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 if (!ctrlr->admin_qpair) { 229 SPDK_ERRLOG("Inactive admin qpair\n"); 230 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 231 rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 232 qpair->connect_req = NULL; 233 qpair->ctrlr = NULL; 234 spdk_nvmf_request_complete(req); 235 return; 236 } 237 238 assert(ctrlr->admin_qpair->group->thread == spdk_get_thread()); 239 240 if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) { 241 if (qpair->connect_req != NULL) { 242 SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid); 243 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 244 rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER; 245 qpair->connect_req = NULL; 246 qpair->ctrlr = NULL; 247 spdk_nvmf_request_complete(req); 248 } else { 249 SPDK_WARNLOG("Duplicate QID detected, re-check in %dus\n", 250 DUPLICATE_QID_RETRY_US); 251 qpair->connect_req = req; 252 /* Set qpair->ctrlr here so that we'll have it when the poller expires. */ 253 qpair->ctrlr = ctrlr; 254 req->poller = SPDK_POLLER_REGISTER(_retry_qid_check, qpair, 255 DUPLICATE_QID_RETRY_US); 256 } 257 return; 258 } 259 260 qpair->ctrlr = ctrlr; 261 spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid); 262 263 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 264 rsp->status_code_specific.success.cntlid = ctrlr->cntlid; 265 SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n", 266 rsp->status_code_specific.success.cntlid); 267 spdk_nvmf_request_complete(req); 268 269 SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn, 270 ctrlr->hostnqn); 271 } 272 273 static int 274 _retry_qid_check(void *ctx) 275 { 276 struct spdk_nvmf_qpair *qpair = ctx; 277 struct spdk_nvmf_request *req = qpair->connect_req; 278 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 279 280 spdk_poller_unregister(&req->poller); 281 ctrlr_add_qpair_and_send_rsp(qpair, ctrlr, req); 282 return SPDK_POLLER_BUSY; 283 } 284 285 static void 286 _nvmf_ctrlr_add_admin_qpair(void *ctx) 287 { 288 struct spdk_nvmf_request *req = ctx; 289 struct spdk_nvmf_qpair *qpair = req->qpair; 290 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 291 292 ctrlr->admin_qpair = qpair; 293 ctrlr->association_timeout = qpair->transport->opts.association_timeout; 294 nvmf_ctrlr_start_keep_alive_timer(ctrlr); 295 ctrlr_add_qpair_and_send_rsp(qpair, ctrlr, req); 296 } 297 298 static void 299 _nvmf_subsystem_add_ctrlr(void *ctx) 300 { 301 struct spdk_nvmf_request *req = ctx; 302 struct spdk_nvmf_qpair *qpair = req->qpair; 303 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 304 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 305 306 if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) { 307 SPDK_ERRLOG("Unable to add controller to subsystem\n"); 308 spdk_bit_array_free(&ctrlr->qpair_mask); 309 free(ctrlr); 310 qpair->ctrlr = NULL; 311 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 312 spdk_nvmf_request_complete(req); 313 return; 314 } 315 316 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req); 317 } 318 319 static void 320 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem, 321 struct spdk_nvmf_ctrlr_data *cdata) 322 { 323 cdata->aerl = SPDK_NVMF_MAX_ASYNC_EVENTS - 1; 324 cdata->kas = KAS_DEFAULT_VALUE; 325 cdata->vid = SPDK_PCI_VID_INTEL; 326 cdata->ssvid = SPDK_PCI_VID_INTEL; 327 /* INTEL OUI */ 328 cdata->ieee[0] = 0xe4; 329 cdata->ieee[1] = 0xd2; 330 cdata->ieee[2] = 0x5c; 331 cdata->oncs.compare = 1; 332 cdata->oncs.reservations = 1; 333 cdata->fuses.compare_and_write = 1; 334 cdata->oncs.copy = 1; 335 cdata->sgls.supported = 1; 336 cdata->sgls.keyed_sgl = 1; 337 cdata->sgls.sgl_offset = 1; 338 cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16; 339 cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16; 340 cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16; 341 cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */ 342 cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC; 343 cdata->nvmf_specific.msdbd = 1; 344 345 if (transport->ops->cdata_init) { 346 transport->ops->cdata_init(transport, subsystem, cdata); 347 } 348 } 349 350 static bool 351 nvmf_subsys_has_multi_iocs(struct spdk_nvmf_subsystem *subsystem) 352 { 353 struct spdk_nvmf_ns *ns; 354 uint32_t i; 355 356 for (i = 0; i < subsystem->max_nsid; i++) { 357 ns = subsystem->ns[i]; 358 if (ns && ns->bdev && spdk_bdev_is_zoned(ns->bdev)) { 359 return true; 360 } 361 } 362 return false; 363 } 364 365 static void 366 nvmf_ctrlr_init_visible_ns(struct spdk_nvmf_ctrlr *ctrlr) 367 { 368 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 369 struct spdk_nvmf_ns *ns; 370 371 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 372 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 373 if (ns->always_visible || nvmf_ns_find_host(ns, ctrlr->hostnqn) != NULL) { 374 spdk_bit_array_set(ctrlr->visible_ns, ns->nsid - 1); 375 } 376 } 377 } 378 379 static struct spdk_nvmf_ctrlr * 380 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem, 381 struct spdk_nvmf_request *req, 382 struct spdk_nvmf_fabric_connect_cmd *connect_cmd, 383 struct spdk_nvmf_fabric_connect_data *connect_data) 384 { 385 struct spdk_nvmf_ctrlr *ctrlr; 386 struct spdk_nvmf_transport *transport = req->qpair->transport; 387 struct spdk_nvme_transport_id listen_trid = {}; 388 bool subsys_has_multi_iocs = false; 389 390 ctrlr = calloc(1, sizeof(*ctrlr)); 391 if (ctrlr == NULL) { 392 SPDK_ERRLOG("Memory allocation failed\n"); 393 return NULL; 394 } 395 396 if (spdk_nvme_trtype_is_fabrics(transport->ops->type)) { 397 ctrlr->dynamic_ctrlr = true; 398 } else { 399 ctrlr->cntlid = connect_data->cntlid; 400 } 401 402 SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_create, ctrlr, subsystem->subnqn, 403 spdk_thread_get_id(req->qpair->group->thread)); 404 405 STAILQ_INIT(&ctrlr->async_events); 406 TAILQ_INIT(&ctrlr->log_head); 407 ctrlr->subsys = subsystem; 408 ctrlr->thread = req->qpair->group->thread; 409 ctrlr->disconnect_in_progress = false; 410 411 ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr); 412 if (!ctrlr->qpair_mask) { 413 SPDK_ERRLOG("Failed to allocate controller qpair mask\n"); 414 goto err_qpair_mask; 415 } 416 417 nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata); 418 419 /* 420 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units. 421 * If this field is cleared to 0h, then Keep Alive is not supported. 422 */ 423 if (ctrlr->cdata.kas) { 424 ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato, 425 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) * 426 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS; 427 } 428 429 ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1; 430 if (ctrlr->subsys->flags.ana_reporting) { 431 ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1; 432 } 433 ctrlr->feat.volatile_write_cache.bits.wce = 1; 434 /* Coalescing Disable */ 435 ctrlr->feat.interrupt_vector_configuration.bits.cd = 1; 436 437 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 438 /* 439 * If keep-alive timeout is not set, discovery controllers use some 440 * arbitrary high value in order to cleanup stale discovery sessions 441 * 442 * From the 1.0a nvme-of spec: 443 * "The Keep Alive command is reserved for 444 * Discovery controllers. A transport may specify a 445 * fixed Discovery controller activity timeout value 446 * (e.g., 2 minutes). If no commands are received 447 * by a Discovery controller within that time 448 * period, the controller may perform the 449 * actions for Keep Alive Timer expiration". 450 * 451 * From the 1.1 nvme-of spec: 452 * "A host requests an explicit persistent connection 453 * to a Discovery controller and Asynchronous Event Notifications from 454 * the Discovery controller on that persistent connection by specifying 455 * a non-zero Keep Alive Timer value in the Connect command." 456 * 457 * In case non-zero KATO is used, we enable discovery_log_change_notice 458 * otherwise we disable it and use default discovery controller KATO. 459 * KATO is in millisecond. 460 */ 461 if (ctrlr->feat.keep_alive_timer.bits.kato == 0) { 462 ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS; 463 ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0; 464 } else { 465 ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1; 466 } 467 } 468 469 /* Subtract 1 for admin queue, 1 for 0's based */ 470 ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 - 471 1; 472 ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 - 473 1; 474 475 spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid); 476 memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn)); 477 478 ctrlr->visible_ns = spdk_bit_array_create(subsystem->max_nsid); 479 if (!ctrlr->visible_ns) { 480 SPDK_ERRLOG("Failed to allocate visible namespace array\n"); 481 goto err_visible_ns; 482 } 483 nvmf_ctrlr_init_visible_ns(ctrlr); 484 485 ctrlr->vcprop.cap.raw = 0; 486 ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */ 487 ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth - 488 1; /* max queue depth */ 489 ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */ 490 /* ready timeout - 500 msec units */ 491 ctrlr->vcprop.cap.bits.to = NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS / 500; 492 ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */ 493 subsys_has_multi_iocs = nvmf_subsys_has_multi_iocs(subsystem); 494 if (subsys_has_multi_iocs) { 495 ctrlr->vcprop.cap.bits.css = 496 SPDK_NVME_CAP_CSS_IOCS; /* One or more I/O command sets supported */ 497 } else { 498 ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */ 499 } 500 501 ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */ 502 ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */ 503 504 /* Version Supported: 1.3 */ 505 ctrlr->vcprop.vs.bits.mjr = 1; 506 ctrlr->vcprop.vs.bits.mnr = 3; 507 ctrlr->vcprop.vs.bits.ter = 0; 508 509 ctrlr->vcprop.cc.raw = 0; 510 ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */ 511 if (subsys_has_multi_iocs) { 512 ctrlr->vcprop.cc.bits.css = 513 SPDK_NVME_CC_CSS_IOCS; /* All supported I/O Command Sets */ 514 } 515 516 ctrlr->vcprop.csts.raw = 0; 517 ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */ 518 519 SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw); 520 SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw); 521 SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw); 522 SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw); 523 524 ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip; 525 526 if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) { 527 if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) { 528 SPDK_ERRLOG("Could not get listener transport ID\n"); 529 goto err_listener; 530 } 531 532 ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid); 533 if (!ctrlr->listener) { 534 SPDK_ERRLOG("Listener was not found\n"); 535 goto err_listener; 536 } 537 } 538 539 req->qpair->ctrlr = ctrlr; 540 spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req); 541 542 return ctrlr; 543 err_listener: 544 spdk_bit_array_free(&ctrlr->visible_ns); 545 err_visible_ns: 546 spdk_bit_array_free(&ctrlr->qpair_mask); 547 err_qpair_mask: 548 free(ctrlr); 549 return NULL; 550 } 551 552 static void 553 _nvmf_ctrlr_destruct(void *ctx) 554 { 555 struct spdk_nvmf_ctrlr *ctrlr = ctx; 556 struct spdk_nvmf_reservation_log *log, *log_tmp; 557 struct spdk_nvmf_async_event_completion *event, *event_tmp; 558 559 SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_destruct, ctrlr, ctrlr->subsys->subnqn, 560 spdk_thread_get_id(ctrlr->thread)); 561 562 assert(spdk_get_thread() == ctrlr->thread); 563 assert(ctrlr->in_destruct); 564 565 SPDK_DEBUGLOG(nvmf, "Destroy ctrlr 0x%hx\n", ctrlr->cntlid); 566 if (ctrlr->disconnect_in_progress) { 567 SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n"); 568 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr); 569 return; 570 } 571 572 nvmf_ctrlr_stop_keep_alive_timer(ctrlr); 573 nvmf_ctrlr_stop_association_timer(ctrlr); 574 spdk_bit_array_free(&ctrlr->qpair_mask); 575 576 TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) { 577 TAILQ_REMOVE(&ctrlr->log_head, log, link); 578 free(log); 579 } 580 STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) { 581 STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link); 582 free(event); 583 } 584 spdk_bit_array_free(&ctrlr->visible_ns); 585 free(ctrlr); 586 } 587 588 void 589 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr) 590 { 591 nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr); 592 593 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr); 594 } 595 596 static void 597 nvmf_ctrlr_add_io_qpair(void *ctx) 598 { 599 struct spdk_nvmf_request *req = ctx; 600 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 601 struct spdk_nvmf_qpair *qpair = req->qpair; 602 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 603 struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair; 604 struct spdk_nvmf_poll_group *admin_qpair_group = NULL; 605 enum spdk_nvmf_qpair_state admin_qpair_state = SPDK_NVMF_QPAIR_UNINITIALIZED; 606 607 SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_io_qpair, ctrlr, req->qpair, req->qpair->qid, 608 spdk_thread_get_id(ctrlr->thread)); 609 610 /* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect. 611 * For error case, the value should be NULL. So set it to NULL at first. 612 */ 613 qpair->ctrlr = NULL; 614 615 /* Make sure the controller is not being destroyed. */ 616 if (ctrlr->in_destruct) { 617 SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n"); 618 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 619 goto end; 620 } 621 622 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 623 SPDK_ERRLOG("I/O connect not allowed on discovery controller\n"); 624 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 625 goto end; 626 } 627 628 if (!ctrlr->vcprop.cc.bits.en) { 629 SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n"); 630 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 631 goto end; 632 } 633 634 if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) { 635 SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n", 636 ctrlr->vcprop.cc.bits.iosqes); 637 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 638 goto end; 639 } 640 641 if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) { 642 SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n", 643 ctrlr->vcprop.cc.bits.iocqes); 644 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 645 goto end; 646 } 647 648 /* There is a chance that admin qpair was destroyed. This is an issue that was observed only with ESX initiators */ 649 if (admin_qpair) { 650 admin_qpair_group = admin_qpair->group; 651 admin_qpair_state = admin_qpair->state; 652 } 653 654 if (admin_qpair_state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair_group == NULL) { 655 /* There is a chance that admin qpair was destroyed or is being destroyed at this moment due to e.g. 656 * expired keep alive timer. Part of the qpair destruction process is change of qpair's 657 * state to DEACTIVATING and removing it from poll group */ 658 SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair_state, admin_qpair_group); 659 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 660 goto end; 661 } 662 663 /* check if we would exceed ctrlr connection limit */ 664 if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) { 665 SPDK_ERRLOG("Requested QID %u but Max QID is %u\n", 666 qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1); 667 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 668 rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER; 669 goto end; 670 } 671 672 ctrlr_add_qpair_and_send_rsp(qpair, ctrlr, req); 673 return; 674 end: 675 spdk_nvmf_request_complete(req); 676 } 677 678 static void 679 _nvmf_ctrlr_add_io_qpair(void *ctx) 680 { 681 struct spdk_nvmf_request *req = ctx; 682 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 683 struct spdk_nvmf_fabric_connect_data *data; 684 struct spdk_nvmf_ctrlr *ctrlr; 685 struct spdk_nvmf_qpair *qpair = req->qpair; 686 struct spdk_nvmf_qpair *admin_qpair; 687 struct spdk_nvmf_tgt *tgt = qpair->transport->tgt; 688 struct spdk_nvmf_subsystem *subsystem; 689 struct spdk_nvme_transport_id listen_trid = {}; 690 const struct spdk_nvmf_subsystem_listener *listener; 691 struct spdk_nvmf_poll_group *admin_qpair_group = NULL; 692 enum spdk_nvmf_qpair_state admin_qpair_state = SPDK_NVMF_QPAIR_UNINITIALIZED; 693 694 assert(req->iovcnt == 1); 695 696 data = req->iov[0].iov_base; 697 698 SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid); 699 700 subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn); 701 /* We already checked this in spdk_nvmf_ctrlr_connect */ 702 assert(subsystem != NULL); 703 704 ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid); 705 if (ctrlr == NULL) { 706 SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid); 707 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid); 708 spdk_nvmf_request_complete(req); 709 return; 710 } 711 712 /* fail before passing a message to the controller thread. */ 713 if (ctrlr->in_destruct) { 714 SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n"); 715 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 716 spdk_nvmf_request_complete(req); 717 return; 718 } 719 720 /* If ANA reporting is enabled, check if I/O connect is on the same listener. */ 721 if (subsystem->flags.ana_reporting) { 722 if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) { 723 SPDK_ERRLOG("Could not get listener transport ID\n"); 724 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 725 spdk_nvmf_request_complete(req); 726 return; 727 } 728 729 listener = nvmf_subsystem_find_listener(subsystem, &listen_trid); 730 if (listener != ctrlr->listener) { 731 SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n"); 732 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 733 spdk_nvmf_request_complete(req); 734 return; 735 } 736 } 737 738 admin_qpair = ctrlr->admin_qpair; 739 740 /* There is a chance that admin qpair was destroyed. This is an issue that was observed only with ESX initiators */ 741 if (admin_qpair) { 742 admin_qpair_group = admin_qpair->group; 743 admin_qpair_state = admin_qpair->state; 744 } 745 746 if (admin_qpair_state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair_group == NULL) { 747 /* There is a chance that admin qpair was destroyed or is being destroyed at this moment due to e.g. 748 * expired keep alive timer. Part of the qpair destruction process is change of qpair's 749 * state to DEACTIVATING and removing it from poll group */ 750 SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair_state, admin_qpair_group); 751 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid); 752 spdk_nvmf_request_complete(req); 753 return; 754 } 755 qpair->ctrlr = ctrlr; 756 spdk_thread_send_msg(admin_qpair_group->thread, nvmf_ctrlr_add_io_qpair, req); 757 } 758 759 static bool 760 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem, 761 const char *hostnqn) 762 { 763 struct spdk_nvme_transport_id listen_trid = {}; 764 765 if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) { 766 SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn); 767 return false; 768 } 769 770 if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) { 771 SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n", 772 subsystem->subnqn); 773 return false; 774 } 775 776 if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) { 777 SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n", 778 subsystem->subnqn, hostnqn); 779 return false; 780 } 781 782 return true; 783 } 784 785 static int 786 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req) 787 { 788 struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base; 789 struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd; 790 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 791 struct spdk_nvmf_qpair *qpair = req->qpair; 792 struct spdk_nvmf_transport *transport = qpair->transport; 793 struct spdk_nvmf_ctrlr *ctrlr; 794 struct spdk_nvmf_subsystem *subsystem; 795 796 SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n", 797 cmd->recfmt, cmd->qid, cmd->sqsize); 798 799 SPDK_DEBUGLOG(nvmf, "Connect data:\n"); 800 SPDK_DEBUGLOG(nvmf, " cntlid: 0x%04x\n", data->cntlid); 801 SPDK_DEBUGLOG(nvmf, " hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n", 802 ntohl(*(uint32_t *)&data->hostid[0]), 803 ntohs(*(uint16_t *)&data->hostid[4]), 804 ntohs(*(uint16_t *)&data->hostid[6]), 805 data->hostid[8], 806 data->hostid[9], 807 ntohs(*(uint16_t *)&data->hostid[10]), 808 ntohl(*(uint32_t *)&data->hostid[12])); 809 SPDK_DEBUGLOG(nvmf, " subnqn: \"%s\"\n", data->subnqn); 810 SPDK_DEBUGLOG(nvmf, " hostnqn: \"%s\"\n", data->hostnqn); 811 812 subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn); 813 if (!subsystem) { 814 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn); 815 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 816 } 817 818 if (cmd->recfmt != 0) { 819 SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt); 820 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 821 rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT; 822 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 823 } 824 825 /* 826 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and 827 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues). 828 */ 829 if (cmd->sqsize == 0) { 830 SPDK_ERRLOG("Invalid SQSIZE = 0\n"); 831 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize); 832 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 833 } 834 835 if (cmd->qid == 0) { 836 if (cmd->sqsize >= transport->opts.max_aq_depth) { 837 SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n", 838 cmd->sqsize, transport->opts.max_aq_depth - 1); 839 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize); 840 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 841 } 842 } else if (cmd->sqsize >= transport->opts.max_queue_depth) { 843 SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n", 844 cmd->sqsize, transport->opts.max_queue_depth - 1); 845 SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize); 846 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 847 } 848 849 qpair->sq_head_max = cmd->sqsize; 850 qpair->qid = cmd->qid; 851 qpair->connect_received = true; 852 853 pthread_mutex_lock(&qpair->group->mutex); 854 qpair->group->current_unassociated_qpairs--; 855 pthread_mutex_unlock(&qpair->group->mutex); 856 857 if (0 == qpair->qid) { 858 qpair->group->stat.admin_qpairs++; 859 qpair->group->stat.current_admin_qpairs++; 860 } else { 861 qpair->group->stat.io_qpairs++; 862 qpair->group->stat.current_io_qpairs++; 863 } 864 865 if (cmd->qid == 0) { 866 SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid); 867 868 if (spdk_nvme_trtype_is_fabrics(transport->ops->type) && data->cntlid != 0xFFFF) { 869 /* This NVMf target only supports dynamic mode. */ 870 SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid); 871 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid); 872 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 873 } 874 875 /* Establish a new ctrlr */ 876 ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data); 877 if (!ctrlr) { 878 SPDK_ERRLOG("nvmf_ctrlr_create() failed\n"); 879 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 880 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 881 } else { 882 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 883 } 884 } else { 885 spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req); 886 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 887 } 888 } 889 890 static struct spdk_nvmf_subsystem_poll_group * 891 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req) 892 { 893 struct spdk_nvmf_fabric_connect_data *data; 894 struct spdk_nvmf_subsystem *subsystem; 895 struct spdk_nvmf_tgt *tgt; 896 897 assert(nvmf_request_is_fabric_connect(req)); 898 assert(req->qpair->ctrlr == NULL); 899 assert(req->iovcnt == 1); 900 901 data = req->iov[0].iov_base; 902 tgt = req->qpair->transport->tgt; 903 904 subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn); 905 if (subsystem == NULL) { 906 return NULL; 907 } 908 909 return &req->qpair->group->sgroups[subsystem->id]; 910 } 911 912 int 913 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req) 914 { 915 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 916 struct spdk_nvmf_subsystem_poll_group *sgroup; 917 struct spdk_nvmf_qpair *qpair = req->qpair; 918 enum spdk_nvmf_request_exec_status status; 919 920 if (req->iovcnt > 1) { 921 SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt); 922 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 923 status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 924 goto out; 925 } 926 927 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 928 if (!sgroup) { 929 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn); 930 status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 931 goto out; 932 } 933 934 sgroup->mgmt_io_outstanding++; 935 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 936 937 status = _nvmf_ctrlr_connect(req); 938 939 out: 940 if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 941 _nvmf_request_complete(req); 942 } 943 944 return status; 945 } 946 947 static int 948 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req) 949 { 950 struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base; 951 struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp; 952 struct spdk_nvmf_transport *transport = req->qpair->transport; 953 struct spdk_nvmf_subsystem *subsystem; 954 955 if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) { 956 SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length); 957 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 958 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 959 } 960 961 if (req->iovcnt > 1) { 962 SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt); 963 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 964 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 965 } 966 967 subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn); 968 if (!subsystem) { 969 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn); 970 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 971 } 972 973 if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) || 974 (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) || 975 (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) || 976 (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) { 977 struct spdk_nvmf_subsystem_poll_group *sgroup; 978 979 /* Subsystem is not ready to handle a connect. Decrement 980 * the mgmt_io_outstanding to avoid the subsystem waiting 981 * for this command to complete before unpausing. Queued 982 * requests get retried when subsystem resumes. 983 */ 984 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 985 assert(sgroup != NULL); 986 sgroup->mgmt_io_outstanding--; 987 TAILQ_REMOVE(&req->qpair->outstanding, req, link); 988 TAILQ_INSERT_TAIL(&sgroup->queued, req, link); 989 SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn); 990 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 991 } 992 993 /* Ensure that hostnqn is null terminated */ 994 if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) { 995 SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n"); 996 SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn); 997 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 998 } 999 1000 if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) { 1001 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1002 rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST; 1003 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1004 } 1005 1006 return _nvmf_ctrlr_connect(req); 1007 } 1008 1009 static int 1010 nvmf_ctrlr_association_remove(void *ctx) 1011 { 1012 struct spdk_nvmf_ctrlr *ctrlr = ctx; 1013 int rc; 1014 1015 nvmf_ctrlr_stop_association_timer(ctrlr); 1016 1017 if (ctrlr->in_destruct) { 1018 return SPDK_POLLER_IDLE; 1019 } 1020 SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n", 1021 ctrlr->subsys->subnqn); 1022 1023 if (ctrlr->admin_qpair) { 1024 rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL); 1025 if (rc < 0 && rc != -EINPROGRESS) { 1026 SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n"); 1027 assert(false); 1028 } 1029 } 1030 1031 return SPDK_POLLER_BUSY; 1032 } 1033 1034 static int 1035 _nvmf_ctrlr_cc_reset_shn_done(void *ctx) 1036 { 1037 struct spdk_nvmf_ctrlr *ctrlr = ctx; 1038 uint64_t now = spdk_get_ticks(); 1039 uint32_t count; 1040 1041 if (ctrlr->cc_timer) { 1042 spdk_poller_unregister(&ctrlr->cc_timer); 1043 } 1044 1045 count = spdk_bit_array_count_set(ctrlr->qpair_mask); 1046 SPDK_DEBUGLOG(nvmf, "ctrlr %p active queue count %u\n", ctrlr, count); 1047 1048 if (count > 1) { 1049 if (now < ctrlr->cc_timeout_tsc) { 1050 /* restart cc timer */ 1051 ctrlr->cc_timer = SPDK_POLLER_REGISTER(_nvmf_ctrlr_cc_reset_shn_done, ctrlr, 100 * 1000); 1052 return SPDK_POLLER_IDLE; 1053 } else { 1054 /* controller fatal status */ 1055 SPDK_WARNLOG("IO timeout, ctrlr %p is in fatal status\n", ctrlr); 1056 nvmf_ctrlr_set_fatal_status(ctrlr); 1057 } 1058 } 1059 1060 spdk_poller_unregister(&ctrlr->cc_timeout_timer); 1061 1062 if (ctrlr->disconnect_is_shn) { 1063 ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE; 1064 ctrlr->disconnect_is_shn = false; 1065 } else { 1066 /* Only a subset of the registers are cleared out on a reset */ 1067 ctrlr->vcprop.cc.raw = 0; 1068 ctrlr->vcprop.csts.raw = 0; 1069 } 1070 1071 /* After CC.EN transitions to 0 (due to shutdown or reset), the association 1072 * between the host and controller shall be preserved for at least 2 minutes */ 1073 if (ctrlr->association_timer) { 1074 SPDK_DEBUGLOG(nvmf, "Association timer already set\n"); 1075 nvmf_ctrlr_stop_association_timer(ctrlr); 1076 } 1077 if (ctrlr->association_timeout) { 1078 ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr, 1079 ctrlr->association_timeout * 1000); 1080 } 1081 ctrlr->disconnect_in_progress = false; 1082 return SPDK_POLLER_BUSY; 1083 } 1084 1085 static void 1086 nvmf_ctrlr_cc_reset_shn_done(struct spdk_io_channel_iter *i, int status) 1087 { 1088 struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i); 1089 1090 if (status < 0) { 1091 SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n"); 1092 assert(false); 1093 } 1094 1095 _nvmf_ctrlr_cc_reset_shn_done((void *)ctrlr); 1096 } 1097 1098 static void 1099 nvmf_bdev_complete_reset(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 1100 { 1101 SPDK_NOTICELOG("Resetting bdev done with %s\n", success ? "success" : "failure"); 1102 1103 spdk_bdev_free_io(bdev_io); 1104 } 1105 1106 1107 static int 1108 nvmf_ctrlr_cc_timeout(void *ctx) 1109 { 1110 struct spdk_nvmf_ctrlr *ctrlr = ctx; 1111 struct spdk_nvmf_poll_group *group; 1112 struct spdk_nvmf_ns *ns; 1113 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 1114 1115 spdk_poller_unregister(&ctrlr->cc_timeout_timer); 1116 SPDK_DEBUGLOG(nvmf, "Ctrlr %p reset or shutdown timeout\n", ctrlr); 1117 1118 if (!ctrlr->admin_qpair) { 1119 SPDK_NOTICELOG("Ctrlr %p admin qpair disconnected\n", ctrlr); 1120 return SPDK_POLLER_IDLE; 1121 } 1122 1123 group = ctrlr->admin_qpair->group; 1124 assert(group != NULL && group->sgroups != NULL); 1125 1126 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL; 1127 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 1128 if (ns->bdev == NULL) { 1129 continue; 1130 } 1131 ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[ns->opts.nsid - 1]; 1132 SPDK_NOTICELOG("Ctrlr %p resetting NSID %u\n", ctrlr, ns->opts.nsid); 1133 spdk_bdev_reset(ns->desc, ns_info->channel, nvmf_bdev_complete_reset, NULL); 1134 } 1135 1136 return SPDK_POLLER_BUSY; 1137 } 1138 1139 const struct spdk_nvmf_registers * 1140 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr) 1141 { 1142 return &ctrlr->vcprop; 1143 } 1144 1145 void 1146 nvmf_ctrlr_set_fatal_status(struct spdk_nvmf_ctrlr *ctrlr) 1147 { 1148 ctrlr->vcprop.csts.bits.cfs = 1; 1149 } 1150 1151 static uint64_t 1152 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr) 1153 { 1154 return ctrlr->vcprop.cap.raw; 1155 } 1156 1157 static uint64_t 1158 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr) 1159 { 1160 return ctrlr->vcprop.vs.raw; 1161 } 1162 1163 static uint64_t 1164 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr) 1165 { 1166 return ctrlr->vcprop.cc.raw; 1167 } 1168 1169 static bool 1170 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1171 { 1172 union spdk_nvme_cc_register cc, diff; 1173 uint32_t cc_timeout_ms; 1174 1175 cc.raw = value; 1176 1177 SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw); 1178 SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw); 1179 1180 /* 1181 * Calculate which bits changed between the current and new CC. 1182 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed. 1183 */ 1184 diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw; 1185 1186 if (diff.bits.en) { 1187 if (cc.bits.en) { 1188 SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n"); 1189 nvmf_ctrlr_stop_association_timer(ctrlr); 1190 1191 ctrlr->vcprop.cc.bits.en = 1; 1192 ctrlr->vcprop.csts.bits.rdy = 1; 1193 } else { 1194 SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n"); 1195 if (ctrlr->disconnect_in_progress) { 1196 SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n"); 1197 return true; 1198 } 1199 1200 ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr, 1201 NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000); 1202 /* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */ 1203 cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2; 1204 ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000; 1205 1206 ctrlr->vcprop.cc.bits.en = 0; 1207 ctrlr->disconnect_in_progress = true; 1208 ctrlr->disconnect_is_shn = false; 1209 spdk_for_each_channel(ctrlr->subsys->tgt, 1210 nvmf_ctrlr_disconnect_io_qpairs_on_pg, 1211 ctrlr, 1212 nvmf_ctrlr_cc_reset_shn_done); 1213 } 1214 diff.bits.en = 0; 1215 } 1216 1217 if (diff.bits.shn) { 1218 if (cc.bits.shn == SPDK_NVME_SHN_NORMAL || 1219 cc.bits.shn == SPDK_NVME_SHN_ABRUPT) { 1220 SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n", 1221 cc.bits.shn >> 1, cc.bits.shn & 1); 1222 if (ctrlr->disconnect_in_progress) { 1223 SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n"); 1224 return true; 1225 } 1226 1227 ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr, 1228 NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000); 1229 /* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */ 1230 cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2; 1231 ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000; 1232 1233 ctrlr->vcprop.cc.bits.shn = cc.bits.shn; 1234 ctrlr->disconnect_in_progress = true; 1235 ctrlr->disconnect_is_shn = true; 1236 spdk_for_each_channel(ctrlr->subsys->tgt, 1237 nvmf_ctrlr_disconnect_io_qpairs_on_pg, 1238 ctrlr, 1239 nvmf_ctrlr_cc_reset_shn_done); 1240 1241 /* From the time a shutdown is initiated the controller shall disable 1242 * Keep Alive timer */ 1243 nvmf_ctrlr_stop_keep_alive_timer(ctrlr); 1244 } else if (cc.bits.shn == 0) { 1245 ctrlr->vcprop.cc.bits.shn = 0; 1246 } else { 1247 SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n", 1248 cc.bits.shn >> 1, cc.bits.shn & 1); 1249 return false; 1250 } 1251 diff.bits.shn = 0; 1252 } 1253 1254 if (diff.bits.iosqes) { 1255 SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n", 1256 cc.bits.iosqes, 1u << cc.bits.iosqes); 1257 ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes; 1258 diff.bits.iosqes = 0; 1259 } 1260 1261 if (diff.bits.iocqes) { 1262 SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n", 1263 cc.bits.iocqes, 1u << cc.bits.iocqes); 1264 ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes; 1265 diff.bits.iocqes = 0; 1266 } 1267 1268 if (diff.bits.ams) { 1269 SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams); 1270 return false; 1271 } 1272 1273 if (diff.bits.mps) { 1274 SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps))); 1275 return false; 1276 } 1277 1278 if (diff.bits.css) { 1279 if (cc.bits.css > SPDK_NVME_CC_CSS_IOCS) { 1280 SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css); 1281 return false; 1282 } 1283 diff.bits.css = 0; 1284 } 1285 1286 if (diff.raw != 0) { 1287 /* Print an error message, but don't fail the command in this case. 1288 * If we did want to fail in this case, we'd need to ensure we acted 1289 * on no other bits or the initiator gets confused. */ 1290 SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw); 1291 } 1292 1293 return true; 1294 } 1295 1296 static uint64_t 1297 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr) 1298 { 1299 return ctrlr->vcprop.csts.raw; 1300 } 1301 1302 static uint64_t 1303 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr) 1304 { 1305 return ctrlr->vcprop.aqa.raw; 1306 } 1307 1308 static bool 1309 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1310 { 1311 union spdk_nvme_aqa_register aqa; 1312 1313 aqa.raw = value; 1314 1315 /* 1316 * We don't need to explicitly check for maximum size, as the fields are 1317 * limited to 12 bits (4096). 1318 */ 1319 if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 || 1320 aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 || 1321 aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) { 1322 return false; 1323 } 1324 1325 ctrlr->vcprop.aqa.raw = value; 1326 1327 return true; 1328 } 1329 1330 static uint64_t 1331 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr) 1332 { 1333 return ctrlr->vcprop.asq; 1334 } 1335 1336 static bool 1337 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1338 { 1339 ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value; 1340 1341 return true; 1342 } 1343 1344 static bool 1345 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1346 { 1347 ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL); 1348 1349 return true; 1350 } 1351 1352 static uint64_t 1353 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr) 1354 { 1355 return ctrlr->vcprop.acq; 1356 } 1357 1358 static bool 1359 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1360 { 1361 ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value; 1362 1363 return true; 1364 } 1365 1366 static bool 1367 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value) 1368 { 1369 ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL); 1370 1371 return true; 1372 } 1373 1374 struct nvmf_prop { 1375 uint32_t ofst; 1376 uint8_t size; 1377 char name[11]; 1378 uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr); 1379 bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value); 1380 bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value); 1381 }; 1382 1383 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \ 1384 { \ 1385 offsetof(struct spdk_nvme_registers, field), \ 1386 size, \ 1387 #field, \ 1388 get_cb, set_cb, set_upper_cb \ 1389 } 1390 1391 static const struct nvmf_prop nvmf_props[] = { 1392 PROP(cap, 8, nvmf_prop_get_cap, NULL, NULL), 1393 PROP(vs, 4, nvmf_prop_get_vs, NULL, NULL), 1394 PROP(cc, 4, nvmf_prop_get_cc, nvmf_prop_set_cc, NULL), 1395 PROP(csts, 4, nvmf_prop_get_csts, NULL, NULL), 1396 PROP(aqa, 4, nvmf_prop_get_aqa, nvmf_prop_set_aqa, NULL), 1397 PROP(asq, 8, nvmf_prop_get_asq, nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper), 1398 PROP(acq, 8, nvmf_prop_get_acq, nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper), 1399 }; 1400 1401 static const struct nvmf_prop * 1402 find_prop(uint32_t ofst, uint8_t size) 1403 { 1404 size_t i; 1405 1406 for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) { 1407 const struct nvmf_prop *prop = &nvmf_props[i]; 1408 1409 if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) { 1410 return prop; 1411 } 1412 } 1413 1414 return NULL; 1415 } 1416 1417 static int 1418 nvmf_property_get(struct spdk_nvmf_request *req) 1419 { 1420 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1421 struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd; 1422 struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp; 1423 const struct nvmf_prop *prop; 1424 uint8_t size; 1425 1426 response->status.sc = 0; 1427 response->value.u64 = 0; 1428 1429 SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n", 1430 cmd->attrib.size, cmd->ofst); 1431 1432 switch (cmd->attrib.size) { 1433 case SPDK_NVMF_PROP_SIZE_4: 1434 size = 4; 1435 break; 1436 case SPDK_NVMF_PROP_SIZE_8: 1437 size = 8; 1438 break; 1439 default: 1440 SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size); 1441 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1442 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1443 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1444 } 1445 1446 prop = find_prop(cmd->ofst, size); 1447 if (prop == NULL || prop->get_cb == NULL) { 1448 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1449 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1450 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1451 } 1452 1453 SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name); 1454 1455 response->value.u64 = prop->get_cb(ctrlr); 1456 1457 if (size != prop->size) { 1458 /* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */ 1459 assert(size == 4); 1460 assert(prop->size == 8); 1461 1462 if (cmd->ofst == prop->ofst) { 1463 /* Keep bottom 4 bytes only */ 1464 response->value.u64 &= 0xFFFFFFFF; 1465 } else { 1466 /* Keep top 4 bytes only */ 1467 response->value.u64 >>= 32; 1468 } 1469 } 1470 1471 SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64); 1472 1473 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1474 } 1475 1476 static int 1477 nvmf_property_set(struct spdk_nvmf_request *req) 1478 { 1479 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1480 struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd; 1481 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1482 const struct nvmf_prop *prop; 1483 uint64_t value; 1484 uint8_t size; 1485 bool ret; 1486 1487 SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n", 1488 cmd->attrib.size, cmd->ofst, cmd->value.u64); 1489 1490 switch (cmd->attrib.size) { 1491 case SPDK_NVMF_PROP_SIZE_4: 1492 size = 4; 1493 break; 1494 case SPDK_NVMF_PROP_SIZE_8: 1495 size = 8; 1496 break; 1497 default: 1498 SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size); 1499 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1500 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1501 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1502 } 1503 1504 prop = find_prop(cmd->ofst, size); 1505 if (prop == NULL || prop->set_cb == NULL) { 1506 SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst); 1507 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1508 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1509 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1510 } 1511 1512 SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name); 1513 1514 value = cmd->value.u64; 1515 1516 if (prop->size == 4) { 1517 ret = prop->set_cb(ctrlr, (uint32_t)value); 1518 } else if (size != prop->size) { 1519 /* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */ 1520 assert(size == 4); 1521 assert(prop->size == 8); 1522 1523 if (cmd->ofst == prop->ofst) { 1524 ret = prop->set_cb(ctrlr, (uint32_t)value); 1525 } else { 1526 ret = prop->set_upper_cb(ctrlr, (uint32_t)value); 1527 } 1528 } else { 1529 ret = prop->set_cb(ctrlr, (uint32_t)value); 1530 if (ret) { 1531 ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32)); 1532 } 1533 } 1534 1535 if (!ret) { 1536 SPDK_ERRLOG("prop set_cb failed\n"); 1537 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1538 response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM; 1539 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1540 } 1541 1542 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1543 } 1544 1545 static int 1546 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req) 1547 { 1548 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1549 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1550 1551 SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11); 1552 1553 ctrlr->feat.arbitration.raw = cmd->cdw11; 1554 ctrlr->feat.arbitration.bits.reserved = 0; 1555 1556 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1557 } 1558 1559 static int 1560 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req) 1561 { 1562 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1563 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1564 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1565 1566 SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11); 1567 1568 /* Only PS = 0 is allowed, since we report NPSS = 0 */ 1569 if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) { 1570 SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps); 1571 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1572 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1573 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1574 } 1575 1576 ctrlr->feat.power_management.raw = cmd->cdw11; 1577 ctrlr->feat.power_management.bits.reserved = 0; 1578 1579 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1580 } 1581 1582 static bool 1583 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts) 1584 { 1585 /* 1586 * Valid TMPSEL values: 1587 * 0000b - 1000b: temperature sensors 1588 * 1111b: set all implemented temperature sensors 1589 */ 1590 if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) { 1591 /* 1001b - 1110b: reserved */ 1592 SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel); 1593 return false; 1594 } 1595 1596 /* 1597 * Valid THSEL values: 1598 * 00b: over temperature threshold 1599 * 01b: under temperature threshold 1600 */ 1601 if (opts->bits.thsel > 1) { 1602 /* 10b - 11b: reserved */ 1603 SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel); 1604 return false; 1605 } 1606 1607 return true; 1608 } 1609 1610 static int 1611 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req) 1612 { 1613 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1614 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1615 1616 SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11); 1617 1618 if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) { 1619 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1620 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1621 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1622 } 1623 1624 /* TODO: no sensors implemented - ignore new values */ 1625 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1626 } 1627 1628 static int 1629 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req) 1630 { 1631 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1632 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1633 1634 SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11); 1635 1636 if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) { 1637 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1638 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1639 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1640 } 1641 1642 /* TODO: no sensors implemented - return 0 for all thresholds */ 1643 rsp->cdw0 = 0; 1644 1645 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1646 } 1647 1648 static int 1649 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req) 1650 { 1651 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1652 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1653 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1654 union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {}; 1655 1656 SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11); 1657 1658 iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv; 1659 iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd; 1660 rsp->cdw0 = iv_conf.raw; 1661 1662 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1663 } 1664 1665 static int 1666 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req) 1667 { 1668 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1669 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1670 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1671 1672 SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11); 1673 1674 if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) { 1675 /* 1676 * Host is not allowed to set this bit, since we don't advertise it in 1677 * Identify Namespace. 1678 */ 1679 SPDK_ERRLOG("Host set unsupported DULBE bit\n"); 1680 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1681 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1682 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1683 } 1684 1685 ctrlr->feat.error_recovery.raw = cmd->cdw11; 1686 ctrlr->feat.error_recovery.bits.reserved = 0; 1687 1688 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1689 } 1690 1691 static int 1692 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req) 1693 { 1694 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1695 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1696 1697 SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11); 1698 1699 ctrlr->feat.volatile_write_cache.raw = cmd->cdw11; 1700 ctrlr->feat.volatile_write_cache.bits.reserved = 0; 1701 1702 SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n", 1703 ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled"); 1704 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1705 } 1706 1707 static int 1708 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req) 1709 { 1710 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1711 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1712 1713 SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11); 1714 1715 ctrlr->feat.write_atomicity.raw = cmd->cdw11; 1716 ctrlr->feat.write_atomicity.bits.reserved = 0; 1717 1718 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1719 } 1720 1721 static int 1722 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req) 1723 { 1724 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1725 1726 SPDK_ERRLOG("Set Features - Host Identifier not allowed\n"); 1727 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 1728 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1729 } 1730 1731 static int 1732 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req) 1733 { 1734 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1735 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1736 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1737 struct spdk_iov_xfer ix; 1738 1739 SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n"); 1740 1741 if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) { 1742 /* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */ 1743 SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n"); 1744 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1745 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1746 } 1747 1748 if (req->iovcnt < 1 || req->length < sizeof(ctrlr->hostid)) { 1749 SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n"); 1750 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1751 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1752 } 1753 1754 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 1755 spdk_iov_xfer_from_buf(&ix, &ctrlr->hostid, sizeof(ctrlr->hostid)); 1756 1757 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1758 } 1759 1760 static int 1761 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req) 1762 { 1763 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1764 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1765 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1766 struct spdk_nvmf_ns *ns; 1767 1768 SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notification Mask\n"); 1769 1770 if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1771 SPDK_ERRLOG("get Features - Invalid Namespace ID\n"); 1772 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1773 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1774 } 1775 1776 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1777 if (ns == NULL) { 1778 SPDK_ERRLOG("get Features - Invalid Namespace ID\n"); 1779 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1780 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1781 } 1782 rsp->cdw0 = ns->mask; 1783 1784 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1785 } 1786 1787 static int 1788 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req) 1789 { 1790 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1791 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 1792 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1793 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1794 struct spdk_nvmf_ns *ns; 1795 1796 SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notification Mask\n"); 1797 1798 if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1799 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 1800 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 1801 ns->mask = cmd->cdw11; 1802 } 1803 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1804 } 1805 1806 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1807 if (ns == NULL) { 1808 SPDK_ERRLOG("Set Features - Invalid Namespace ID\n"); 1809 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1810 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1811 } 1812 ns->mask = cmd->cdw11; 1813 1814 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1815 } 1816 1817 static int 1818 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req) 1819 { 1820 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1821 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1822 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1823 struct spdk_nvmf_ns *ns; 1824 1825 SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n"); 1826 1827 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1828 /* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */ 1829 if (ns == NULL) { 1830 SPDK_ERRLOG("Get Features - Invalid Namespace ID\n"); 1831 response->status.sct = SPDK_NVME_SCT_GENERIC; 1832 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1833 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1834 } 1835 1836 response->cdw0 = ns->ptpl_activated; 1837 1838 response->status.sct = SPDK_NVME_SCT_GENERIC; 1839 response->status.sc = SPDK_NVME_SC_SUCCESS; 1840 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1841 } 1842 1843 static int 1844 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req) 1845 { 1846 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1847 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1848 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1849 struct spdk_nvmf_ns *ns; 1850 bool ptpl; 1851 1852 SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n"); 1853 1854 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 1855 ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl; 1856 1857 if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && nvmf_ns_is_ptpl_capable(ns)) { 1858 ns->ptpl_activated = ptpl; 1859 } else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 1860 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns; 1861 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 1862 if (nvmf_ns_is_ptpl_capable(ns)) { 1863 ns->ptpl_activated = ptpl; 1864 } 1865 } 1866 } else { 1867 SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n"); 1868 response->status.sct = SPDK_NVME_SCT_GENERIC; 1869 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1870 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1871 } 1872 1873 /* TODO: Feature not changeable for now */ 1874 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 1875 response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE; 1876 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1877 } 1878 1879 static int 1880 nvmf_ctrlr_get_features_host_behavior_support(struct spdk_nvmf_request *req) 1881 { 1882 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1883 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1884 struct spdk_nvme_host_behavior host_behavior = {}; 1885 struct spdk_iov_xfer ix; 1886 1887 SPDK_DEBUGLOG(nvmf, "Get Features - Host Behavior Support\n"); 1888 1889 if (req->iovcnt < 1 || req->length < sizeof(struct spdk_nvme_host_behavior)) { 1890 SPDK_ERRLOG("invalid data buffer for Host Behavior Support\n"); 1891 response->status.sct = SPDK_NVME_SCT_GENERIC; 1892 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1893 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1894 } 1895 1896 host_behavior.acre = ctrlr->acre_enabled; 1897 1898 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 1899 spdk_iov_xfer_from_buf(&ix, &host_behavior, sizeof(host_behavior)); 1900 1901 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1902 } 1903 1904 static int 1905 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req) 1906 { 1907 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1908 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1909 struct spdk_nvme_host_behavior *host_behavior; 1910 1911 SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n"); 1912 if (req->iovcnt != 1) { 1913 SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt); 1914 response->status.sct = SPDK_NVME_SCT_GENERIC; 1915 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1916 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1917 } 1918 if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) { 1919 SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len); 1920 response->status.sct = SPDK_NVME_SCT_GENERIC; 1921 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1922 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1923 } 1924 1925 host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base; 1926 if (host_behavior->acre == 0) { 1927 ctrlr->acre_enabled = false; 1928 } else if (host_behavior->acre == 1) { 1929 ctrlr->acre_enabled = true; 1930 } else { 1931 SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre); 1932 response->status.sct = SPDK_NVME_SCT_GENERIC; 1933 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1934 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1935 } 1936 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1937 } 1938 1939 static int 1940 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req) 1941 { 1942 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1943 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1944 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1945 1946 SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11); 1947 1948 /* 1949 * if attempts to disable keep alive by setting kato to 0h 1950 * a status value of keep alive invalid shall be returned 1951 */ 1952 if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) { 1953 rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID; 1954 } else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) { 1955 ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS; 1956 } else { 1957 /* round up to milliseconds */ 1958 ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up( 1959 cmd->cdw11_bits.feat_keep_alive_timer.bits.kato, 1960 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) * 1961 KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS; 1962 } 1963 1964 /* 1965 * if change the keep alive timeout value successfully 1966 * update the keep alive poller. 1967 */ 1968 if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) { 1969 if (ctrlr->keep_alive_poller != NULL) { 1970 spdk_poller_unregister(&ctrlr->keep_alive_poller); 1971 } 1972 ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr, 1973 ctrlr->feat.keep_alive_timer.bits.kato * 1000); 1974 } 1975 1976 SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n", 1977 ctrlr->feat.keep_alive_timer.bits.kato); 1978 1979 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1980 } 1981 1982 static int 1983 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req) 1984 { 1985 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 1986 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 1987 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1988 uint32_t count; 1989 1990 SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n", 1991 req->cmd->nvme_cmd.cdw11); 1992 1993 if (cmd->cdw11_bits.feat_num_of_queues.bits.ncqr == UINT16_MAX || 1994 cmd->cdw11_bits.feat_num_of_queues.bits.nsqr == UINT16_MAX) { 1995 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 1996 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1997 } 1998 1999 count = spdk_bit_array_count_set(ctrlr->qpair_mask); 2000 /* verify that the controller is ready to process commands */ 2001 if (count > 1) { 2002 SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n"); 2003 rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 2004 } else { 2005 /* 2006 * Ignore the value requested by the host - 2007 * always return the pre-configured value based on max_qpairs_allowed. 2008 */ 2009 rsp->cdw0 = ctrlr->feat.number_of_queues.raw; 2010 } 2011 2012 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2013 } 2014 2015 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ctrlr) == 4928, 2016 "Please check migration fields that need to be added or not"); 2017 2018 static void 2019 nvmf_ctrlr_migr_data_copy(struct spdk_nvmf_ctrlr_migr_data *data, 2020 const struct spdk_nvmf_ctrlr_migr_data *data_src, size_t data_size) 2021 { 2022 assert(data); 2023 assert(data_src); 2024 assert(data_size); 2025 2026 memcpy(&data->regs, &data_src->regs, spdk_min(data->regs_size, data_src->regs_size)); 2027 memcpy(&data->feat, &data_src->feat, spdk_min(data->feat_size, data_src->feat_size)); 2028 2029 #define SET_FIELD(field) \ 2030 if (offsetof(struct spdk_nvmf_ctrlr_migr_data, field) + sizeof(data->field) <= data_size) { \ 2031 data->field = data_src->field; \ 2032 } \ 2033 2034 SET_FIELD(cntlid); 2035 SET_FIELD(acre); 2036 SET_FIELD(num_aer_cids); 2037 SET_FIELD(num_async_events); 2038 SET_FIELD(notice_aen_mask); 2039 #undef SET_FIELD 2040 2041 #define SET_ARRAY(arr) \ 2042 if (offsetof(struct spdk_nvmf_ctrlr_migr_data, arr) + sizeof(data->arr) <= data_size) { \ 2043 memcpy(&data->arr, &data_src->arr, sizeof(data->arr)); \ 2044 } \ 2045 2046 SET_ARRAY(async_events); 2047 SET_ARRAY(aer_cids); 2048 #undef SET_ARRAY 2049 } 2050 2051 int 2052 spdk_nvmf_ctrlr_save_migr_data(struct spdk_nvmf_ctrlr *ctrlr, 2053 struct spdk_nvmf_ctrlr_migr_data *data) 2054 { 2055 struct spdk_nvmf_async_event_completion *event, *event_tmp; 2056 uint32_t i; 2057 struct spdk_nvmf_ctrlr_migr_data data_local = { 2058 .data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused), 2059 .regs_size = sizeof(struct spdk_nvmf_registers), 2060 .feat_size = sizeof(struct spdk_nvmf_ctrlr_feat) 2061 }; 2062 2063 assert(data->data_size <= sizeof(data_local)); 2064 assert(spdk_get_thread() == ctrlr->thread); 2065 2066 memcpy(&data_local.regs, &ctrlr->vcprop, sizeof(struct spdk_nvmf_registers)); 2067 memcpy(&data_local.feat, &ctrlr->feat, sizeof(struct spdk_nvmf_ctrlr_feat)); 2068 2069 data_local.cntlid = ctrlr->cntlid; 2070 data_local.acre = ctrlr->acre_enabled; 2071 data_local.num_aer_cids = ctrlr->nr_aer_reqs; 2072 2073 STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) { 2074 if (data_local.num_async_events + 1 > SPDK_NVMF_MIGR_MAX_PENDING_AERS) { 2075 SPDK_ERRLOG("ctrlr %p has too many pending AERs\n", ctrlr); 2076 break; 2077 } 2078 2079 data_local.async_events[data_local.num_async_events++].raw = event->event.raw; 2080 } 2081 2082 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 2083 struct spdk_nvmf_request *req = ctrlr->aer_req[i]; 2084 data_local.aer_cids[i] = req->cmd->nvme_cmd.cid; 2085 } 2086 data_local.notice_aen_mask = ctrlr->notice_aen_mask; 2087 2088 nvmf_ctrlr_migr_data_copy(data, &data_local, spdk_min(data->data_size, data_local.data_size)); 2089 return 0; 2090 } 2091 2092 int 2093 spdk_nvmf_ctrlr_restore_migr_data(struct spdk_nvmf_ctrlr *ctrlr, 2094 const struct spdk_nvmf_ctrlr_migr_data *data) 2095 { 2096 uint32_t i; 2097 struct spdk_nvmf_ctrlr_migr_data data_local = { 2098 .data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused), 2099 .regs_size = sizeof(struct spdk_nvmf_registers), 2100 .feat_size = sizeof(struct spdk_nvmf_ctrlr_feat) 2101 }; 2102 2103 assert(data->data_size <= sizeof(data_local)); 2104 assert(spdk_get_thread() == ctrlr->thread); 2105 2106 /* local version of data should have defaults set before copy */ 2107 nvmf_ctrlr_migr_data_copy(&data_local, data, spdk_min(data->data_size, data_local.data_size)); 2108 memcpy(&ctrlr->vcprop, &data_local.regs, sizeof(struct spdk_nvmf_registers)); 2109 memcpy(&ctrlr->feat, &data_local.feat, sizeof(struct spdk_nvmf_ctrlr_feat)); 2110 2111 ctrlr->cntlid = data_local.cntlid; 2112 ctrlr->acre_enabled = data_local.acre; 2113 2114 for (i = 0; i < data_local.num_async_events; i++) { 2115 struct spdk_nvmf_async_event_completion *event; 2116 2117 event = calloc(1, sizeof(*event)); 2118 if (!event) { 2119 return -ENOMEM; 2120 } 2121 2122 event->event.raw = data_local.async_events[i].raw; 2123 STAILQ_INSERT_TAIL(&ctrlr->async_events, event, link); 2124 } 2125 ctrlr->notice_aen_mask = data_local.notice_aen_mask; 2126 2127 return 0; 2128 } 2129 2130 static int 2131 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req) 2132 { 2133 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2134 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2135 2136 SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n", 2137 cmd->cdw11); 2138 ctrlr->feat.async_event_configuration.raw = cmd->cdw11; 2139 ctrlr->feat.async_event_configuration.bits.reserved1 = 0; 2140 ctrlr->feat.async_event_configuration.bits.reserved2 = 0; 2141 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2142 } 2143 2144 static int 2145 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req) 2146 { 2147 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2148 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 2149 struct spdk_nvmf_async_event_completion *pending_event; 2150 2151 SPDK_DEBUGLOG(nvmf, "Async Event Request\n"); 2152 2153 /* Four asynchronous events are supported for now */ 2154 if (ctrlr->nr_aer_reqs >= SPDK_NVMF_MAX_ASYNC_EVENTS) { 2155 SPDK_DEBUGLOG(nvmf, "AERL exceeded\n"); 2156 rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 2157 rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED; 2158 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2159 } 2160 2161 if (!STAILQ_EMPTY(&ctrlr->async_events)) { 2162 pending_event = STAILQ_FIRST(&ctrlr->async_events); 2163 rsp->cdw0 = pending_event->event.raw; 2164 STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link); 2165 free(pending_event); 2166 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2167 } 2168 2169 ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req; 2170 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 2171 } 2172 2173 static void 2174 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length) 2175 { 2176 struct spdk_nvme_firmware_page fw_page; 2177 size_t copy_len; 2178 struct spdk_iov_xfer ix; 2179 2180 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2181 2182 memset(&fw_page, 0, sizeof(fw_page)); 2183 fw_page.afi.active_slot = 1; 2184 fw_page.afi.next_reset_slot = 0; 2185 spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' '); 2186 2187 if (offset < sizeof(fw_page)) { 2188 copy_len = spdk_min(sizeof(fw_page) - offset, length); 2189 if (copy_len > 0) { 2190 spdk_iov_xfer_from_buf(&ix, (const char *)&fw_page + offset, copy_len); 2191 } 2192 } 2193 } 2194 2195 /* 2196 * Asynchronous Event Mask Bit 2197 */ 2198 enum spdk_nvme_async_event_mask_bit { 2199 /* Mask Namespace Change Notification */ 2200 SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT = 0, 2201 /* Mask Asymmetric Namespace Access Change Notification */ 2202 SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT = 1, 2203 /* Mask Discovery Log Change Notification */ 2204 SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT = 2, 2205 /* Mask Reservation Log Page Available Notification */ 2206 SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT = 3, 2207 /* Mask Error Event */ 2208 SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT = 4, 2209 /* 4 - 63 Reserved */ 2210 }; 2211 2212 static inline void 2213 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr, 2214 enum spdk_nvme_async_event_mask_bit mask) 2215 { 2216 ctrlr->notice_aen_mask &= ~(1 << mask); 2217 } 2218 2219 static inline bool 2220 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr, 2221 enum spdk_nvme_async_event_mask_bit mask) 2222 { 2223 if (ctrlr->notice_aen_mask & (1 << mask)) { 2224 return false; 2225 } else { 2226 ctrlr->notice_aen_mask |= (1 << mask); 2227 return true; 2228 } 2229 } 2230 2231 /* we have to use the typedef in the function declaration to appease astyle. */ 2232 typedef enum spdk_nvme_ana_state spdk_nvme_ana_state_t; 2233 2234 static inline spdk_nvme_ana_state_t 2235 nvmf_ctrlr_get_ana_state(struct spdk_nvmf_ctrlr *ctrlr, uint32_t anagrpid) 2236 { 2237 if (!ctrlr->subsys->flags.ana_reporting) { 2238 return SPDK_NVME_ANA_OPTIMIZED_STATE; 2239 } 2240 2241 if (spdk_unlikely(ctrlr->listener == NULL)) { 2242 return SPDK_NVME_ANA_INACCESSIBLE_STATE; 2243 } 2244 2245 assert(anagrpid - 1 < ctrlr->subsys->max_nsid); 2246 return ctrlr->listener->ana_state[anagrpid - 1]; 2247 } 2248 2249 static spdk_nvme_ana_state_t 2250 nvmf_ctrlr_get_ana_state_from_nsid(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid) 2251 { 2252 struct spdk_nvmf_ns *ns; 2253 2254 /* We do not have NVM subsystem specific ANA state. Hence if NSID is either 2255 * SPDK_NVMF_GLOBAL_NS_TAG, invalid, or for inactive namespace, return 2256 * the optimized state. 2257 */ 2258 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 2259 if (ns == NULL) { 2260 return SPDK_NVME_ANA_OPTIMIZED_STATE; 2261 } 2262 2263 return nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid); 2264 } 2265 2266 static void 2267 nvmf_get_error_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt, 2268 uint64_t offset, uint32_t length, uint32_t rae) 2269 { 2270 if (!rae) { 2271 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT); 2272 } 2273 2274 /* TODO: actually fill out log page data */ 2275 } 2276 2277 static void 2278 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt, 2279 uint64_t offset, uint32_t length, uint32_t rae) 2280 { 2281 struct spdk_nvme_ana_page ana_hdr; 2282 struct spdk_nvme_ana_group_descriptor ana_desc; 2283 size_t copy_len, copied_len; 2284 uint32_t num_anagrp = 0, anagrpid; 2285 struct spdk_nvmf_ns *ns; 2286 struct spdk_iov_xfer ix; 2287 2288 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2289 2290 if (length == 0) { 2291 goto done; 2292 } 2293 2294 if (offset >= sizeof(ana_hdr)) { 2295 offset -= sizeof(ana_hdr); 2296 } else { 2297 for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) { 2298 if (ctrlr->subsys->ana_group[anagrpid - 1] > 0) { 2299 num_anagrp++; 2300 } 2301 } 2302 2303 memset(&ana_hdr, 0, sizeof(ana_hdr)); 2304 2305 ana_hdr.num_ana_group_desc = num_anagrp; 2306 /* TODO: Support Change Count. */ 2307 ana_hdr.change_count = 0; 2308 2309 copy_len = spdk_min(sizeof(ana_hdr) - offset, length); 2310 copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_hdr + offset, copy_len); 2311 assert(copied_len == copy_len); 2312 length -= copied_len; 2313 offset = 0; 2314 } 2315 2316 if (length == 0) { 2317 goto done; 2318 } 2319 2320 for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) { 2321 if (ctrlr->subsys->ana_group[anagrpid - 1] == 0) { 2322 continue; 2323 } 2324 2325 if (offset >= sizeof(ana_desc)) { 2326 offset -= sizeof(ana_desc); 2327 } else { 2328 memset(&ana_desc, 0, sizeof(ana_desc)); 2329 2330 ana_desc.ana_group_id = anagrpid; 2331 ana_desc.num_of_nsid = ctrlr->subsys->ana_group[anagrpid - 1]; 2332 ana_desc.ana_state = nvmf_ctrlr_get_ana_state(ctrlr, anagrpid); 2333 2334 copy_len = spdk_min(sizeof(ana_desc) - offset, length); 2335 copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_desc + offset, 2336 copy_len); 2337 assert(copied_len == copy_len); 2338 length -= copied_len; 2339 offset = 0; 2340 2341 if (length == 0) { 2342 goto done; 2343 } 2344 } 2345 2346 /* TODO: Revisit here about O(n^2) cost if we have subsystem with 2347 * many namespaces in the future. 2348 */ 2349 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL; 2350 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 2351 if (ns->anagrpid != anagrpid) { 2352 continue; 2353 } 2354 2355 if (offset >= sizeof(uint32_t)) { 2356 offset -= sizeof(uint32_t); 2357 continue; 2358 } 2359 2360 copy_len = spdk_min(sizeof(uint32_t) - offset, length); 2361 copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ns->nsid + offset, 2362 copy_len); 2363 assert(copied_len == copy_len); 2364 length -= copied_len; 2365 offset = 0; 2366 2367 if (length == 0) { 2368 goto done; 2369 } 2370 } 2371 } 2372 2373 done: 2374 if (!rae) { 2375 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT); 2376 } 2377 } 2378 2379 void 2380 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid) 2381 { 2382 uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list); 2383 uint16_t i; 2384 bool found = false; 2385 2386 for (i = 0; i < ctrlr->changed_ns_list_count; i++) { 2387 if (ctrlr->changed_ns_list.ns_list[i] == nsid) { 2388 /* nsid is already in the list */ 2389 found = true; 2390 break; 2391 } 2392 } 2393 2394 if (!found) { 2395 if (ctrlr->changed_ns_list_count == max_changes) { 2396 /* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */ 2397 ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu; 2398 for (i = 1; i < max_changes; i++) { 2399 ctrlr->changed_ns_list.ns_list[i] = 0; 2400 } 2401 } else { 2402 ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid; 2403 } 2404 } 2405 } 2406 2407 static void 2408 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr, 2409 struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae) 2410 { 2411 size_t copy_length; 2412 struct spdk_iov_xfer ix; 2413 2414 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2415 2416 if (offset < sizeof(ctrlr->changed_ns_list)) { 2417 copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset); 2418 if (copy_length) { 2419 spdk_iov_xfer_from_buf(&ix, (char *)&ctrlr->changed_ns_list + offset, copy_length); 2420 } 2421 } 2422 2423 /* Clear log page each time it is read */ 2424 ctrlr->changed_ns_list_count = 0; 2425 memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list)); 2426 2427 if (!rae) { 2428 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT); 2429 } 2430 } 2431 2432 /* The structure can be modified if we provide support for other commands in future */ 2433 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = { 2434 .admin_cmds_supported = { 2435 /* CSUPP, LBCC, NCC, NIC, CCC, CSE */ 2436 /* Get Log Page */ 2437 [SPDK_NVME_OPC_GET_LOG_PAGE] = {1, 0, 0, 0, 0, 0, 0, 0}, 2438 /* Identify */ 2439 [SPDK_NVME_OPC_IDENTIFY] = {1, 0, 0, 0, 0, 0, 0, 0}, 2440 /* Abort */ 2441 [SPDK_NVME_OPC_ABORT] = {1, 0, 0, 0, 0, 0, 0, 0}, 2442 /* Set Features */ 2443 [SPDK_NVME_OPC_SET_FEATURES] = {1, 0, 0, 0, 0, 0, 0, 0}, 2444 /* Get Features */ 2445 [SPDK_NVME_OPC_GET_FEATURES] = {1, 0, 0, 0, 0, 0, 0, 0}, 2446 /* Async Event Request */ 2447 [SPDK_NVME_OPC_ASYNC_EVENT_REQUEST] = {1, 0, 0, 0, 0, 0, 0, 0}, 2448 /* Keep Alive */ 2449 [SPDK_NVME_OPC_KEEP_ALIVE] = {1, 0, 0, 0, 0, 0, 0, 0}, 2450 }, 2451 .io_cmds_supported = { 2452 /* FLUSH */ 2453 [SPDK_NVME_OPC_FLUSH] = {1, 1, 0, 0, 0, 0, 0, 0}, 2454 /* WRITE */ 2455 [SPDK_NVME_OPC_WRITE] = {1, 1, 0, 0, 0, 0, 0, 0}, 2456 /* READ */ 2457 [SPDK_NVME_OPC_READ] = {1, 0, 0, 0, 0, 0, 0, 0}, 2458 /* WRITE ZEROES */ 2459 [SPDK_NVME_OPC_WRITE_ZEROES] = {1, 1, 0, 0, 0, 0, 0, 0}, 2460 /* DATASET MANAGEMENT */ 2461 [SPDK_NVME_OPC_DATASET_MANAGEMENT] = {1, 1, 0, 0, 0, 0, 0, 0}, 2462 /* COMPARE */ 2463 [SPDK_NVME_OPC_COMPARE] = {1, 0, 0, 0, 0, 0, 0, 0}, 2464 /* ZONE MANAGEMENT SEND */ 2465 [SPDK_NVME_OPC_ZONE_MGMT_SEND] = {1, 1, 0, 0, 0, 0, 0, 0}, 2466 /* ZONE MANAGEMENT RECEIVE */ 2467 [SPDK_NVME_OPC_ZONE_MGMT_RECV] = {1, 0, 0, 0, 0, 0, 0, 0}, 2468 /* COPY */ 2469 [SPDK_NVME_OPC_COPY] = {1, 1, 0, 0, 0, 0, 0, 0}, 2470 }, 2471 }; 2472 2473 static void 2474 nvmf_get_cmds_and_effects_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt, 2475 uint64_t offset, uint32_t length) 2476 { 2477 uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page); 2478 size_t copy_len = 0; 2479 struct spdk_nvme_cmds_and_effect_log_page cmds_and_effect_log_page = g_cmds_and_effect_log_page; 2480 struct spdk_nvme_cmds_and_effect_entry csupp_and_lbcc_effect_entry = {1, 1, 0, 0, 0, 0, 0, 0}; 2481 struct spdk_iov_xfer ix; 2482 2483 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2484 2485 if (offset < page_size) { 2486 if (ctrlr->subsys->zone_append_supported) { 2487 cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_ZONE_APPEND] = 2488 csupp_and_lbcc_effect_entry; 2489 } 2490 copy_len = spdk_min(page_size - offset, length); 2491 spdk_iov_xfer_from_buf(&ix, (char *)(&cmds_and_effect_log_page) + offset, copy_len); 2492 } 2493 } 2494 2495 static void 2496 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr, 2497 struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae) 2498 { 2499 uint32_t unit_log_len, avail_log_len, next_pos, copy_len; 2500 struct spdk_nvmf_reservation_log *log, *log_tmp; 2501 struct spdk_iov_xfer ix; 2502 2503 spdk_iov_xfer_init(&ix, iovs, iovcnt); 2504 2505 unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log); 2506 /* No available log, return zeroed log pages */ 2507 if (!ctrlr->num_avail_log_pages) { 2508 return; 2509 } 2510 2511 avail_log_len = ctrlr->num_avail_log_pages * unit_log_len; 2512 if (offset >= avail_log_len) { 2513 return; 2514 } 2515 2516 next_pos = 0; 2517 TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) { 2518 TAILQ_REMOVE(&ctrlr->log_head, log, link); 2519 ctrlr->num_avail_log_pages--; 2520 2521 next_pos += unit_log_len; 2522 if (next_pos > offset) { 2523 copy_len = spdk_min(next_pos - offset, length); 2524 spdk_iov_xfer_from_buf(&ix, &log->log, copy_len); 2525 length -= copy_len; 2526 offset += copy_len; 2527 } 2528 free(log); 2529 2530 if (length == 0) { 2531 break; 2532 } 2533 } 2534 2535 if (!rae) { 2536 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT); 2537 } 2538 return; 2539 } 2540 2541 static int 2542 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req) 2543 { 2544 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 2545 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2546 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 2547 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 2548 struct spdk_nvme_transport_id cmd_source_trid; 2549 uint64_t offset, len; 2550 uint32_t rae, numdl, numdu; 2551 uint8_t lid; 2552 2553 if (req->iovcnt < 1) { 2554 SPDK_DEBUGLOG(nvmf, "get log command with no buffer\n"); 2555 response->status.sct = SPDK_NVME_SCT_GENERIC; 2556 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2557 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2558 } 2559 2560 offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32); 2561 if (offset & 3) { 2562 SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset); 2563 response->status.sct = SPDK_NVME_SCT_GENERIC; 2564 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2565 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2566 } 2567 2568 rae = cmd->cdw10_bits.get_log_page.rae; 2569 numdl = cmd->cdw10_bits.get_log_page.numdl; 2570 numdu = cmd->cdw11_bits.get_log_page.numdu; 2571 len = ((numdu << 16) + numdl + (uint64_t)1) * 4; 2572 if (len > req->length) { 2573 SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n", 2574 len, req->length); 2575 response->status.sct = SPDK_NVME_SCT_GENERIC; 2576 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2577 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2578 } 2579 2580 lid = cmd->cdw10_bits.get_log_page.lid; 2581 SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n", 2582 lid, offset, len, rae); 2583 2584 if (spdk_nvmf_subsystem_is_discovery(subsystem)) { 2585 switch (lid) { 2586 case SPDK_NVME_LOG_DISCOVERY: 2587 if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &cmd_source_trid)) { 2588 SPDK_ERRLOG("Failed to get LOG_DISCOVERY source trid\n"); 2589 response->status.sct = SPDK_NVME_SCT_GENERIC; 2590 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 2591 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2592 } 2593 nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt, 2594 offset, len, &cmd_source_trid); 2595 if (!rae) { 2596 nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT); 2597 } 2598 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2599 default: 2600 goto invalid_log_page; 2601 } 2602 } else { 2603 if (offset > len) { 2604 SPDK_ERRLOG("Get log page: offset (%" PRIu64 ") > len (%" PRIu64 ")\n", 2605 offset, len); 2606 response->status.sct = SPDK_NVME_SCT_GENERIC; 2607 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2608 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2609 } 2610 2611 switch (lid) { 2612 case SPDK_NVME_LOG_ERROR: 2613 nvmf_get_error_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2614 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2615 case SPDK_NVME_LOG_HEALTH_INFORMATION: 2616 /* TODO: actually fill out log page data */ 2617 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2618 case SPDK_NVME_LOG_FIRMWARE_SLOT: 2619 nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len); 2620 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2621 case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS: 2622 if (subsystem->flags.ana_reporting) { 2623 nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2624 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2625 } else { 2626 goto invalid_log_page; 2627 } 2628 case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG: 2629 nvmf_get_cmds_and_effects_log_page(ctrlr, req->iov, req->iovcnt, offset, len); 2630 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2631 case SPDK_NVME_LOG_CHANGED_NS_LIST: 2632 nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2633 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2634 case SPDK_NVME_LOG_RESERVATION_NOTIFICATION: 2635 nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae); 2636 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2637 default: 2638 goto invalid_log_page; 2639 } 2640 } 2641 2642 invalid_log_page: 2643 SPDK_INFOLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid); 2644 response->status.sct = SPDK_NVME_SCT_GENERIC; 2645 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 2646 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2647 } 2648 2649 static struct spdk_nvmf_ns * 2650 _nvmf_ctrlr_get_ns_safe(struct spdk_nvmf_ctrlr *ctrlr, 2651 uint32_t nsid, 2652 struct spdk_nvme_cpl *rsp) 2653 { 2654 struct spdk_nvmf_ns *ns; 2655 if (nsid == 0 || nsid > ctrlr->subsys->max_nsid) { 2656 SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", nsid); 2657 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2658 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 2659 return NULL; 2660 } 2661 2662 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 2663 if (ns == NULL || ns->bdev == NULL) { 2664 /* 2665 * Inactive namespaces should return a zero filled data structure. 2666 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(), 2667 * so we can just return early here. 2668 */ 2669 SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", nsid); 2670 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2671 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2672 return NULL; 2673 } 2674 return ns; 2675 } 2676 2677 int 2678 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr, 2679 struct spdk_nvme_cmd *cmd, 2680 struct spdk_nvme_cpl *rsp, 2681 struct spdk_nvme_ns_data *nsdata) 2682 { 2683 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2684 struct spdk_nvmf_ns *ns; 2685 uint32_t max_num_blocks, format_index; 2686 enum spdk_nvme_ana_state ana_state; 2687 2688 ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp); 2689 if (ns == NULL) { 2690 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2691 } 2692 2693 nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip); 2694 2695 assert(ctrlr->admin_qpair); 2696 2697 format_index = spdk_nvme_ns_get_format_index(nsdata); 2698 2699 /* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */ 2700 max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size / 2701 (1U << nsdata->lbaf[format_index].lbads); 2702 if (nsdata->noiob > max_num_blocks) { 2703 nsdata->noiob = max_num_blocks; 2704 } 2705 2706 /* Set NOWS equal to Controller MDTS */ 2707 if (nsdata->nsfeat.optperf) { 2708 nsdata->nows = max_num_blocks - 1; 2709 } 2710 2711 if (subsystem->flags.ana_reporting) { 2712 assert(ns->anagrpid - 1 < subsystem->max_nsid); 2713 nsdata->anagrpid = ns->anagrpid; 2714 2715 ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid); 2716 if (ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE || 2717 ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) { 2718 nsdata->nuse = 0; 2719 } 2720 } 2721 2722 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2723 } 2724 2725 static void 2726 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr, 2727 struct spdk_nvme_ctrlr_data *cdata) 2728 { 2729 cdata->oacs = ctrlr->cdata.oacs; 2730 2731 cdata->oacs.virtualization_management = 2732 g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL; 2733 cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL 2734 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL; 2735 cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL 2736 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL; 2737 cdata->oacs.device_self_test = 2738 g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL; 2739 cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL 2740 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL; 2741 cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr != 2742 NULL 2743 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL; 2744 cdata->oacs.format = 2745 g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL; 2746 cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL 2747 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL; 2748 cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr != 2749 NULL; 2750 } 2751 2752 int 2753 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata) 2754 { 2755 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 2756 struct spdk_nvmf_transport *transport; 2757 2758 /* 2759 * Common fields for discovery and NVM subsystems 2760 */ 2761 assert(ctrlr->admin_qpair); 2762 transport = ctrlr->admin_qpair->transport; 2763 spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' '); 2764 assert((transport->opts.max_io_size % 4096) == 0); 2765 cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096); 2766 cdata->cntlid = ctrlr->cntlid; 2767 cdata->ver = ctrlr->vcprop.vs; 2768 cdata->aerl = ctrlr->cdata.aerl; 2769 cdata->lpa.edlp = 1; 2770 cdata->elpe = 127; 2771 cdata->maxcmd = transport->opts.max_queue_depth; 2772 cdata->sgls = ctrlr->cdata.sgls; 2773 cdata->fuses = ctrlr->cdata.fuses; 2774 cdata->acwu = 0; /* ACWU is 0-based. */ 2775 if (subsystem->flags.ana_reporting) { 2776 cdata->mnan = subsystem->max_nsid; 2777 } 2778 spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0'); 2779 2780 SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd); 2781 SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls)); 2782 2783 2784 if (spdk_nvmf_subsystem_is_discovery(subsystem)) { 2785 /* 2786 * NVM Discovery subsystem fields 2787 */ 2788 cdata->oaes.discovery_log_change_notices = 1; 2789 } else { 2790 cdata->vid = ctrlr->cdata.vid; 2791 cdata->ssvid = ctrlr->cdata.ssvid; 2792 cdata->ieee[0] = ctrlr->cdata.ieee[0]; 2793 cdata->ieee[1] = ctrlr->cdata.ieee[1]; 2794 cdata->ieee[2] = ctrlr->cdata.ieee[2]; 2795 2796 /* 2797 * NVM subsystem fields (reserved for discovery subsystems) 2798 */ 2799 spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' '); 2800 spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' '); 2801 cdata->kas = ctrlr->cdata.kas; 2802 2803 cdata->rab = 6; 2804 cdata->cmic.multi_port = 1; 2805 cdata->cmic.multi_ctrlr = 1; 2806 cdata->oaes.ns_attribute_notices = 1; 2807 cdata->ctratt.bits.host_id_exhid_supported = 1; 2808 /* We do not have any actual limitation to the number of abort commands. 2809 * We follow the recommendation by the NVMe specification. 2810 */ 2811 cdata->acl = NVMF_ABORT_COMMAND_LIMIT; 2812 cdata->frmw.slot1_ro = 1; 2813 cdata->frmw.num_slots = 1; 2814 2815 cdata->lpa.celp = 1; /* Command Effects log page supported */ 2816 2817 cdata->sqes.min = 6; 2818 cdata->sqes.max = 6; 2819 cdata->cqes.min = 4; 2820 cdata->cqes.max = 4; 2821 cdata->nn = subsystem->max_nsid; 2822 cdata->vwc.present = 1; 2823 cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED; 2824 2825 cdata->nvmf_specific = ctrlr->cdata.nvmf_specific; 2826 2827 cdata->oncs.compare = ctrlr->cdata.oncs.compare; 2828 cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr); 2829 cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr); 2830 cdata->oncs.reservations = ctrlr->cdata.oncs.reservations; 2831 cdata->oncs.copy = ctrlr->cdata.oncs.copy; 2832 cdata->ocfs.copy_format0 = cdata->oncs.copy; 2833 if (subsystem->flags.ana_reporting) { 2834 /* Asymmetric Namespace Access Reporting is supported. */ 2835 cdata->cmic.ana_reporting = 1; 2836 cdata->oaes.ana_change_notices = 1; 2837 2838 cdata->anatt = ANA_TRANSITION_TIME_IN_SEC; 2839 /* ANA Change state is not used, and ANA Persistent Loss state 2840 * is not supported for now. 2841 */ 2842 cdata->anacap.ana_optimized_state = 1; 2843 cdata->anacap.ana_non_optimized_state = 1; 2844 cdata->anacap.ana_inaccessible_state = 1; 2845 /* ANAGRPID does not change while namespace is attached to controller */ 2846 cdata->anacap.no_change_anagrpid = 1; 2847 cdata->anagrpmax = subsystem->max_nsid; 2848 cdata->nanagrpid = subsystem->max_nsid; 2849 } 2850 2851 nvmf_ctrlr_populate_oacs(ctrlr, cdata); 2852 2853 assert(subsystem->tgt != NULL); 2854 cdata->crdt[0] = subsystem->tgt->crdt[0]; 2855 cdata->crdt[1] = subsystem->tgt->crdt[1]; 2856 cdata->crdt[2] = subsystem->tgt->crdt[2]; 2857 2858 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n", 2859 cdata->nvmf_specific.ioccsz); 2860 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n", 2861 cdata->nvmf_specific.iorcsz); 2862 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n", 2863 cdata->nvmf_specific.icdoff); 2864 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n", 2865 *(uint8_t *)&cdata->nvmf_specific.ctrattr); 2866 SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n", 2867 cdata->nvmf_specific.msdbd); 2868 } 2869 2870 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2871 } 2872 2873 static int 2874 nvmf_ns_identify_iocs_zns(struct spdk_nvmf_ns *ns, 2875 struct spdk_nvme_cmd *cmd, 2876 struct spdk_nvme_cpl *rsp, 2877 struct spdk_nvme_zns_ns_data *nsdata_zns) 2878 { 2879 nsdata_zns->zoc.variable_zone_capacity = 0; 2880 nsdata_zns->zoc.zone_active_excursions = 0; 2881 nsdata_zns->ozcs.read_across_zone_boundaries = 1; 2882 /* Underflowing the zero based mar and mor bdev helper results in the correct 2883 value of FFFFFFFFh. */ 2884 nsdata_zns->mar = spdk_bdev_get_max_active_zones(ns->bdev) - 1; 2885 nsdata_zns->mor = spdk_bdev_get_max_open_zones(ns->bdev) - 1; 2886 nsdata_zns->rrl = 0; 2887 nsdata_zns->frl = 0; 2888 nsdata_zns->lbafe[0].zsze = spdk_bdev_get_zone_size(ns->bdev); 2889 2890 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2891 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2892 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2893 } 2894 2895 int 2896 spdk_nvmf_ns_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr, 2897 struct spdk_nvme_cmd *cmd, 2898 struct spdk_nvme_cpl *rsp, 2899 void *nsdata, 2900 size_t nsdata_size) 2901 { 2902 uint8_t csi = cmd->cdw11_bits.identify.csi; 2903 struct spdk_nvmf_ns *ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp); 2904 2905 memset(nsdata, 0, nsdata_size); 2906 2907 if (ns == NULL) { 2908 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2909 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 2910 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2911 } 2912 2913 switch (csi) { 2914 case SPDK_NVME_CSI_ZNS: 2915 return nvmf_ns_identify_iocs_zns(ns, cmd, rsp, nsdata); 2916 default: 2917 break; 2918 } 2919 2920 SPDK_DEBUGLOG(nvmf, 2921 "Returning zero filled struct for the iocs specific ns " 2922 "identify command and CSI 0x%02x\n", 2923 csi); 2924 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2925 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2926 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2927 } 2928 2929 static int 2930 nvmf_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ctrlr *ctrlr, 2931 struct spdk_nvme_cmd *cmd, 2932 struct spdk_nvme_cpl *rsp, 2933 struct spdk_nvme_nvm_ctrlr_data *cdata_nvm) 2934 { 2935 /* The unit of max_write_zeroes_size_kib is KiB. 2936 * The unit of wzsl is the minimum memory page size(2 ^ (12 + CAP.MPSMIN) bytes) 2937 * and is reported as a power of two (2^n). 2938 */ 2939 cdata_nvm->wzsl = spdk_u64log2(ctrlr->subsys->max_write_zeroes_size_kib >> 2940 (2 + ctrlr->vcprop.cap.bits.mpsmin)); 2941 2942 /* The unit of max_discard_size_kib is KiB. 2943 * The dmrsl indicates the maximum number of logical blocks for 2944 * dataset management command. 2945 */ 2946 cdata_nvm->dmrsl = ctrlr->subsys->max_discard_size_kib << 1; 2947 cdata_nvm->dmrl = 1; 2948 2949 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2950 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2951 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2952 } 2953 2954 static int 2955 nvmf_ctrlr_identify_iocs_zns(struct spdk_nvmf_ctrlr *ctrlr, 2956 struct spdk_nvme_cmd *cmd, 2957 struct spdk_nvme_cpl *rsp, 2958 struct spdk_nvme_zns_ctrlr_data *cdata_zns) 2959 { 2960 /* The unit of max_zone_append_size_kib is KiB. 2961 The unit of zasl is the minimum memory page size 2962 (2 ^ (12 + CAP.MPSMIN) KiB) 2963 and is reported as a power of two (2^n). */ 2964 cdata_zns->zasl = spdk_u64log2(ctrlr->subsys->max_zone_append_size_kib >> 2965 (12 + ctrlr->vcprop.cap.bits.mpsmin)); 2966 2967 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2968 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2969 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2970 } 2971 2972 int 2973 spdk_nvmf_ctrlr_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr, 2974 struct spdk_nvme_cmd *cmd, 2975 struct spdk_nvme_cpl *rsp, 2976 void *cdata, 2977 size_t cdata_size) 2978 { 2979 uint8_t csi = cmd->cdw11_bits.identify.csi; 2980 2981 memset(cdata, 0, cdata_size); 2982 2983 switch (csi) { 2984 case SPDK_NVME_CSI_NVM: 2985 return nvmf_ctrlr_identify_iocs_nvm(ctrlr, cmd, rsp, cdata); 2986 case SPDK_NVME_CSI_ZNS: 2987 return nvmf_ctrlr_identify_iocs_zns(ctrlr, cmd, rsp, cdata); 2988 default: 2989 break; 2990 } 2991 2992 SPDK_DEBUGLOG(nvmf, 2993 "Returning zero filled struct for the iocs specific ctrlr " 2994 "identify command and CSI 0x%02x\n", 2995 csi); 2996 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 2997 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 2998 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 2999 } 3000 3001 static int 3002 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_ctrlr *ctrlr, 3003 struct spdk_nvme_cmd *cmd, 3004 struct spdk_nvme_cpl *rsp, 3005 struct spdk_nvme_ns_list *ns_list) 3006 { 3007 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 3008 struct spdk_nvmf_ns *ns; 3009 uint32_t count = 0; 3010 3011 if (cmd->nsid >= 0xfffffffeUL) { 3012 SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid); 3013 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 3014 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3015 } 3016 3017 memset(ns_list, 0, sizeof(*ns_list)); 3018 3019 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 3020 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 3021 if (ns->opts.nsid <= cmd->nsid || !nvmf_ctrlr_ns_is_visible(ctrlr, ns->opts.nsid)) { 3022 continue; 3023 } 3024 3025 ns_list->ns_list[count++] = ns->opts.nsid; 3026 if (count == SPDK_COUNTOF(ns_list->ns_list)) { 3027 break; 3028 } 3029 } 3030 3031 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3032 } 3033 3034 static void 3035 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain, 3036 enum spdk_nvme_nidt type, 3037 const void *data, size_t data_size) 3038 { 3039 struct spdk_nvme_ns_id_desc *desc; 3040 size_t desc_size = sizeof(*desc) + data_size; 3041 3042 /* 3043 * These should never fail in practice, since all valid NS ID descriptors 3044 * should be defined so that they fit in the available 4096-byte buffer. 3045 */ 3046 assert(data_size > 0); 3047 assert(data_size <= UINT8_MAX); 3048 assert(desc_size < *buf_remain); 3049 if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) { 3050 return; 3051 } 3052 3053 desc = *buf_ptr; 3054 desc->nidt = type; 3055 desc->nidl = data_size; 3056 memcpy(desc->nid, data, data_size); 3057 3058 *buf_ptr += desc_size; 3059 *buf_remain -= desc_size; 3060 } 3061 3062 static int 3063 nvmf_ctrlr_identify_ns_id_descriptor_list( 3064 struct spdk_nvmf_ctrlr *ctrlr, 3065 struct spdk_nvme_cmd *cmd, 3066 struct spdk_nvme_cpl *rsp, 3067 void *id_desc_list, size_t id_desc_list_size) 3068 { 3069 struct spdk_nvmf_ns *ns; 3070 size_t buf_remain = id_desc_list_size; 3071 void *buf_ptr = id_desc_list; 3072 3073 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 3074 if (ns == NULL || ns->bdev == NULL) { 3075 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3076 rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 3077 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3078 } 3079 3080 #define ADD_ID_DESC(type, data, size) \ 3081 do { \ 3082 if (!spdk_mem_all_zero(data, size)) { \ 3083 _add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \ 3084 } \ 3085 } while (0) 3086 3087 ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64)); 3088 ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid)); 3089 ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid)); 3090 ADD_ID_DESC(SPDK_NVME_NIDT_CSI, &ns->csi, sizeof(uint8_t)); 3091 3092 /* 3093 * The list is automatically 0-terminated, both in the temporary buffer 3094 * used by nvmf_ctrlr_identify(), and the eventual iov destination - 3095 * controller to host buffers in admin commands always get zeroed in 3096 * nvmf_ctrlr_process_admin_cmd(). 3097 */ 3098 3099 #undef ADD_ID_DESC 3100 3101 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3102 } 3103 3104 static int 3105 nvmf_ctrlr_identify_iocs(struct spdk_nvmf_ctrlr *ctrlr, 3106 struct spdk_nvme_cmd *cmd, 3107 struct spdk_nvme_cpl *rsp, 3108 void *cdata, size_t cdata_size) 3109 { 3110 struct spdk_nvme_iocs_vector *vector; 3111 struct spdk_nvmf_ns *ns; 3112 3113 if (cdata_size < sizeof(struct spdk_nvme_iocs_vector)) { 3114 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3115 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3116 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3117 } 3118 3119 /* For now we only support this command sent to the current 3120 * controller. 3121 */ 3122 if (cmd->cdw10_bits.identify.cntid != 0xFFFF && 3123 cmd->cdw10_bits.identify.cntid != ctrlr->cntlid) { 3124 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3125 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3126 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3127 } 3128 memset(cdata, 0, cdata_size); 3129 3130 vector = cdata; 3131 vector->nvm = 1; 3132 for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL; 3133 ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) { 3134 if (ns->bdev == NULL) { 3135 continue; 3136 } 3137 if (spdk_bdev_is_zoned(ns->bdev)) { 3138 vector->zns = 1; 3139 } 3140 } 3141 3142 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3143 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 3144 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3145 } 3146 3147 static int 3148 nvmf_ctrlr_identify(struct spdk_nvmf_request *req) 3149 { 3150 uint8_t cns; 3151 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3152 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3153 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 3154 struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys; 3155 int ret = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3156 char tmpbuf[SPDK_NVME_IDENTIFY_BUFLEN] = ""; 3157 struct spdk_iov_xfer ix; 3158 3159 if (req->iovcnt < 1 || req->length < SPDK_NVME_IDENTIFY_BUFLEN) { 3160 SPDK_DEBUGLOG(nvmf, "identify command with invalid buffer\n"); 3161 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3162 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3163 return ret; 3164 } 3165 3166 cns = cmd->cdw10_bits.identify.cns; 3167 3168 if (spdk_nvmf_subsystem_is_discovery(subsystem) && 3169 cns != SPDK_NVME_IDENTIFY_CTRLR) { 3170 /* Discovery controllers only support Identify Controller */ 3171 goto invalid_cns; 3172 } 3173 3174 /* 3175 * We must use a temporary buffer: it's entirely possible the out buffer 3176 * is split across more than one IOV. 3177 */ 3178 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 3179 3180 SPDK_DEBUGLOG(nvmf, "Received identify command with CNS 0x%02x\n", cns); 3181 3182 switch (cns) { 3183 case SPDK_NVME_IDENTIFY_NS: 3184 ret = spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, (void *)&tmpbuf); 3185 break; 3186 case SPDK_NVME_IDENTIFY_CTRLR: 3187 ret = spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, (void *)&tmpbuf); 3188 break; 3189 case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST: 3190 ret = nvmf_ctrlr_identify_active_ns_list(ctrlr, cmd, rsp, (void *)&tmpbuf); 3191 break; 3192 case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST: 3193 ret = nvmf_ctrlr_identify_ns_id_descriptor_list(ctrlr, cmd, rsp, 3194 tmpbuf, req->length); 3195 break; 3196 case SPDK_NVME_IDENTIFY_NS_IOCS: 3197 ret = spdk_nvmf_ns_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length); 3198 break; 3199 case SPDK_NVME_IDENTIFY_CTRLR_IOCS: 3200 ret = spdk_nvmf_ctrlr_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length); 3201 break; 3202 case SPDK_NVME_IDENTIFY_IOCS: 3203 ret = nvmf_ctrlr_identify_iocs(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length); 3204 break; 3205 default: 3206 goto invalid_cns; 3207 } 3208 3209 if (ret == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 3210 spdk_iov_xfer_from_buf(&ix, tmpbuf, sizeof(tmpbuf)); 3211 } 3212 3213 return ret; 3214 3215 invalid_cns: 3216 SPDK_DEBUGLOG(nvmf, "Identify command with unsupported CNS 0x%02x\n", cns); 3217 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3218 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3219 return ret; 3220 } 3221 3222 static bool 3223 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid) 3224 { 3225 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 3226 struct spdk_nvmf_request *req; 3227 int i; 3228 3229 if (!nvmf_qpair_is_admin_queue(qpair)) { 3230 return false; 3231 } 3232 3233 assert(spdk_get_thread() == ctrlr->thread); 3234 3235 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 3236 if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) { 3237 SPDK_DEBUGLOG(nvmf, "Aborting AER request\n"); 3238 req = ctrlr->aer_req[i]; 3239 ctrlr->aer_req[i] = NULL; 3240 ctrlr->nr_aer_reqs--; 3241 3242 /* Move the last req to the aborting position for making aer_reqs 3243 * in continuous 3244 */ 3245 if (i < ctrlr->nr_aer_reqs) { 3246 ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs]; 3247 ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL; 3248 } 3249 3250 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3251 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3252 _nvmf_request_complete(req); 3253 return true; 3254 } 3255 } 3256 3257 return false; 3258 } 3259 3260 void 3261 nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair) 3262 { 3263 struct spdk_nvmf_request *req, *tmp; 3264 3265 TAILQ_FOREACH_SAFE(req, &qpair->outstanding, link, tmp) { 3266 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) { 3267 /* Zero-copy requests are kept on the outstanding queue from the moment 3268 * zcopy_start is sent until a zcopy_end callback is received. Therefore, 3269 * we can't remove them from the outstanding queue here, but need to rely on 3270 * the transport to do a zcopy_end to release their buffers and, in turn, 3271 * remove them from the queue. 3272 */ 3273 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3274 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3275 nvmf_transport_req_free(req); 3276 } 3277 } 3278 } 3279 3280 static void 3281 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req) 3282 { 3283 uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 3284 3285 if (nvmf_qpair_abort_aer(qpair, cid)) { 3286 SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n", 3287 qpair->ctrlr, qpair->qid, cid); 3288 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */ 3289 3290 spdk_nvmf_request_complete(req); 3291 return; 3292 } 3293 3294 nvmf_transport_qpair_abort_request(qpair, req); 3295 } 3296 3297 static void 3298 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status) 3299 { 3300 struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i); 3301 3302 if (status == 0) { 3303 /* There was no qpair whose ID matches SQID of the abort command. 3304 * Hence call _nvmf_request_complete() here. 3305 */ 3306 _nvmf_request_complete(req); 3307 } 3308 } 3309 3310 static void 3311 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i) 3312 { 3313 struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i); 3314 struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i); 3315 struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch); 3316 uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid; 3317 struct spdk_nvmf_qpair *qpair; 3318 3319 TAILQ_FOREACH(qpair, &group->qpairs, link) { 3320 if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) { 3321 /* Found the qpair */ 3322 3323 nvmf_qpair_abort_request(qpair, req); 3324 3325 /* Return -1 for the status so the iteration across threads stops. */ 3326 spdk_for_each_channel_continue(i, -1); 3327 return; 3328 } 3329 } 3330 3331 spdk_for_each_channel_continue(i, 0); 3332 } 3333 3334 static int 3335 nvmf_ctrlr_abort(struct spdk_nvmf_request *req) 3336 { 3337 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 3338 3339 rsp->cdw0 = 1U; /* Command not aborted */ 3340 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 3341 rsp->status.sc = SPDK_NVME_SC_SUCCESS; 3342 3343 /* Send a message to each poll group, searching for this ctrlr, sqid, and command. */ 3344 spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt, 3345 nvmf_ctrlr_abort_on_pg, 3346 req, 3347 nvmf_ctrlr_abort_done 3348 ); 3349 3350 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 3351 } 3352 3353 int 3354 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req) 3355 { 3356 struct spdk_nvmf_request *req_to_abort = req->req_to_abort; 3357 struct spdk_bdev *bdev; 3358 struct spdk_bdev_desc *desc; 3359 struct spdk_io_channel *ch; 3360 int rc; 3361 3362 assert(req_to_abort != NULL); 3363 3364 if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr && 3365 nvmf_qpair_is_admin_queue(req_to_abort->qpair)) { 3366 return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req); 3367 } 3368 3369 rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort, 3370 &bdev, &desc, &ch); 3371 if (rc != 0) { 3372 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3373 } 3374 3375 return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort); 3376 } 3377 3378 static int 3379 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0) 3380 { 3381 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 3382 3383 rsp->cdw0 = cdw0; 3384 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3385 } 3386 3387 /* we have to use the typedef in the function declaration to appease astyle. */ 3388 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t; 3389 3390 static spdk_nvme_path_status_code_t 3391 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state) 3392 { 3393 switch (ana_state) { 3394 case SPDK_NVME_ANA_INACCESSIBLE_STATE: 3395 return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE; 3396 case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE: 3397 return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS; 3398 case SPDK_NVME_ANA_CHANGE_STATE: 3399 return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION; 3400 default: 3401 return SPDK_NVME_SC_INTERNAL_PATH_ERROR; 3402 } 3403 } 3404 3405 static int 3406 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req) 3407 { 3408 uint8_t feature; 3409 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3410 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3411 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 3412 enum spdk_nvme_ana_state ana_state; 3413 3414 feature = cmd->cdw10_bits.get_features.fid; 3415 3416 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 3417 /* 3418 * Features supported by Discovery controller 3419 */ 3420 switch (feature) { 3421 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3422 return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw); 3423 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3424 return get_features_generic(req, ctrlr->feat.async_event_configuration.raw); 3425 default: 3426 SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature); 3427 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3428 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3429 } 3430 } 3431 /* 3432 * Process Get Features command for non-discovery controller 3433 */ 3434 ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid); 3435 switch (ana_state) { 3436 case SPDK_NVME_ANA_INACCESSIBLE_STATE: 3437 case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE: 3438 case SPDK_NVME_ANA_CHANGE_STATE: 3439 switch (feature) { 3440 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3441 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3442 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3443 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3444 response->status.sct = SPDK_NVME_SCT_PATH; 3445 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 3446 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3447 default: 3448 break; 3449 } 3450 break; 3451 default: 3452 break; 3453 } 3454 3455 switch (feature) { 3456 case SPDK_NVME_FEAT_ARBITRATION: 3457 return get_features_generic(req, ctrlr->feat.arbitration.raw); 3458 case SPDK_NVME_FEAT_POWER_MANAGEMENT: 3459 return get_features_generic(req, ctrlr->feat.power_management.raw); 3460 case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD: 3461 return nvmf_ctrlr_get_features_temperature_threshold(req); 3462 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3463 return get_features_generic(req, ctrlr->feat.error_recovery.raw); 3464 case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE: 3465 return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw); 3466 case SPDK_NVME_FEAT_NUMBER_OF_QUEUES: 3467 return get_features_generic(req, ctrlr->feat.number_of_queues.raw); 3468 case SPDK_NVME_FEAT_INTERRUPT_COALESCING: 3469 return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw); 3470 case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION: 3471 return nvmf_ctrlr_get_features_interrupt_vector_configuration(req); 3472 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3473 return get_features_generic(req, ctrlr->feat.write_atomicity.raw); 3474 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3475 return get_features_generic(req, ctrlr->feat.async_event_configuration.raw); 3476 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3477 return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw); 3478 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 3479 return nvmf_ctrlr_get_features_host_identifier(req); 3480 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3481 return nvmf_ctrlr_get_features_reservation_notification_mask(req); 3482 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3483 return nvmf_ctrlr_get_features_reservation_persistence(req); 3484 case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT: 3485 return nvmf_ctrlr_get_features_host_behavior_support(req); 3486 default: 3487 SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature); 3488 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3489 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3490 } 3491 } 3492 3493 static int 3494 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req) 3495 { 3496 uint8_t feature, save; 3497 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3498 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3499 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 3500 enum spdk_nvme_ana_state ana_state; 3501 /* 3502 * Features are not saveable by the controller as indicated by 3503 * ONCS field of the Identify Controller data. 3504 * */ 3505 save = cmd->cdw10_bits.set_features.sv; 3506 if (save) { 3507 response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE; 3508 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 3509 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3510 } 3511 3512 feature = cmd->cdw10_bits.set_features.fid; 3513 3514 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 3515 /* 3516 * Features supported by Discovery controller 3517 */ 3518 switch (feature) { 3519 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3520 return nvmf_ctrlr_set_features_keep_alive_timer(req); 3521 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3522 return nvmf_ctrlr_set_features_async_event_configuration(req); 3523 default: 3524 SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature); 3525 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3526 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3527 } 3528 } 3529 /* 3530 * Process Set Features command for non-discovery controller 3531 */ 3532 ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid); 3533 switch (ana_state) { 3534 case SPDK_NVME_ANA_INACCESSIBLE_STATE: 3535 case SPDK_NVME_ANA_CHANGE_STATE: 3536 if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) { 3537 response->status.sct = SPDK_NVME_SCT_PATH; 3538 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 3539 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3540 } else { 3541 switch (feature) { 3542 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3543 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3544 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3545 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3546 response->status.sct = SPDK_NVME_SCT_PATH; 3547 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 3548 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3549 default: 3550 break; 3551 } 3552 } 3553 break; 3554 case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE: 3555 response->status.sct = SPDK_NVME_SCT_PATH; 3556 response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS; 3557 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3558 default: 3559 break; 3560 } 3561 3562 switch (feature) { 3563 case SPDK_NVME_FEAT_ARBITRATION: 3564 return nvmf_ctrlr_set_features_arbitration(req); 3565 case SPDK_NVME_FEAT_POWER_MANAGEMENT: 3566 return nvmf_ctrlr_set_features_power_management(req); 3567 case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD: 3568 return nvmf_ctrlr_set_features_temperature_threshold(req); 3569 case SPDK_NVME_FEAT_ERROR_RECOVERY: 3570 return nvmf_ctrlr_set_features_error_recovery(req); 3571 case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE: 3572 return nvmf_ctrlr_set_features_volatile_write_cache(req); 3573 case SPDK_NVME_FEAT_NUMBER_OF_QUEUES: 3574 return nvmf_ctrlr_set_features_number_of_queues(req); 3575 case SPDK_NVME_FEAT_INTERRUPT_COALESCING: 3576 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 3577 response->status.sc = SPDK_NVME_SC_FEATURE_NOT_CHANGEABLE; 3578 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3579 case SPDK_NVME_FEAT_WRITE_ATOMICITY: 3580 return nvmf_ctrlr_set_features_write_atomicity(req); 3581 case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION: 3582 return nvmf_ctrlr_set_features_async_event_configuration(req); 3583 case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER: 3584 return nvmf_ctrlr_set_features_keep_alive_timer(req); 3585 case SPDK_NVME_FEAT_HOST_IDENTIFIER: 3586 return nvmf_ctrlr_set_features_host_identifier(req); 3587 case SPDK_NVME_FEAT_HOST_RESERVE_MASK: 3588 return nvmf_ctrlr_set_features_reservation_notification_mask(req); 3589 case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST: 3590 return nvmf_ctrlr_set_features_reservation_persistence(req); 3591 case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT: 3592 return nvmf_ctrlr_set_features_host_behavior_support(req); 3593 default: 3594 SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature); 3595 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3596 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3597 } 3598 } 3599 3600 static int 3601 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req) 3602 { 3603 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3604 3605 SPDK_DEBUGLOG(nvmf, "Keep Alive\n"); 3606 /* 3607 * To handle keep alive just clear or reset the 3608 * ctrlr based keep alive duration counter. 3609 * When added, a separate timer based process 3610 * will monitor if the time since last recorded 3611 * keep alive has exceeded the max duration and 3612 * take appropriate action. 3613 */ 3614 ctrlr->last_keep_alive_tick = spdk_get_ticks(); 3615 3616 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3617 } 3618 3619 int 3620 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req) 3621 { 3622 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 3623 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 3624 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 3625 struct spdk_nvmf_subsystem_poll_group *sgroup; 3626 int rc; 3627 3628 if (ctrlr == NULL) { 3629 SPDK_ERRLOG("Admin command sent before CONNECT\n"); 3630 response->status.sct = SPDK_NVME_SCT_GENERIC; 3631 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 3632 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3633 } 3634 3635 if (cmd->opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST) { 3636 /* We do not want to treat AERs as outstanding commands, 3637 * so decrement mgmt_io_outstanding here to offset 3638 * the increment that happened prior to this call. 3639 */ 3640 sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id]; 3641 assert(sgroup != NULL); 3642 sgroup->mgmt_io_outstanding--; 3643 } 3644 3645 assert(spdk_get_thread() == ctrlr->thread); 3646 3647 if (cmd->fuse != 0) { 3648 /* Fused admin commands are not supported. */ 3649 response->status.sct = SPDK_NVME_SCT_GENERIC; 3650 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 3651 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3652 } 3653 3654 if (ctrlr->vcprop.cc.bits.en != 1) { 3655 SPDK_ERRLOG("Admin command sent to disabled controller\n"); 3656 response->status.sct = SPDK_NVME_SCT_GENERIC; 3657 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 3658 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3659 } 3660 3661 if (req->iovcnt && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 3662 spdk_iov_memset(req->iov, req->iovcnt, 0); 3663 } 3664 3665 if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) { 3666 /* Discovery controllers only support these admin OPS. */ 3667 switch (cmd->opc) { 3668 case SPDK_NVME_OPC_IDENTIFY: 3669 case SPDK_NVME_OPC_GET_LOG_PAGE: 3670 case SPDK_NVME_OPC_KEEP_ALIVE: 3671 case SPDK_NVME_OPC_SET_FEATURES: 3672 case SPDK_NVME_OPC_GET_FEATURES: 3673 case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST: 3674 break; 3675 default: 3676 goto invalid_opcode; 3677 } 3678 } 3679 3680 /* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */ 3681 if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) { 3682 rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req); 3683 if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 3684 /* The handler took care of this command */ 3685 return rc; 3686 } 3687 } 3688 3689 switch (cmd->opc) { 3690 case SPDK_NVME_OPC_GET_LOG_PAGE: 3691 return nvmf_ctrlr_get_log_page(req); 3692 case SPDK_NVME_OPC_IDENTIFY: 3693 return nvmf_ctrlr_identify(req); 3694 case SPDK_NVME_OPC_ABORT: 3695 return nvmf_ctrlr_abort(req); 3696 case SPDK_NVME_OPC_GET_FEATURES: 3697 return nvmf_ctrlr_get_features(req); 3698 case SPDK_NVME_OPC_SET_FEATURES: 3699 return nvmf_ctrlr_set_features(req); 3700 case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST: 3701 return nvmf_ctrlr_async_event_request(req); 3702 case SPDK_NVME_OPC_KEEP_ALIVE: 3703 return nvmf_ctrlr_keep_alive(req); 3704 3705 case SPDK_NVME_OPC_CREATE_IO_SQ: 3706 case SPDK_NVME_OPC_CREATE_IO_CQ: 3707 case SPDK_NVME_OPC_DELETE_IO_SQ: 3708 case SPDK_NVME_OPC_DELETE_IO_CQ: 3709 /* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */ 3710 goto invalid_opcode; 3711 3712 default: 3713 goto invalid_opcode; 3714 } 3715 3716 invalid_opcode: 3717 SPDK_INFOLOG(nvmf, "Unsupported admin opcode 0x%x\n", cmd->opc); 3718 response->status.sct = SPDK_NVME_SCT_GENERIC; 3719 response->status.sc = SPDK_NVME_SC_INVALID_OPCODE; 3720 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3721 } 3722 3723 static int 3724 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req) 3725 { 3726 struct spdk_nvmf_qpair *qpair = req->qpair; 3727 struct spdk_nvmf_capsule_cmd *cap_hdr; 3728 3729 cap_hdr = &req->cmd->nvmf_cmd; 3730 3731 if (qpair->ctrlr == NULL) { 3732 /* No ctrlr established yet; the only valid command is Connect */ 3733 if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) { 3734 return nvmf_ctrlr_cmd_connect(req); 3735 } else { 3736 SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n", 3737 cap_hdr->fctype); 3738 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3739 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 3740 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3741 } 3742 } else if (nvmf_qpair_is_admin_queue(qpair)) { 3743 /* 3744 * Controller session is established, and this is an admin queue. 3745 * Disallow Connect and allow other fabrics commands. 3746 */ 3747 switch (cap_hdr->fctype) { 3748 case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET: 3749 return nvmf_property_set(req); 3750 case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET: 3751 return nvmf_property_get(req); 3752 default: 3753 SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n", 3754 cap_hdr->fctype); 3755 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3756 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 3757 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3758 } 3759 } else { 3760 /* Controller session is established, and this is an I/O queue */ 3761 /* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */ 3762 SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype); 3763 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3764 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 3765 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 3766 } 3767 } 3768 3769 static inline void 3770 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr, 3771 union spdk_nvme_async_event_completion *event) 3772 { 3773 struct spdk_nvmf_async_event_completion *nvmf_event; 3774 3775 nvmf_event = calloc(1, sizeof(*nvmf_event)); 3776 if (!nvmf_event) { 3777 SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n"); 3778 return; 3779 } 3780 nvmf_event->event.raw = event->raw; 3781 STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link); 3782 } 3783 3784 static inline int 3785 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr, 3786 union spdk_nvme_async_event_completion *event) 3787 { 3788 struct spdk_nvmf_request *req; 3789 struct spdk_nvme_cpl *rsp; 3790 3791 assert(spdk_get_thread() == ctrlr->thread); 3792 3793 /* If there is no outstanding AER request, queue the event. Then 3794 * if an AER is later submitted, this event can be sent as a 3795 * response. 3796 */ 3797 if (ctrlr->nr_aer_reqs == 0) { 3798 nvmf_ctrlr_queue_pending_async_event(ctrlr, event); 3799 return 0; 3800 } 3801 3802 req = ctrlr->aer_req[--ctrlr->nr_aer_reqs]; 3803 rsp = &req->rsp->nvme_cpl; 3804 3805 rsp->cdw0 = event->raw; 3806 3807 _nvmf_request_complete(req); 3808 ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL; 3809 3810 return 0; 3811 } 3812 3813 int 3814 nvmf_ctrlr_async_event_ns_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.ns_attr_notice) { 3820 return 0; 3821 } 3822 3823 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_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_NS_ATTR_CHANGED; 3829 event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST; 3830 3831 return nvmf_ctrlr_async_event_notification(ctrlr, &event); 3832 } 3833 3834 int 3835 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr) 3836 { 3837 union spdk_nvme_async_event_completion event = {0}; 3838 3839 /* Users may disable the event notification */ 3840 if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) { 3841 return 0; 3842 } 3843 3844 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) { 3845 return 0; 3846 } 3847 3848 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE; 3849 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE; 3850 event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS; 3851 3852 return nvmf_ctrlr_async_event_notification(ctrlr, &event); 3853 } 3854 3855 void 3856 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr) 3857 { 3858 union spdk_nvme_async_event_completion event = {0}; 3859 3860 if (!ctrlr->num_avail_log_pages) { 3861 return; 3862 } 3863 3864 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) { 3865 return; 3866 } 3867 3868 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO; 3869 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL; 3870 event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION; 3871 3872 nvmf_ctrlr_async_event_notification(ctrlr, &event); 3873 } 3874 3875 void 3876 nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx) 3877 { 3878 union spdk_nvme_async_event_completion event = {0}; 3879 struct spdk_nvmf_ctrlr *ctrlr = ctx; 3880 3881 /* Users may disable the event notification manually or 3882 * it may not be enabled due to keep alive timeout 3883 * not being set in connect command to discovery controller. 3884 */ 3885 if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) { 3886 return; 3887 } 3888 3889 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) { 3890 return; 3891 } 3892 3893 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE; 3894 event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE; 3895 event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY; 3896 3897 nvmf_ctrlr_async_event_notification(ctrlr, &event); 3898 } 3899 3900 int 3901 spdk_nvmf_ctrlr_async_event_error_event(struct spdk_nvmf_ctrlr *ctrlr, 3902 enum spdk_nvme_async_event_info_error info) 3903 { 3904 union spdk_nvme_async_event_completion event; 3905 3906 if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT)) { 3907 return 0; 3908 } 3909 3910 if (info > SPDK_NVME_ASYNC_EVENT_FW_IMAGE_LOAD) { 3911 return 0; 3912 } 3913 3914 event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_ERROR; 3915 event.bits.log_page_identifier = SPDK_NVME_LOG_ERROR; 3916 event.bits.async_event_info = info; 3917 3918 return nvmf_ctrlr_async_event_notification(ctrlr, &event); 3919 } 3920 3921 void 3922 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair) 3923 { 3924 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 3925 int i; 3926 3927 if (ctrlr == NULL || !nvmf_qpair_is_admin_queue(qpair)) { 3928 return; 3929 } 3930 3931 assert(spdk_get_thread() == ctrlr->thread); 3932 3933 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 3934 spdk_nvmf_request_free(ctrlr->aer_req[i]); 3935 ctrlr->aer_req[i] = NULL; 3936 } 3937 3938 ctrlr->nr_aer_reqs = 0; 3939 } 3940 3941 void 3942 spdk_nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr) 3943 { 3944 struct spdk_nvmf_request *req; 3945 int i; 3946 3947 assert(spdk_get_thread() == ctrlr->thread); 3948 3949 if (!ctrlr->nr_aer_reqs) { 3950 return; 3951 } 3952 3953 for (i = 0; i < ctrlr->nr_aer_reqs; i++) { 3954 req = ctrlr->aer_req[i]; 3955 3956 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3957 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 3958 _nvmf_request_complete(req); 3959 3960 ctrlr->aer_req[i] = NULL; 3961 } 3962 3963 ctrlr->nr_aer_reqs = 0; 3964 } 3965 3966 static void 3967 _nvmf_ctrlr_add_reservation_log(void *ctx) 3968 { 3969 struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx; 3970 struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr; 3971 3972 ctrlr->log_page_count++; 3973 3974 /* Maximum number of queued log pages is 255 */ 3975 if (ctrlr->num_avail_log_pages == 0xff) { 3976 struct spdk_nvmf_reservation_log *entry; 3977 entry = TAILQ_LAST(&ctrlr->log_head, log_page_head); 3978 entry->log.log_page_count = ctrlr->log_page_count; 3979 free(log); 3980 return; 3981 } 3982 3983 log->log.log_page_count = ctrlr->log_page_count; 3984 log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++; 3985 TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link); 3986 3987 nvmf_ctrlr_async_event_reservation_notification(ctrlr); 3988 } 3989 3990 void 3991 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr, 3992 struct spdk_nvmf_ns *ns, 3993 enum spdk_nvme_reservation_notification_log_page_type type) 3994 { 3995 struct spdk_nvmf_reservation_log *log; 3996 3997 switch (type) { 3998 case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY: 3999 return; 4000 case SPDK_NVME_REGISTRATION_PREEMPTED: 4001 if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) { 4002 return; 4003 } 4004 break; 4005 case SPDK_NVME_RESERVATION_RELEASED: 4006 if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) { 4007 return; 4008 } 4009 break; 4010 case SPDK_NVME_RESERVATION_PREEMPTED: 4011 if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) { 4012 return; 4013 } 4014 break; 4015 default: 4016 return; 4017 } 4018 4019 log = calloc(1, sizeof(*log)); 4020 if (!log) { 4021 SPDK_ERRLOG("Alloc log page failed, ignore the log\n"); 4022 return; 4023 } 4024 log->ctrlr = ctrlr; 4025 log->log.type = type; 4026 log->log.nsid = ns->nsid; 4027 4028 spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log); 4029 } 4030 4031 /* Check from subsystem poll group's namespace information data structure */ 4032 static bool 4033 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info, 4034 struct spdk_nvmf_ctrlr *ctrlr) 4035 { 4036 uint32_t i; 4037 4038 for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) { 4039 if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) { 4040 return true; 4041 } 4042 } 4043 4044 return false; 4045 } 4046 4047 /* 4048 * Check the NVMe command is permitted or not for current controller(Host). 4049 */ 4050 static int 4051 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info, 4052 struct spdk_nvmf_ctrlr *ctrlr, 4053 struct spdk_nvmf_request *req) 4054 { 4055 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 4056 enum spdk_nvme_reservation_type rtype = ns_info->rtype; 4057 uint8_t status = SPDK_NVME_SC_SUCCESS; 4058 uint8_t racqa; 4059 bool is_registrant; 4060 4061 /* No valid reservation */ 4062 if (!rtype) { 4063 return 0; 4064 } 4065 4066 is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr); 4067 /* All registrants type and current ctrlr is a valid registrant */ 4068 if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS || 4069 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) { 4070 return 0; 4071 } else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) { 4072 return 0; 4073 } 4074 4075 /* Non-holder for current controller */ 4076 switch (cmd->opc) { 4077 case SPDK_NVME_OPC_READ: 4078 case SPDK_NVME_OPC_COMPARE: 4079 if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 4080 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4081 goto exit; 4082 } 4083 if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY || 4084 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) { 4085 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4086 } 4087 break; 4088 case SPDK_NVME_OPC_FLUSH: 4089 case SPDK_NVME_OPC_WRITE: 4090 case SPDK_NVME_OPC_WRITE_UNCORRECTABLE: 4091 case SPDK_NVME_OPC_WRITE_ZEROES: 4092 case SPDK_NVME_OPC_DATASET_MANAGEMENT: 4093 if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE || 4094 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) { 4095 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4096 goto exit; 4097 } 4098 if (!is_registrant) { 4099 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4100 } 4101 break; 4102 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 4103 racqa = cmd->cdw10_bits.resv_acquire.racqa; 4104 if (racqa == SPDK_NVME_RESERVE_ACQUIRE) { 4105 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4106 goto exit; 4107 } 4108 if (!is_registrant) { 4109 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4110 } 4111 break; 4112 case SPDK_NVME_OPC_RESERVATION_RELEASE: 4113 if (!is_registrant) { 4114 status = SPDK_NVME_SC_RESERVATION_CONFLICT; 4115 } 4116 break; 4117 default: 4118 break; 4119 } 4120 4121 exit: 4122 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4123 req->rsp->nvme_cpl.status.sc = status; 4124 if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) { 4125 return -EPERM; 4126 } 4127 4128 return 0; 4129 } 4130 4131 static int 4132 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 4133 struct spdk_bdev_desc *desc, struct spdk_io_channel *ch) 4134 { 4135 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 4136 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 4137 struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req; 4138 int rc; 4139 4140 if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) { 4141 /* first fused operation (should be compare) */ 4142 if (first_fused_req != NULL) { 4143 struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl; 4144 4145 SPDK_ERRLOG("Wrong sequence of fused operations\n"); 4146 4147 /* abort req->qpair->first_fused_request and continue with new fused command */ 4148 fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4149 fused_response->status.sct = SPDK_NVME_SCT_GENERIC; 4150 _nvmf_request_complete(first_fused_req); 4151 } else if (cmd->opc != SPDK_NVME_OPC_COMPARE) { 4152 SPDK_ERRLOG("Wrong op code of fused operations\n"); 4153 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4154 rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE; 4155 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4156 } 4157 4158 req->qpair->first_fused_req = req; 4159 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 4160 } else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) { 4161 /* second fused operation (should be write) */ 4162 if (first_fused_req == NULL) { 4163 SPDK_ERRLOG("Wrong sequence of fused operations\n"); 4164 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4165 rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4166 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4167 } else if (cmd->opc != SPDK_NVME_OPC_WRITE) { 4168 struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl; 4169 4170 SPDK_ERRLOG("Wrong op code of fused operations\n"); 4171 4172 /* abort req->qpair->first_fused_request and fail current command */ 4173 fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4174 fused_response->status.sct = SPDK_NVME_SCT_GENERIC; 4175 _nvmf_request_complete(first_fused_req); 4176 4177 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4178 rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE; 4179 req->qpair->first_fused_req = NULL; 4180 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4181 } 4182 4183 /* save request of first command to generate response later */ 4184 req->first_fused_req = first_fused_req; 4185 req->first_fused = true; 4186 req->qpair->first_fused_req = NULL; 4187 } else { 4188 SPDK_ERRLOG("Invalid fused command fuse field.\n"); 4189 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4190 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 4191 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4192 } 4193 4194 rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req); 4195 4196 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 4197 if (spdk_nvme_cpl_is_error(rsp)) { 4198 struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl; 4199 4200 fused_response->status = rsp->status; 4201 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 4202 rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED; 4203 /* Complete first of fused commands. Second will be completed by upper layer */ 4204 _nvmf_request_complete(first_fused_req); 4205 req->first_fused_req = NULL; 4206 req->first_fused = false; 4207 } 4208 } 4209 4210 return rc; 4211 } 4212 4213 bool 4214 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req) 4215 { 4216 struct spdk_nvmf_transport *transport = req->qpair->transport; 4217 struct spdk_nvmf_ns *ns; 4218 4219 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE); 4220 4221 if (!transport->opts.zcopy) { 4222 return false; 4223 } 4224 4225 if (nvmf_qpair_is_admin_queue(req->qpair)) { 4226 /* Admin queue */ 4227 return false; 4228 } 4229 4230 if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) && 4231 (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) { 4232 /* Not a READ or WRITE command */ 4233 return false; 4234 } 4235 4236 if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) { 4237 /* Fused commands dont use zcopy buffers */ 4238 return false; 4239 } 4240 4241 ns = nvmf_ctrlr_get_ns(req->qpair->ctrlr, req->cmd->nvme_cmd.nsid); 4242 if (ns == NULL || ns->bdev == NULL || !ns->zcopy) { 4243 return false; 4244 } 4245 4246 req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT; 4247 return true; 4248 } 4249 4250 void 4251 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req) 4252 { 4253 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT); 4254 4255 /* Set iovcnt to be the maximum number of iovs that the ZCOPY can use */ 4256 req->iovcnt = NVMF_REQ_MAX_BUFFERS; 4257 4258 spdk_nvmf_request_exec(req); 4259 } 4260 4261 void 4262 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req, bool commit) 4263 { 4264 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE); 4265 req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING; 4266 4267 nvmf_bdev_ctrlr_zcopy_end(req, commit); 4268 } 4269 4270 int 4271 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req) 4272 { 4273 uint32_t nsid; 4274 struct spdk_nvmf_ns *ns; 4275 struct spdk_bdev *bdev; 4276 struct spdk_bdev_desc *desc; 4277 struct spdk_io_channel *ch; 4278 struct spdk_nvmf_poll_group *group = req->qpair->group; 4279 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 4280 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 4281 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 4282 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4283 enum spdk_nvme_ana_state ana_state; 4284 4285 /* pre-set response details for this command */ 4286 response->status.sc = SPDK_NVME_SC_SUCCESS; 4287 nsid = cmd->nsid; 4288 4289 if (spdk_unlikely(ctrlr == NULL)) { 4290 SPDK_ERRLOG("I/O command sent before CONNECT\n"); 4291 response->status.sct = SPDK_NVME_SCT_GENERIC; 4292 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 4293 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4294 } 4295 4296 if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) { 4297 SPDK_ERRLOG("I/O command sent to disabled controller\n"); 4298 response->status.sct = SPDK_NVME_SCT_GENERIC; 4299 response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 4300 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4301 } 4302 4303 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 4304 if (spdk_unlikely(ns == NULL || ns->bdev == NULL)) { 4305 SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid); 4306 response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4307 response->status.dnr = 1; 4308 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4309 } 4310 4311 ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid); 4312 if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE && 4313 ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) { 4314 SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n", 4315 ana_state); 4316 response->status.sct = SPDK_NVME_SCT_PATH; 4317 response->status.sc = _nvme_ana_state_to_path_status(ana_state); 4318 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4319 } 4320 4321 if (spdk_likely(ctrlr->listener != NULL)) { 4322 SPDK_DTRACE_PROBE3_TICKS(nvmf_request_io_exec_path, req, 4323 ctrlr->listener->trid->traddr, 4324 ctrlr->listener->trid->trsvcid); 4325 } 4326 4327 /* scan-build falsely reporting dereference of null pointer */ 4328 assert(group != NULL && group->sgroups != NULL); 4329 ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1]; 4330 if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) { 4331 SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n", 4332 cmd->nsid, cmd->opc); 4333 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4334 } 4335 4336 bdev = ns->bdev; 4337 desc = ns->desc; 4338 ch = ns_info->channel; 4339 4340 if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) { 4341 return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch); 4342 } else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) { 4343 struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl; 4344 4345 SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n"); 4346 4347 /* abort req->qpair->first_fused_request and continue with new command */ 4348 fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED; 4349 fused_response->status.sct = SPDK_NVME_SCT_GENERIC; 4350 _nvmf_request_complete(req->qpair->first_fused_req); 4351 req->qpair->first_fused_req = NULL; 4352 } 4353 4354 if (spdk_nvmf_request_using_zcopy(req)) { 4355 assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT); 4356 return nvmf_bdev_ctrlr_zcopy_start(bdev, desc, ch, req); 4357 } else { 4358 switch (cmd->opc) { 4359 case SPDK_NVME_OPC_READ: 4360 return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req); 4361 case SPDK_NVME_OPC_WRITE: 4362 return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req); 4363 case SPDK_NVME_OPC_COMPARE: 4364 return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req); 4365 case SPDK_NVME_OPC_WRITE_ZEROES: 4366 return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req); 4367 case SPDK_NVME_OPC_FLUSH: 4368 return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req); 4369 case SPDK_NVME_OPC_DATASET_MANAGEMENT: 4370 return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req); 4371 case SPDK_NVME_OPC_RESERVATION_REGISTER: 4372 case SPDK_NVME_OPC_RESERVATION_ACQUIRE: 4373 case SPDK_NVME_OPC_RESERVATION_RELEASE: 4374 case SPDK_NVME_OPC_RESERVATION_REPORT: 4375 spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req); 4376 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 4377 case SPDK_NVME_OPC_COPY: 4378 return nvmf_bdev_ctrlr_copy_cmd(bdev, desc, ch, req); 4379 default: 4380 return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req); 4381 } 4382 } 4383 } 4384 4385 static void 4386 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair) 4387 { 4388 if (spdk_unlikely(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING)) { 4389 assert(qpair->state_cb != NULL); 4390 4391 if (TAILQ_EMPTY(&qpair->outstanding)) { 4392 qpair->state_cb(qpair->state_cb_arg, 0); 4393 } 4394 } 4395 } 4396 4397 int 4398 spdk_nvmf_request_free(struct spdk_nvmf_request *req) 4399 { 4400 struct spdk_nvmf_qpair *qpair = req->qpair; 4401 4402 TAILQ_REMOVE(&qpair->outstanding, req, link); 4403 if (spdk_unlikely(nvmf_transport_req_free(req))) { 4404 SPDK_ERRLOG("Unable to free transport level request resources.\n"); 4405 } 4406 4407 nvmf_qpair_request_cleanup(qpair); 4408 4409 return 0; 4410 } 4411 4412 static void 4413 _nvmf_request_complete(void *ctx) 4414 { 4415 struct spdk_nvmf_request *req = ctx; 4416 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 4417 struct spdk_nvmf_qpair *qpair; 4418 struct spdk_nvmf_subsystem_poll_group *sgroup = NULL; 4419 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4420 bool is_aer = false; 4421 uint32_t nsid; 4422 bool paused; 4423 uint8_t opcode; 4424 4425 rsp->sqid = 0; 4426 rsp->status.p = 0; 4427 rsp->cid = req->cmd->nvme_cmd.cid; 4428 nsid = req->cmd->nvme_cmd.nsid; 4429 opcode = req->cmd->nvmf_cmd.opcode; 4430 4431 qpair = req->qpair; 4432 if (spdk_likely(qpair->ctrlr)) { 4433 sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id]; 4434 assert(sgroup != NULL); 4435 is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST; 4436 if (spdk_likely(qpair->qid != 0)) { 4437 qpair->group->stat.completed_nvme_io++; 4438 } 4439 4440 /* 4441 * Set the crd value. 4442 * If the the IO has any error, and dnr (DoNotRetry) is not 1, 4443 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT. 4444 */ 4445 if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp) && 4446 rsp->status.dnr == 0 && 4447 qpair->ctrlr->acre_enabled)) { 4448 rsp->status.crd = 1; 4449 } 4450 } else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) { 4451 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 4452 } 4453 4454 if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) { 4455 spdk_nvme_print_completion(qpair->qid, rsp); 4456 } 4457 4458 switch (req->zcopy_phase) { 4459 case NVMF_ZCOPY_PHASE_NONE: 4460 TAILQ_REMOVE(&qpair->outstanding, req, link); 4461 break; 4462 case NVMF_ZCOPY_PHASE_INIT: 4463 if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) { 4464 req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED; 4465 TAILQ_REMOVE(&qpair->outstanding, req, link); 4466 } else { 4467 req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE; 4468 } 4469 break; 4470 case NVMF_ZCOPY_PHASE_EXECUTE: 4471 break; 4472 case NVMF_ZCOPY_PHASE_END_PENDING: 4473 TAILQ_REMOVE(&qpair->outstanding, req, link); 4474 req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE; 4475 break; 4476 default: 4477 SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase); 4478 break; 4479 } 4480 4481 if (spdk_unlikely(nvmf_transport_req_complete(req))) { 4482 SPDK_ERRLOG("Transport request completion error!\n"); 4483 } 4484 4485 /* AER cmd is an exception */ 4486 if (spdk_likely(sgroup && !is_aer)) { 4487 if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC || 4488 nvmf_qpair_is_admin_queue(qpair))) { 4489 assert(sgroup->mgmt_io_outstanding > 0); 4490 sgroup->mgmt_io_outstanding--; 4491 } else { 4492 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE || 4493 req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE || 4494 req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT_FAILED) { 4495 /* End of request */ 4496 4497 /* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */ 4498 if (spdk_likely(nsid - 1 < sgroup->num_ns)) { 4499 sgroup->ns_info[nsid - 1].io_outstanding--; 4500 } 4501 } 4502 } 4503 4504 if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING && 4505 sgroup->mgmt_io_outstanding == 0)) { 4506 paused = true; 4507 for (nsid = 0; nsid < sgroup->num_ns; nsid++) { 4508 ns_info = &sgroup->ns_info[nsid]; 4509 4510 if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING && 4511 ns_info->io_outstanding > 0) { 4512 paused = false; 4513 break; 4514 } 4515 } 4516 4517 if (paused) { 4518 sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED; 4519 sgroup->cb_fn(sgroup->cb_arg, 0); 4520 sgroup->cb_fn = NULL; 4521 sgroup->cb_arg = NULL; 4522 } 4523 } 4524 4525 } 4526 4527 nvmf_qpair_request_cleanup(qpair); 4528 } 4529 4530 int 4531 spdk_nvmf_request_complete(struct spdk_nvmf_request *req) 4532 { 4533 struct spdk_nvmf_qpair *qpair = req->qpair; 4534 4535 spdk_thread_exec_msg(qpair->group->thread, _nvmf_request_complete, req); 4536 4537 return 0; 4538 } 4539 4540 SPDK_LOG_DEPRECATION_REGISTER(nvmf_request_exec_fabrics, "spdk_nvmf_request_exec_fabrics()", 4541 "v24.09", 1); 4542 void 4543 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req) 4544 { 4545 SPDK_LOG_DEPRECATED(nvmf_request_exec_fabrics); 4546 4547 return spdk_nvmf_request_exec(req); 4548 } 4549 4550 static bool 4551 nvmf_check_subsystem_active(struct spdk_nvmf_request *req) 4552 { 4553 struct spdk_nvmf_qpair *qpair = req->qpair; 4554 struct spdk_nvmf_subsystem_poll_group *sgroup = NULL; 4555 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4556 uint32_t nsid; 4557 4558 if (spdk_likely(qpair->ctrlr)) { 4559 sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id]; 4560 assert(sgroup != NULL); 4561 } else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) { 4562 sgroup = nvmf_subsystem_pg_from_connect_cmd(req); 4563 } 4564 4565 /* Check if the subsystem is paused (if there is a subsystem) */ 4566 if (spdk_likely(sgroup != NULL)) { 4567 if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC || 4568 nvmf_qpair_is_admin_queue(qpair))) { 4569 if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) { 4570 /* The subsystem is not currently active. Queue this request. */ 4571 TAILQ_INSERT_TAIL(&sgroup->queued, req, link); 4572 return false; 4573 } 4574 sgroup->mgmt_io_outstanding++; 4575 } else { 4576 nsid = req->cmd->nvme_cmd.nsid; 4577 4578 /* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */ 4579 if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) { 4580 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4581 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4582 req->rsp->nvme_cpl.status.dnr = 1; 4583 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4584 _nvmf_request_complete(req); 4585 return false; 4586 } 4587 4588 ns_info = &sgroup->ns_info[nsid - 1]; 4589 if (spdk_unlikely(ns_info->channel == NULL)) { 4590 /* This can can happen if host sends I/O to a namespace that is 4591 * in the process of being added, but before the full addition 4592 * process is complete. Report invalid namespace in that case. 4593 */ 4594 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4595 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4596 req->rsp->nvme_cpl.status.dnr = 1; 4597 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4598 ns_info->io_outstanding++; 4599 _nvmf_request_complete(req); 4600 return false; 4601 } 4602 4603 if (spdk_unlikely(ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE)) { 4604 /* The namespace is not currently active. Queue this request. */ 4605 TAILQ_INSERT_TAIL(&sgroup->queued, req, link); 4606 return false; 4607 } 4608 4609 ns_info->io_outstanding++; 4610 } 4611 4612 if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) { 4613 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 4614 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR; 4615 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4616 _nvmf_request_complete(req); 4617 return false; 4618 } 4619 } 4620 4621 return true; 4622 } 4623 4624 void 4625 spdk_nvmf_request_exec(struct spdk_nvmf_request *req) 4626 { 4627 struct spdk_nvmf_qpair *qpair = req->qpair; 4628 enum spdk_nvmf_request_exec_status status; 4629 4630 if (spdk_unlikely(!nvmf_check_subsystem_active(req))) { 4631 return; 4632 } 4633 4634 if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) { 4635 spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd); 4636 } 4637 4638 /* Place the request on the outstanding list so we can keep track of it */ 4639 TAILQ_INSERT_TAIL(&qpair->outstanding, req, link); 4640 4641 if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) { 4642 status = nvmf_ctrlr_process_fabrics_cmd(req); 4643 } else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) { 4644 status = nvmf_ctrlr_process_admin_cmd(req); 4645 } else { 4646 status = nvmf_ctrlr_process_io_cmd(req); 4647 } 4648 4649 if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 4650 _nvmf_request_complete(req); 4651 } 4652 } 4653 4654 static bool 4655 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd, 4656 struct spdk_dif_ctx *dif_ctx) 4657 { 4658 struct spdk_nvmf_ns *ns; 4659 struct spdk_bdev *bdev; 4660 4661 if (ctrlr == NULL || cmd == NULL) { 4662 return false; 4663 } 4664 4665 ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid); 4666 if (ns == NULL || ns->bdev == NULL) { 4667 return false; 4668 } 4669 4670 bdev = ns->bdev; 4671 4672 switch (cmd->opc) { 4673 case SPDK_NVME_OPC_READ: 4674 case SPDK_NVME_OPC_WRITE: 4675 case SPDK_NVME_OPC_COMPARE: 4676 return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx); 4677 default: 4678 break; 4679 } 4680 4681 return false; 4682 } 4683 4684 bool 4685 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx) 4686 { 4687 struct spdk_nvmf_qpair *qpair = req->qpair; 4688 struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr; 4689 4690 if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) { 4691 return false; 4692 } 4693 4694 if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) { 4695 return false; 4696 } 4697 4698 if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) { 4699 return false; 4700 } 4701 4702 if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) { 4703 return false; 4704 } 4705 4706 return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx); 4707 } 4708 4709 void 4710 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr) 4711 { 4712 g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr; 4713 } 4714 4715 static int 4716 nvmf_passthru_admin_cmd_for_bdev_nsid(struct spdk_nvmf_request *req, uint32_t bdev_nsid) 4717 { 4718 struct spdk_bdev *bdev; 4719 struct spdk_bdev_desc *desc; 4720 struct spdk_io_channel *ch; 4721 struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req); 4722 int rc; 4723 4724 rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch); 4725 if (rc) { 4726 response->status.sct = SPDK_NVME_SCT_GENERIC; 4727 response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4728 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4729 } 4730 return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL); 4731 } 4732 4733 static int 4734 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req) 4735 { 4736 struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req); 4737 uint32_t bdev_nsid; 4738 4739 if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid != 0) { 4740 bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid; 4741 } else { 4742 bdev_nsid = cmd->nsid; 4743 } 4744 4745 return nvmf_passthru_admin_cmd_for_bdev_nsid(req, bdev_nsid); 4746 } 4747 4748 int 4749 nvmf_passthru_admin_cmd_for_ctrlr(struct spdk_nvmf_request *req, struct spdk_nvmf_ctrlr *ctrlr) 4750 { 4751 struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req); 4752 struct spdk_nvmf_ns *ns; 4753 4754 ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); 4755 if (ns == NULL) { 4756 /* Is there a better sc to use here? */ 4757 response->status.sct = SPDK_NVME_SCT_GENERIC; 4758 response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT; 4759 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 4760 } 4761 4762 return nvmf_passthru_admin_cmd_for_bdev_nsid(req, ns->nsid); 4763 } 4764 4765 void 4766 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid) 4767 { 4768 g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd; 4769 g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid; 4770 } 4771 4772 int 4773 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req, 4774 struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch) 4775 { 4776 struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr; 4777 struct spdk_nvmf_ns *ns; 4778 struct spdk_nvmf_poll_group *group = req->qpair->group; 4779 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 4780 4781 *bdev = NULL; 4782 *desc = NULL; 4783 *ch = NULL; 4784 4785 ns = nvmf_ctrlr_get_ns(ctrlr, nsid); 4786 if (ns == NULL || ns->bdev == NULL) { 4787 return -EINVAL; 4788 } 4789 4790 assert(group != NULL && group->sgroups != NULL); 4791 ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1]; 4792 *bdev = ns->bdev; 4793 *desc = ns->desc; 4794 *ch = ns_info->channel; 4795 4796 return 0; 4797 } 4798 4799 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req) 4800 { 4801 return req->qpair->ctrlr; 4802 } 4803 4804 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req) 4805 { 4806 return &req->cmd->nvme_cmd; 4807 } 4808 4809 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req) 4810 { 4811 return &req->rsp->nvme_cpl; 4812 } 4813 4814 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req) 4815 { 4816 return req->qpair->ctrlr->subsys; 4817 } 4818 4819 size_t 4820 spdk_nvmf_request_copy_from_buf(struct spdk_nvmf_request *req, 4821 void *buf, size_t buflen) 4822 { 4823 struct spdk_iov_xfer ix; 4824 4825 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 4826 return spdk_iov_xfer_from_buf(&ix, buf, buflen); 4827 } 4828 4829 size_t 4830 spdk_nvmf_request_copy_to_buf(struct spdk_nvmf_request *req, 4831 void *buf, size_t buflen) 4832 { 4833 struct spdk_iov_xfer ix; 4834 4835 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 4836 return spdk_iov_xfer_to_buf(&ix, buf, buflen); 4837 } 4838 4839 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr) 4840 { 4841 return ctrlr->subsys; 4842 } 4843 4844 uint16_t 4845 spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr) 4846 { 4847 return ctrlr->cntlid; 4848 } 4849 4850 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req) 4851 { 4852 return req->req_to_abort; 4853 } 4854