, *temp_qpair; 124 struct spdk_io_channel *ch; 125 struct spdk_nvmf_poll_group *group; 126 127 ctrlr = spdk_io_channel_iter_get_ctx(i); 128 ch = spdk_io_channel_iter_get_channel(i); 129 group = spdk_io_channel_get_ctx(ch); 130 131 TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, temp_qpair) { 132 if (qpair->ctrlr == ctrlr && (include_admin || !nvmf_qpair_is_admin_queue(qpair))) { 133 rc = spdk_nvmf_qpair_disconnect(qpair, NULL, NULL); 134 if (rc) { 135 if (rc == -EINPROGRESS) { 136 rc = 0; 137 } else { 138 SPDK_ERRLOG("Qpair disconnect failed\n"); 139 return rc; 140 } 141 } 142 } 143 } 144 145 return rc; 146 } 147 148 static void 149 nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i) 150 { 151 spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, true)); 152 } 153 154 static void 155 nvmf_ctrlr_disconnect_io_qpairs_on_pg(struct spdk_io_channel_iter *i) 156 { 157 spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, false)); 158 } 159 160 static int 161 nvmf_ctrlr_keep_alive_poll(void *ctx) 162 { 163 uint64_t keep_alive_timeout_tick; 164 uint64_t now = spdk_get_ticks(); 165 struct spdk_nvmf_ctrlr *ctrlr = ctx; 166 167 if (ctrlr->in_destruct) { 168 nvmf_ctrlr_stop_keep_alive_timer(ctrlr); 169 return SPDK_POLLER_IDLE; 170 } 171 172 SPDK_DEBUGLOG(nvmf, "Polling ctrlr keep alive timeout\n"); 173 174 /* If the Keep alive feature is in use and the timer expires */ 175 keep_alive_timeout_tick = ctrlr->last_keep_alive_tick + 176 ctrlr->feat.keep_alive_timer.bits.kato * spdk_get_ticks_hz() / UINT64_C(1000); 177 if (now > keep_alive_timeout_tick) { 178 SPDK_NOTICELOG("Disconnecting host %s from subsystem %s due to keep alive timeout.\n", 179 ctrlr->hostnqn, ctrlr->subsys->subnqn); 180 /* set the Controller Fatal Status bit to '1' */ 181 if (ctrlr->vcprop.csts.bits.cfs == 0) { 182 nvmf_ctrlr_set_fatal_status(ctrlr); 183 184 /* 185 * disconnect qpairs, terminate Transport connection 186 * destroy ctrlr, break the host to controller association 187 * disconnect qpairs with qpair->ctrlr == ctrlr 188 */ 189 spdk_for_each_channel(ctrlr->subsys->tgt, 190 nvmf_ctrlr_disconnect_qpairs_on_pg, 191 ctrlr, 192 nvmf_ctrlr_disconnect_qpairs_done); 193 return SPDK_POLLER_BUSY; 194 } 195 } 196 197 return SPDK_POLLER_IDLE; 198 } 199 200 static void 201 nvmf_ctrlr_start_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr) 202 { 203 if (!ctrlr) { 204 SPDK_ERRLOG("Controller is NULL\n"); 205 return; 206 } 207 208 /* if cleared to 0 then the Keep Alive Timer is disabled */ 209 if (ctrlr->feat.keep_alive_timer.bits.kato != 0) { 210 211 ctrlr->last_keep_alive_tick = spdk_get_ticks(); 212 213 SPDK_DEBUGLOG(nvmf, "Ctrlr add keep alive poller\n"); 214 ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr, 215 ctrlr->feat.keep_alive_timer.bits.kato * 1000); 216 } 217 } 218 219 static int _retry_qid_check(void *ctx); 220 221 static void 222 ctrlr_add_qpair_and_send_rsp(struct spdk_nvmf_qpair *qpair, 223 struct spdk_nvmf_ctrlr *ctrlr, 224 struct spdk_nvmf_request *req) 225 { 226 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 227 228 assert(ctrlr->admin_qpair->group->thread == spdk_get_thread()); 229 230 if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) { 231 if (qpair->connect_req != NULL) { 232 SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid); 233 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 234 rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER; 235 qpair->connect_req = NULL; 236 qpair->ctrlr = NULL; 237 spdk_nvmf_request_complete(req); 238 } else { 239 SPDK_WARNLOG("Duplicate QID detected, re-check in %dus\n", 240 DUPLICATE_QID_RETRY_US); 241 qpair->connect_req = req; 242 /* Set qpair->ctrlr here so that we'll have it when the poller expires. */ 243 qpair->ctrlr = ctrlr; 244 req->poller = SPDK_POLLER_REGISTER(_retry_qid_check, qpair, 245 DUPLICATE_QID_RETRY_US); 246 } 247 return; 248 } 249 250 qpair->ctrlr = ctrlr; 251 spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid); 252 253 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 254 rsp->status_code_specific.success.cntlid = ctrlr->cntlid; 255 SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n", 256 rsp->status_code_specific.success.cntlid); 257 spdk_nvmf_request_complete(req); 258 259 SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn, 260 ctrlr->hostnqn); 261 } 262 263 static int 264 _retry_qid_check(void *ctx) 265 { 266 struct spdk_nvmf_qpair *qpair = ctx; 267 struct spdk_nvmf_request *req = qpair->connect_req; 268 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 269 270 spdk_poller_unregister(&req->poller); 271 ctrlr_add_qpair_and_send_rsp(qpair, ctrlr, req); 272 return SPDK_POLLER_BUSY; 273 } 274 275 static void 276 _nvmf_ctrlr_add_admin_qpair(void *ctx) 277 { 278 struct spdk_nvmf_request *req = ctx; 279 struct spdk_nvmf_qpair *qpair = req->qpair; 280 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 281 282 ctrlr->admin_qpair = qpair; 283 ctrlr->association_timeout = qpair->transport->opts.association_timeout; 284 nvmf_ctrlr_start_keep_alive_timer(ctrlr); 285 ctrlr_add_qpair_and_send_rsp(qpair, ctrlr, req); 286 } 287 288 static void 289 _nvmf_subsystem_add_ctrlr(void *ctx) 290 { 291 struct spdk_nvmf_request *req = ctx; 292 struct spdk_nvmf_qpair *qpair = req->qpair; 293 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 294 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 295 296 if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) { 297 SPDK_ERRLOG("Unable to add controller to subsystem\n"); 298 spdk_bit_array_free(&ctrlr->qpair_mask); 299 free(ctrlr); 300 qpair->ctrlr = NULL; 301 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 302 spdk_nvmf_request_complete(req); 303 return; 304 } 305 306 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req); 307 } 308 309 static void 310 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem, 311 struct spdk_nvmf_ctrlr_data *cdata) 312 { 313 cdata->aerl = SPDK_NVMF_MAX_ASYNC_EVENTS - 1; 314 cdata->kas = KAS_DEFAULT_VALUE; 315 cdata->vid = SPDK_PCI_VID_INTEL; 316 cdata->ssvid = SPDK_PCI_VID_INTEL; 317 /* INTEL OUI */ 318 cdata->ieee[0] = 0xe4; 319 cdata->ieee[1] = 0xd2; 320 cdata->ieee[2] = 0x5c; 321 cdata->oncs.compare = 1; 322 cdata->oncs.reservations = 1; 323 cdata->fuses.compare_and_write = 1; 324 cdata->oncs.copy = 1; 325 cdata->sgls.supported = 1; 326 cdata->sgls.keyed_sgl = 1; 327 cdata->sgls.sgl_offset = 1; 328 cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16; 329 cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16; 330 cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16; 331 cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */ 332 cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC; 333 cdata->nvmf_specific.msdbd = 1; 334 335 if (transport->ops->cdata_init) { 336 transport->ops->cdata_init(transport, subsystem, cdata); 337 } 338 } 339 340 static bool 341 nvmf_subsys_has_multi_iocs(struct spdk_nvmf_subsystem *subsystem) 342 { 343 struct spdk_nvmf_ns *ns; 344 uint32_t i; 345 346 for (i = 0; i < subsystem->max_nsid; i++) { 347 ns = subsystem->ns[i]; 348 if (ns && ns->bdev && spdk_bdev_is_zoned(ns->bdev)) { 349 return true; 350 } 351 } 352 return false; 353 } 354 355 static struct spdk_nvmf_ctrlr * 356 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem, 357 struct spdk_nvmf_request *req, 358 struct spdk_nvmf_fabric_connect_cmd *connect_cmd, 359 struct spdk_nvmf_fabric_connect_data *connect_data) 360 { 361 struct spdk_nvmf_ctrlr *ctrlr; 362 struct spdk_nvmf_transport *transport = req->qpair->transport; 363 struct spdk_nvme_transport_id listen_trid = {}; 364 bool subsys_has_multi_iocs = false; 365 366 ctrlr = calloc(1, sizeof(*ctrlr)); 367 if (ctrlr == NULL) { 368 SPDK_ERRLOG("Memory allocation failed\n"); 369 return NULL; 370 } 371 372 if (spdk_nvme_trtype_is_fabrics(transport->ops->type)) { 373 ctrlr->dynamic_ctrlr = true; 374 } else { 375 ctrlr->cntlid = connect_data->cntlid; 376 } 377 378 SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_create, ctrlr, subsystem->subnqn, 379 spdk_thread_get_id(req->qpair->group->thread)); 380 381 STAILQ_INIT(&ctrlr->async_events); 382 TAILQ_INIT(&ctrlr->log_head); 383 ctrlr->subsys = subsystem; 384 ctrlr->thread = req->qpair->group->thread; 385 ctrlr->disconnect_in_progress = false; 386 387 ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr); 388 if (!ctrlr->qpair_mask) { 389 SPDK_ERRLOG("Failed to allocate controller qpair mask\n"); 390 goto err_qpair_mask; 391 } 392 393 nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata); 394 395 /* 396 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units. 397 * If this field is cleared to 0h, then Keep Alive is not supported. 398 */ 399 if (ctrlr->cdata.kas) { 400 ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato, 401 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) * 402 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS; 403 } 404 405 ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1; 406 if (ctrlr->subsys->flags.ana_reporting) { 407 ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1; 408 } 409 ctrlr->feat.volatile_write_cache.bits.wce = 1; 410 /* Coalescing Disable */ 411 ctrlr->feat.interrupt_vector_configuration.bits.cd = 1; 412 413 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 414 /* 415 * If keep-alive timeout is not set, discovery controllers use some 416 * arbitrary high value in order to cleanup stale discovery sessions 417 * 418 * From the 1.0a nvme-of spec: 419 * "The Keep Alive command is reserved for 420 * Discovery controllers. A transport may specify a 421 * fixed Discovery controller activity timeout value 422 * (e.g., 2 minutes). If no commands are received 423 * by a Discovery controller within that time 424 * period, the controller may perform the 425 * actions for Keep Alive Timer expiration". 426 * 427 * From the 1.1 nvme-of spec: 428 * "A host requests an explicit persistent connection 429 * to a Discovery controller and Asynchronous Event Notifications from 430 * the Discovery controller on that persistent connection by specifying 431 * a non-zero Keep Alive Timer value in the Connect command." 432 * 433 * In case non-zero KATO is used, we enable discovery_log_change_notice 434 * otherwise we disable it and use default discovery controller KATO. 435 * KATO is in millisecond. 436 */ 437 if (ctrlr->feat.keep_alive_timer.bits.kato == 0) { 438 ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS; 439 ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0; 440 } else { 441 ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1; 442 } 443 } 444 445 /* Subtract 1 for admin queue, 1 for 0's based */ 446 ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 - 447 1; 448 ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 - 449 1; 450 451 spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid); 452 memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn)); 453 454 ctrlr->vcprop.cap.raw = 0; 455 ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */ 456 ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth - 457 1; /* max queue depth */ 458 ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */ 459 /* ready timeout - 500 msec units */ 460 ctrlr->vcprop.cap.bits.to = NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS / 500; 461 ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */ 462 subsys_has_multi_iocs = nvmf_subsys_has_multi_iocs(subsystem); 463 if (subsys_has_multi_iocs) { 464 ctrlr->vcprop.cap.bits.css = 465 SPDK_NVME_CAP_CSS_IOCS; /* One or more I/O command sets supported */ 466 } else { 467 ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */ 468 } 469 470 ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */ 471 ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */ 472 473 /* Version Supported: 1.3 */ 474 ctrlr->vcprop.vs.bits.mjr = 1; 475 ctrlr->vcprop.vs.bits.mnr = 3; 476 ctrlr->vcprop.vs.bits.ter = 0; 477 478 ctrlr->vcprop.cc.raw = 0; 479 ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */ 480 if (subsys_has_multi_iocs) { 481 ctrlr->vcprop.cc.bits.css = 482 SPDK_NVME_CC_CSS_IOCS; /* All supported I/O Command Sets */ 483 } 484 485 ctrlr->vcprop.csts.raw = 0; 486 ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */ 487 488 SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw); 489 SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw); 490 SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw); 491 SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw); 492 493 ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip; 494 495 if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) { 496 if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) { 497 SPDK_ERRLOG("Could not get listener transport ID\n"); 498 goto err_listener; 499 } 500 501 ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid); 502 if (!ctrlr->listener) { 503 SPDK_ERRLOG("Listener was not found\n"); 504 goto err_listener; 505 } 506 } 507 508 req->qpair->ctrlr = ctrlr; 509 spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req); 510 511 return ctrlr; 512 err_listener: 513 spdk_bit_array_free(&ctrlr->qpair_mask); 514 err_qpair_mask: 515 free(ctrlr); 516 return NULL; 517 } 518 519 static void 520 _nvmf_ctrlr_destruct(void *ctx) 521 { 522 struct spdk_nvmf_ctrlr *ctrlr = ctx; 523 struct spdk_nvmf_reservation_log *log, *log_tmp; 524 struct spdk_nvmf_async_event_completion *event, *event_tmp; 525 526 SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_destruct, ctrlr, ctrlr->subsys->subnqn, 527 spdk_thread_get_id(ctrlr->thread)); 528 529 assert(spdk_get_thread() == ctrlr->thread); 530 assert(ctrlr->in_destruct); 531 532 SPDK_DEBUGLOG(nvmf, "Destroy ctrlr 0x%hx\n", ctrlr->cntlid); 533 if (ctrlr->disconnect_in_progress) { 534 SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n"); 535 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr); 536 return; 537 } 538 539 nvmf_ctrlr_stop_keep_alive_timer(ctrlr); 540 nvmf_ctrlr_stop_association_timer(ctrlr); 541 spdk_bit_array_free(&ctrlr->qpair_mask); 542 543 TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) { 544 TAILQ_REMOVE(&ctrlr->log_head, log, link); 545 free(log); 546 } 547 STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) { 548 STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link); 549 free(event); 550 } 551 free(ctrlr); 552 } 553 554 void 555 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr) 556 { 557 nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr); 558 559 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr); 560 } 561 562 static void 563 nvmf_ctrlr_add_io_qpair(void *ctx) 564 { 565 struct spdk_nvmf_request *req = ctx; 566 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 567 struct spdk_nvmf_qpair *qpair = req->qpair; 568 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 569 struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair; 570 571 SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_io_qpair, ctrlr, req->qpair, req->qpair->qid, 572 spdk_thread_get_id(ctrlr->thread)); 573 574 /* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect. 575 * For error case, the value should be NULL. So set it to NULL at first. 576 */ 577 qpair->ctrlr = NULL; 578 579 /* Make sure the controller is not being destroyed. */ 580 if (ctrlr->in_destruct) { 581 SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n"); 582 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 583 goto end; 584 } 585 586 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 587 SPDK_ERRLOG("I/O connect not allowed on discovery controller\n"); 588 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 589 goto end; 590 } 591 592 if (!ctrlr->vcprop.cc.bits.en) { 593 SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n"); 594 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 595 goto end; 596 } 597 598 if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) { 599 SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n", 600 ctrlr->vcprop.cc.bits.iosqes); 601 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 602 goto end; 603 } 604 605 if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) { 606 SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n", 607 ctrlr->vcprop.cc.bits.iocqes); 608 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 609 goto end; 610 } 611 612 if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) { 613 /* There is a chance that admin qpair is being destroyed at this moment due to e.g. 614 * expired keep alive timer. Part of the qpair destruction process is change of qpair's 615 * state to DEACTIVATING and removing it from poll group */ 616 SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group); 617 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 618 goto end; 619 } 620 621 /* check if we would exceed ctrlr connection limit */ 622 if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) { 623 SPDK_ERRLOG("Requested QID %u but Max QID is %u\n", 624 qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1); 625 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 626 rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER; 627 goto end; 628 } 629 630 ctrlr_add_qpair_and_send_rsp(qpair, ctrlr, req); 631 return; 632 end: 633 spdk_nvmf_request_complete(req); 634 } 635 636 static void 637 _nvmf_ctrlr_add_io_qpair(void *ctx) 638 { 639 struct spdk_nvmf_request *req = ctx; 640 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 641 struct spdk_nvmf_fabric_connect_data *data; 642 struct spdk_nvmf_ctrlr *ctrlr; 643 struct spdk_nvmf_qpair *qpair = req->qpair; 644 struct spdk_nvmf_qpair *admin_qpair; 645 struct spdk_nvmf_tgt *tgt = qpair->transport->tgt; 646 struct spdk_nvmf_subsystem *subsystem; 647 struct spdk_nvme_transport_id listen_trid = {}; 648 const struct spdk_nvmf_subsystem_listener *listener; 649 650 assert(req->iovcnt == 1); 651 652 data = req->iov[0].iov_base; 653 654 SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid); 655 656 subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn); 657 /* We already checked this in spdk_nvmf_ctrlr_connect */ 658 assert(subsystem != NULL); 659 660 ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid); 661 if (ctrlr == NULL) { 662 SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid); 663 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid); 664 spdk_nvmf_request_complete(req); 665 return; 666 } 667 668 /* fail before passing a message to the controller thread. */ 669 if (ctrlr->in_destruct) { 670 SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n"); 671 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 672 spdk_nvmf_request_complete(req); 673 return; 674 } 675 676 /* If ANA reporting is enabled, check if I/O connect is on the same listener. */ 677 if (subsystem->flags.ana_reporting) { 678 if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) { 679 SPDK_ERRLOG("Could not get listener transport ID\n"); 680 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 681 spdk_nvmf_request_complete(req); 682 return; 683 } 684 685 listener = nvmf_subsystem_find_listener(subsystem, &listen_trid); 686 if (listener != ctrlr->listener) { 687 SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n"); 688 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 689 spdk_nvmf_request_complete(req); 690 return; 691 } 692 } 693 694 admin_qpair = ctrlr->admin_qpair; 695 if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) { 696 /* There is a chance that admin qpair is being destroyed at this moment due to e.g. 697 * expired keep alive timer. Part of the qpair destruction process is change of qpair's 698 * state to DEACTIVATING and removing it from poll group */ 699 SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group); 700 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 701 spdk_nvmf_request_complete(req); 702 return; 703 } 704 qpair->ctrlr = ctrlr; 705 spdk_thread_send_msg(admin_qpair->group->thread, nvmf_ctrlr_add_io_qpair, req); 706 } 707 708 static bool 709 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem, 710 const char *hostnqn) 711 { 712 struct spdk_nvme_transport_id listen_trid = {}; 713 714 if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) { 715 SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn); 716 return false; 717 } 718 719 if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) { 720 SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n", 721 subsystem->subnqn); 722 return false; 723 } 724 725 if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) { 726 SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n", 727 subsystem->subnqn, hostnqn); 728 return false; 729 } 730 731 return true; 732 } 733 734 static int 735 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req) 736 { 737 struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base; 738 struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd; 739 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 740 struct spdk_nvmf_qpair *qpair = req->qpair; 741 struct spdk_nvmf_transport *transport = qpair->transport; 742 struct spdk_nvmf_ctrlr *ctrlr; 743 struct spdk_nvmf_subsystem *subsystem; 744 745 SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n", 746 cmd->recfmt, cmd->qid, cmd->sqsize); 747 748 SPDK_DEBUGLOG(nvmf, "Connect data:\n"); 749 SPDK_DEBUGLOG(nvmf, " cntlid: 0x%04x\n", data->cntlid); 750 SPDK_DEBUGLOG(nvmf, " hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n", 751 ntohl(*(uint32_t *)&data->hostid[0]), 752 ntohs(*(uint16_t *)&data->hostid[4]), 753 ntohs(*(uint16_t *)&data->hostid[6]), 754 data->hostid[8], 755 data->hostid[9], 756 ntohs(*(uint16_t *)&data->hostid[10]), 757 ntohl(*(uint32_t *)&data->hostid[12])); 758 SPDK_DEBUGLOG(nvmf, " subnqn: \"%s\"\n", data->subnqn); 759 SPDK_DEBUGLOG(nvmf, " hostnqn: \"%s\"\n", data->hostnqn); 760 761 subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn); 762 if (!subsystem) { 763 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn); 764 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 765 } 766 767 if (cmd->recfmt != 0) { 768 SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt); 769 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 770 rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT; 771 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 772 } 773 774 /* 775 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and 776 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues). 777 */ 778 if (cmd->sqsize == 0) { 779 SPDK_ERRLOG("Invalid SQSIZE = 0\n"); 780 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize); 781 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 782 } 783 784 if (cmd->qid == 0) { 785 if (cmd->sqsize >= transport->opts.max_aq_depth) { 786 SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n", 787 cmd->sqsize, transport->opts.max_aq_depth - 1); 788 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize); 789 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 790 } 791 } else if (cmd->sqsize >= transport->opts.max_queue_depth) { 792 SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n", 793 cmd->sqsize, transport->opts.max_queue_depth - 1); 794 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize); 795 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 796 } 797 798 qpair->sq_head_max = cmd->sqsize; 799 qpair->qid = cmd->qid; 800 qpair->connect_received = true; 801 802 pthread_mutex_lock(&qpair->group->mutex); 803 qpair->group->current_unassociated_qpairs--; 804 pthread_mutex_unlock(&qpair->group->mutex); 805 806 if (0 == qpair->qid) { 807 qpair->group->stat.admin_qpairs++; 808 qpair->group->stat.current_admin_qpairs++; 809 } else { 810 qpair->group->stat.io_qpairs++; 811 qpair->group->stat.current_io_qpairs++; 812 } 813 814 if (cmd->qid == 0) { 815 SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid); 816 817 if (spdk_nvme_trtype_is_fabrics(transport->ops->type) && data->cntlid != 0xFFFF) { 818 /* This NVMf target only supports dynamic mode. */ 819 SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid); 820 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid); 821 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 822 } 823 824 /* Establish a new ctrlr */ 825 ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data); 826 if (!ctrlr) { 827 SPDK_ERRLOG("nvmf_ctrlr_create() failed\n"); 828 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 829 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 830 } else { 831 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 832 } 833 } else { 834 spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req); 835 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 836 } 837 } 838 839 static inline bool 840 nvmf_request_is_fabric_connect(struct spdk_nvmf_request *req) 841 { 842 return req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC && 843 req->cmd->nvmf_cmd.fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT; 844 } 845 846 static struct spdk_nvmf_subsystem_poll_group * 847 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req) 848 { 849 struct spdk_nvmf_fabric_connect_data *data; 850 struct spdk_nvmf_subsystem *subsystem; 851 struct spdk_nvmf_tgt *tgt; 852 853 assert(nvmf_request_is_fabric_connect(req)); 854 assert(req->qpair->ctrlr == NULL); 855 assert(req->iovcnt == 1); 856 857 data = req->iov[0].iov_base; 858 tgt = req->qpair->transport->tgt; 859 860 subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn); 861 if (subsystem == NULL) { 862 return NULL; 863 } 864 865 return &req->qpair->group->sgroups[subsystem->id]; 866 } 867 868 int 869 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req) 870 { 871 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 872 struct spdk_nvmf_subsystem_poll_group *sgroup; 873 struct spdk_nvmf_qpair *qpair = req->qpair; 874 enum spdk_nvmf_request_exec_status status; 875 876 if (req->iovcnt > 1) { 877 SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt); 878 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 879 status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 880 goto out; 881 } 882 883 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 884 if (!sgroup) { 885 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn); 886 status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 887 goto out; 888 } 889 890 sgroup->mgmt_io_outstanding++; 891 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 892 893 status = _nvmf_ctrlr_connect(req); 894 895 out: 896 if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 897 _nvmf_request_complete(req); 898 } 899 900 return status; 901 } 902 903 static int nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req); 904 905 static int 906 retry_connect(void *arg) 907 { 908 struct spdk_nvmf_request *req = arg; 909 struct spdk_nvmf_subsystem_poll_group *sgroup; 910 int rc; 911 912 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 913 /* subsystem may be deleted during the retry interval, so we need to check sgroup */ 914 if (sgroup != NULL) { 915 sgroup->mgmt_io_outstanding++; 916 } 917 spdk_poller_unregister(&req->poller); 918 rc = nvmf_ctrlr_cmd_connect(req); 919 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 920 _nvmf_request_complete(req); 921 } 922 return SPDK_POLLER_BUSY; 923 } 924 925 static int 926 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req) 927 { 928 struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base; 929 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 930 struct spdk_nvmf_transport *transport = req->qpair->transport; 931 struct spdk_nvmf_subsystem *subsystem; 932 933 if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) { 934 SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length); 935 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 936 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 937 } 938 939 if (req->iovcnt > 1) { 940 SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt); 941 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 942 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 943 } 944 945 subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn); 946 if (!subsystem) { 947 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn); 948 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 949 } 950 951 if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) || 952 (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) || 953 (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) || 954 (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) { 955 struct spdk_nvmf_subsystem_poll_group *sgroup; 956 957 if (req->timeout_tsc == 0) { 958 /* We will only retry the request up to 1 second. */ 959 req->timeout_tsc = spdk_get_ticks() + spdk_get_ticks_hz(); 960 } else if (spdk_get_ticks() > req->timeout_tsc) { 961 SPDK_ERRLOG("Subsystem '%s' was not ready for 1 second\n", subsystem->subnqn); 962 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 963 rsp->status.sc = SPDK_NVMF_FABRIC_SC_CONTROLLER_BUSY; 964 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 965 } 966 967 /* Subsystem is not ready to handle a connect. Use a poller to retry it 968 * again later. Decrement the mgmt_io_outstanding to avoid the 969 * subsystem waiting for this command to complete before unpausing. 970 */ 971 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 972 assert(sgroup != NULL); 973 sgroup->mgmt_io_outstanding--; 974 SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn); 975 req->poller = SPDK_POLLER_REGISTER(retry_connect, req, 100); 976 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 977 } 978 979 /* Ensure that hostnqn is null terminated */ 980 if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) { 981 SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n"); 982 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn); 983 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 984 } 985 986 if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) { 987 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 988 rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST; 989 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 990 } 991 992 return _nvmf_ctrlr_connect(req); 993 } 994 995 static int 996 nvmf_ctrlr_association_remove(void *ctx) 997 { 998 struct spdk_nvmf_ctrlr *ctrlr = ctx; 999 int rc; 1000 1001 nvmf_ctrlr_stop_association_timer(ctrlr); 1002 1003 if (ctrlr->in_destruct) { 1004 return SPDK_POLLER_IDLE; 1005 } 1006 SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n", 1007 ctrlr->subsys->subnqn); 1008 1009 if (ctrlr->admin_qpair) { 1010 rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL); 1011 if (rc < 0 && rc != -EINPROGRESS) { 1012 SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n"); 1013 assert(false); 1014 } 1015 } 1016 1017 return SPDK_POLLER_BUSY; 1018 } 1019 1020 static int 1021 _nvmf_ctrlr_cc_reset_shn_done(void *ctx) 1022 { 1023 struct spdk_nvmf_ctrlr *ctrlr = ctx; 1024 uint64_t now = spdk_get_ticks(); 1025 uint32_t count; 1026 1027 if (ctrlr->cc_timer) { 1028 spdk_poller_unregister(&ctrlr->cc_timer); 1029 } 1030 1031 count = spdk_bit_array_count_set(ctrlr->qpair_mask); 1032 SPDK_DEBUGLOG(nvmf, "ctrlr %p active queue count %u\n", ctrlr, count); 1033 1034 if (count > 1) { 1035 if (now < ctrlr->cc_timeout_tsc) { 1036 /* restart cc timer */ 1037 ctrlr->cc_timer = SPDK_POLLER_REGISTER(_nvmf_ctrlr_cc_reset_shn_done, ctrlr, 100 * 1000); 1038 return SPDK_POLLER_IDLE; 1039 } else { 1040 /* controller fatal status */ 1041 SPDK_WARNLOG("IO timeout, ctrlr %p is in fatal status\n", ctrlr); 1042 nvmf_ctrlr_set_fatal_status(ctrlr); 1043 } 1044 } 1045 1046 spdk_poller_unregister(&ctrlr->cc_timeout_timer); 1047 1048 if (ctrlr->disconnect_is_shn) { 1049 ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE; 1050 ctrlr->disconnect_is_shn = false; 1051 } else { 1052 /* Only a subset of the registers are cleared out on a reset */ 1053 ctrlr->vcprop.cc.raw = 0; 1054 ctrlr->vcprop.csts.raw = 0; 1055 } 1056 1057 /* After CC.EN transitions to 0 (due to shutdown or reset), the association 1058 * between the host and controller shall be preserved for at least 2 minutes */ 1059 if (ctrlr->association_timer) { 1060 SPDK_DEBUGLOG(nvmf, "Association timer already set\n"); 1061 nvmf_ctrlr_stop_association_timer(ctrlr); 1062 } 1063 if (ctrlr->association_timeout) { 1064 ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr, 1065 ctrlr->association_timeout * 1000); 1066 } 1067 ctrlr->disconnect_in_progress = false; 1068 return SPDK_POLLER_BUSY; 1069 } 1070 1071 static void 1072 nvmf_ctrlr_cc_reset_shn_done(struct spdk_io_channel_iter *i, int status) 1073 { 1074 struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i); 1075 1076 if (status < 0) { 1077 SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n"); 1078 assert(false); 1079 } 1080 1081 _nvmf_ctrlr_cc_reset_shn_done((void *)ctrlr); 1082 } 1083 1084 static void 1085 nvmf_bdev_complete_reset(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 1086 { 1087 SPDK_NOTICELOG("Resetting bdev done with %s\n", success ? "success" : "failure"); 1088 1089 spdk_bdev_free_io(bdev_io); 1090 } 1091 1092 1093 static int 1094 nvmf_ctrlr_cc_timeout(void *ctx) 1095 { 1096 struct spdk_nvmf_ctrlr *ctrlr = ctx; 1097 struct spdk_nvmf_poll_group *group = ctrlr->admin_qpair->group; 1098 struct spdk_nvmf_ns *ns; 1099 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 1100 1101 assert(group != NULL && group->sgroups != NULL); 1102 spdk_poller_unregister(&ctrlr->cc_timeout_timer); 1103 SPDK_DEBUGLOG(nvmf, "Ctrlr %p reset or shutdown timeout\n", ctrlr); 1104 1105 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL; 1106 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 1107 if (ns->bdev == NULL) { 1108 continue; 1109 } 1110 ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[ns->opts.nsid - 1]; 1111 SPDK_NOTICELOG("Ctrlr %p resetting NSID %u\n", ctrlr, ns->opts.nsid); 1112 spdk_bdev_reset(ns->desc, ns_info->channel, nvmf_bdev_complete_reset, NULL); 1113 } 1114 1115 return SPDK_POLLER_BUSY; 1116 } 1117 1118 const struct spdk_nvmf_registers * 1119 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr) 1120 { 1121 return &ctrlr->vcprop; 1122 } 1123 1124 void 1125 nvmf_ctrlr_set_fatal_status(struct spdk_nvmf_ctrlr *ctrlr) 1126 { 1127 ctrlr->vcprop.csts.bits.cfs = 1; 1128 } 1129 1130 static uint64_t 1131 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr) 1132 { 1133 return ctrlr->vcprop.cap.raw; 1134 } 1135 1136 static uint64_t 1137 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr) 1138 { 1139 return ctrlr->vcprop.vs.raw; 1140 } 1141 1142 static uint64_t 1143 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr) 1144 { 1145 return ctrlr->vcprop.cc.raw; 1146 } 1147 1148 static bool 1149 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1150 { 1151 union spdk_nvme_cc_register cc, diff; 1152 uint32_t cc_timeout_ms; 1153 1154 cc.raw = value; 1155 1156 SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw); 1157 SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw); 1158 1159 /* 1160 * Calculate which bits changed between the current and new CC. 1161 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed. 1162 */ 1163 diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw; 1164 1165 if (diff.bits.en) { 1166 if (cc.bits.en) { 1167 SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n"); 1168 nvmf_ctrlr_stop_association_timer(ctrlr); 1169 1170 ctrlr->vcprop.cc.bits.en = 1; 1171 ctrlr->vcprop.csts.bits.rdy = 1; 1172 } else { 1173 SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n"); 1174 if (ctrlr->disconnect_in_progress) { 1175 SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n"); 1176 return true; 1177 } 1178 1179 ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr, 1180 NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000); 1181 /* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */ 1182 cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2; 1183 ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000; 1184 1185 ctrlr->vcprop.cc.bits.en = 0; 1186 ctrlr->disconnect_in_progress = true; 1187 ctrlr->disconnect_is_shn = false; 1188 spdk_for_each_channel(ctrlr->subsys->tgt, 1189 nvmf_ctrlr_disconnect_io_qpairs_on_pg, 1190 ctrlr, 1191 nvmf_ctrlr_cc_reset_shn_done); 1192 } 1193 diff.bits.en = 0; 1194 } 1195 1196 if (diff.bits.shn) { 1197 if (cc.bits.shn == SPDK_NVME_SHN_NORMAL || 1198 cc.bits.shn == SPDK_NVME_SHN_ABRUPT) { 1199 SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n", 1200 cc.bits.shn >> 1, cc.bits.shn & 1); 1201 if (ctrlr->disconnect_in_progress) { 1202 SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n"); 1203 return true; 1204 } 1205 1206 ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr, 1207 NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000); 1208 /* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */ 1209 cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2; 1210 ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000; 1211 1212 ctrlr->vcprop.cc.bits.shn = cc.bits.shn; 1213 ctrlr->disconnect_in_progress = true; 1214 ctrlr->disconnect_is_shn = true; 1215 spdk_for_each_channel(ctrlr->subsys->tgt, 1216 nvmf_ctrlr_disconnect_io_qpairs_on_pg, 1217 ctrlr, 1218 nvmf_ctrlr_cc_reset_shn_done); 1219 1220 /* From the time a shutdown is initiated the controller shall disable 1221 * Keep Alive timer */ 1222 nvmf_ctrlr_stop_keep_alive_timer(ctrlr); 1223 } else if (cc.bits.shn == 0) { 1224 ctrlr->vcprop.cc.bits.shn = 0; 1225 } else { 1226 SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n", 1227 cc.bits.shn >> 1, cc.bits.shn & 1); 1228 return false; 1229 } 1230 diff.bits.shn = 0; 1231 } 1232 1233 if (diff.bits.iosqes) { 1234 SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n", 1235 cc.bits.iosqes, 1u << cc.bits.iosqes); 1236 ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes; 1237 diff.bits.iosqes = 0; 1238 } 1239 1240 if (diff.bits.iocqes) { 1241 SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n", 1242 cc.bits.iocqes, 1u << cc.bits.iocqes); 1243 ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes; 1244 diff.bits.iocqes = 0; 1245 } 1246 1247 if (diff.bits.ams) { 1248 SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams); 1249 return false; 1250 } 1251 1252 if (diff.bits.mps) { 1253 SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps))); 1254 return false; 1255 } 1256 1257 if (diff.bits.css) { 1258 if (cc.bits.css > SPDK_NVME_CC_CSS_IOCS) { 1259 SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css); 1260 return false; 1261 } 1262 diff.bits.css = 0; 1263 } 1264 1265 if (diff.raw != 0) { 1266 /* Print an error message, but don't fail the command in this case. 1267 * If we did want to fail in this case, we'd need to ensure we acted 1268 * on no other bits or the initiator gets confused. */ 1269 SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw); 1270 } 1271 1272 return true; 1273 } 1274 1275 static uint64_t 1276 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr) 1277 { 1278 return ctrlr->vcprop.csts.raw; 1279 } 1280 1281 static uint64_t 1282 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr) 1283 { 1284 return ctrlr->vcprop.aqa.raw; 1285 } 1286 1287 static bool 1288 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1289 { 1290 union spdk_nvme_aqa_register aqa; 1291 1292 aqa.raw = value; 1293 1294 /* 1295 * We don't need to explicitly check for maximum size, as the fields are 1296 * limited to 12 bits (4096). 1297 */ 1298 if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 || 1299 aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 || 1300 aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) { 1301 return false; 1302 } 1303 1304 ctrlr->vcprop.aqa.raw = value; 1305 1306 return true; 1307 } 1308 1309 static uint64_t 1310 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr) 1311 { 1312 return ctrlr->vcprop.asq; 1313 } 1314 1315 static bool 1316 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1317 { 1318 ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value; 1319 1320 return true; 1321 } 1322 1323 static bool 1324 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1325 { 1326 ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL); 1327 1328 return true; 1329 } 1330 1331 static uint64_t 1332 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr) 1333 { 1334 return ctrlr->vcprop.acq; 1335 } 1336 1337 static bool 1338 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1339 { 1340 ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value; 1341 1342 return true; 1343 } 1344 1345 static bool 1346 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1347 { 1348 ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL); 1349 1350 return true; 1351 } 1352 1353 struct nvmf_prop { 1354 uint32_t ofst; 1355 uint8_t size; 1356 char name[11]; 1357 uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr); 1358 bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value); 1359 bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value); 1360 }; 1361 1362 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \ 1363 { \ 1364 offsetof(struct spdk_nvme_registers, field), \ 1365 size, \ 1366 #field, \ 1367 get_cb, set_cb, set_upper_cb \ 1368 } 1369 1370 static const struct nvmf_prop nvmf_props[] = { 1371 PROP(cap, 8, nvmf_prop_get_cap, NULL, NULL), 1372 PROP(vs, 4, nvmf_prop_get_vs, NULL, NULL), 1373 PROP(cc, 4, nvmf_prop_get_cc, nvmf_prop_set_cc, NULL), 1374 PROP(csts, 4, nvmf_prop_get_csts, NULL, NULL), 1375 PROP(aqa, 4, nvmf_prop_get_aqa, nvmf_prop_set_aqa, NULL), 1376 PROP(asq, 8, nvmf_prop_get_asq, nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper), 1377 PROP(acq, 8, nvmf_prop_get_acq, nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper), 1378 }; 1379 1380 static const struct nvmf_prop * 1381 find_prop(uint32_t ofst, uint8_t size) 1382 { 1383 size_t i; 1384 1385 for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) { 1386 const struct nvmf_prop *prop = &nvmf_props[i]; 1387 1388 if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) { 1389 return prop; 1390 } 1391 } 1392 1393 return NULL; 1394 } 1395 1396 static int 1397 nvmf_property_get(struct spdk_nvmf_request *req) 1398 { 1399 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1400 struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd; 1401 struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp; 1402 const struct nvmf_prop *prop; 1403 uint8_t size; 1404 1405 response->status.sc = 0; 1406 response->value.u64 = 0; 1407 1408 SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n", 1409 cmd->attrib.size, cmd->ofst); 1410 1411 switch (cmd->attrib.size) { 1412 case SPDK_NVMF_PROP_SIZE_4: 1413 size = 4; 1414 break; 1415 case SPDK_NVMF_PROP_SIZE_8: 1416 size = 8; 1417 break; 1418 default: 1419 SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size); 1420 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1421 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1422 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1423 } 1424 1425 prop = find_prop(cmd->ofst, size); 1426 if (prop == NULL || prop->get_cb == NULL) { 1427 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1428 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1429 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1430 } 1431 1432 SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name); 1433 1434 response->value.u64 = prop->get_cb(ctrlr); 1435 1436 if (size != prop->size) { 1437 /* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */ 1438 assert(size == 4); 1439 assert(prop->size == 8); 1440 1441 if (cmd->ofst == prop->ofst) { 1442 /* Keep bottom 4 bytes only */ 1443 response->value.u64 &= 0xFFFFFFFF; 1444 } else { 1445 /* Keep top 4 bytes only */ 1446 response->value.u64 >>= 32; 1447 } 1448 } 1449 1450 SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64); 1451 1452 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1453 } 1454 1455 static int 1456 nvmf_property_set(struct spdk_nvmf_request *req) 1457 { 1458 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1459 struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd; 1460 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1461 const struct nvmf_prop *prop; 1462 uint64_t value; 1463 uint8_t size; 1464 bool ret; 1465 1466 SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n", 1467 cmd->attrib.size, cmd->ofst, cmd->value.u64); 1468 1469 switch (cmd->attrib.size) { 1470 case SPDK_NVMF_PROP_SIZE_4: 1471 size = 4; 1472 break; 1473 case SPDK_NVMF_PROP_SIZE_8: 1474 size = 8; 1475 break; 1476 default: 1477 SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size); 1478 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1479 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1480 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1481 } 1482 1483 prop = find_prop(cmd->ofst, size); 1484 if (prop == NULL || prop->set_cb == NULL) { 1485 SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst); 1486 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1487 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1488 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1489 } 1490 1491 SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name); 1492 1493 value = cmd->value.u64; 1494 1495 if (prop->size == 4) { 1496 ret = prop->set_cb(ctrlr, (uint32_t)value); 1497 } else if (size != prop->size) { 1498 /* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */ 1499 assert(size == 4); 1500 assert(prop->size == 8); 1501 1502 if (cmd->ofst == prop->ofst) { 1503 ret = prop->set_cb(ctrlr, (uint32_t)value); 1504 } else { 1505 ret = prop->set_upper_cb(ctrlr, (uint32_t)value); 1506 } 1507 } else { 1508 ret = prop->set_cb(ctrlr, (uint32_t)value); 1509 if (ret) { 1510 ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32)); 1511 } 1512 } 1513 1514 if (!ret) { 1515 SPDK_ERRLOG("prop set_cb failed\n"); 1516 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1517 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1518 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1519 } 1520 1521 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1522 } 1523 1524 static int 1525 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req) 1526 { 1527 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1528 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1529 1530 SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11); 1531 1532 ctrlr->feat.arbitration.raw = cmd->cdw11; 1533 ctrlr->feat.arbitration.bits.reserved = 0; 1534 1535 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1536 } 1537 1538 static int 1539 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req) 1540 { 1541 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1542 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1543 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1544 1545 SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11); 1546 1547 /* Only PS = 0 is allowed, since we report NPSS = 0 */ 1548 if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) { 1549 SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps); 1550 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1551 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1552 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1553 } 1554 1555 ctrlr->feat.power_management.raw = cmd->cdw11; 1556 ctrlr->feat.power_management.bits.reserved = 0; 1557 1558 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1559 } 1560 1561 static bool 1562 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts) 1563 { 1564 /* 1565 * Valid TMPSEL values: 1566 * 0000b - 1000b: temperature sensors 1567 * 1111b: set all implemented temperature sensors 1568 */ 1569 if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) { 1570 /* 1001b - 1110b: reserved */ 1571 SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel); 1572 return false; 1573 } 1574 1575 /* 1576 * Valid THSEL values: 1577 * 00b: over temperature threshold 1578 * 01b: under temperature threshold 1579 */ 1580 if (opts->bits.thsel > 1) { 1581 /* 10b - 11b: reserved */ 1582 SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel); 1583 return false; 1584 } 1585 1586 return true; 1587 } 1588 1589 static int 1590 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req) 1591 { 1592 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1593 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1594 1595 SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11); 1596 1597 if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) { 1598 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1599 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1600 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1601 } 1602 1603 /* TODO: no sensors implemented - ignore new values */ 1604 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1605 } 1606 1607 static int 1608 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req) 1609 { 1610 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1611 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1612 1613 SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11); 1614 1615 if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) { 1616 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1617 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1618 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1619 } 1620 1621 /* TODO: no sensors implemented - return 0 for all thresholds */ 1622 rsp->cdw0 = 0; 1623 1624 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1625 } 1626 1627 static int 1628 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req) 1629 { 1630 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1631 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1632 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1633 union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {}; 1634 1635 SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11); 1636 1637 iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv; 1638 iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd; 1639 rsp->cdw0 = iv_conf.raw; 1640 1641 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1642 } 1643 1644 static int 1645 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req) 1646 { 1647 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1648 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1649 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1650 1651 SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11); 1652 1653 if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) { 1654 /* 1655 * Host is not allowed to set this bit, since we don't advertise it in 1656 * Identify Namespace. 1657 */ 1658 SPDK_ERRLOG("Host set unsupported DULBE bit\n"); 1659 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1660 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1661 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1662 } 1663 1664 ctrlr->feat.error_recovery.raw = cmd->cdw11; 1665 ctrlr->feat.error_recovery.bits.reserved = 0; 1666 1667 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1668 } 1669 1670 static int 1671 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req) 1672 { 1673 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1674 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1675 1676 SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11); 1677 1678 ctrlr->feat.volatile_write_cache.raw = cmd->cdw11; 1679 ctrlr->feat.volatile_write_cache.bits.reserved = 0; 1680 1681 SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n", 1682 ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled"); 1683 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1684 } 1685 1686 static int 1687 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req) 1688 { 1689 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1690 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1691 1692 SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11); 1693 1694 ctrlr->feat.write_atomicity.raw = cmd->cdw11; 1695 ctrlr->feat.write_atomicity.bits.reserved = 0; 1696 1697 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1698 } 1699 1700 static int 1701 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req) 1702 { 1703 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1704 1705 SPDK_ERRLOG("Set Features - Host Identifier not allowed\n"); 1706 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 1707 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1708 } 1709 1710 static int 1711 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req) 1712 { 1713 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1714 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1715 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1716 struct spdk_iov_xfer ix; 1717 1718 SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n"); 1719 1720 if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) { 1721 /* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */ 1722 SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n"); 1723 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1724 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1725 } 1726 1727 if (req->iovcnt < 1 || req->length < sizeof(ctrlr->hostid)) { 1728 SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n"); 1729 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1730 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1731 } 1732 1733 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 1734 spdk_iov_xfer_from_buf(&ix, &ctrlr->hostid, sizeof(ctrlr->hostid)); 1735 1736 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1737 } 1738 1739 static int 1740 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req) 1741 { 1742 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1743 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1744 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1745 struct spdk_nvmf_ns *ns; 1746 1747 SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notification Mask\n"); 1748 1749 if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1750 SPDK_ERRLOG("get Features - Invalid Namespace ID\n"); 1751 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1752 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1753 } 1754 1755 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1756 if (ns == NULL) { 1757 SPDK_ERRLOG("get Features - Invalid Namespace ID\n"); 1758 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1759 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1760 } 1761 rsp->cdw0 = ns->mask; 1762 1763 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1764 } 1765 1766 static int 1767 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req) 1768 { 1769 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1770 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 1771 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1772 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1773 struct spdk_nvmf_ns *ns; 1774 1775 SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notification Mask\n"); 1776 1777 if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1778 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 1779 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 1780 ns->mask = cmd->cdw11; 1781 } 1782 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1783 } 1784 1785 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1786 if (ns == NULL) { 1787 SPDK_ERRLOG("Set Features - Invalid Namespace ID\n"); 1788 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1789 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1790 } 1791 ns->mask = cmd->cdw11; 1792 1793 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1794 } 1795 1796 static int 1797 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req) 1798 { 1799 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1800 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1801 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1802 struct spdk_nvmf_ns *ns; 1803 1804 SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n"); 1805 1806 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1807 /* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */ 1808 if (ns == NULL) { 1809 SPDK_ERRLOG("Get Features - Invalid Namespace ID\n"); 1810 response->status.sct = SPDK_NVME_SCT_GENERIC; 1811 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1812 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1813 } 1814 1815 response->cdw0 = ns->ptpl_activated; 1816 1817 response->status.sct = SPDK_NVME_SCT_GENERIC; 1818 response->status.sc = SPDK_NVME_SC_SUCCESS; 1819 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1820 } 1821 1822 static int 1823 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req) 1824 { 1825 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1826 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1827 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1828 struct spdk_nvmf_ns *ns; 1829 bool ptpl; 1830 1831 SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n"); 1832 1833 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1834 ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl; 1835 1836 if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && nvmf_ns_is_ptpl_capable(ns)) { 1837 ns->ptpl_activated = ptpl; 1838 } else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1839 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns; 1840 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 1841 if (nvmf_ns_is_ptpl_capable(ns)) { 1842 ns->ptpl_activated = ptpl; 1843 } 1844 } 1845 } else { 1846 SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n"); 1847 response->status.sct = SPDK_NVME_SCT_GENERIC; 1848 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1849 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1850 } 1851 1852 /* TODO: Feature not changeable for now */ 1853 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1854 response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE; 1855 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1856 } 1857 1858 static int 1859 nvmf_ctrlr_get_features_host_behavior_support(struct spdk_nvmf_request *req) 1860 { 1861 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1862 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1863 struct spdk_nvme_host_behavior host_behavior = {}; 1864 struct spdk_iov_xfer ix; 1865 1866 SPDK_DEBUGLOG(nvmf, "Get Features - Host Behavior Support\n"); 1867 1868 if (req->iovcnt < 1 || req->length < sizeof(struct spdk_nvme_host_behavior)) { 1869 SPDK_ERRLOG("invalid data buffer for Host Behavior Support\n"); 1870 response->status.sct = SPDK_NVME_SCT_GENERIC; 1871 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1872 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1873 } 1874 1875 host_behavior.acre = ctrlr->acre_enabled; 1876 1877 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 1878 spdk_iov_xfer_from_buf(&ix, &host_behavior, sizeof(host_behavior)); 1879 1880 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1881 } 1882 1883 static int 1884 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req) 1885 { 1886 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1887 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1888 struct spdk_nvme_host_behavior *host_behavior; 1889 1890 SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n"); 1891 if (req->iovcnt != 1) { 1892 SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt); 1893 response->status.sct = SPDK_NVME_SCT_GENERIC; 1894 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1895 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1896 } 1897 if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) { 1898 SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len); 1899 response->status.sct = SPDK_NVME_SCT_GENERIC; 1900 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1901 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1902 } 1903 1904 host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base; 1905 if (host_behavior->acre == 0) { 1906 ctrlr->acre_enabled = false; 1907 } else if (host_behavior->acre == 1) { 1908 ctrlr->acre_enabled = true; 1909 } else { 1910 SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre); 1911 response->status.sct = SPDK_NVME_SCT_GENERIC; 1912 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1913 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1914 } 1915 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1916 } 1917 1918 static int 1919 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req) 1920 { 1921 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1922 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1923 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1924 1925 SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11); 1926 1927 /* 1928 * if attempts to disable keep alive by setting kato to 0h 1929 * a status value of keep alive invalid shall be returned 1930 */ 1931 if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) { 1932 rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID; 1933 } else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) { 1934 ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS; 1935 } else { 1936 /* round up to milliseconds */ 1937 ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up( 1938 cmd->cdw11_bits.feat_keep_alive_timer.bits.kato, 1939 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) * 1940 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS; 1941 } 1942 1943 /* 1944 * if change the keep alive timeout value successfully 1945 * update the keep alive poller. 1946 */ 1947 if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) { 1948 if (ctrlr->keep_alive_poller != NULL) { 1949 spdk_poller_unregister(&ctrlr->keep_alive_poller); 1950 } 1951 ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr, 1952 ctrlr->feat.keep_alive_timer.bits.kato * 1000); 1953 } 1954 1955 SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n", 1956 ctrlr->feat.keep_alive_timer.bits.kato); 1957 1958 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1959 } 1960 1961 static int 1962 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req) 1963 { 1964 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1965 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1966 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1967 uint32_t count; 1968 1969 SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n", 1970 req->cmd->nvme_cmd.cdw11); 1971 1972 if (cmd->cdw11_bits.feat_num_of_queues.bits.ncqr == UINT16_MAX || 1973 cmd->cdw11_bits.feat_num_of_queues.bits.nsqr == UINT16_MAX) { 1974 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1975 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1976 } 1977 1978 count = spdk_bit_array_count_set(ctrlr->qpair_mask); 1979 /* verify that the controller is ready to process commands */ 1980 if (count > 1) { 1981 SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n"); 1982 rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 1983 } else { 1984 /* 1985 * Ignore the value requested by the host - 1986 * always return the pre-configured value based on max_qpairs_allowed. 1987 */ 1988 rsp->cdw0 = ctrlr->feat.number_of_queues.raw; 1989 } 1990 1991 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1992 } 1993 1994 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ctrlr) == 4920, 1995 "Please check migration fields that need to be added or not"); 1996 1997 static void 1998 nvmf_ctrlr_migr_data_copy(struct spdk_nvmf_ctrlr_migr_data *data, 1999 const struct spdk_nvmf_ctrlr_migr_data *data_src, size_t data_size) 2000 { 2001 assert(data); 2002 assert(data_src); 2003 assert(data_size); 2004 2005 memcpy(&data->regs, &data_src->regs, spdk_min(data->regs_size, data_src->regs_size)); 2006 memcpy(&data->feat, &data_src->feat, spdk_min(data->feat_size, data_src->feat_size)); 2007 2008 #define SET_FIELD(field) \ 2009 if (offsetof(struct spdk_nvmf_ctrlr_migr_data, field) + sizeof(data->field) <= data_size) { \ 2010 data->field = data_src->field; \ 2011 } \ 2012 2013 SET_FIELD(cntlid); 2014 SET_FIELD(acre); 2015 SET_FIELD(num_aer_cids); 2016 SET_FIELD(num_async_events); 2017 SET_FIELD(notice_aen_mask); 2018 #undef SET_FIELD 2019 2020 #define SET_ARRAY(arr) \ 2021 if (offsetof(struct spdk_nvmf_ctrlr_migr_data, arr) + sizeof(data->arr) <= data_size) { \ 2022 memcpy(&data->arr, &data_src->arr, sizeof(data->arr)); \ 2023 } \ 2024 2025 SET_ARRAY(async_events); 2026 SET_ARRAY(aer_cids); 2027 #undef SET_ARRAY 2028 } 2029 2030 int 2031 spdk_nvmf_ctrlr_save_migr_data(struct spdk_nvmf_ctrlr *ctrlr, 2032 struct spdk_nvmf_ctrlr_migr_data *data) 2033 { 2034 struct spdk_nvmf_async_event_completion *event, *event_tmp; 2035 uint32_t i; 2036 struct spdk_nvmf_ctrlr_migr_data data_local = { 2037 .data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused), 2038 .regs_size = sizeof(struct spdk_nvmf_registers), 2039 .feat_size = sizeof(struct spdk_nvmf_ctrlr_feat) 2040 }; 2041 2042 assert(data->data_size <= sizeof(data_local)); 2043 assert(spdk_get_thread() == ctrlr->thread); 2044 2045 memcpy(&data_local.regs, &ctrlr->vcprop, sizeof(struct spdk_nvmf_registers)); 2046 memcpy(&data_local.feat, &ctrlr->feat, sizeof(struct spdk_nvmf_ctrlr_feat)); 2047 2048 data_local.cntlid = ctrlr->cntlid; 2049 data_local.acre = ctrlr->acre_enabled; 2050 data_local.num_aer_cids = ctrlr->nr_aer_reqs; 2051 2052 STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) { 2053 if (data_local.num_async_events + 1 > SPDK_NVMF_MIGR_MAX_PENDING_AERS) { 2054 SPDK_ERRLOG("ctrlr %p has too many pending AERs\n", ctrlr); 2055 break; 2056 } 2057 2058 data_local.async_events[data_local.num_async_events++].raw = event->event.raw; 2059 } 2060 2061 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 2062 struct spdk_nvmf_request *req = ctrlr->aer_req[i]; 2063 data_local.aer_cids[i] = req->cmd->nvme_cmd.cid; 2064 } 2065 data_local.notice_aen_mask = ctrlr->notice_aen_mask; 2066 2067 nvmf_ctrlr_migr_data_copy(data, &data_local, spdk_min(data->data_size, data_local.data_size)); 2068 return 0; 2069 } 2070 2071 int 2072 spdk_nvmf_ctrlr_restore_migr_data(struct spdk_nvmf_ctrlr *ctrlr, 2073 const struct spdk_nvmf_ctrlr_migr_data *data) 2074 { 2075 uint32_t i; 2076 struct spdk_nvmf_ctrlr_migr_data data_local = { 2077 .data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused), 2078 .regs_size = sizeof(struct spdk_nvmf_registers), 2079 .feat_size = sizeof(struct spdk_nvmf_ctrlr_feat) 2080 }; 2081 2082 assert(data->data_size <= sizeof(data_local)); 2083 assert(spdk_get_thread() == ctrlr->thread); 2084 2085 /* local version of data should have defaults set before copy */ 2086 nvmf_ctrlr_migr_data_copy(&data_local, data, spdk_min(data->data_size, data_local.data_size)); 2087 memcpy(&ctrlr->vcprop, &data_local.regs, sizeof(struct spdk_nvmf_registers)); 2088 memcpy(&ctrlr->feat, &data_local.feat, sizeof(struct spdk_nvmf_ctrlr_feat)); 2089 2090 ctrlr->cntlid = data_local.cntlid; 2091 ctrlr->acre_enabled = data_local.acre; 2092 2093 for (i = 0; i < data_local.num_async_events; i++) { 2094 struct spdk_nvmf_async_event_completion *event; 2095 2096 event = calloc(1, sizeof(*event)); 2097 if (!event) { 2098 return -ENOMEM; 2099 } 2100 2101 event->event.raw = data_local.async_events[i].raw; 2102 STAILQ_INSERT_TAIL(&ctrlr->async_events, event, link); 2103 } 2104 ctrlr->notice_aen_mask = data_local.notice_aen_mask; 2105 2106 return 0; 2107 } 2108 2109 static int 2110 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req) 2111 { 2112 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2113 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2114 2115 SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n", 2116 cmd->cdw11); 2117 ctrlr->feat.async_event_configuration.raw = cmd->cdw11; 2118 ctrlr->feat.async_event_configuration.bits.reserved1 = 0; 2119 ctrlr->feat.async_event_configuration.bits.reserved2 = 0; 2120 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2121 } 2122 2123 static int 2124 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req) 2125 { 2126 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2127 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 2128 struct spdk_nvmf_async_event_completion *pending_event; 2129 2130 SPDK_DEBUGLOG(nvmf, "Async Event Request\n"); 2131 2132 /* Four asynchronous events are supported for now */ 2133 if (ctrlr->nr_aer_reqs >= SPDK_NVMF_MAX_ASYNC_EVENTS) { 2134 SPDK_DEBUGLOG(nvmf, "AERL exceeded\n"); 2135 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 2136 rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED; 2137 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2138 } 2139 2140 if (!STAILQ_EMPTY(&ctrlr->async_events)) { 2141 pending_event = STAILQ_FIRST(&ctrlr->async_events); 2142 rsp->cdw0 = pending_event->event.raw; 2143 STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link); 2144 free(pending_event); 2145 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2146 } 2147 2148 ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req; 2149 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 2150 } 2151 2152 static void 2153 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length) 2154 { 2155 struct spdk_nvme_firmware_page fw_page; 2156 size_t copy_len; 2157 struct spdk_iov_xfer ix; 2158 2159 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2160 2161 memset(&fw_page, 0, sizeof(fw_page)); 2162 fw_page.afi.active_slot = 1; 2163 fw_page.afi.next_reset_slot = 0; 2164 spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' '); 2165 2166 if (offset < sizeof(fw_page)) { 2167 copy_len = spdk_min(sizeof(fw_page) - offset, length); 2168 if (copy_len > 0) { 2169 spdk_iov_xfer_from_buf(&ix, (const char *)&fw_page + offset, copy_len); 2170 } 2171 } 2172 } 2173 2174 /* 2175 * Asynchronous Event Mask Bit 2176 */ 2177 enum spdk_nvme_async_event_mask_bit { 2178 /* Mask Namespace Change Notification */ 2179 SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT = 0, 2180 /* Mask Asymmetric Namespace Access Change Notification */ 2181 SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT = 1, 2182 /* Mask Discovery Log Change Notification */ 2183 SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT = 2, 2184 /* Mask Reservation Log Page Available Notification */ 2185 SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT = 3, 2186 /* Mask Error Event */ 2187 SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT = 4, 2188 /* 4 - 63 Reserved */ 2189 }; 2190 2191 static inline void 2192 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr, 2193 enum spdk_nvme_async_event_mask_bit mask) 2194 { 2195 ctrlr->notice_aen_mask &= ~(1 << mask); 2196 } 2197 2198 static inline bool 2199 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr, 2200 enum spdk_nvme_async_event_mask_bit mask) 2201 { 2202 if (ctrlr->notice_aen_mask & (1 << mask)) { 2203 return false; 2204 } else { 2205 ctrlr->notice_aen_mask |= (1 << mask); 2206 return true; 2207 } 2208 } 2209 2210 /* we have to use the typedef in the function declaration to appease astyle. */ 2211 typedef enum spdk_nvme_ana_state spdk_nvme_ana_state_t; 2212 2213 static inline spdk_nvme_ana_state_t 2214 nvmf_ctrlr_get_ana_state(struct spdk_nvmf_ctrlr *ctrlr, uint32_t anagrpid) 2215 { 2216 if (!ctrlr->subsys->flags.ana_reporting) { 2217 return SPDK_NVME_ANA_OPTIMIZED_STATE; 2218 } 2219 2220 if (spdk_unlikely(ctrlr->listener == NULL)) { 2221 return SPDK_NVME_ANA_INACCESSIBLE_STATE; 2222 } 2223 2224 assert(anagrpid - 1 < ctrlr->subsys->max_nsid); 2225 return ctrlr->listener->ana_state[anagrpid - 1]; 2226 } 2227 2228 static spdk_nvme_ana_state_t 2229 nvmf_ctrlr_get_ana_state_from_nsid(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid) 2230 { 2231 struct spdk_nvmf_ns *ns; 2232 2233 /* We do not have NVM subsystem specific ANA state. Hence if NSID is either 2234 * SPDK_NVMF_GLOBAL_NS_TAG, invalid, or for inactive namespace, return 2235 * the optimized state. 2236 */ 2237 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 2238 if (ns == NULL) { 2239 return SPDK_NVME_ANA_OPTIMIZED_STATE; 2240 } 2241 2242 return nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid); 2243 } 2244 2245 static void 2246 nvmf_get_error_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt, 2247 uint64_t offset, uint32_t length, uint32_t rae) 2248 { 2249 if (!rae) { 2250 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT); 2251 } 2252 2253 /* TODO: actually fill out log page data */ 2254 } 2255 2256 static void 2257 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt, 2258 uint64_t offset, uint32_t length, uint32_t rae) 2259 { 2260 struct spdk_nvme_ana_page ana_hdr; 2261 struct spdk_nvme_ana_group_descriptor ana_desc; 2262 size_t copy_len, copied_len; 2263 uint32_t num_anagrp = 0, anagrpid; 2264 struct spdk_nvmf_ns *ns; 2265 struct spdk_iov_xfer ix; 2266 2267 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2268 2269 if (length == 0) { 2270 goto done; 2271 } 2272 2273 if (offset >= sizeof(ana_hdr)) { 2274 offset -= sizeof(ana_hdr); 2275 } else { 2276 for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) { 2277 if (ctrlr->subsys->ana_group[anagrpid - 1] > 0) { 2278 num_anagrp++; 2279 } 2280 } 2281 2282 memset(&ana_hdr, 0, sizeof(ana_hdr)); 2283 2284 ana_hdr.num_ana_group_desc = num_anagrp; 2285 /* TODO: Support Change Count. */ 2286 ana_hdr.change_count = 0; 2287 2288 copy_len = spdk_min(sizeof(ana_hdr) - offset, length); 2289 copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_hdr + offset, copy_len); 2290 assert(copied_len == copy_len); 2291 length -= copied_len; 2292 offset = 0; 2293 } 2294 2295 if (length == 0) { 2296 goto done; 2297 } 2298 2299 for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) { 2300 if (ctrlr->subsys->ana_group[anagrpid - 1] == 0) { 2301 continue; 2302 } 2303 2304 if (offset >= sizeof(ana_desc)) { 2305 offset -= sizeof(ana_desc); 2306 } else { 2307 memset(&ana_desc, 0, sizeof(ana_desc)); 2308 2309 ana_desc.ana_group_id = anagrpid; 2310 ana_desc.num_of_nsid = ctrlr->subsys->ana_group[anagrpid - 1]; 2311 ana_desc.ana_state = nvmf_ctrlr_get_ana_state(ctrlr, anagrpid); 2312 2313 copy_len = spdk_min(sizeof(ana_desc) - offset, length); 2314 copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_desc + offset, 2315 copy_len); 2316 assert(copied_len == copy_len); 2317 length -= copied_len; 2318 offset = 0; 2319 2320 if (length == 0) { 2321 goto done; 2322 } 2323 } 2324 2325 /* TODO: Revisit here about O(n^2) cost if we have subsystem with 2326 * many namespaces in the future. 2327 */ 2328 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL; 2329 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 2330 if (ns->anagrpid != anagrpid) { 2331 continue; 2332 } 2333 2334 if (offset >= sizeof(uint32_t)) { 2335 offset -= sizeof(uint32_t); 2336 continue; 2337 } 2338 2339 copy_len = spdk_min(sizeof(uint32_t) - offset, length); 2340 copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ns->nsid + offset, 2341 copy_len); 2342 assert(copied_len == copy_len); 2343 length -= copied_len; 2344 offset = 0; 2345 2346 if (length == 0) { 2347 goto done; 2348 } 2349 } 2350 } 2351 2352 done: 2353 if (!rae) { 2354 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT); 2355 } 2356 } 2357 2358 void 2359 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid) 2360 { 2361 uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list); 2362 uint16_t i; 2363 bool found = false; 2364 2365 for (i = 0; i < ctrlr->changed_ns_list_count; i++) { 2366 if (ctrlr->changed_ns_list.ns_list[i] == nsid) { 2367 /* nsid is already in the list */ 2368 found = true; 2369 break; 2370 } 2371 } 2372 2373 if (!found) { 2374 if (ctrlr->changed_ns_list_count == max_changes) { 2375 /* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */ 2376 ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu; 2377 for (i = 1; i < max_changes; i++) { 2378 ctrlr->changed_ns_list.ns_list[i] = 0; 2379 } 2380 } else { 2381 ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid; 2382 } 2383 } 2384 } 2385 2386 static void 2387 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr, 2388 struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae) 2389 { 2390 size_t copy_length; 2391 struct spdk_iov_xfer ix; 2392 2393 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2394 2395 if (offset < sizeof(ctrlr->changed_ns_list)) { 2396 copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset); 2397 if (copy_length) { 2398 spdk_iov_xfer_from_buf(&ix, (char *)&ctrlr->changed_ns_list + offset, copy_length); 2399 } 2400 } 2401 2402 /* Clear log page each time it is read */ 2403 ctrlr->changed_ns_list_count = 0; 2404 memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list)); 2405 2406 if (!rae) { 2407 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT); 2408 } 2409 } 2410 2411 /* The structure can be modified if we provide support for other commands in future */ 2412 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = { 2413 .admin_cmds_supported = { 2414 /* CSUPP, LBCC, NCC, NIC, CCC, CSE */ 2415 /* Get Log Page */ 2416 [SPDK_NVME_OPC_GET_LOG_PAGE] = {1, 0, 0, 0, 0, 0, 0, 0}, 2417 /* Identify */ 2418 [SPDK_NVME_OPC_IDENTIFY] = {1, 0, 0, 0, 0, 0, 0, 0}, 2419 /* Abort */ 2420 [SPDK_NVME_OPC_ABORT] = {1, 0, 0, 0, 0, 0, 0, 0}, 2421 /* Set Features */ 2422 [SPDK_NVME_OPC_SET_FEATURES] = {1, 0, 0, 0, 0, 0, 0, 0}, 2423 /* Get Features */ 2424 [SPDK_NVME_OPC_GET_FEATURES] = {1, 0, 0, 0, 0, 0, 0, 0}, 2425 /* Async Event Request */ 2426 [SPDK_NVME_OPC_ASYNC_EVENT_REQUEST] = {1, 0, 0, 0, 0, 0, 0, 0}, 2427 /* Keep Alive */ 2428 [SPDK_NVME_OPC_KEEP_ALIVE] = {1, 0, 0, 0, 0, 0, 0, 0}, 2429 }, 2430 .io_cmds_supported = { 2431 /* FLUSH */ 2432 [SPDK_NVME_OPC_FLUSH] = {1, 1, 0, 0, 0, 0, 0, 0}, 2433 /* WRITE */ 2434 [SPDK_NVME_OPC_WRITE] = {1, 1, 0, 0, 0, 0, 0, 0}, 2435 /* READ */ 2436 [SPDK_NVME_OPC_READ] = {1, 0, 0, 0, 0, 0, 0, 0}, 2437 /* WRITE ZEROES */ 2438 [SPDK_NVME_OPC_WRITE_ZEROES] = {1, 1, 0, 0, 0, 0, 0, 0}, 2439 /* DATASET MANAGEMENT */ 2440 [SPDK_NVME_OPC_DATASET_MANAGEMENT] = {1, 1, 0, 0, 0, 0, 0, 0}, 2441 /* COMPARE */ 2442 [SPDK_NVME_OPC_COMPARE] = {1, 0, 0, 0, 0, 0, 0, 0}, 2443 /* ZONE MANAGEMENT SEND */ 2444 [SPDK_NVME_OPC_ZONE_MGMT_SEND] = {1, 1, 0, 0, 0, 0, 0, 0}, 2445 /* ZONE MANAGEMENT RECEIVE */ 2446 [SPDK_NVME_OPC_ZONE_MGMT_RECV] = {1, 0, 0, 0, 0, 0, 0, 0}, 2447 /* COPY */ 2448 [SPDK_NVME_OPC_COPY] = {1, 1, 0, 0, 0, 0, 0, 0}, 2449 }, 2450 }; 2451 2452 static void 2453 nvmf_get_cmds_and_effects_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt, 2454 uint64_t offset, uint32_t length) 2455 { 2456 uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page); 2457 size_t copy_len = 0; 2458 struct spdk_nvme_cmds_and_effect_log_page cmds_and_effect_log_page = g_cmds_and_effect_log_page; 2459 struct spdk_nvme_cmds_and_effect_entry csupp_and_lbcc_effect_entry = {1, 1, 0, 0, 0, 0, 0, 0}; 2460 struct spdk_iov_xfer ix; 2461 2462 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2463 2464 if (offset < page_size) { 2465 if (ctrlr->subsys->zone_append_supported) { 2466 cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_ZONE_APPEND] = 2467 csupp_and_lbcc_effect_entry; 2468 } 2469 copy_len = spdk_min(page_size - offset, length); 2470 spdk_iov_xfer_from_buf(&ix, (char *)(&cmds_and_effect_log_page) + offset, copy_len); 2471 } 2472 } 2473 2474 static void 2475 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr, 2476 struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae) 2477 { 2478 uint32_t unit_log_len, avail_log_len, next_pos, copy_len; 2479 struct spdk_nvmf_reservation_log *log, *log_tmp; 2480 struct spdk_iov_xfer ix; 2481 2482 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2483 2484 unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log); 2485 /* No available log, return zeroed log pages */ 2486 if (!ctrlr->num_avail_log_pages) { 2487 return; 2488 } 2489 2490 avail_log_len = ctrlr->num_avail_log_pages * unit_log_len; 2491 if (offset >= avail_log_len) { 2492 return; 2493 } 2494 2495 next_pos = 0; 2496 TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) { 2497 TAILQ_REMOVE(&ctrlr->log_head, log, link); 2498 ctrlr->num_avail_log_pages--; 2499 2500 next_pos += unit_log_len; 2501 if (next_pos > offset) { 2502 copy_len = spdk_min(next_pos - offset, length); 2503 spdk_iov_xfer_from_buf(&ix, &log->log, copy_len); 2504 length -= copy_len; 2505 offset += copy_len; 2506 } 2507 free(log); 2508 2509 if (length == 0) { 2510 break; 2511 } 2512 } 2513 2514 if (!rae) { 2515 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT); 2516 } 2517 return; 2518 } 2519 2520 static int 2521 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req) 2522 { 2523 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2524 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2525 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2526 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 2527 struct spdk_nvme_transport_id cmd_source_trid; 2528 uint64_t offset, len; 2529 uint32_t rae, numdl, numdu; 2530 uint8_t lid; 2531 2532 if (req->iovcnt < 1) { 2533 SPDK_DEBUGLOG(nvmf, "get log command with no buffer\n"); 2534 response->status.sct = SPDK_NVME_SCT_GENERIC; 2535 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2536 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2537 } 2538 2539 offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32); 2540 if (offset & 3) { 2541 SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset); 2542 response->status.sct = SPDK_NVME_SCT_GENERIC; 2543 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2544 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2545 } 2546 2547 rae = cmd->cdw10_bits.get_log_page.rae; 2548 numdl = cmd->cdw10_bits.get_log_page.numdl; 2549 numdu = cmd->cdw11_bits.get_log_page.numdu; 2550 len = ((numdu << 16) + numdl + (uint64_t)1) * 4; 2551 if (len > req->length) { 2552 SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n", 2553 len, req->length); 2554 response->status.sct = SPDK_NVME_SCT_GENERIC; 2555 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2556 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2557 } 2558 2559 lid = cmd->cdw10_bits.get_log_page.lid; 2560 SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n", 2561 lid, offset, len, rae); 2562 2563 if (spdk_nvmf_subsystem_is_discovery(subsystem)) { 2564 switch (lid) { 2565 case SPDK_NVME_LOG_DISCOVERY: 2566 if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &cmd_source_trid)) { 2567 SPDK_ERRLOG("Failed to get LOG_DISCOVERY source trid\n"); 2568 response->status.sct = SPDK_NVME_SCT_GENERIC; 2569 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2570 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2571 } 2572 nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt, 2573 offset, len, &cmd_source_trid); 2574 if (!rae) { 2575 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT); 2576 } 2577 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2578 default: 2579 goto invalid_log_page; 2580 } 2581 } else { 2582 if (offset > len) { 2583 SPDK_ERRLOG("Get log page: offset (%" PRIu64 ") > len (%" PRIu64 ")\n", 2584 offset, len); 2585 response->status.sct = SPDK_NVME_SCT_GENERIC; 2586 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2587 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2588 } 2589 2590 switch (lid) { 2591 case SPDK_NVME_LOG_ERROR: 2592 nvmf_get_error_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2593 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2594 case SPDK_NVME_LOG_HEALTH_INFORMATION: 2595 /* TODO: actually fill out log page data */ 2596 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2597 case SPDK_NVME_LOG_FIRMWARE_SLOT: 2598 nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len); 2599 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2600 case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS: 2601 if (subsystem->flags.ana_reporting) { 2602 nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2603 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2604 } else { 2605 goto invalid_log_page; 2606 } 2607 case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG: 2608 nvmf_get_cmds_and_effects_log_page(ctrlr, req->iov, req->iovcnt, offset, len); 2609 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2610 case SPDK_NVME_LOG_CHANGED_NS_LIST: 2611 nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2612 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2613 case SPDK_NVME_LOG_RESERVATION_NOTIFICATION: 2614 nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2615 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2616 default: 2617 goto invalid_log_page; 2618 } 2619 } 2620 2621 invalid_log_page: 2622 SPDK_INFOLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid); 2623 response->status.sct = SPDK_NVME_SCT_GENERIC; 2624 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2625 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2626 } 2627 2628 static struct spdk_nvmf_ns * 2629 _nvmf_ctrlr_get_ns_safe(struct spdk_nvmf_ctrlr *ctrlr, 2630 uint32_t nsid, 2631 struct spdk_nvme_cpl *rsp) 2632 { 2633 struct spdk_nvmf_ns *ns; 2634 if (nsid == 0 || nsid > ctrlr->subsys->max_nsid) { 2635 SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", nsid); 2636 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2637 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 2638 return NULL; 2639 } 2640 2641 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 2642 if (ns == NULL || ns->bdev == NULL) { 2643 /* 2644 * Inactive namespaces should return a zero filled data structure. 2645 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(), 2646 * so we can just return early here. 2647 */ 2648 SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", nsid); 2649 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2650 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2651 return NULL; 2652 } 2653 return ns; 2654 } 2655 2656 int 2657 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr, 2658 struct spdk_nvme_cmd *cmd, 2659 struct spdk_nvme_cpl *rsp, 2660 struct spdk_nvme_ns_data *nsdata) 2661 { 2662 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2663 struct spdk_nvmf_ns *ns; 2664 uint32_t max_num_blocks, format_index; 2665 enum spdk_nvme_ana_state ana_state; 2666 2667 ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp); 2668 if (ns == NULL) { 2669 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2670 } 2671 2672 nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip); 2673 2674 assert(ctrlr->admin_qpair); 2675 2676 format_index = spdk_nvme_ns_get_format_index(nsdata); 2677 2678 /* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */ 2679 max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size / 2680 (1U << nsdata->lbaf[format_index].lbads); 2681 if (nsdata->noiob > max_num_blocks) { 2682 nsdata->noiob = max_num_blocks; 2683 } 2684 2685 /* Set NOWS equal to Controller MDTS */ 2686 if (nsdata->nsfeat.optperf) { 2687 nsdata->nows = max_num_blocks - 1; 2688 } 2689 2690 if (subsystem->flags.ana_reporting) { 2691 assert(ns->anagrpid - 1 < subsystem->max_nsid); 2692 nsdata->anagrpid = ns->anagrpid; 2693 2694 ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid); 2695 if (ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE || 2696 ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) { 2697 nsdata->nuse = 0; 2698 } 2699 } 2700 2701 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2702 } 2703 2704 static void 2705 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr, 2706 struct spdk_nvme_ctrlr_data *cdata) 2707 { 2708 cdata->oacs = ctrlr->cdata.oacs; 2709 2710 cdata->oacs.virtualization_management = 2711 g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL; 2712 cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL 2713 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL; 2714 cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL 2715 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL; 2716 cdata->oacs.device_self_test = 2717 g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL; 2718 cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL 2719 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL; 2720 cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr != 2721 NULL 2722 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL; 2723 cdata->oacs.format = 2724 g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL; 2725 cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL 2726 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL; 2727 cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr != 2728 NULL; 2729 } 2730 2731 int 2732 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata) 2733 { 2734 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2735 struct spdk_nvmf_transport *transport; 2736 2737 /* 2738 * Common fields for discovery and NVM subsystems 2739 */ 2740 assert(ctrlr->admin_qpair); 2741 transport = ctrlr->admin_qpair->transport; 2742 spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' '); 2743 assert((transport->opts.max_io_size % 4096) == 0); 2744 cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096); 2745 cdata->cntlid = ctrlr->cntlid; 2746 cdata->ver = ctrlr->vcprop.vs; 2747 cdata->aerl = ctrlr->cdata.aerl; 2748 cdata->lpa.edlp = 1; 2749 cdata->elpe = 127; 2750 cdata->maxcmd = transport->opts.max_queue_depth; 2751 cdata->sgls = ctrlr->cdata.sgls; 2752 cdata->fuses = ctrlr->cdata.fuses; 2753 cdata->acwu = 0; /* ACWU is 0-based. */ 2754 if (subsystem->flags.ana_reporting) { 2755 cdata->mnan = subsystem->max_nsid; 2756 } 2757 spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0'); 2758 2759 SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd); 2760 SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls)); 2761 2762 2763 if (spdk_nvmf_subsystem_is_discovery(subsystem)) { 2764 /* 2765 * NVM Discovery subsystem fields 2766 */ 2767 cdata->oaes.discovery_log_change_notices = 1; 2768 } else { 2769 cdata->vid = ctrlr->cdata.vid; 2770 cdata->ssvid = ctrlr->cdata.ssvid; 2771 cdata->ieee[0] = ctrlr->cdata.ieee[0]; 2772 cdata->ieee[1] = ctrlr->cdata.ieee[1]; 2773 cdata->ieee[2] = ctrlr->cdata.ieee[2]; 2774 2775 /* 2776 * NVM subsystem fields (reserved for discovery subsystems) 2777 */ 2778 spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' '); 2779 spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' '); 2780 cdata->kas = ctrlr->cdata.kas; 2781 2782 cdata->rab = 6; 2783 cdata->cmic.multi_port = 1; 2784 cdata->cmic.multi_ctrlr = 1; 2785 cdata->oaes.ns_attribute_notices = 1; 2786 cdata->ctratt.host_id_exhid_supported = 1; 2787 /* We do not have any actual limitation to the number of abort commands. 2788 * We follow the recommendation by the NVMe specification. 2789 */ 2790 cdata->acl = NVMF_ABORT_COMMAND_LIMIT; 2791 cdata->frmw.slot1_ro = 1; 2792 cdata->frmw.num_slots = 1; 2793 2794 cdata->lpa.celp = 1; /* Command Effects log page supported */ 2795 2796 cdata->sqes.min = 6; 2797 cdata->sqes.max = 6; 2798 cdata->cqes.min = 4; 2799 cdata->cqes.max = 4; 2800 cdata->nn = subsystem->max_nsid; 2801 cdata->vwc.present = 1; 2802 cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED; 2803 2804 cdata->nvmf_specific = ctrlr->cdata.nvmf_specific; 2805 2806 cdata->oncs.compare = ctrlr->cdata.oncs.compare; 2807 cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr); 2808 cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr); 2809 cdata->oncs.reservations = ctrlr->cdata.oncs.reservations; 2810 cdata->oncs.copy = ctrlr->cdata.oncs.copy; 2811 cdata->ocfs.copy_format0 = cdata->oncs.copy; 2812 if (subsystem->flags.ana_reporting) { 2813 /* Asymmetric Namespace Access Reporting is supported. */ 2814 cdata->cmic.ana_reporting = 1; 2815 cdata->oaes.ana_change_notices = 1; 2816 2817 cdata->anatt = ANA_TRANSITION_TIME_IN_SEC; 2818 /* ANA Change state is not used, and ANA Persistent Loss state 2819 * is not supported for now. 2820 */ 2821 cdata->anacap.ana_optimized_state = 1; 2822 cdata->anacap.ana_non_optimized_state = 1; 2823 cdata->anacap.ana_inaccessible_state = 1; 2824 /* ANAGRPID does not change while namespace is attached to controller */ 2825 cdata->anacap.no_change_anagrpid = 1; 2826 cdata->anagrpmax = subsystem->max_nsid; 2827 cdata->nanagrpid = subsystem->max_nsid; 2828 } 2829 2830 nvmf_ctrlr_populate_oacs(ctrlr, cdata); 2831 2832 assert(subsystem->tgt != NULL); 2833 cdata->crdt[0] = subsystem->tgt->crdt[0]; 2834 cdata->crdt[1] = subsystem->tgt->crdt[1]; 2835 cdata->crdt[2] = subsystem->tgt->crdt[2]; 2836 2837 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n", 2838 cdata->nvmf_specific.ioccsz); 2839 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n", 2840 cdata->nvmf_specific.iorcsz); 2841 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n", 2842 cdata->nvmf_specific.icdoff); 2843 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n", 2844 *(uint8_t *)&cdata->nvmf_specific.ctrattr); 2845 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n", 2846 cdata->nvmf_specific.msdbd); 2847 } 2848 2849 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2850 } 2851 2852 static int 2853 nvmf_ns_identify_iocs_zns(struct spdk_nvmf_ns *ns, 2854 struct spdk_nvme_cmd *cmd, 2855 struct spdk_nvme_cpl *rsp, 2856 struct spdk_nvme_zns_ns_data *nsdata_zns) 2857 { 2858 nsdata_zns->zoc.variable_zone_capacity = 0; 2859 nsdata_zns->zoc.zone_active_excursions = 0; 2860 nsdata_zns->ozcs.read_across_zone_boundaries = 1; 2861 /* Underflowing the zero based mar and mor bdev helper results in the correct 2862 value of FFFFFFFFh. */ 2863 nsdata_zns->mar = spdk_bdev_get_max_active_zones(ns->bdev) - 1; 2864 nsdata_zns->mor = spdk_bdev_get_max_open_zones(ns->bdev) - 1; 2865 nsdata_zns->rrl = 0; 2866 nsdata_zns->frl = 0; 2867 nsdata_zns->lbafe[0].zsze = spdk_bdev_get_zone_size(ns->bdev); 2868 2869 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2870 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2871 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2872 } 2873 2874 int 2875 spdk_nvmf_ns_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr, 2876 struct spdk_nvme_cmd *cmd, 2877 struct spdk_nvme_cpl *rsp, 2878 void *nsdata, 2879 size_t nsdata_size) 2880 { 2881 uint8_t csi = cmd->cdw11_bits.identify.csi; 2882 struct spdk_nvmf_ns *ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp); 2883 2884 memset(nsdata, 0, nsdata_size); 2885 2886 if (ns == NULL) { 2887 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2888 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 2889 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2890 } 2891 2892 switch (csi) { 2893 case SPDK_NVME_CSI_ZNS: 2894 return nvmf_ns_identify_iocs_zns(ns, cmd, rsp, nsdata); 2895 default: 2896 break; 2897 } 2898 2899 SPDK_DEBUGLOG(nvmf, 2900 "Returning zero filled struct for the iocs specific ns " 2901 "identify command and CSI 0x%02x\n", 2902 csi); 2903 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2904 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2905 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2906 } 2907 2908 static int 2909 nvmf_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ctrlr *ctrlr, 2910 struct spdk_nvme_cmd *cmd, 2911 struct spdk_nvme_cpl *rsp, 2912 struct spdk_nvme_nvm_ctrlr_data *cdata_nvm) 2913 { 2914 /* The unit of max_write_zeroes_size_kib is KiB. 2915 * The unit of wzsl is the minimum memory page size(2 ^ (12 + CAP.MPSMIN) bytes) 2916 * and is reported as a power of two (2^n). 2917 */ 2918 cdata_nvm->wzsl = spdk_u64log2(ctrlr->subsys->max_write_zeroes_size_kib >> 2919 (2 + ctrlr->vcprop.cap.bits.mpsmin)); 2920 2921 /* The unit of max_discard_size_kib is KiB. 2922 * The dmrsl indicates the maximum number of logical blocks for 2923 * dataset management command. 2924 */ 2925 cdata_nvm->dmrsl = ctrlr->subsys->max_discard_size_kib << 1; 2926 cdata_nvm->dmrl = 1; 2927 2928 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2929 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2930 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2931 } 2932 2933 static int 2934 nvmf_ctrlr_identify_iocs_zns(struct spdk_nvmf_ctrlr *ctrlr, 2935 struct spdk_nvme_cmd *cmd, 2936 struct spdk_nvme_cpl *rsp, 2937 struct spdk_nvme_zns_ctrlr_data *cdata_zns) 2938 { 2939 /* The unit of max_zone_append_size_kib is KiB. 2940 The unit of zasl is the minimum memory page size 2941 (2 ^ (12 + CAP.MPSMIN) KiB) 2942 and is reported as a power of two (2^n). */ 2943 cdata_zns->zasl = spdk_u64log2(ctrlr->subsys->max_zone_append_size_kib >> 2944 (12 + ctrlr->vcprop.cap.bits.mpsmin)); 2945 2946 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2947 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2948 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2949 } 2950 2951 int 2952 spdk_nvmf_ctrlr_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr, 2953 struct spdk_nvme_cmd *cmd, 2954 struct spdk_nvme_cpl *rsp, 2955 void *cdata, 2956 size_t cdata_size) 2957 { 2958 uint8_t csi = cmd->cdw11_bits.identify.csi; 2959 2960 memset(cdata, 0, cdata_size); 2961 2962 switch (csi) { 2963 case SPDK_NVME_CSI_NVM: 2964 return nvmf_ctrlr_identify_iocs_nvm(ctrlr, cmd, rsp, cdata); 2965 case SPDK_NVME_CSI_ZNS: 2966 return nvmf_ctrlr_identify_iocs_zns(ctrlr, cmd, rsp, cdata); 2967 default: 2968 break; 2969 } 2970 2971 SPDK_DEBUGLOG(nvmf, 2972 "Returning zero filled struct for the iocs specific ctrlr " 2973 "identify command and CSI 0x%02x\n", 2974 csi); 2975 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2976 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2977 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2978 } 2979 2980 static int 2981 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_ctrlr *ctrlr, 2982 struct spdk_nvme_cmd *cmd, 2983 struct spdk_nvme_cpl *rsp, 2984 struct spdk_nvme_ns_list *ns_list) 2985 { 2986 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2987 struct spdk_nvmf_ns *ns; 2988 uint32_t count = 0; 2989 2990 if (cmd->nsid >= 0xfffffffeUL) { 2991 SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid); 2992 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 2993 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2994 } 2995 2996 memset(ns_list, 0, sizeof(*ns_list)); 2997 2998 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 2999 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 3000 if (ns->opts.nsid <= cmd->nsid) { 3001 continue; 3002 } 3003 3004 ns_list->ns_list[count++] = ns->opts.nsid; 3005 if (count == SPDK_COUNTOF(ns_list->ns_list)) { 3006 break; 3007 } 3008 } 3009 3010 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3011 } 3012 3013 static void 3014 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain, 3015 enum spdk_nvme_nidt type, 3016 const void *data, size_t data_size) 3017 { 3018 struct spdk_nvme_ns_id_desc *desc; 3019 size_t desc_size = sizeof(*desc) + data_size; 3020 3021 /* 3022 * These should never fail in practice, since all valid NS ID descriptors 3023 * should be defined so that they fit in the available 4096-byte buffer. 3024 */ 3025 assert(data_size > 0); 3026 assert(data_size <= UINT8_MAX); 3027 assert(desc_size < *buf_remain); 3028 if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) { 3029 return; 3030 } 3031 3032 desc = *buf_ptr; 3033 desc->nidt = type; 3034 desc->nidl = data_size; 3035 memcpy(desc->nid, data, data_size); 3036 3037 *buf_ptr += desc_size; 3038 *buf_remain -= desc_size; 3039 } 3040 3041 static int 3042 nvmf_ctrlr_identify_ns_id_descriptor_list( 3043 struct spdk_nvmf_ctrlr *ctrlr, 3044 struct spdk_nvme_cmd *cmd, 3045 struct spdk_nvme_cpl *rsp, 3046 void *id_desc_list, size_t id_desc_list_size) 3047 { 3048 struct spdk_nvmf_ns *ns; 3049 size_t buf_remain = id_desc_list_size; 3050 void *buf_ptr = id_desc_list; 3051 3052 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 3053 if (ns == NULL || ns->bdev == NULL) { 3054 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3055 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 3056 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3057 } 3058 3059 #define ADD_ID_DESC(type, data, size) \ 3060 do { \ 3061 if (!spdk_mem_all_zero(data, size)) { \ 3062 _add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \ 3063 } \ 3064 } while (0) 3065 3066 ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64)); 3067 ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid)); 3068 ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid)); 3069 ADD_ID_DESC(SPDK_NVME_NIDT_CSI, &ns->csi, sizeof(uint8_t)); 3070 3071 /* 3072 * The list is automatically 0-terminated, both in the temporary buffer 3073 * used by nvmf_ctrlr_identify(), and the eventual iov destination - 3074 * controller to host buffers in admin commands always get zeroed in 3075 * nvmf_ctrlr_process_admin_cmd(). 3076 */ 3077 3078 #undef ADD_ID_DESC 3079 3080 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3081 } 3082 3083 static int 3084 nvmf_ctrlr_identify_iocs(struct spdk_nvmf_ctrlr *ctrlr, 3085 struct spdk_nvme_cmd *cmd, 3086 struct spdk_nvme_cpl *rsp, 3087 void *cdata, size_t cdata_size) 3088 { 3089 struct spdk_nvme_iocs_vector *vector; 3090 struct spdk_nvmf_ns *ns; 3091 3092 if (cdata_size < sizeof(struct spdk_nvme_iocs_vector)) { 3093 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3094 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3095 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3096 } 3097 3098 /* For now we only support this command sent to the current 3099 * controller. 3100 */ 3101 if (cmd->cdw10_bits.identify.cntid != 0xFFFF && 3102 cmd->cdw10_bits.identify.cntid != ctrlr->cntlid) { 3103 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3104 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3105 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3106 } 3107 memset(cdata, 0, cdata_size); 3108 3109 vector = cdata; 3110 vector->nvm = 1; 3111 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL; 3112 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 3113 if (ns->bdev == NULL) { 3114 continue; 3115 } 3116 if (spdk_bdev_is_zoned(ns->bdev)) { 3117 vector->zns = 1; 3118 } 3119 } 3120 3121 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3122 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 3123 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3124 } 3125 3126 static int 3127 nvmf_ctrlr_identify(struct spdk_nvmf_request *req) 3128 { 3129 uint8_t cns; 3130 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3131 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3132 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 3133 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 3134 int ret = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3135 char tmpbuf[SPDK_NVME_IDENTIFY_BUFLEN] = ""; 3136 struct spdk_iov_xfer ix; 3137 3138 if (req->iovcnt < 1 || req->length < SPDK_NVME_IDENTIFY_BUFLEN) { 3139 SPDK_DEBUGLOG(nvmf, "identify command with invalid buffer\n"); 3140 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3141 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3142 return ret; 3143 } 3144 3145 cns = cmd->cdw10_bits.identify.cns; 3146 3147 if (spdk_nvmf_subsystem_is_discovery(subsystem) && 3148 cns != SPDK_NVME_IDENTIFY_CTRLR) { 3149 /* Discovery controllers only support Identify Controller */ 3150 goto invalid_cns; 3151 } 3152 3153 /* 3154 * We must use a temporary buffer: it's entirely possible the out buffer 3155 * is split across more than one IOV. 3156 */ 3157 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 3158 3159 SPDK_DEBUGLOG(nvmf, "Received identify command with CNS 0x%02x\n", cns); 3160 3161 switch (cns) { 3162 case SPDK_NVME_IDENTIFY_NS: 3163 ret = spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, (void *)&tmpbuf); 3164 break; 3165 case SPDK_NVME_IDENTIFY_CTRLR: 3166 ret = spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, (void *)&tmpbuf); 3167 break; 3168 case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST: 3169 ret = nvmf_ctrlr_identify_active_ns_list(ctrlr, cmd, rsp, (void *)&tmpbuf); 3170 break; 3171 case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST: 3172 ret = nvmf_ctrlr_identify_ns_id_descriptor_list(ctrlr, cmd, rsp, 3173 tmpbuf, req->length); 3174 break; 3175 case SPDK_NVME_IDENTIFY_NS_IOCS: 3176 ret = spdk_nvmf_ns_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length); 3177 break; 3178 case SPDK_NVME_IDENTIFY_CTRLR_IOCS: 3179 ret = spdk_nvmf_ctrlr_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length); 3180 break; 3181 case SPDK_NVME_IDENTIFY_IOCS: 3182 ret = nvmf_ctrlr_identify_iocs(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length); 3183 break; 3184 default: 3185 goto invalid_cns; 3186 } 3187 3188 if (ret == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 3189 spdk_iov_xfer_from_buf(&ix, tmpbuf, sizeof(tmpbuf)); 3190 } 3191 3192 return ret; 3193 3194 invalid_cns: 3195 SPDK_DEBUGLOG(nvmf, "Identify command with unsupported CNS 0x%02x\n", cns); 3196 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3197 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3198 return ret; 3199 } 3200 3201 static bool 3202 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid) 3203 { 3204 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 3205 struct spdk_nvmf_request *req; 3206 int i; 3207 3208 if (!nvmf_qpair_is_admin_queue(qpair)) { 3209 return false; 3210 } 3211 3212 assert(spdk_get_thread() == ctrlr->thread); 3213 3214 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 3215 if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) { 3216 SPDK_DEBUGLOG(nvmf, "Aborting AER request\n"); 3217 req = ctrlr->aer_req[i]; 3218 ctrlr->aer_req[i] = NULL; 3219 ctrlr->nr_aer_reqs--; 3220 3221 /* Move the last req to the aborting position for making aer_reqs 3222 * in continuous 3223 */ 3224 if (i < ctrlr->nr_aer_reqs) { 3225 ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs]; 3226 ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL; 3227 } 3228 3229 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3230 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3231 _nvmf_request_complete(req); 3232 return true; 3233 } 3234 } 3235 3236 return false; 3237 } 3238 3239 void 3240 nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair) 3241 { 3242 struct spdk_nvmf_request *req, *tmp; 3243 3244 TAILQ_FOREACH_SAFE(req, &qpair->outstanding, link, tmp) { 3245 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) { 3246 /* Zero-copy requests are kept on the outstanding queue from the moment 3247 * zcopy_start is sent until a zcopy_end callback is received. Therefore, 3248 * we can't remove them from the outstanding queue here, but need to rely on 3249 * the transport to do a zcopy_end to release their buffers and, in turn, 3250 * remove them from the queue. 3251 */ 3252 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3253 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3254 nvmf_transport_req_free(req); 3255 } 3256 } 3257 } 3258 3259 static void 3260 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req) 3261 { 3262 uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3263 3264 if (nvmf_qpair_abort_aer(qpair, cid)) { 3265 SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n", 3266 qpair->ctrlr, qpair->qid, cid); 3267 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */ 3268 3269 spdk_nvmf_request_complete(req); 3270 return; 3271 } 3272 3273 nvmf_transport_qpair_abort_request(qpair, req); 3274 } 3275 3276 static void 3277 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status) 3278 { 3279 struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i); 3280 3281 if (status == 0) { 3282 /* There was no qpair whose ID matches SQID of the abort command. 3283 * Hence call _nvmf_request_complete() here. 3284 */ 3285 _nvmf_request_complete(req); 3286 } 3287 } 3288 3289 static void 3290 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i) 3291 { 3292 struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i); 3293 struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i); 3294 struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch); 3295 uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid; 3296 struct spdk_nvmf_qpair *qpair; 3297 3298 TAILQ_FOREACH(qpair, &group->qpairs, link) { 3299 if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) { 3300 /* Found the qpair */ 3301 3302 nvmf_qpair_abort_request(qpair, req); 3303 3304 /* Return -1 for the status so the iteration across threads stops. */ 3305 spdk_for_each_channel_continue(i, -1); 3306 return; 3307 } 3308 } 3309 3310 spdk_for_each_channel_continue(i, 0); 3311 } 3312 3313 static int 3314 nvmf_ctrlr_abort(struct spdk_nvmf_request *req) 3315 { 3316 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 3317 3318 rsp->cdw0 = 1U; /* Command not aborted */ 3319 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3320 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 3321 3322 /* Send a message to each poll group, searching for this ctrlr, sqid, and command. */ 3323 spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt, 3324 nvmf_ctrlr_abort_on_pg, 3325 req, 3326 nvmf_ctrlr_abort_done 3327 ); 3328 3329 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 3330 } 3331 3332 int 3333 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req) 3334 { 3335 struct spdk_nvmf_request *req_to_abort = req->req_to_abort; 3336 struct spdk_bdev *bdev; 3337 struct spdk_bdev_desc *desc; 3338 struct spdk_io_channel *ch; 3339 int rc; 3340 3341 assert(req_to_abort != NULL); 3342 3343 if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr && 3344 nvmf_qpair_is_admin_queue(req_to_abort->qpair)) { 3345 return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req); 3346 } 3347 3348 rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort, 3349 &bdev, &desc, &ch); 3350 if (rc != 0) { 3351 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3352 } 3353 3354 return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort); 3355 } 3356 3357 static int 3358 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0) 3359 { 3360 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 3361 3362 rsp->cdw0 = cdw0; 3363 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3364 } 3365 3366 /* we have to use the typedef in the function declaration to appease astyle. */ 3367 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t; 3368 3369 static spdk_nvme_path_status_code_t 3370 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state) 3371 { 3372 switch (ana_state) { 3373 case SPDK_NVME_ANA_INACCESSIBLE_STATE: 3374 return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE; 3375 case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE: 3376 return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS; 3377 case SPDK_NVME_ANA_CHANGE_STATE: 3378 return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION; 3379 default: 3380 return SPDK_NVME_SC_INTERNAL_PATH_ERROR; 3381 } 3382 } 3383 3384 static int 3385 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req) 3386 { 3387 uint8_t feature; 3388 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3389 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3390 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 3391 enum spdk_nvme_ana_state ana_state; 3392 3393 feature = cmd->cdw10_bits.get_features.fid; 3394 3395 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 3396 /* 3397 * Features supported by Discovery controller 3398 */ 3399 switch (feature) { 3400 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3401 return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw); 3402 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3403 return get_features_generic(req, ctrlr->feat.async_event_configuration.raw); 3404 default: 3405 SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature); 3406 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3407 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3408 } 3409 } 3410 /* 3411 * Process Get Features command for non-discovery controller 3412 */ 3413 ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid); 3414 switch (ana_state) { 3415 case SPDK_NVME_ANA_INACCESSIBLE_STATE: 3416 case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE: 3417 case SPDK_NVME_ANA_CHANGE_STATE: 3418 switch (feature) { 3419 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3420 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3421 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3422 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3423 response->status.sct = SPDK_NVME_SCT_PATH; 3424 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 3425 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3426 default: 3427 break; 3428 } 3429 break; 3430 default: 3431 break; 3432 } 3433 3434 switch (feature) { 3435 case SPDK_NVME_FEAT_ARBITRATION: 3436 return get_features_generic(req, ctrlr->feat.arbitration.raw); 3437 case SPDK_NVME_FEAT_POWER_MANAGEMENT: 3438 return get_features_generic(req, ctrlr->feat.power_management.raw); 3439 case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD: 3440 return nvmf_ctrlr_get_features_temperature_threshold(req); 3441 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3442 return get_features_generic(req, ctrlr->feat.error_recovery.raw); 3443 case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE: 3444 return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw); 3445 case SPDK_NVME_FEAT_NUMBER_OF_QUEUES: 3446 return get_features_generic(req, ctrlr->feat.number_of_queues.raw); 3447 case SPDK_NVME_FEAT_INTERRUPT_COALESCING: 3448 return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw); 3449 case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION: 3450 return nvmf_ctrlr_get_features_interrupt_vector_configuration(req); 3451 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3452 return get_features_generic(req, ctrlr->feat.write_atomicity.raw); 3453 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3454 return get_features_generic(req, ctrlr->feat.async_event_configuration.raw); 3455 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3456 return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw); 3457 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 3458 return nvmf_ctrlr_get_features_host_identifier(req); 3459 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3460 return nvmf_ctrlr_get_features_reservation_notification_mask(req); 3461 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3462 return nvmf_ctrlr_get_features_reservation_persistence(req); 3463 case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT: 3464 return nvmf_ctrlr_get_features_host_behavior_support(req); 3465 default: 3466 SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature); 3467 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3468 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3469 } 3470 } 3471 3472 static int 3473 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req) 3474 { 3475 uint8_t feature, save; 3476 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3477 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3478 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 3479 enum spdk_nvme_ana_state ana_state; 3480 /* 3481 * Features are not saveable by the controller as indicated by 3482 * ONCS field of the Identify Controller data. 3483 * */ 3484 save = cmd->cdw10_bits.set_features.sv; 3485 if (save) { 3486 response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE; 3487 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 3488 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3489 } 3490 3491 feature = cmd->cdw10_bits.set_features.fid; 3492 3493 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 3494 /* 3495 * Features supported by Discovery controller 3496 */ 3497 switch (feature) { 3498 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3499 return nvmf_ctrlr_set_features_keep_alive_timer(req); 3500 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3501 return nvmf_ctrlr_set_features_async_event_configuration(req); 3502 default: 3503 SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature); 3504 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3505 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3506 } 3507 } 3508 /* 3509 * Process Set Features command for non-discovery controller 3510 */ 3511 ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid); 3512 switch (ana_state) { 3513 case SPDK_NVME_ANA_INACCESSIBLE_STATE: 3514 case SPDK_NVME_ANA_CHANGE_STATE: 3515 if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 3516 response->status.sct = SPDK_NVME_SCT_PATH; 3517 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 3518 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3519 } else { 3520 switch (feature) { 3521 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3522 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3523 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3524 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3525 response->status.sct = SPDK_NVME_SCT_PATH; 3526 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 3527 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3528 default: 3529 break; 3530 } 3531 } 3532 break; 3533 case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE: 3534 response->status.sct = SPDK_NVME_SCT_PATH; 3535 response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS; 3536 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3537 default: 3538 break; 3539 } 3540 3541 switch (feature) { 3542 case SPDK_NVME_FEAT_ARBITRATION: 3543 return nvmf_ctrlr_set_features_arbitration(req); 3544 case SPDK_NVME_FEAT_POWER_MANAGEMENT: 3545 return nvmf_ctrlr_set_features_power_management(req); 3546 case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD: 3547 return nvmf_ctrlr_set_features_temperature_threshold(req); 3548 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3549 return nvmf_ctrlr_set_features_error_recovery(req); 3550 case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE: 3551 return nvmf_ctrlr_set_features_volatile_write_cache(req); 3552 case SPDK_NVME_FEAT_NUMBER_OF_QUEUES: 3553 return nvmf_ctrlr_set_features_number_of_queues(req); 3554 case SPDK_NVME_FEAT_INTERRUPT_COALESCING: 3555 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 3556 response->status.sc = SPDK_NVME_SC_FEATURE_NOT_CHANGEABLE; 3557 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3558 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3559 return nvmf_ctrlr_set_features_write_atomicity(req); 3560 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3561 return nvmf_ctrlr_set_features_async_event_configuration(req); 3562 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3563 return nvmf_ctrlr_set_features_keep_alive_timer(req); 3564 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 3565 return nvmf_ctrlr_set_features_host_identifier(req); 3566 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3567 return nvmf_ctrlr_set_features_reservation_notification_mask(req); 3568 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3569 return nvmf_ctrlr_set_features_reservation_persistence(req); 3570 case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT: 3571 return nvmf_ctrlr_set_features_host_behavior_support(req); 3572 default: 3573 SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature); 3574 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3575 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3576 } 3577 } 3578 3579 static int 3580 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req) 3581 { 3582 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3583 3584 SPDK_DEBUGLOG(nvmf, "Keep Alive\n"); 3585 /* 3586 * To handle keep alive just clear or reset the 3587 * ctrlr based keep alive duration counter. 3588 * When added, a separate timer based process 3589 * will monitor if the time since last recorded 3590 * keep alive has exceeded the max duration and 3591 * take appropriate action. 3592 */ 3593 ctrlr->last_keep_alive_tick = spdk_get_ticks(); 3594 3595 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3596 } 3597 3598 int 3599 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req) 3600 { 3601 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3602 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3603 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 3604 struct spdk_nvmf_subsystem_poll_group *sgroup; 3605 int rc; 3606 3607 if (ctrlr == NULL) { 3608 SPDK_ERRLOG("Admin command sent before CONNECT\n"); 3609 response->status.sct = SPDK_NVME_SCT_GENERIC; 3610 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 3611 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3612 } 3613 3614 if (cmd->opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST) { 3615 /* We do not want to treat AERs as outstanding commands, 3616 * so decrement mgmt_io_outstanding here to offset 3617 * the increment that happened prior to this call. 3618 */ 3619 sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id]; 3620 assert(sgroup != NULL); 3621 sgroup->mgmt_io_outstanding--; 3622 } 3623 3624 assert(spdk_get_thread() == ctrlr->thread); 3625 3626 if (cmd->fuse != 0) { 3627 /* Fused admin commands are not supported. */ 3628 response->status.sct = SPDK_NVME_SCT_GENERIC; 3629 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3630 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3631 } 3632 3633 if (ctrlr->vcprop.cc.bits.en != 1) { 3634 SPDK_ERRLOG("Admin command sent to disabled controller\n"); 3635 response->status.sct = SPDK_NVME_SCT_GENERIC; 3636 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 3637 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3638 } 3639 3640 if (req->iovcnt && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 3641 spdk_iov_memset(req->iov, req->iovcnt, 0); 3642 } 3643 3644 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 3645 /* Discovery controllers only support these admin OPS. */ 3646 switch (cmd->opc) { 3647 case SPDK_NVME_OPC_IDENTIFY: 3648 case SPDK_NVME_OPC_GET_LOG_PAGE: 3649 case SPDK_NVME_OPC_KEEP_ALIVE: 3650 case SPDK_NVME_OPC_SET_FEATURES: 3651 case SPDK_NVME_OPC_GET_FEATURES: 3652 case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST: 3653 break; 3654 default: 3655 goto invalid_opcode; 3656 } 3657 } 3658 3659 /* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */ 3660 if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) { 3661 rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req); 3662 if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 3663 /* The handler took care of this command */ 3664 return rc; 3665 } 3666 } 3667 3668 switch (cmd->opc) { 3669 case SPDK_NVME_OPC_GET_LOG_PAGE: 3670 return nvmf_ctrlr_get_log_page(req); 3671 case SPDK_NVME_OPC_IDENTIFY: 3672 return nvmf_ctrlr_identify(req); 3673 case SPDK_NVME_OPC_ABORT: 3674 return nvmf_ctrlr_abort(req); 3675 case SPDK_NVME_OPC_GET_FEATURES: 3676 return nvmf_ctrlr_get_features(req); 3677 case SPDK_NVME_OPC_SET_FEATURES: 3678 return nvmf_ctrlr_set_features(req); 3679 case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST: 3680 return nvmf_ctrlr_async_event_request(req); 3681 case SPDK_NVME_OPC_KEEP_ALIVE: 3682 return nvmf_ctrlr_keep_alive(req); 3683 3684 case SPDK_NVME_OPC_CREATE_IO_SQ: 3685 case SPDK_NVME_OPC_CREATE_IO_CQ: 3686 case SPDK_NVME_OPC_DELETE_IO_SQ: 3687 case SPDK_NVME_OPC_DELETE_IO_CQ: 3688 /* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */ 3689 goto invalid_opcode; 3690 3691 default: 3692 goto invalid_opcode; 3693 } 3694 3695 invalid_opcode: 3696 SPDK_INFOLOG(nvmf, "Unsupported admin opcode 0x%x\n", cmd->opc); 3697 response->status.sct = SPDK_NVME_SCT_GENERIC; 3698 response->status.sc = SPDK_NVME_SC_INVALID_OPCODE; 3699 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3700 } 3701 3702 static int 3703 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req) 3704 { 3705 struct spdk_nvmf_qpair *qpair = req->qpair; 3706 struct spdk_nvmf_capsule_cmd *cap_hdr; 3707 3708 cap_hdr = &req->cmd->nvmf_cmd; 3709 3710 if (qpair->ctrlr == NULL) { 3711 /* No ctrlr established yet; the only valid command is Connect */ 3712 if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) { 3713 return nvmf_ctrlr_cmd_connect(req); 3714 } else { 3715 SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n", 3716 cap_hdr->fctype); 3717 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3718 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 3719 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3720 } 3721 } else if (nvmf_qpair_is_admin_queue(qpair)) { 3722 /* 3723 * Controller session is established, and this is an admin queue. 3724 * Disallow Connect and allow other fabrics commands. 3725 */ 3726 switch (cap_hdr->fctype) { 3727 case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET: 3728 return nvmf_property_set(req); 3729 case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET: 3730 return nvmf_property_get(req); 3731 default: 3732 SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n", 3733 cap_hdr->fctype); 3734 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3735 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 3736 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3737 } 3738 } else { 3739 /* Controller session is established, and this is an I/O queue */ 3740 /* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */ 3741 SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype); 3742 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3743 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 3744 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3745 } 3746 } 3747 3748 static inline void 3749 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr, 3750 union spdk_nvme_async_event_completion *event) 3751 { 3752 struct spdk_nvmf_async_event_completion *nvmf_event; 3753 3754 nvmf_event = calloc(1, sizeof(*nvmf_event)); 3755 if (!nvmf_event) { 3756 SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n"); 3757 return; 3758 } 3759 nvmf_event->event.raw = event->raw; 3760 STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link); 3761 } 3762 3763 static inline int 3764 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr, 3765 union spdk_nvme_async_event_completion *event) 3766 { 3767 struct spdk_nvmf_request *req; 3768 struct spdk_nvme_cpl *rsp; 3769 3770 assert(spdk_get_thread() == ctrlr->thread); 3771 3772 /* If there is no outstanding AER request, queue the event. Then 3773 * if an AER is later submitted, this event can be sent as a 3774 * response. 3775 */ 3776 if (ctrlr->nr_aer_reqs == 0) { 3777 nvmf_ctrlr_queue_pending_async_event(ctrlr, event); 3778 return 0; 3779 } 3780 3781 req = ctrlr->aer_req[--ctrlr->nr_aer_reqs]; 3782 rsp = &req->rsp->nvme_cpl; 3783 3784 rsp->cdw0 = event->raw; 3785 3786 _nvmf_request_complete(req); 3787 ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL; 3788 3789 return 0; 3790 } 3791 3792 int 3793 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr) 3794 { 3795 union spdk_nvme_async_event_completion event = {0}; 3796 3797 /* Users may disable the event notification */ 3798 if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) { 3799 return 0; 3800 } 3801 3802 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) { 3803 return 0; 3804 } 3805 3806 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE; 3807 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED; 3808 event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST; 3809 3810 return nvmf_ctrlr_async_event_notification(ctrlr, &event); 3811 } 3812 3813 int 3814 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr) 3815 { 3816 union spdk_nvme_async_event_completion event = {0}; 3817 3818 /* Users may disable the event notification */ 3819 if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) { 3820 return 0; 3821 } 3822 3823 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) { 3824 return 0; 3825 } 3826 3827 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE; 3828 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE; 3829 event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS; 3830 3831 return nvmf_ctrlr_async_event_notification(ctrlr, &event); 3832 } 3833 3834 void 3835 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr) 3836 { 3837 union spdk_nvme_async_event_completion event = {0}; 3838 3839 if (!ctrlr->num_avail_log_pages) { 3840 return; 3841 } 3842 3843 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) { 3844 return; 3845 } 3846 3847 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO; 3848 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL; 3849 event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION; 3850 3851 nvmf_ctrlr_async_event_notification(ctrlr, &event); 3852 } 3853 3854 void 3855 nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx) 3856 { 3857 union spdk_nvme_async_event_completion event = {0}; 3858 struct spdk_nvmf_ctrlr *ctrlr = ctx; 3859 3860 /* Users may disable the event notification manually or 3861 * it may not be enabled due to keep alive timeout 3862 * not being set in connect command to discovery controller. 3863 */ 3864 if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) { 3865 return; 3866 } 3867 3868 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) { 3869 return; 3870 } 3871 3872 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE; 3873 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE; 3874 event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY; 3875 3876 nvmf_ctrlr_async_event_notification(ctrlr, &event); 3877 } 3878 3879 int 3880 spdk_nvmf_ctrlr_async_event_error_event(struct spdk_nvmf_ctrlr *ctrlr, 3881 enum spdk_nvme_async_event_info_error info) 3882 { 3883 union spdk_nvme_async_event_completion event; 3884 3885 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT)) { 3886 return 0; 3887 } 3888 3889 if (info > SPDK_NVME_ASYNC_EVENT_FW_IMAGE_LOAD) { 3890 return 0; 3891 } 3892 3893 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_ERROR; 3894 event.bits.log_page_identifier = SPDK_NVME_LOG_ERROR; 3895 event.bits.async_event_info = info; 3896 3897 return nvmf_ctrlr_async_event_notification(ctrlr, &event); 3898 } 3899 3900 void 3901 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair) 3902 { 3903 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 3904 int i; 3905 3906 if (!nvmf_qpair_is_admin_queue(qpair)) { 3907 return; 3908 } 3909 3910 assert(spdk_get_thread() == ctrlr->thread); 3911 3912 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 3913 spdk_nvmf_request_free(ctrlr->aer_req[i]); 3914 ctrlr->aer_req[i] = NULL; 3915 } 3916 3917 ctrlr->nr_aer_reqs = 0; 3918 } 3919 3920 void 3921 spdk_nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr) 3922 { 3923 struct spdk_nvmf_request *req; 3924 int i; 3925 3926 assert(spdk_get_thread() == ctrlr->thread); 3927 3928 if (!ctrlr->nr_aer_reqs) { 3929 return; 3930 } 3931 3932 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 3933 req = ctrlr->aer_req[i]; 3934 3935 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3936 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3937 _nvmf_request_complete(req); 3938 3939 ctrlr->aer_req[i] = NULL; 3940 } 3941 3942 ctrlr->nr_aer_reqs = 0; 3943 } 3944 3945 static void 3946 _nvmf_ctrlr_add_reservation_log(void *ctx) 3947 { 3948 struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx; 3949 struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr; 3950 3951 ctrlr->log_page_count++; 3952 3953 /* Maximum number of queued log pages is 255 */ 3954 if (ctrlr->num_avail_log_pages == 0xff) { 3955 struct spdk_nvmf_reservation_log *entry; 3956 entry = TAILQ_LAST(&ctrlr->log_head, log_page_head); 3957 entry->log.log_page_count = ctrlr->log_page_count; 3958 free(log); 3959 return; 3960 } 3961 3962 log->log.log_page_count = ctrlr->log_page_count; 3963 log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++; 3964 TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link); 3965 3966 nvmf_ctrlr_async_event_reservation_notification(ctrlr); 3967 } 3968 3969 void 3970 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr, 3971 struct spdk_nvmf_ns *ns, 3972 enum spdk_nvme_reservation_notification_log_page_type type) 3973 { 3974 struct spdk_nvmf_reservation_log *log; 3975 3976 switch (type) { 3977 case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY: 3978 return; 3979 case SPDK_NVME_REGISTRATION_PREEMPTED: 3980 if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) { 3981 return; 3982 } 3983 break; 3984 case SPDK_NVME_RESERVATION_RELEASED: 3985 if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) { 3986 return; 3987 } 3988 break; 3989 case SPDK_NVME_RESERVATION_PREEMPTED: 3990 if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) { 3991 return; 3992 } 3993 break; 3994 default: 3995 return; 3996 } 3997 3998 log = calloc(1, sizeof(*log)); 3999 if (!log) { 4000 SPDK_ERRLOG("Alloc log page failed, ignore the log\n"); 4001 return; 4002 } 4003 log->ctrlr = ctrlr; 4004 log->log.type = type; 4005 log->log.nsid = ns->nsid; 4006 4007 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log); 4008 } 4009 4010 /* Check from subsystem poll group's namespace information data structure */ 4011 static bool 4012 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info, 4013 struct spdk_nvmf_ctrlr *ctrlr) 4014 { 4015 uint32_t i; 4016 4017 for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) { 4018 if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) { 4019 return true; 4020 } 4021 } 4022 4023 return false; 4024 } 4025 4026 /* 4027 * Check the NVMe command is permitted or not for current controller(Host). 4028 */ 4029 static int 4030 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info, 4031 struct spdk_nvmf_ctrlr *ctrlr, 4032 struct spdk_nvmf_request *req) 4033 { 4034 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 4035 enum spdk_nvme_reservation_type rtype = ns_info->rtype; 4036 uint8_t status = SPDK_NVME_SC_SUCCESS; 4037 uint8_t racqa; 4038 bool is_registrant; 4039 4040 /* No valid reservation */ 4041 if (!rtype) { 4042 return 0; 4043 } 4044 4045 is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr); 4046 /* All registrants type and current ctrlr is a valid registrant */ 4047 if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS || 4048 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) { 4049 return 0; 4050 } else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) { 4051 return 0; 4052 } 4053 4054 /* Non-holder for current controller */ 4055 switch (cmd->opc) { 4056 case SPDK_NVME_OPC_READ: 4057 case SPDK_NVME_OPC_COMPARE: 4058 if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 4059 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4060 goto exit; 4061 } 4062 if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY || 4063 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) { 4064 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4065 } 4066 break; 4067 case SPDK_NVME_OPC_FLUSH: 4068 case SPDK_NVME_OPC_WRITE: 4069 case SPDK_NVME_OPC_WRITE_UNCORRECTABLE: 4070 case SPDK_NVME_OPC_WRITE_ZEROES: 4071 case SPDK_NVME_OPC_DATASET_MANAGEMENT: 4072 if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE || 4073 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 4074 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4075 goto exit; 4076 } 4077 if (!is_registrant) { 4078 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4079 } 4080 break; 4081 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 4082 racqa = cmd->cdw10_bits.resv_acquire.racqa; 4083 if (racqa == SPDK_NVME_RESERVE_ACQUIRE) { 4084 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4085 goto exit; 4086 } 4087 if (!is_registrant) { 4088 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4089 } 4090 break; 4091 case SPDK_NVME_OPC_RESERVATION_RELEASE: 4092 if (!is_registrant) { 4093 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4094 } 4095 break; 4096 default: 4097 break; 4098 } 4099 4100 exit: 4101 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4102 req->rsp->nvme_cpl.status.sc = status; 4103 if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) { 4104 return -EPERM; 4105 } 4106 4107 return 0; 4108 } 4109 4110 static int 4111 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 4112 struct spdk_bdev_desc *desc, struct spdk_io_channel *ch) 4113 { 4114 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 4115 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 4116 struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req; 4117 int rc; 4118 4119 if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) { 4120 /* first fused operation (should be compare) */ 4121 if (first_fused_req != NULL) { 4122 struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl; 4123 4124 SPDK_ERRLOG("Wrong sequence of fused operations\n"); 4125 4126 /* abort req->qpair->first_fused_request and continue with new fused command */ 4127 fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4128 fused_response->status.sct = SPDK_NVME_SCT_GENERIC; 4129 _nvmf_request_complete(first_fused_req); 4130 } else if (cmd->opc != SPDK_NVME_OPC_COMPARE) { 4131 SPDK_ERRLOG("Wrong op code of fused operations\n"); 4132 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4133 rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE; 4134 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4135 } 4136 4137 req->qpair->first_fused_req = req; 4138 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 4139 } else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) { 4140 /* second fused operation (should be write) */ 4141 if (first_fused_req == NULL) { 4142 SPDK_ERRLOG("Wrong sequence of fused operations\n"); 4143 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4144 rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4145 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4146 } else if (cmd->opc != SPDK_NVME_OPC_WRITE) { 4147 struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl; 4148 4149 SPDK_ERRLOG("Wrong op code of fused operations\n"); 4150 4151 /* abort req->qpair->first_fused_request and fail current command */ 4152 fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4153 fused_response->status.sct = SPDK_NVME_SCT_GENERIC; 4154 _nvmf_request_complete(first_fused_req); 4155 4156 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4157 rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE; 4158 req->qpair->first_fused_req = NULL; 4159 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4160 } 4161 4162 /* save request of first command to generate response later */ 4163 req->first_fused_req = first_fused_req; 4164 req->first_fused = true; 4165 req->qpair->first_fused_req = NULL; 4166 } else { 4167 SPDK_ERRLOG("Invalid fused command fuse field.\n"); 4168 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4169 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 4170 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4171 } 4172 4173 rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req); 4174 4175 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 4176 if (spdk_nvme_cpl_is_error(rsp)) { 4177 struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl; 4178 4179 fused_response->status = rsp->status; 4180 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4181 rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED; 4182 /* Complete first of fused commands. Second will be completed by upper layer */ 4183 _nvmf_request_complete(first_fused_req); 4184 req->first_fused_req = NULL; 4185 req->first_fused = false; 4186 } 4187 } 4188 4189 return rc; 4190 } 4191 4192 bool 4193 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req) 4194 { 4195 struct spdk_nvmf_transport *transport = req->qpair->transport; 4196 struct spdk_nvmf_ns *ns; 4197 4198 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE); 4199 4200 if (!transport->opts.zcopy) { 4201 return false; 4202 } 4203 4204 if (nvmf_qpair_is_admin_queue(req->qpair)) { 4205 /* Admin queue */ 4206 return false; 4207 } 4208 4209 if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) && 4210 (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) { 4211 /* Not a READ or WRITE command */ 4212 return false; 4213 } 4214 4215 if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 4216 /* Fused commands dont use zcopy buffers */ 4217 return false; 4218 } 4219 4220 ns = nvmf_ctrlr_get_ns(req->qpair->ctrlr, req->cmd->nvme_cmd.nsid); 4221 if (ns == NULL || ns->bdev == NULL || !ns->zcopy) { 4222 return false; 4223 } 4224 4225 req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT; 4226 return true; 4227 } 4228 4229 void 4230 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req) 4231 { 4232 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT); 4233 4234 /* Set iovcnt to be the maximum number of iovs that the ZCOPY can use */ 4235 req->iovcnt = NVMF_REQ_MAX_BUFFERS; 4236 4237 spdk_nvmf_request_exec(req); 4238 } 4239 4240 void 4241 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req, bool commit) 4242 { 4243 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE); 4244 req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING; 4245 4246 nvmf_bdev_ctrlr_zcopy_end(req, commit); 4247 } 4248 4249 int 4250 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req) 4251 { 4252 uint32_t nsid; 4253 struct spdk_nvmf_ns *ns; 4254 struct spdk_bdev *bdev; 4255 struct spdk_bdev_desc *desc; 4256 struct spdk_io_channel *ch; 4257 struct spdk_nvmf_poll_group *group = req->qpair->group; 4258 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 4259 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 4260 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 4261 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4262 enum spdk_nvme_ana_state ana_state; 4263 4264 /* pre-set response details for this command */ 4265 response->status.sc = SPDK_NVME_SC_SUCCESS; 4266 nsid = cmd->nsid; 4267 4268 if (spdk_unlikely(ctrlr == NULL)) { 4269 SPDK_ERRLOG("I/O command sent before CONNECT\n"); 4270 response->status.sct = SPDK_NVME_SCT_GENERIC; 4271 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 4272 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4273 } 4274 4275 if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) { 4276 SPDK_ERRLOG("I/O command sent to disabled controller\n"); 4277 response->status.sct = SPDK_NVME_SCT_GENERIC; 4278 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 4279 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4280 } 4281 4282 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 4283 if (spdk_unlikely(ns == NULL || ns->bdev == NULL)) { 4284 SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid); 4285 response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4286 response->status.dnr = 1; 4287 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4288 } 4289 4290 ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid); 4291 if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE && 4292 ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) { 4293 SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n", 4294 ana_state); 4295 response->status.sct = SPDK_NVME_SCT_PATH; 4296 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 4297 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4298 } 4299 4300 if (spdk_likely(ctrlr->listener != NULL)) { 4301 SPDK_DTRACE_PROBE3_TICKS(nvmf_request_io_exec_path, req, 4302 ctrlr->listener->trid->traddr, 4303 ctrlr->listener->trid->trsvcid); 4304 } 4305 4306 /* scan-build falsely reporting dereference of null pointer */ 4307 assert(group != NULL && group->sgroups != NULL); 4308 ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1]; 4309 if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) { 4310 SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n", 4311 cmd->nsid, cmd->opc); 4312 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4313 } 4314 4315 bdev = ns->bdev; 4316 desc = ns->desc; 4317 ch = ns_info->channel; 4318 4319 if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) { 4320 return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch); 4321 } else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) { 4322 struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl; 4323 4324 SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n"); 4325 4326 /* abort req->qpair->first_fused_request and continue with new command */ 4327 fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4328 fused_response->status.sct = SPDK_NVME_SCT_GENERIC; 4329 _nvmf_request_complete(req->qpair->first_fused_req); 4330 req->qpair->first_fused_req = NULL; 4331 } 4332 4333 if (spdk_nvmf_request_using_zcopy(req)) { 4334 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT); 4335 return nvmf_bdev_ctrlr_zcopy_start(bdev, desc, ch, req); 4336 } else { 4337 switch (cmd->opc) { 4338 case SPDK_NVME_OPC_READ: 4339 return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req); 4340 case SPDK_NVME_OPC_WRITE: 4341 return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req); 4342 case SPDK_NVME_OPC_COMPARE: 4343 return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req); 4344 case SPDK_NVME_OPC_WRITE_ZEROES: 4345 return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req); 4346 case SPDK_NVME_OPC_FLUSH: 4347 return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req); 4348 case SPDK_NVME_OPC_DATASET_MANAGEMENT: 4349 return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req); 4350 case SPDK_NVME_OPC_RESERVATION_REGISTER: 4351 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 4352 case SPDK_NVME_OPC_RESERVATION_RELEASE: 4353 case SPDK_NVME_OPC_RESERVATION_REPORT: 4354 spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req); 4355 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 4356 case SPDK_NVME_OPC_COPY: 4357 return nvmf_bdev_ctrlr_copy_cmd(bdev, desc, ch, req); 4358 default: 4359 return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req); 4360 } 4361 } 4362 } 4363 4364 static void 4365 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair) 4366 { 4367 if (spdk_unlikely(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING)) { 4368 assert(qpair->state_cb != NULL); 4369 4370 if (TAILQ_EMPTY(&qpair->outstanding)) { 4371 qpair->state_cb(qpair->state_cb_arg, 0); 4372 } 4373 } 4374 } 4375 4376 int 4377 spdk_nvmf_request_free(struct spdk_nvmf_request *req) 4378 { 4379 struct spdk_nvmf_qpair *qpair = req->qpair; 4380 4381 TAILQ_REMOVE(&qpair->outstanding, req, link); 4382 if (spdk_unlikely(nvmf_transport_req_free(req))) { 4383 SPDK_ERRLOG("Unable to free transport level request resources.\n"); 4384 } 4385 4386 nvmf_qpair_request_cleanup(qpair); 4387 4388 return 0; 4389 } 4390 4391 static void 4392 _nvmf_request_complete(void *ctx) 4393 { 4394 struct spdk_nvmf_request *req = ctx; 4395 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 4396 struct spdk_nvmf_qpair *qpair; 4397 struct spdk_nvmf_subsystem_poll_group *sgroup = NULL; 4398 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4399 bool is_aer = false; 4400 uint32_t nsid; 4401 bool paused; 4402 uint8_t opcode; 4403 4404 rsp->sqid = 0; 4405 rsp->status.p = 0; 4406 rsp->cid = req->cmd->nvme_cmd.cid; 4407 nsid = req->cmd->nvme_cmd.nsid; 4408 opcode = req->cmd->nvmf_cmd.opcode; 4409 4410 qpair = req->qpair; 4411 if (spdk_likely(qpair->ctrlr)) { 4412 sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id]; 4413 assert(sgroup != NULL); 4414 is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST; 4415 if (spdk_likely(qpair->qid != 0)) { 4416 qpair->group->stat.completed_nvme_io++; 4417 } 4418 4419 /* 4420 * Set the crd value. 4421 * If the the IO has any error, and dnr (DoNotRetry) is not 1, 4422 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT. 4423 */ 4424 if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp) && 4425 rsp->status.dnr == 0 && 4426 qpair->ctrlr->acre_enabled)) { 4427 rsp->status.crd = 1; 4428 } 4429 } else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) { 4430 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 4431 } 4432 4433 if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) { 4434 spdk_nvme_print_completion(qpair->qid, rsp); 4435 } 4436 4437 switch (req->zcopy_phase) { 4438 case NVMF_ZCOPY_PHASE_NONE: 4439 TAILQ_REMOVE(&qpair->outstanding, req, link); 4440 break; 4441 case NVMF_ZCOPY_PHASE_INIT: 4442 if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) { 4443 req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED; 4444 TAILQ_REMOVE(&qpair->outstanding, req, link); 4445 } else { 4446 req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE; 4447 } 4448 break; 4449 case NVMF_ZCOPY_PHASE_EXECUTE: 4450 break; 4451 case NVMF_ZCOPY_PHASE_END_PENDING: 4452 TAILQ_REMOVE(&qpair->outstanding, req, link); 4453 req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE; 4454 break; 4455 default: 4456 SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase); 4457 break; 4458 } 4459 4460 if (spdk_unlikely(nvmf_transport_req_complete(req))) { 4461 SPDK_ERRLOG("Transport request completion error!\n"); 4462 } 4463 4464 /* AER cmd is an exception */ 4465 if (spdk_likely(sgroup && !is_aer)) { 4466 if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC || 4467 nvmf_qpair_is_admin_queue(qpair))) { 4468 assert(sgroup->mgmt_io_outstanding > 0); 4469 sgroup->mgmt_io_outstanding--; 4470 } else { 4471 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE || 4472 req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE || 4473 req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT_FAILED) { 4474 /* End of request */ 4475 4476 /* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */ 4477 if (spdk_likely(nsid - 1 < sgroup->num_ns)) { 4478 sgroup->ns_info[nsid - 1].io_outstanding--; 4479 } 4480 } 4481 } 4482 4483 if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING && 4484 sgroup->mgmt_io_outstanding == 0)) { 4485 paused = true; 4486 for (nsid = 0; nsid < sgroup->num_ns; nsid++) { 4487 ns_info = &sgroup->ns_info[nsid]; 4488 4489 if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING && 4490 ns_info->io_outstanding > 0) { 4491 paused = false; 4492 break; 4493 } 4494 } 4495 4496 if (paused) { 4497 sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED; 4498 sgroup->cb_fn(sgroup->cb_arg, 0); 4499 sgroup->cb_fn = NULL; 4500 sgroup->cb_arg = NULL; 4501 } 4502 } 4503 4504 } 4505 4506 nvmf_qpair_request_cleanup(qpair); 4507 } 4508 4509 int 4510 spdk_nvmf_request_complete(struct spdk_nvmf_request *req) 4511 { 4512 struct spdk_nvmf_qpair *qpair = req->qpair; 4513 4514 spdk_thread_exec_msg(qpair->group->thread, _nvmf_request_complete, req); 4515 4516 return 0; 4517 } 4518 4519 void 4520 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req) 4521 { 4522 struct spdk_nvmf_qpair *qpair = req->qpair; 4523 struct spdk_nvmf_subsystem_poll_group *sgroup = NULL; 4524 enum spdk_nvmf_request_exec_status status; 4525 4526 if (qpair->ctrlr) { 4527 sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id]; 4528 } else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) { 4529 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 4530 } 4531 4532 assert(sgroup != NULL); 4533 sgroup->mgmt_io_outstanding++; 4534 4535 /* Place the request on the outstanding list so we can keep track of it */ 4536 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4537 4538 assert(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC); 4539 status = nvmf_ctrlr_process_fabrics_cmd(req); 4540 4541 if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 4542 _nvmf_request_complete(req); 4543 } 4544 } 4545 4546 static bool 4547 nvmf_check_subsystem_active(struct spdk_nvmf_request *req) 4548 { 4549 struct spdk_nvmf_qpair *qpair = req->qpair; 4550 struct spdk_nvmf_subsystem_poll_group *sgroup = NULL; 4551 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4552 uint32_t nsid; 4553 4554 if (spdk_likely(qpair->ctrlr)) { 4555 sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id]; 4556 assert(sgroup != NULL); 4557 } else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) { 4558 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 4559 } 4560 4561 /* Check if the subsystem is paused (if there is a subsystem) */ 4562 if (spdk_likely(sgroup != NULL)) { 4563 if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC || 4564 nvmf_qpair_is_admin_queue(qpair))) { 4565 if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) { 4566 /* The subsystem is not currently active. Queue this request. */ 4567 TAILQ_INSERT_TAIL(&sgroup->queued, req, link); 4568 return false; 4569 } 4570 sgroup->mgmt_io_outstanding++; 4571 } else { 4572 nsid = req->cmd->nvme_cmd.nsid; 4573 4574 /* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */ 4575 if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) { 4576 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4577 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4578 req->rsp->nvme_cpl.status.dnr = 1; 4579 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4580 _nvmf_request_complete(req); 4581 return false; 4582 } 4583 4584 ns_info = &sgroup->ns_info[nsid - 1]; 4585 if (spdk_unlikely(ns_info->channel == NULL)) { 4586 /* This can can happen if host sends I/O to a namespace that is 4587 * in the process of being added, but before the full addition 4588 * process is complete. Report invalid namespace in that case. 4589 */ 4590 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4591 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4592 req->rsp->nvme_cpl.status.dnr = 1; 4593 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4594 ns_info->io_outstanding++; 4595 _nvmf_request_complete(req); 4596 return false; 4597 } 4598 4599 if (spdk_unlikely(ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE)) { 4600 /* The namespace is not currently active. Queue this request. */ 4601 TAILQ_INSERT_TAIL(&sgroup->queued, req, link); 4602 return false; 4603 } 4604 4605 ns_info->io_outstanding++; 4606 } 4607 4608 if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) { 4609 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4610 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 4611 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4612 _nvmf_request_complete(req); 4613 return false; 4614 } 4615 } 4616 4617 return true; 4618 } 4619 4620 void 4621 spdk_nvmf_request_exec(struct spdk_nvmf_request *req) 4622 { 4623 struct spdk_nvmf_qpair *qpair = req->qpair; 4624 struct spdk_nvmf_transport *transport = qpair->transport; 4625 enum spdk_nvmf_request_exec_status status; 4626 4627 if (spdk_unlikely(!nvmf_check_subsystem_active(req))) { 4628 return; 4629 } 4630 4631 if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) { 4632 spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd); 4633 } 4634 4635 /* Place the request on the outstanding list so we can keep track of it */ 4636 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4637 4638 if (spdk_unlikely((req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC) && 4639 spdk_nvme_trtype_is_fabrics(transport->ops->type))) { 4640 status = nvmf_ctrlr_process_fabrics_cmd(req); 4641 } else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) { 4642 status = nvmf_ctrlr_process_admin_cmd(req); 4643 } else { 4644 status = nvmf_ctrlr_process_io_cmd(req); 4645 } 4646 4647 if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 4648 _nvmf_request_complete(req); 4649 } 4650 } 4651 4652 static bool 4653 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd, 4654 struct spdk_dif_ctx *dif_ctx) 4655 { 4656 struct spdk_nvmf_ns *ns; 4657 struct spdk_bdev *bdev; 4658 4659 if (ctrlr == NULL || cmd == NULL) { 4660 return false; 4661 } 4662 4663 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 4664 if (ns == NULL || ns->bdev == NULL) { 4665 return false; 4666 } 4667 4668 bdev = ns->bdev; 4669 4670 switch (cmd->opc) { 4671 case SPDK_NVME_OPC_READ: 4672 case SPDK_NVME_OPC_WRITE: 4673 case SPDK_NVME_OPC_COMPARE: 4674 return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx); 4675 default: 4676 break; 4677 } 4678 4679 return false; 4680 } 4681 4682 bool 4683 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx) 4684 { 4685 struct spdk_nvmf_qpair *qpair = req->qpair; 4686 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 4687 4688 if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) { 4689 return false; 4690 } 4691 4692 if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) { 4693 return false; 4694 } 4695 4696 if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) { 4697 return false; 4698 } 4699 4700 if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) { 4701 return false; 4702 } 4703 4704 return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx); 4705 } 4706 4707 void 4708 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr) 4709 { 4710 g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr; 4711 } 4712 4713 static int 4714 nvmf_passthru_admin_cmd_for_bdev_nsid(struct spdk_nvmf_request *req, uint32_t bdev_nsid) 4715 { 4716 struct spdk_bdev *bdev; 4717 struct spdk_bdev_desc *desc; 4718 struct spdk_io_channel *ch; 4719 struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req); 4720 int rc; 4721 4722 rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch); 4723 if (rc) { 4724 response->status.sct = SPDK_NVME_SCT_GENERIC; 4725 response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4726 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4727 } 4728 return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL); 4729 } 4730 4731 static int 4732 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req) 4733 { 4734 struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req); 4735 uint32_t bdev_nsid; 4736 4737 if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid != 0) { 4738 bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid; 4739 } else { 4740 bdev_nsid = cmd->nsid; 4741 } 4742 4743 return nvmf_passthru_admin_cmd_for_bdev_nsid(req, bdev_nsid); 4744 } 4745 4746 int 4747 nvmf_passthru_admin_cmd_for_ctrlr(struct spdk_nvmf_request *req, struct spdk_nvmf_ctrlr *ctrlr) 4748 { 4749 struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req); 4750 struct spdk_nvmf_ns *ns; 4751 4752 ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); 4753 if (ns == NULL) { 4754 /* Is there a better sc to use here? */ 4755 response->status.sct = SPDK_NVME_SCT_GENERIC; 4756 response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4757 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4758 } 4759 4760 return nvmf_passthru_admin_cmd_for_bdev_nsid(req, ns->nsid); 4761 } 4762 4763 void 4764 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid) 4765 { 4766 g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd; 4767 g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid; 4768 } 4769 4770 int 4771 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req, 4772 struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch) 4773 { 4774 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 4775 struct spdk_nvmf_ns *ns; 4776 struct spdk_nvmf_poll_group *group = req->qpair->group; 4777 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4778 4779 *bdev = NULL; 4780 *desc = NULL; 4781 *ch = NULL; 4782 4783 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 4784 if (ns == NULL || ns->bdev == NULL) { 4785 return -EINVAL; 4786 } 4787 4788 assert(group != NULL && group->sgroups != NULL); 4789 ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1]; 4790 *bdev = ns->bdev; 4791 *desc = ns->desc; 4792 *ch = ns_info->channel; 4793 4794 return 0; 4795 } 4796 4797 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req) 4798 { 4799 return req->qpair->ctrlr; 4800 } 4801 4802 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req) 4803 { 4804 return &req->cmd->nvme_cmd; 4805 } 4806 4807 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req) 4808 { 4809 return &req->rsp->nvme_cpl; 4810 } 4811 4812 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req) 4813 { 4814 return req->qpair->ctrlr->subsys; 4815 } 4816 4817 size_t 4818 spdk_nvmf_request_copy_from_buf(struct spdk_nvmf_request *req, 4819 void *buf, size_t buflen) 4820 { 4821 struct spdk_iov_xfer ix; 4822 4823 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 4824 return spdk_iov_xfer_from_buf(&ix, buf, buflen); 4825 } 4826 4827 size_t 4828 spdk_nvmf_request_copy_to_buf(struct spdk_nvmf_request *req, 4829 void *buf, size_t buflen) 4830 { 4831 struct spdk_iov_xfer ix; 4832 4833 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 4834 return spdk_iov_xfer_to_buf(&ix, buf, buflen); 4835 } 4836 4837 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr) 4838 { 4839 return ctrlr->subsys; 4840 } 4841 4842 uint16_t 4843 spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr) 4844 { 4845 return ctrlr->cntlid; 4846 } 4847 4848 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req) 4849 { 4850 return req->req_to_abort; 4851 } 4852