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