xref: /spdk/lib/nvmf/ctrlr.c (revision 91fcde065a5883d85ca1034a9a1b254e1eadbcad)
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);
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 void
220 nvmf_qpair_set_ctrlr(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_ctrlr *ctrlr)
221 {
222 	if (qpair->ctrlr != NULL) {
223 		/* Admin queues will call this function twice. */
224 		assert(qpair->ctrlr == ctrlr);
225 		return;
226 	}
227 
228 	qpair->ctrlr = ctrlr;
229 	spdk_trace_owner_append_description(qpair->trace_id,
230 					    spdk_nvmf_subsystem_get_nqn(ctrlr->subsys));
231 }
232 
233 static int _retry_qid_check(void *ctx);
234 
235 static void
236 nvmf_ctrlr_send_connect_rsp(void *ctx)
237 {
238 	struct spdk_nvmf_request *req = ctx;
239 	struct spdk_nvmf_qpair *qpair = req->qpair;
240 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
241 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
242 	int rc;
243 
244 	/* The qpair might have been disconnected in the meantime */
245 	assert(qpair->state == SPDK_NVMF_QPAIR_CONNECTING ||
246 	       qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
247 	if (qpair->state == SPDK_NVMF_QPAIR_CONNECTING) {
248 		if (nvmf_subsystem_host_auth_required(ctrlr->subsys, ctrlr->hostnqn)) {
249 			rc = nvmf_qpair_auth_init(qpair);
250 			if (rc != 0) {
251 				rsp->status.sct = SPDK_NVME_SCT_GENERIC;
252 				rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
253 				spdk_nvmf_request_complete(req);
254 				spdk_nvmf_qpair_disconnect(qpair);
255 				return;
256 			}
257 			rsp->status_code_specific.success.authreq.atr = 1;
258 			nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_AUTHENTICATING);
259 		} else {
260 			nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ENABLED);
261 		}
262 	}
263 
264 	SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n", ctrlr->cntlid);
265 
266 	assert(spdk_get_thread() == qpair->group->thread);
267 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
268 	rsp->status_code_specific.success.cntlid = ctrlr->cntlid;
269 	spdk_nvmf_request_complete(req);
270 }
271 
272 static void
273 nvmf_ctrlr_add_qpair(struct spdk_nvmf_qpair *qpair,
274 		     struct spdk_nvmf_ctrlr *ctrlr,
275 		     struct spdk_nvmf_request *req)
276 {
277 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
278 
279 	if (!ctrlr->admin_qpair) {
280 		SPDK_ERRLOG("Inactive admin qpair\n");
281 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
282 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
283 		qpair->connect_req = NULL;
284 		qpair->ctrlr = NULL;
285 		spdk_nvmf_request_complete(req);
286 		return;
287 	}
288 
289 	assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
290 
291 	if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) {
292 		if (qpair->connect_req != NULL) {
293 			SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid);
294 			rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
295 			rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
296 			qpair->connect_req = NULL;
297 			qpair->ctrlr = NULL;
298 			spdk_nvmf_request_complete(req);
299 		} else {
300 			SPDK_WARNLOG("Duplicate QID (%d) detected, re-check in %dus\n",
301 				     qpair->qid, DUPLICATE_QID_RETRY_US);
302 			qpair->connect_req = req;
303 			/* Set qpair->ctrlr here so that we'll have it when the poller expires. */
304 			nvmf_qpair_set_ctrlr(qpair, ctrlr);
305 			req->poller = SPDK_POLLER_REGISTER(_retry_qid_check, qpair,
306 							   DUPLICATE_QID_RETRY_US);
307 		}
308 		return;
309 	}
310 
311 	SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn,
312 				 ctrlr->hostnqn);
313 	nvmf_qpair_set_ctrlr(qpair, ctrlr);
314 	spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid);
315 
316 	spdk_thread_send_msg(qpair->group->thread, nvmf_ctrlr_send_connect_rsp, req);
317 }
318 
319 static int
320 _retry_qid_check(void *ctx)
321 {
322 	struct spdk_nvmf_qpair *qpair = ctx;
323 	struct spdk_nvmf_request *req = qpair->connect_req;
324 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
325 
326 	spdk_poller_unregister(&req->poller);
327 	SPDK_WARNLOG("Retrying adding qpair, qid:%d\n", qpair->qid);
328 	nvmf_ctrlr_add_qpair(qpair, ctrlr, req);
329 	return SPDK_POLLER_BUSY;
330 }
331 
332 static void
333 _nvmf_ctrlr_add_admin_qpair(void *ctx)
334 {
335 	struct spdk_nvmf_request *req = ctx;
336 	struct spdk_nvmf_qpair *qpair = req->qpair;
337 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
338 
339 	ctrlr->admin_qpair = qpair;
340 	ctrlr->association_timeout = qpair->transport->opts.association_timeout;
341 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
342 	nvmf_ctrlr_add_qpair(qpair, ctrlr, req);
343 }
344 
345 static void
346 _nvmf_subsystem_add_ctrlr(void *ctx)
347 {
348 	struct spdk_nvmf_request *req = ctx;
349 	struct spdk_nvmf_qpair *qpair = req->qpair;
350 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
351 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
352 
353 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
354 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
355 		spdk_bit_array_free(&ctrlr->qpair_mask);
356 		free(ctrlr);
357 		qpair->ctrlr = NULL;
358 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
359 		spdk_nvmf_request_complete(req);
360 		return;
361 	}
362 
363 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
364 }
365 
366 static void
367 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
368 		      struct spdk_nvmf_ctrlr_data *cdata)
369 {
370 	cdata->aerl = SPDK_NVMF_MAX_ASYNC_EVENTS - 1;
371 	cdata->kas = KAS_DEFAULT_VALUE;
372 	cdata->vid = SPDK_PCI_VID_INTEL;
373 	cdata->ssvid = SPDK_PCI_VID_INTEL;
374 	/* INTEL OUI */
375 	cdata->ieee[0] = 0xe4;
376 	cdata->ieee[1] = 0xd2;
377 	cdata->ieee[2] = 0x5c;
378 	cdata->oncs.compare = 1;
379 	cdata->oncs.reservations = 1;
380 	cdata->fuses.compare_and_write = 1;
381 	cdata->oncs.copy = 1;
382 	cdata->sgls.supported = 1;
383 	cdata->sgls.keyed_sgl = 1;
384 	cdata->sgls.sgl_offset = 1;
385 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
386 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
387 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
388 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
389 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
390 	cdata->nvmf_specific.msdbd = 1;
391 
392 	if (transport->ops->cdata_init) {
393 		transport->ops->cdata_init(transport, subsystem, cdata);
394 	}
395 }
396 
397 static bool
398 nvmf_subsys_has_multi_iocs(struct spdk_nvmf_subsystem *subsystem)
399 {
400 	struct spdk_nvmf_ns *ns;
401 	uint32_t i;
402 
403 	for (i = 0; i < subsystem->max_nsid; i++) {
404 		ns = subsystem->ns[i];
405 		if (ns && ns->bdev && spdk_bdev_is_zoned(ns->bdev)) {
406 			return true;
407 		}
408 	}
409 	return false;
410 }
411 
412 static void
413 nvmf_ctrlr_init_visible_ns(struct spdk_nvmf_ctrlr *ctrlr)
414 {
415 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
416 	struct spdk_nvmf_ns *ns;
417 
418 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
419 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
420 		if (ns->always_visible || nvmf_ns_find_host(ns, ctrlr->hostnqn) != NULL) {
421 			spdk_bit_array_set(ctrlr->visible_ns, ns->nsid - 1);
422 		}
423 	}
424 }
425 
426 static struct spdk_nvmf_ctrlr *
427 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
428 		  struct spdk_nvmf_request *req,
429 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
430 		  struct spdk_nvmf_fabric_connect_data *connect_data)
431 {
432 	struct spdk_nvmf_ctrlr *ctrlr;
433 	struct spdk_nvmf_transport *transport = req->qpair->transport;
434 	struct spdk_nvme_transport_id listen_trid = {};
435 	bool subsys_has_multi_iocs = false;
436 
437 	ctrlr = calloc(1, sizeof(*ctrlr));
438 	if (ctrlr == NULL) {
439 		SPDK_ERRLOG("Memory allocation failed\n");
440 		return NULL;
441 	}
442 
443 	if (spdk_nvme_trtype_is_fabrics(transport->ops->type)) {
444 		ctrlr->dynamic_ctrlr = true;
445 	} else {
446 		ctrlr->cntlid = connect_data->cntlid;
447 	}
448 
449 	SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_create, ctrlr, subsystem->subnqn,
450 				 spdk_thread_get_id(req->qpair->group->thread));
451 
452 	STAILQ_INIT(&ctrlr->async_events);
453 	TAILQ_INIT(&ctrlr->log_head);
454 	ctrlr->subsys = subsystem;
455 	ctrlr->thread = req->qpair->group->thread;
456 	ctrlr->disconnect_in_progress = false;
457 
458 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
459 	if (!ctrlr->qpair_mask) {
460 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
461 		goto err_qpair_mask;
462 	}
463 
464 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
465 
466 	/*
467 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
468 	 * If this field is cleared to 0h, then Keep Alive is not supported.
469 	 */
470 	if (ctrlr->cdata.kas) {
471 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
472 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
473 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
474 	}
475 
476 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
477 	if (ctrlr->subsys->flags.ana_reporting) {
478 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
479 	}
480 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
481 	/* Coalescing Disable */
482 	ctrlr->feat.interrupt_vector_configuration.bits.cd = 1;
483 
484 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
485 		/*
486 		 * If keep-alive timeout is not set, discovery controllers use some
487 		 * arbitrary high value in order to cleanup stale discovery sessions
488 		 *
489 		 * From the 1.0a nvme-of spec:
490 		 * "The Keep Alive command is reserved for
491 		 * Discovery controllers. A transport may specify a
492 		 * fixed Discovery controller activity timeout value
493 		 * (e.g., 2 minutes). If no commands are received
494 		 * by a Discovery controller within that time
495 		 * period, the controller may perform the
496 		 * actions for Keep Alive Timer expiration".
497 		 *
498 		 * From the 1.1 nvme-of spec:
499 		 * "A host requests an explicit persistent connection
500 		 * to a Discovery controller and Asynchronous Event Notifications from
501 		 * the Discovery controller on that persistent connection by specifying
502 		 * a non-zero Keep Alive Timer value in the Connect command."
503 		 *
504 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
505 		 * otherwise we disable it and use default discovery controller KATO.
506 		 * KATO is in millisecond.
507 		 */
508 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
509 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
510 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
511 		} else {
512 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
513 		}
514 	}
515 
516 	/* Subtract 1 for admin queue, 1 for 0's based */
517 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
518 			1;
519 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
520 			1;
521 
522 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
523 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, SPDK_NVMF_NQN_MAX_LEN);
524 
525 	ctrlr->visible_ns = spdk_bit_array_create(subsystem->max_nsid);
526 	if (!ctrlr->visible_ns) {
527 		SPDK_ERRLOG("Failed to allocate visible namespace array\n");
528 		goto err_visible_ns;
529 	}
530 	nvmf_ctrlr_init_visible_ns(ctrlr);
531 
532 	ctrlr->vcprop.cap.raw = 0;
533 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
534 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
535 				      1; /* max queue depth */
536 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
537 	/* ready timeout - 500 msec units */
538 	ctrlr->vcprop.cap.bits.to = NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS / 500;
539 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
540 	subsys_has_multi_iocs = nvmf_subsys_has_multi_iocs(subsystem);
541 	if (subsys_has_multi_iocs) {
542 		ctrlr->vcprop.cap.bits.css =
543 			SPDK_NVME_CAP_CSS_IOCS; /* One or more I/O command sets supported */
544 	} else {
545 		ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
546 	}
547 
548 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
549 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
550 
551 	/* Version Supported: 1.3 */
552 	ctrlr->vcprop.vs.bits.mjr = 1;
553 	ctrlr->vcprop.vs.bits.mnr = 3;
554 	ctrlr->vcprop.vs.bits.ter = 0;
555 
556 	ctrlr->vcprop.cc.raw = 0;
557 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
558 	if (subsys_has_multi_iocs) {
559 		ctrlr->vcprop.cc.bits.css =
560 			SPDK_NVME_CC_CSS_IOCS; /* All supported I/O Command Sets */
561 	}
562 
563 	ctrlr->vcprop.csts.raw = 0;
564 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
565 
566 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
567 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
568 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
569 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
570 
571 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
572 
573 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
574 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
575 			SPDK_ERRLOG("Could not get listener transport ID\n");
576 			goto err_listener;
577 		}
578 
579 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid);
580 		if (!ctrlr->listener) {
581 			SPDK_ERRLOG("Listener was not found\n");
582 			goto err_listener;
583 		}
584 	}
585 
586 	nvmf_qpair_set_ctrlr(req->qpair, ctrlr);
587 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
588 
589 	return ctrlr;
590 err_listener:
591 	spdk_bit_array_free(&ctrlr->visible_ns);
592 err_visible_ns:
593 	spdk_bit_array_free(&ctrlr->qpair_mask);
594 err_qpair_mask:
595 	free(ctrlr);
596 	return NULL;
597 }
598 
599 static void
600 _nvmf_ctrlr_destruct(void *ctx)
601 {
602 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
603 	struct spdk_nvmf_reservation_log *log, *log_tmp;
604 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
605 
606 	SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_destruct, ctrlr, ctrlr->subsys->subnqn,
607 				 spdk_thread_get_id(ctrlr->thread));
608 
609 	assert(spdk_get_thread() == ctrlr->thread);
610 	assert(ctrlr->in_destruct);
611 
612 	SPDK_DEBUGLOG(nvmf, "Destroy ctrlr 0x%hx\n", ctrlr->cntlid);
613 	if (ctrlr->disconnect_in_progress) {
614 		SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n");
615 		spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
616 		return;
617 	}
618 
619 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
620 	nvmf_ctrlr_stop_association_timer(ctrlr);
621 	spdk_bit_array_free(&ctrlr->qpair_mask);
622 
623 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
624 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
625 		free(log);
626 	}
627 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
628 		STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link);
629 		free(event);
630 	}
631 	spdk_bit_array_free(&ctrlr->visible_ns);
632 	free(ctrlr);
633 }
634 
635 void
636 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
637 {
638 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
639 
640 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
641 }
642 
643 static void
644 nvmf_ctrlr_add_io_qpair(void *ctx)
645 {
646 	struct spdk_nvmf_request *req = ctx;
647 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
648 	struct spdk_nvmf_qpair *qpair = req->qpair;
649 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
650 	struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair;
651 	struct spdk_nvmf_poll_group *admin_qpair_group = NULL;
652 	enum spdk_nvmf_qpair_state admin_qpair_state = SPDK_NVMF_QPAIR_UNINITIALIZED;
653 	bool admin_qpair_active = false;
654 
655 	SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_io_qpair, ctrlr, req->qpair, req->qpair->qid,
656 				 spdk_thread_get_id(ctrlr->thread));
657 
658 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
659 	  * For error case, the value should be NULL. So set it to NULL at first.
660 	  */
661 	qpair->ctrlr = NULL;
662 
663 	/* Make sure the controller is not being destroyed. */
664 	if (ctrlr->in_destruct) {
665 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
666 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
667 		goto end;
668 	}
669 
670 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
671 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
672 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
673 		goto end;
674 	}
675 
676 	if (!ctrlr->vcprop.cc.bits.en) {
677 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
678 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
679 		goto end;
680 	}
681 
682 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
683 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
684 			    ctrlr->vcprop.cc.bits.iosqes);
685 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
686 		goto end;
687 	}
688 
689 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
690 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
691 			    ctrlr->vcprop.cc.bits.iocqes);
692 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
693 		goto end;
694 	}
695 
696 	/* There is a chance that admin qpair was destroyed. This is an issue that was observed only with ESX initiators */
697 	if (admin_qpair) {
698 		admin_qpair_active = spdk_nvmf_qpair_is_active(admin_qpair);
699 		admin_qpair_group = admin_qpair->group;
700 		admin_qpair_state = admin_qpair->state;
701 	}
702 
703 	if (!admin_qpair_active || admin_qpair_group == NULL) {
704 		/* There is a chance that admin qpair was destroyed or is being destroyed at this moment due to e.g.
705 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
706 		 * state to DEACTIVATING and removing it from poll group */
707 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair_state, admin_qpair_group);
708 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
709 		goto end;
710 	}
711 
712 	/* check if we would exceed ctrlr connection limit */
713 	if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) {
714 		SPDK_ERRLOG("Requested QID %u but Max QID is %u\n",
715 			    qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1);
716 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
717 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
718 		goto end;
719 	}
720 
721 	nvmf_ctrlr_add_qpair(qpair, ctrlr, req);
722 	return;
723 end:
724 	spdk_nvmf_request_complete(req);
725 }
726 
727 static void
728 _nvmf_ctrlr_add_io_qpair(void *ctx)
729 {
730 	struct spdk_nvmf_request *req = ctx;
731 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
732 	struct spdk_nvmf_fabric_connect_data *data;
733 	struct spdk_nvmf_ctrlr *ctrlr;
734 	struct spdk_nvmf_qpair *qpair = req->qpair;
735 	struct spdk_nvmf_qpair *admin_qpair;
736 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
737 	struct spdk_nvmf_subsystem *subsystem;
738 	struct spdk_nvme_transport_id listen_trid = {};
739 	const struct spdk_nvmf_subsystem_listener *listener;
740 	struct spdk_nvmf_poll_group *admin_qpair_group = NULL;
741 	enum spdk_nvmf_qpair_state admin_qpair_state = SPDK_NVMF_QPAIR_UNINITIALIZED;
742 	bool admin_qpair_active = false;
743 
744 	assert(req->iovcnt == 1);
745 
746 	data = req->iov[0].iov_base;
747 
748 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
749 
750 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
751 	/* We already checked this in spdk_nvmf_ctrlr_connect */
752 	assert(subsystem != NULL);
753 
754 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
755 	if (ctrlr == NULL) {
756 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
757 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
758 		spdk_nvmf_request_complete(req);
759 		return;
760 	}
761 
762 	/* fail before passing a message to the controller thread. */
763 	if (ctrlr->in_destruct) {
764 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
765 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
766 		spdk_nvmf_request_complete(req);
767 		return;
768 	}
769 
770 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
771 	if (subsystem->flags.ana_reporting) {
772 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
773 			SPDK_ERRLOG("Could not get listener transport ID\n");
774 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
775 			spdk_nvmf_request_complete(req);
776 			return;
777 		}
778 
779 		listener = nvmf_subsystem_find_listener(subsystem, &listen_trid);
780 		if (listener != ctrlr->listener) {
781 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
782 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
783 			spdk_nvmf_request_complete(req);
784 			return;
785 		}
786 	}
787 
788 	admin_qpair = ctrlr->admin_qpair;
789 
790 	/* There is a chance that admin qpair was destroyed. This is an issue that was observed only with ESX initiators */
791 	if (admin_qpair) {
792 		admin_qpair_active = spdk_nvmf_qpair_is_active(admin_qpair);
793 		admin_qpair_group = admin_qpair->group;
794 		admin_qpair_state = admin_qpair->state;
795 	}
796 
797 	if (!admin_qpair_active || admin_qpair_group == NULL) {
798 		/* There is a chance that admin qpair was destroyed or is being destroyed at this moment due to e.g.
799 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
800 		 * state to DEACTIVATING and removing it from poll group */
801 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair_state, admin_qpair_group);
802 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
803 		spdk_nvmf_request_complete(req);
804 		return;
805 	}
806 	qpair->ctrlr = ctrlr;
807 	spdk_thread_send_msg(admin_qpair_group->thread, nvmf_ctrlr_add_io_qpair, req);
808 }
809 
810 static bool
811 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
812 			  const char *hostnqn)
813 {
814 	struct spdk_nvme_transport_id listen_trid = {};
815 
816 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
817 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
818 		return false;
819 	}
820 
821 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
822 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
823 			    subsystem->subnqn);
824 		return false;
825 	}
826 
827 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
828 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
829 			    subsystem->subnqn, hostnqn);
830 		return false;
831 	}
832 
833 	return true;
834 }
835 
836 static int
837 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
838 {
839 	struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base;
840 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
841 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
842 	struct spdk_nvmf_qpair *qpair = req->qpair;
843 	struct spdk_nvmf_transport *transport = qpair->transport;
844 	struct spdk_nvmf_ctrlr *ctrlr;
845 	struct spdk_nvmf_subsystem *subsystem;
846 
847 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
848 		      cmd->recfmt, cmd->qid, cmd->sqsize);
849 
850 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
851 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
852 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
853 		      ntohl(*(uint32_t *)&data->hostid[0]),
854 		      ntohs(*(uint16_t *)&data->hostid[4]),
855 		      ntohs(*(uint16_t *)&data->hostid[6]),
856 		      data->hostid[8],
857 		      data->hostid[9],
858 		      ntohs(*(uint16_t *)&data->hostid[10]),
859 		      ntohl(*(uint32_t *)&data->hostid[12]));
860 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
861 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
862 
863 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
864 	if (!subsystem) {
865 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
866 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
867 	}
868 
869 	if (cmd->recfmt != 0) {
870 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
871 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
872 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
873 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
874 	}
875 
876 	/*
877 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
878 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
879 	 */
880 	if (cmd->sqsize == 0) {
881 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
882 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
883 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
884 	}
885 
886 	if (cmd->qid == 0) {
887 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
888 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
889 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
890 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
891 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
892 		}
893 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
894 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
895 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
896 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
897 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
898 	}
899 
900 	qpair->sq_head_max = cmd->sqsize;
901 	qpair->qid = cmd->qid;
902 	qpair->connect_received = true;
903 
904 	pthread_mutex_lock(&qpair->group->mutex);
905 	qpair->group->current_unassociated_qpairs--;
906 	pthread_mutex_unlock(&qpair->group->mutex);
907 
908 	if (0 == qpair->qid) {
909 		qpair->group->stat.admin_qpairs++;
910 		qpair->group->stat.current_admin_qpairs++;
911 	} else {
912 		qpair->group->stat.io_qpairs++;
913 		qpair->group->stat.current_io_qpairs++;
914 	}
915 
916 	if (cmd->qid == 0) {
917 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
918 
919 		if (spdk_nvme_trtype_is_fabrics(transport->ops->type) && data->cntlid != 0xFFFF) {
920 			/* This NVMf target only supports dynamic mode. */
921 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
922 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
923 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
924 		}
925 
926 		/* Establish a new ctrlr */
927 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
928 		if (!ctrlr) {
929 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
930 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
931 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
932 		} else {
933 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
934 		}
935 	} else {
936 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
937 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
938 	}
939 }
940 
941 static struct spdk_nvmf_subsystem_poll_group *
942 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
943 {
944 	struct spdk_nvmf_fabric_connect_data *data;
945 	struct spdk_nvmf_subsystem *subsystem;
946 	struct spdk_nvmf_tgt *tgt;
947 
948 	assert(nvmf_request_is_fabric_connect(req));
949 	assert(req->qpair->ctrlr == NULL);
950 	assert(req->iovcnt == 1);
951 
952 	data = req->iov[0].iov_base;
953 	tgt = req->qpair->transport->tgt;
954 
955 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
956 	if (subsystem == NULL) {
957 		return NULL;
958 	}
959 
960 	return &req->qpair->group->sgroups[subsystem->id];
961 }
962 
963 int
964 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
965 {
966 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
967 	struct spdk_nvmf_subsystem_poll_group *sgroup;
968 	struct spdk_nvmf_qpair *qpair = req->qpair;
969 	enum spdk_nvmf_request_exec_status status;
970 
971 	if (req->iovcnt > 1) {
972 		SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt);
973 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
974 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
975 		goto out;
976 	}
977 
978 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
979 	if (!sgroup) {
980 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
981 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
982 		goto out;
983 	}
984 
985 	sgroup->mgmt_io_outstanding++;
986 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
987 
988 	status = _nvmf_ctrlr_connect(req);
989 
990 out:
991 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
992 		_nvmf_request_complete(req);
993 	}
994 
995 	return status;
996 }
997 
998 static int
999 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
1000 {
1001 	struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base;
1002 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
1003 	struct spdk_nvmf_transport *transport = req->qpair->transport;
1004 	struct spdk_nvmf_subsystem *subsystem;
1005 
1006 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
1007 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
1008 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1009 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1010 	}
1011 
1012 	if (req->iovcnt > 1) {
1013 		SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt);
1014 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1015 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1016 	}
1017 
1018 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
1019 	if (!subsystem) {
1020 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
1021 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1022 	}
1023 
1024 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
1025 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
1026 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
1027 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
1028 		struct spdk_nvmf_subsystem_poll_group *sgroup;
1029 
1030 		/* Subsystem is not ready to handle a connect. Decrement
1031 		 * the mgmt_io_outstanding to avoid the subsystem waiting
1032 		 * for this command to complete before unpausing. Queued
1033 		 * requests get retried when subsystem resumes.
1034 		 */
1035 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
1036 		assert(sgroup != NULL);
1037 		sgroup->mgmt_io_outstanding--;
1038 		TAILQ_REMOVE(&req->qpair->outstanding, req, link);
1039 		TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
1040 		SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn);
1041 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
1042 	}
1043 
1044 	/* Ensure that hostnqn is null terminated */
1045 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
1046 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
1047 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
1048 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1049 	}
1050 
1051 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
1052 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1053 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
1054 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1055 	}
1056 
1057 	return _nvmf_ctrlr_connect(req);
1058 }
1059 
1060 static int
1061 nvmf_ctrlr_association_remove(void *ctx)
1062 {
1063 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1064 	int rc;
1065 
1066 	nvmf_ctrlr_stop_association_timer(ctrlr);
1067 
1068 	if (ctrlr->in_destruct) {
1069 		return SPDK_POLLER_IDLE;
1070 	}
1071 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
1072 		      ctrlr->subsys->subnqn);
1073 
1074 	if (ctrlr->admin_qpair) {
1075 		rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair);
1076 		if (rc < 0 && rc != -EINPROGRESS) {
1077 			SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
1078 			assert(false);
1079 		}
1080 	}
1081 
1082 	return SPDK_POLLER_BUSY;
1083 }
1084 
1085 static int
1086 _nvmf_ctrlr_cc_reset_shn_done(void *ctx)
1087 {
1088 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1089 	uint64_t now = spdk_get_ticks();
1090 	uint32_t count;
1091 
1092 	if (ctrlr->cc_timer) {
1093 		spdk_poller_unregister(&ctrlr->cc_timer);
1094 	}
1095 
1096 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1097 	SPDK_DEBUGLOG(nvmf, "ctrlr %p active queue count %u\n", ctrlr, count);
1098 
1099 	if (count > 1) {
1100 		if (now < ctrlr->cc_timeout_tsc) {
1101 			/* restart cc timer */
1102 			ctrlr->cc_timer = SPDK_POLLER_REGISTER(_nvmf_ctrlr_cc_reset_shn_done, ctrlr, 100 * 1000);
1103 			return SPDK_POLLER_IDLE;
1104 		} else {
1105 			/* controller fatal status */
1106 			SPDK_WARNLOG("IO timeout, ctrlr %p is in fatal status\n", ctrlr);
1107 			nvmf_ctrlr_set_fatal_status(ctrlr);
1108 		}
1109 	}
1110 
1111 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
1112 
1113 	if (ctrlr->disconnect_is_shn) {
1114 		ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
1115 		ctrlr->disconnect_is_shn = false;
1116 	} else {
1117 		/* Only a subset of the registers are cleared out on a reset */
1118 		ctrlr->vcprop.cc.raw = 0;
1119 		ctrlr->vcprop.csts.raw = 0;
1120 	}
1121 
1122 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
1123 	 * between the host and controller shall be preserved for at least 2 minutes */
1124 	if (ctrlr->association_timer) {
1125 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
1126 		nvmf_ctrlr_stop_association_timer(ctrlr);
1127 	}
1128 	if (ctrlr->association_timeout) {
1129 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
1130 					   ctrlr->association_timeout * 1000);
1131 	}
1132 	ctrlr->disconnect_in_progress = false;
1133 	return SPDK_POLLER_BUSY;
1134 }
1135 
1136 static void
1137 nvmf_ctrlr_cc_reset_shn_done(struct spdk_io_channel_iter *i, int status)
1138 {
1139 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
1140 
1141 	if (status < 0) {
1142 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
1143 		assert(false);
1144 	}
1145 
1146 	_nvmf_ctrlr_cc_reset_shn_done((void *)ctrlr);
1147 }
1148 
1149 static void
1150 nvmf_bdev_complete_reset(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
1151 {
1152 	SPDK_NOTICELOG("Resetting bdev done with %s\n", success ? "success" : "failure");
1153 
1154 	spdk_bdev_free_io(bdev_io);
1155 }
1156 
1157 
1158 static int
1159 nvmf_ctrlr_cc_timeout(void *ctx)
1160 {
1161 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1162 	struct spdk_nvmf_poll_group *group;
1163 	struct spdk_nvmf_ns *ns;
1164 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1165 
1166 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
1167 	SPDK_DEBUGLOG(nvmf, "Ctrlr %p reset or shutdown timeout\n", ctrlr);
1168 
1169 	if (!ctrlr->admin_qpair) {
1170 		SPDK_NOTICELOG("Ctrlr %p admin qpair disconnected\n", ctrlr);
1171 		return SPDK_POLLER_IDLE;
1172 	}
1173 
1174 	group = ctrlr->admin_qpair->group;
1175 	assert(group != NULL && group->sgroups != NULL);
1176 
1177 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1178 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1179 		if (ns->bdev == NULL) {
1180 			continue;
1181 		}
1182 		ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[ns->opts.nsid - 1];
1183 		SPDK_NOTICELOG("Ctrlr %p resetting NSID %u\n", ctrlr, ns->opts.nsid);
1184 		spdk_bdev_reset(ns->desc, ns_info->channel, nvmf_bdev_complete_reset, NULL);
1185 	}
1186 
1187 	return SPDK_POLLER_BUSY;
1188 }
1189 
1190 const struct spdk_nvmf_registers *
1191 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
1192 {
1193 	return &ctrlr->vcprop;
1194 }
1195 
1196 void
1197 nvmf_ctrlr_set_fatal_status(struct spdk_nvmf_ctrlr *ctrlr)
1198 {
1199 	ctrlr->vcprop.csts.bits.cfs = 1;
1200 }
1201 
1202 static uint64_t
1203 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
1204 {
1205 	return ctrlr->vcprop.cap.raw;
1206 }
1207 
1208 static uint64_t
1209 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
1210 {
1211 	return ctrlr->vcprop.vs.raw;
1212 }
1213 
1214 static uint64_t
1215 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
1216 {
1217 	return ctrlr->vcprop.cc.raw;
1218 }
1219 
1220 static bool
1221 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1222 {
1223 	union spdk_nvme_cc_register cc, diff;
1224 	uint32_t cc_timeout_ms;
1225 
1226 	cc.raw = value;
1227 
1228 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
1229 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
1230 
1231 	/*
1232 	 * Calculate which bits changed between the current and new CC.
1233 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
1234 	 */
1235 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
1236 
1237 	if (diff.bits.en) {
1238 		if (cc.bits.en) {
1239 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
1240 			nvmf_ctrlr_stop_association_timer(ctrlr);
1241 
1242 			ctrlr->vcprop.cc.bits.en = 1;
1243 			ctrlr->vcprop.csts.bits.rdy = 1;
1244 		} else {
1245 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
1246 			if (ctrlr->disconnect_in_progress) {
1247 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1248 				return true;
1249 			}
1250 
1251 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1252 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1253 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1254 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1255 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1256 
1257 			ctrlr->vcprop.cc.bits.en = 0;
1258 			ctrlr->disconnect_in_progress = true;
1259 			ctrlr->disconnect_is_shn = false;
1260 			spdk_for_each_channel(ctrlr->subsys->tgt,
1261 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1262 					      ctrlr,
1263 					      nvmf_ctrlr_cc_reset_shn_done);
1264 		}
1265 		diff.bits.en = 0;
1266 	}
1267 
1268 	if (diff.bits.shn) {
1269 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
1270 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
1271 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
1272 				      cc.bits.shn >> 1, cc.bits.shn & 1);
1273 			if (ctrlr->disconnect_in_progress) {
1274 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1275 				return true;
1276 			}
1277 
1278 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1279 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1280 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1281 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1282 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1283 
1284 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
1285 			ctrlr->disconnect_in_progress = true;
1286 			ctrlr->disconnect_is_shn = true;
1287 			spdk_for_each_channel(ctrlr->subsys->tgt,
1288 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1289 					      ctrlr,
1290 					      nvmf_ctrlr_cc_reset_shn_done);
1291 
1292 			/* From the time a shutdown is initiated the controller shall disable
1293 			 * Keep Alive timer */
1294 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
1295 		} else if (cc.bits.shn == 0) {
1296 			ctrlr->vcprop.cc.bits.shn = 0;
1297 		} else {
1298 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
1299 				    cc.bits.shn >> 1, cc.bits.shn & 1);
1300 			return false;
1301 		}
1302 		diff.bits.shn = 0;
1303 	}
1304 
1305 	if (diff.bits.iosqes) {
1306 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
1307 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
1308 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
1309 		diff.bits.iosqes = 0;
1310 	}
1311 
1312 	if (diff.bits.iocqes) {
1313 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
1314 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
1315 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
1316 		diff.bits.iocqes = 0;
1317 	}
1318 
1319 	if (diff.bits.ams) {
1320 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
1321 		return false;
1322 	}
1323 
1324 	if (diff.bits.mps) {
1325 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
1326 		return false;
1327 	}
1328 
1329 	if (diff.bits.css) {
1330 		if (cc.bits.css > SPDK_NVME_CC_CSS_IOCS) {
1331 			SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
1332 			return false;
1333 		}
1334 		diff.bits.css = 0;
1335 	}
1336 
1337 	if (diff.raw != 0) {
1338 		/* Print an error message, but don't fail the command in this case.
1339 		 * If we did want to fail in this case, we'd need to ensure we acted
1340 		 * on no other bits or the initiator gets confused. */
1341 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
1342 	}
1343 
1344 	return true;
1345 }
1346 
1347 static uint64_t
1348 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
1349 {
1350 	return ctrlr->vcprop.csts.raw;
1351 }
1352 
1353 static uint64_t
1354 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
1355 {
1356 	return ctrlr->vcprop.aqa.raw;
1357 }
1358 
1359 static bool
1360 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1361 {
1362 	union spdk_nvme_aqa_register aqa;
1363 
1364 	aqa.raw = value;
1365 
1366 	/*
1367 	 * We don't need to explicitly check for maximum size, as the fields are
1368 	 * limited to 12 bits (4096).
1369 	 */
1370 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1371 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1372 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1373 		return false;
1374 	}
1375 
1376 	ctrlr->vcprop.aqa.raw = value;
1377 
1378 	return true;
1379 }
1380 
1381 static uint64_t
1382 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1383 {
1384 	return ctrlr->vcprop.asq;
1385 }
1386 
1387 static bool
1388 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1389 {
1390 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1391 
1392 	return true;
1393 }
1394 
1395 static bool
1396 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1397 {
1398 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1399 
1400 	return true;
1401 }
1402 
1403 static uint64_t
1404 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1405 {
1406 	return ctrlr->vcprop.acq;
1407 }
1408 
1409 static bool
1410 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1411 {
1412 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1413 
1414 	return true;
1415 }
1416 
1417 static bool
1418 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1419 {
1420 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1421 
1422 	return true;
1423 }
1424 
1425 struct nvmf_prop {
1426 	uint32_t ofst;
1427 	uint8_t size;
1428 	char name[11];
1429 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1430 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1431 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1432 };
1433 
1434 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1435 	{ \
1436 		offsetof(struct spdk_nvme_registers, field), \
1437 		size, \
1438 		#field, \
1439 		get_cb, set_cb, set_upper_cb \
1440 	}
1441 
1442 static const struct nvmf_prop nvmf_props[] = {
1443 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1444 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1445 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1446 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1447 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1448 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1449 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1450 };
1451 
1452 static const struct nvmf_prop *
1453 find_prop(uint32_t ofst, uint8_t size)
1454 {
1455 	size_t i;
1456 
1457 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1458 		const struct nvmf_prop *prop = &nvmf_props[i];
1459 
1460 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1461 			return prop;
1462 		}
1463 	}
1464 
1465 	return NULL;
1466 }
1467 
1468 static int
1469 nvmf_property_get(struct spdk_nvmf_request *req)
1470 {
1471 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1472 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1473 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1474 	const struct nvmf_prop *prop;
1475 	uint8_t size;
1476 
1477 	response->status.sc = 0;
1478 	response->value.u64 = 0;
1479 
1480 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1481 		      cmd->attrib.size, cmd->ofst);
1482 
1483 	switch (cmd->attrib.size) {
1484 	case SPDK_NVMF_PROP_SIZE_4:
1485 		size = 4;
1486 		break;
1487 	case SPDK_NVMF_PROP_SIZE_8:
1488 		size = 8;
1489 		break;
1490 	default:
1491 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1492 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1493 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1494 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1495 	}
1496 
1497 	prop = find_prop(cmd->ofst, size);
1498 	if (prop == NULL || prop->get_cb == NULL) {
1499 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1500 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1501 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1502 	}
1503 
1504 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1505 
1506 	response->value.u64 = prop->get_cb(ctrlr);
1507 
1508 	if (size != prop->size) {
1509 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1510 		assert(size == 4);
1511 		assert(prop->size == 8);
1512 
1513 		if (cmd->ofst == prop->ofst) {
1514 			/* Keep bottom 4 bytes only */
1515 			response->value.u64 &= 0xFFFFFFFF;
1516 		} else {
1517 			/* Keep top 4 bytes only */
1518 			response->value.u64 >>= 32;
1519 		}
1520 	}
1521 
1522 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1523 
1524 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1525 }
1526 
1527 static int
1528 nvmf_property_set(struct spdk_nvmf_request *req)
1529 {
1530 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1531 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1532 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1533 	const struct nvmf_prop *prop;
1534 	uint64_t value;
1535 	uint8_t size;
1536 	bool ret;
1537 
1538 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1539 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1540 
1541 	switch (cmd->attrib.size) {
1542 	case SPDK_NVMF_PROP_SIZE_4:
1543 		size = 4;
1544 		break;
1545 	case SPDK_NVMF_PROP_SIZE_8:
1546 		size = 8;
1547 		break;
1548 	default:
1549 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1550 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1551 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1552 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1553 	}
1554 
1555 	prop = find_prop(cmd->ofst, size);
1556 	if (prop == NULL || prop->set_cb == NULL) {
1557 		SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst);
1558 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1559 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1560 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1561 	}
1562 
1563 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1564 
1565 	value = cmd->value.u64;
1566 
1567 	if (prop->size == 4) {
1568 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1569 	} else if (size != prop->size) {
1570 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1571 		assert(size == 4);
1572 		assert(prop->size == 8);
1573 
1574 		if (cmd->ofst == prop->ofst) {
1575 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1576 		} else {
1577 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1578 		}
1579 	} else {
1580 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1581 		if (ret) {
1582 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1583 		}
1584 	}
1585 
1586 	if (!ret) {
1587 		SPDK_ERRLOG("prop set_cb failed\n");
1588 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1589 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1590 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1591 	}
1592 
1593 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1594 }
1595 
1596 static int
1597 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1598 {
1599 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1600 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1601 
1602 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1603 
1604 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1605 	ctrlr->feat.arbitration.bits.reserved = 0;
1606 
1607 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1608 }
1609 
1610 static int
1611 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1612 {
1613 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1614 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1615 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1616 
1617 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1618 
1619 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1620 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1621 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1622 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1623 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1624 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1625 	}
1626 
1627 	ctrlr->feat.power_management.raw = cmd->cdw11;
1628 	ctrlr->feat.power_management.bits.reserved = 0;
1629 
1630 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1631 }
1632 
1633 static bool
1634 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1635 {
1636 	/*
1637 	 * Valid TMPSEL values:
1638 	 *  0000b - 1000b: temperature sensors
1639 	 *  1111b: set all implemented temperature sensors
1640 	 */
1641 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1642 		/* 1001b - 1110b: reserved */
1643 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1644 		return false;
1645 	}
1646 
1647 	/*
1648 	 * Valid THSEL values:
1649 	 *  00b: over temperature threshold
1650 	 *  01b: under temperature threshold
1651 	 */
1652 	if (opts->bits.thsel > 1) {
1653 		/* 10b - 11b: reserved */
1654 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1655 		return false;
1656 	}
1657 
1658 	return true;
1659 }
1660 
1661 static int
1662 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1663 {
1664 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1665 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1666 
1667 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1668 
1669 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1670 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1671 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1672 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1673 	}
1674 
1675 	/* TODO: no sensors implemented - ignore new values */
1676 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1677 }
1678 
1679 static int
1680 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1681 {
1682 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1683 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1684 
1685 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1686 
1687 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1688 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1689 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1690 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1691 	}
1692 
1693 	/* TODO: no sensors implemented - return 0 for all thresholds */
1694 	rsp->cdw0 = 0;
1695 
1696 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1697 }
1698 
1699 static int
1700 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req)
1701 {
1702 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1703 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1704 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1705 	union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {};
1706 
1707 	SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11);
1708 
1709 	iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv;
1710 	iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd;
1711 	rsp->cdw0 = iv_conf.raw;
1712 
1713 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1714 }
1715 
1716 static int
1717 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1718 {
1719 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1720 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1721 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1722 
1723 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1724 
1725 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1726 		/*
1727 		 * Host is not allowed to set this bit, since we don't advertise it in
1728 		 * Identify Namespace.
1729 		 */
1730 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1731 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1732 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1733 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1734 	}
1735 
1736 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1737 	ctrlr->feat.error_recovery.bits.reserved = 0;
1738 
1739 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1740 }
1741 
1742 static int
1743 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1744 {
1745 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1746 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1747 
1748 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1749 
1750 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1751 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1752 
1753 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1754 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1755 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1756 }
1757 
1758 static int
1759 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1760 {
1761 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1762 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1763 
1764 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1765 
1766 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1767 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1768 
1769 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1770 }
1771 
1772 static int
1773 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1774 {
1775 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1776 
1777 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1778 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1779 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1780 }
1781 
1782 static int
1783 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1784 {
1785 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1786 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1787 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1788 	struct spdk_iov_xfer ix;
1789 
1790 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1791 
1792 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1793 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1794 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1795 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1796 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1797 	}
1798 
1799 	if (req->iovcnt < 1 || req->length < sizeof(ctrlr->hostid)) {
1800 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1801 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1802 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1803 	}
1804 
1805 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
1806 	spdk_iov_xfer_from_buf(&ix, &ctrlr->hostid, sizeof(ctrlr->hostid));
1807 
1808 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1809 }
1810 
1811 static int
1812 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1813 {
1814 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1815 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1816 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1817 	struct spdk_nvmf_ns *ns;
1818 
1819 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notification Mask\n");
1820 
1821 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1822 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1823 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1824 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1825 	}
1826 
1827 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1828 	if (ns == NULL) {
1829 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1830 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1831 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1832 	}
1833 	rsp->cdw0 = ns->mask;
1834 
1835 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1836 }
1837 
1838 static int
1839 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1840 {
1841 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1842 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1843 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1844 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1845 	struct spdk_nvmf_ns *ns;
1846 
1847 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notification Mask\n");
1848 
1849 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1850 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1851 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1852 			ns->mask = cmd->cdw11;
1853 		}
1854 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1855 	}
1856 
1857 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1858 	if (ns == NULL) {
1859 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1860 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1861 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1862 	}
1863 	ns->mask = cmd->cdw11;
1864 
1865 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1866 }
1867 
1868 static int
1869 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1870 {
1871 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1872 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1873 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1874 	struct spdk_nvmf_ns *ns;
1875 
1876 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1877 
1878 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1879 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1880 	if (ns == NULL) {
1881 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1882 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1883 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1884 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1885 	}
1886 
1887 	response->cdw0 = ns->ptpl_activated;
1888 
1889 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1890 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1891 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1892 }
1893 
1894 static int
1895 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req)
1896 {
1897 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1898 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1899 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1900 	struct spdk_nvmf_ns *ns;
1901 	bool ptpl;
1902 
1903 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1904 
1905 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1906 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1907 
1908 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && nvmf_ns_is_ptpl_capable(ns)) {
1909 		ns->ptpl_activated = ptpl;
1910 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1911 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns;
1912 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1913 			if (nvmf_ns_is_ptpl_capable(ns)) {
1914 				ns->ptpl_activated = ptpl;
1915 			}
1916 		}
1917 	} else {
1918 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1919 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1920 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1921 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1922 	}
1923 
1924 	/* TODO: Feature not changeable for now */
1925 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1926 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1927 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1928 }
1929 
1930 static int
1931 nvmf_ctrlr_get_features_host_behavior_support(struct spdk_nvmf_request *req)
1932 {
1933 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1934 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1935 	struct spdk_nvme_host_behavior host_behavior = {};
1936 	struct spdk_iov_xfer ix;
1937 
1938 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Behavior Support\n");
1939 
1940 	if (req->iovcnt < 1 || req->length < sizeof(struct spdk_nvme_host_behavior)) {
1941 		SPDK_ERRLOG("invalid data buffer for Host Behavior Support\n");
1942 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1943 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1944 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1945 	}
1946 
1947 	host_behavior.acre = ctrlr->acre_enabled;
1948 
1949 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
1950 	spdk_iov_xfer_from_buf(&ix, &host_behavior, sizeof(host_behavior));
1951 
1952 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1953 }
1954 
1955 static int
1956 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req)
1957 {
1958 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1959 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1960 	struct spdk_nvme_host_behavior *host_behavior;
1961 
1962 	SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n");
1963 	if (req->iovcnt != 1) {
1964 		SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt);
1965 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1966 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1967 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1968 	}
1969 	if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) {
1970 		SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len);
1971 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1972 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1973 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1974 	}
1975 
1976 	host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base;
1977 	if (host_behavior->acre == 0) {
1978 		ctrlr->acre_enabled = false;
1979 	} else if (host_behavior->acre == 1) {
1980 		ctrlr->acre_enabled = true;
1981 	} else {
1982 		SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre);
1983 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1984 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1985 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1986 	}
1987 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1988 }
1989 
1990 static int
1991 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
1992 {
1993 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1994 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1995 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1996 
1997 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
1998 
1999 	/*
2000 	 * if attempts to disable keep alive by setting kato to 0h
2001 	 * a status value of keep alive invalid shall be returned
2002 	 */
2003 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
2004 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
2005 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
2006 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
2007 	} else {
2008 		/* round up to milliseconds */
2009 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
2010 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
2011 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
2012 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
2013 	}
2014 
2015 	/*
2016 	 * if change the keep alive timeout value successfully
2017 	 * update the keep alive poller.
2018 	 */
2019 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
2020 		if (ctrlr->keep_alive_poller != NULL) {
2021 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
2022 		}
2023 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
2024 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
2025 	}
2026 
2027 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
2028 		      ctrlr->feat.keep_alive_timer.bits.kato);
2029 
2030 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2031 }
2032 
2033 static int
2034 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
2035 {
2036 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2037 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2038 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2039 	uint32_t count;
2040 
2041 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
2042 		      req->cmd->nvme_cmd.cdw11);
2043 
2044 	if (cmd->cdw11_bits.feat_num_of_queues.bits.ncqr == UINT16_MAX ||
2045 	    cmd->cdw11_bits.feat_num_of_queues.bits.nsqr == UINT16_MAX) {
2046 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2047 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2048 	}
2049 
2050 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
2051 	/* verify that the controller is ready to process commands */
2052 	if (count > 1) {
2053 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
2054 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2055 	} else {
2056 		/*
2057 		 * Ignore the value requested by the host -
2058 		 * always return the pre-configured value based on max_qpairs_allowed.
2059 		 */
2060 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
2061 	}
2062 
2063 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2064 }
2065 
2066 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ctrlr) == 4928,
2067 		   "Please check migration fields that need to be added or not");
2068 
2069 static void
2070 nvmf_ctrlr_migr_data_copy(struct spdk_nvmf_ctrlr_migr_data *data,
2071 			  const struct spdk_nvmf_ctrlr_migr_data *data_src, size_t data_size)
2072 {
2073 	assert(data);
2074 	assert(data_src);
2075 	assert(data_size);
2076 
2077 	memcpy(&data->regs, &data_src->regs, spdk_min(data->regs_size, data_src->regs_size));
2078 	memcpy(&data->feat, &data_src->feat, spdk_min(data->feat_size, data_src->feat_size));
2079 
2080 #define SET_FIELD(field) \
2081     if (offsetof(struct spdk_nvmf_ctrlr_migr_data, field) + sizeof(data->field) <= data_size) { \
2082         data->field = data_src->field; \
2083     } \
2084 
2085 	SET_FIELD(cntlid);
2086 	SET_FIELD(acre);
2087 	SET_FIELD(num_aer_cids);
2088 	SET_FIELD(num_async_events);
2089 	SET_FIELD(notice_aen_mask);
2090 #undef SET_FIELD
2091 
2092 #define SET_ARRAY(arr) \
2093     if (offsetof(struct spdk_nvmf_ctrlr_migr_data, arr) + sizeof(data->arr) <= data_size) { \
2094         memcpy(&data->arr, &data_src->arr, sizeof(data->arr)); \
2095     } \
2096 
2097 	SET_ARRAY(async_events);
2098 	SET_ARRAY(aer_cids);
2099 #undef SET_ARRAY
2100 }
2101 
2102 int
2103 spdk_nvmf_ctrlr_save_migr_data(struct spdk_nvmf_ctrlr *ctrlr,
2104 			       struct spdk_nvmf_ctrlr_migr_data *data)
2105 {
2106 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
2107 	uint32_t i;
2108 	struct spdk_nvmf_ctrlr_migr_data data_local = {
2109 		.data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused),
2110 		.regs_size = sizeof(struct spdk_nvmf_registers),
2111 		.feat_size = sizeof(struct spdk_nvmf_ctrlr_feat)
2112 	};
2113 
2114 	assert(data->data_size <= sizeof(data_local));
2115 	assert(spdk_get_thread() == ctrlr->thread);
2116 
2117 	memcpy(&data_local.regs, &ctrlr->vcprop, sizeof(struct spdk_nvmf_registers));
2118 	memcpy(&data_local.feat, &ctrlr->feat, sizeof(struct spdk_nvmf_ctrlr_feat));
2119 
2120 	data_local.cntlid = ctrlr->cntlid;
2121 	data_local.acre = ctrlr->acre_enabled;
2122 	data_local.num_aer_cids = ctrlr->nr_aer_reqs;
2123 
2124 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
2125 		if (data_local.num_async_events + 1 > SPDK_NVMF_MIGR_MAX_PENDING_AERS) {
2126 			SPDK_ERRLOG("ctrlr %p has too many pending AERs\n", ctrlr);
2127 			break;
2128 		}
2129 
2130 		data_local.async_events[data_local.num_async_events++].raw = event->event.raw;
2131 	}
2132 
2133 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2134 		struct spdk_nvmf_request *req = ctrlr->aer_req[i];
2135 		data_local.aer_cids[i] = req->cmd->nvme_cmd.cid;
2136 	}
2137 	data_local.notice_aen_mask = ctrlr->notice_aen_mask;
2138 
2139 	nvmf_ctrlr_migr_data_copy(data, &data_local, spdk_min(data->data_size, data_local.data_size));
2140 	return 0;
2141 }
2142 
2143 int
2144 spdk_nvmf_ctrlr_restore_migr_data(struct spdk_nvmf_ctrlr *ctrlr,
2145 				  const struct spdk_nvmf_ctrlr_migr_data *data)
2146 {
2147 	uint32_t i;
2148 	struct spdk_nvmf_ctrlr_migr_data data_local = {
2149 		.data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused),
2150 		.regs_size = sizeof(struct spdk_nvmf_registers),
2151 		.feat_size = sizeof(struct spdk_nvmf_ctrlr_feat)
2152 	};
2153 
2154 	assert(data->data_size <= sizeof(data_local));
2155 	assert(spdk_get_thread() == ctrlr->thread);
2156 
2157 	/* local version of data should have defaults set before copy */
2158 	nvmf_ctrlr_migr_data_copy(&data_local, data, spdk_min(data->data_size, data_local.data_size));
2159 	memcpy(&ctrlr->vcprop, &data_local.regs, sizeof(struct spdk_nvmf_registers));
2160 	memcpy(&ctrlr->feat, &data_local.feat, sizeof(struct spdk_nvmf_ctrlr_feat));
2161 
2162 	ctrlr->cntlid = data_local.cntlid;
2163 	ctrlr->acre_enabled = data_local.acre;
2164 
2165 	for (i = 0; i < data_local.num_async_events; i++) {
2166 		struct spdk_nvmf_async_event_completion *event;
2167 
2168 		event = calloc(1, sizeof(*event));
2169 		if (!event) {
2170 			return -ENOMEM;
2171 		}
2172 
2173 		event->event.raw = data_local.async_events[i].raw;
2174 		STAILQ_INSERT_TAIL(&ctrlr->async_events, event, link);
2175 	}
2176 	ctrlr->notice_aen_mask = data_local.notice_aen_mask;
2177 
2178 	return 0;
2179 }
2180 
2181 static int
2182 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
2183 {
2184 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2185 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2186 
2187 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
2188 		      cmd->cdw11);
2189 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
2190 	ctrlr->feat.async_event_configuration.bits.reserved1 = 0;
2191 	ctrlr->feat.async_event_configuration.bits.reserved2 = 0;
2192 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2193 }
2194 
2195 static int
2196 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
2197 {
2198 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2199 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2200 	struct spdk_nvmf_async_event_completion *pending_event;
2201 
2202 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
2203 
2204 	/* Four asynchronous events are supported for now */
2205 	if (ctrlr->nr_aer_reqs >= SPDK_NVMF_MAX_ASYNC_EVENTS) {
2206 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
2207 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
2208 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
2209 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2210 	}
2211 
2212 	if (!STAILQ_EMPTY(&ctrlr->async_events)) {
2213 		pending_event = STAILQ_FIRST(&ctrlr->async_events);
2214 		rsp->cdw0 = pending_event->event.raw;
2215 		STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link);
2216 		free(pending_event);
2217 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2218 	}
2219 
2220 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
2221 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2222 }
2223 
2224 static void
2225 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length)
2226 {
2227 	struct spdk_nvme_firmware_page fw_page;
2228 	size_t copy_len;
2229 	struct spdk_iov_xfer ix;
2230 
2231 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2232 
2233 	memset(&fw_page, 0, sizeof(fw_page));
2234 	fw_page.afi.active_slot = 1;
2235 	fw_page.afi.next_reset_slot = 0;
2236 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
2237 
2238 	if (offset < sizeof(fw_page)) {
2239 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
2240 		if (copy_len > 0) {
2241 			spdk_iov_xfer_from_buf(&ix, (const char *)&fw_page + offset, copy_len);
2242 		}
2243 	}
2244 }
2245 
2246 /*
2247  * Asynchronous Event Mask Bit
2248  */
2249 enum spdk_nvme_async_event_mask_bit {
2250 	/* Mask Namespace Change Notification */
2251 	SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT		= 0,
2252 	/* Mask Asymmetric Namespace Access Change Notification */
2253 	SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT		= 1,
2254 	/* Mask Discovery Log Change Notification */
2255 	SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT	= 2,
2256 	/* Mask Reservation Log Page Available Notification */
2257 	SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT	= 3,
2258 	/* Mask Error Event */
2259 	SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT			= 4,
2260 	/* 4 - 63 Reserved */
2261 };
2262 
2263 static inline void
2264 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2265 		      enum spdk_nvme_async_event_mask_bit mask)
2266 {
2267 	ctrlr->notice_aen_mask &= ~(1 << mask);
2268 }
2269 
2270 static inline bool
2271 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2272 		    enum spdk_nvme_async_event_mask_bit mask)
2273 {
2274 	if (ctrlr->notice_aen_mask & (1 << mask)) {
2275 		return false;
2276 	} else {
2277 		ctrlr->notice_aen_mask |= (1 << mask);
2278 		return true;
2279 	}
2280 }
2281 
2282 /* we have to use the typedef in the function declaration to appease astyle. */
2283 typedef enum spdk_nvme_ana_state spdk_nvme_ana_state_t;
2284 
2285 static inline spdk_nvme_ana_state_t
2286 nvmf_ctrlr_get_ana_state(struct spdk_nvmf_ctrlr *ctrlr, uint32_t anagrpid)
2287 {
2288 	if (!ctrlr->subsys->flags.ana_reporting) {
2289 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2290 	}
2291 
2292 	if (spdk_unlikely(ctrlr->listener == NULL)) {
2293 		return SPDK_NVME_ANA_INACCESSIBLE_STATE;
2294 	}
2295 
2296 	assert(anagrpid - 1 < ctrlr->subsys->max_nsid);
2297 	return ctrlr->listener->ana_state[anagrpid - 1];
2298 }
2299 
2300 static spdk_nvme_ana_state_t
2301 nvmf_ctrlr_get_ana_state_from_nsid(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2302 {
2303 	struct spdk_nvmf_ns *ns;
2304 
2305 	/* We do not have NVM subsystem specific ANA state. Hence if NSID is either
2306 	 * SPDK_NVMF_GLOBAL_NS_TAG, invalid, or for inactive namespace, return
2307 	 * the optimized state.
2308 	 */
2309 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
2310 	if (ns == NULL) {
2311 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2312 	}
2313 
2314 	return nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2315 }
2316 
2317 static void
2318 nvmf_get_error_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2319 			uint64_t offset, uint32_t length, uint32_t rae)
2320 {
2321 	if (!rae) {
2322 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT);
2323 	}
2324 
2325 	/* TODO: actually fill out log page data */
2326 }
2327 
2328 static void
2329 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2330 		      uint64_t offset, uint32_t length, uint32_t rae)
2331 {
2332 	struct spdk_nvme_ana_page ana_hdr;
2333 	struct spdk_nvme_ana_group_descriptor ana_desc;
2334 	size_t copy_len, copied_len;
2335 	uint32_t num_anagrp = 0, anagrpid;
2336 	struct spdk_nvmf_ns *ns;
2337 	struct spdk_iov_xfer ix;
2338 
2339 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2340 
2341 	if (length == 0) {
2342 		goto done;
2343 	}
2344 
2345 	if (offset >= sizeof(ana_hdr)) {
2346 		offset -= sizeof(ana_hdr);
2347 	} else {
2348 		for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2349 			if (ctrlr->subsys->ana_group[anagrpid - 1] > 0) {
2350 				num_anagrp++;
2351 			}
2352 		}
2353 
2354 		memset(&ana_hdr, 0, sizeof(ana_hdr));
2355 
2356 		ana_hdr.num_ana_group_desc = num_anagrp;
2357 		/* TODO: Support Change Count. */
2358 		ana_hdr.change_count = 0;
2359 
2360 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
2361 		copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_hdr + offset, copy_len);
2362 		assert(copied_len == copy_len);
2363 		length -= copied_len;
2364 		offset = 0;
2365 	}
2366 
2367 	if (length == 0) {
2368 		goto done;
2369 	}
2370 
2371 	for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2372 		if (ctrlr->subsys->ana_group[anagrpid - 1] == 0) {
2373 			continue;
2374 		}
2375 
2376 		if (offset >= sizeof(ana_desc)) {
2377 			offset -= sizeof(ana_desc);
2378 		} else {
2379 			memset(&ana_desc, 0, sizeof(ana_desc));
2380 
2381 			ana_desc.ana_group_id = anagrpid;
2382 			ana_desc.num_of_nsid = ctrlr->subsys->ana_group[anagrpid - 1];
2383 			ana_desc.ana_state = nvmf_ctrlr_get_ana_state(ctrlr, anagrpid);
2384 
2385 			copy_len = spdk_min(sizeof(ana_desc) - offset, length);
2386 			copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_desc + offset,
2387 							    copy_len);
2388 			assert(copied_len == copy_len);
2389 			length -= copied_len;
2390 			offset = 0;
2391 
2392 			if (length == 0) {
2393 				goto done;
2394 			}
2395 		}
2396 
2397 		/* TODO: Revisit here about O(n^2) cost if we have subsystem with
2398 		 * many namespaces in the future.
2399 		 */
2400 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
2401 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
2402 			if (ns->anagrpid != anagrpid) {
2403 				continue;
2404 			}
2405 
2406 			if (offset >= sizeof(uint32_t)) {
2407 				offset -= sizeof(uint32_t);
2408 				continue;
2409 			}
2410 
2411 			copy_len = spdk_min(sizeof(uint32_t) - offset, length);
2412 			copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ns->nsid + offset,
2413 							    copy_len);
2414 			assert(copied_len == copy_len);
2415 			length -= copied_len;
2416 			offset = 0;
2417 
2418 			if (length == 0) {
2419 				goto done;
2420 			}
2421 		}
2422 	}
2423 
2424 done:
2425 	if (!rae) {
2426 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT);
2427 	}
2428 }
2429 
2430 void
2431 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2432 {
2433 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
2434 	uint16_t i;
2435 	bool found = false;
2436 
2437 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
2438 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
2439 			/* nsid is already in the list */
2440 			found = true;
2441 			break;
2442 		}
2443 	}
2444 
2445 	if (!found) {
2446 		if (ctrlr->changed_ns_list_count == max_changes) {
2447 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
2448 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
2449 			for (i = 1; i < max_changes; i++) {
2450 				ctrlr->changed_ns_list.ns_list[i] = 0;
2451 			}
2452 		} else {
2453 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
2454 		}
2455 	}
2456 }
2457 
2458 static void
2459 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2460 				  struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2461 {
2462 	size_t copy_length;
2463 	struct spdk_iov_xfer ix;
2464 
2465 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2466 
2467 	if (offset < sizeof(ctrlr->changed_ns_list)) {
2468 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
2469 		if (copy_length) {
2470 			spdk_iov_xfer_from_buf(&ix, (char *)&ctrlr->changed_ns_list + offset, copy_length);
2471 		}
2472 	}
2473 
2474 	/* Clear log page each time it is read */
2475 	ctrlr->changed_ns_list_count = 0;
2476 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
2477 
2478 	if (!rae) {
2479 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT);
2480 	}
2481 }
2482 
2483 /* The structure can be modified if we provide support for other commands in future */
2484 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
2485 	.admin_cmds_supported = {
2486 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
2487 		/* Get Log Page */
2488 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2489 		/* Identify */
2490 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
2491 		/* Abort */
2492 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
2493 		/* Set Features */
2494 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2495 		/* Get Features */
2496 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2497 		/* Async Event Request */
2498 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
2499 		/* Keep Alive */
2500 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2501 	},
2502 	.io_cmds_supported = {
2503 		/* FLUSH */
2504 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
2505 		/* WRITE */
2506 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
2507 		/* READ */
2508 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
2509 		/* WRITE ZEROES */
2510 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
2511 		/* DATASET MANAGEMENT */
2512 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
2513 		/* COMPARE */
2514 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
2515 		/* ZONE MANAGEMENT SEND */
2516 		[SPDK_NVME_OPC_ZONE_MGMT_SEND]		= {1, 1, 0, 0, 0, 0, 0, 0},
2517 		/* ZONE MANAGEMENT RECEIVE */
2518 		[SPDK_NVME_OPC_ZONE_MGMT_RECV]		= {1, 0, 0, 0, 0, 0, 0, 0},
2519 		/* COPY */
2520 		[SPDK_NVME_OPC_COPY]			= {1, 1, 0, 0, 0, 0, 0, 0},
2521 	},
2522 };
2523 
2524 static void
2525 nvmf_get_cmds_and_effects_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2526 				   uint64_t offset, uint32_t length)
2527 {
2528 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
2529 	size_t copy_len = 0;
2530 	struct spdk_nvme_cmds_and_effect_log_page cmds_and_effect_log_page = g_cmds_and_effect_log_page;
2531 	struct spdk_nvme_cmds_and_effect_entry csupp_and_lbcc_effect_entry = {1, 1, 0, 0, 0, 0, 0, 0};
2532 	struct spdk_iov_xfer ix;
2533 
2534 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2535 
2536 	if (offset < page_size) {
2537 		if (ctrlr->subsys->zone_append_supported) {
2538 			cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_ZONE_APPEND] =
2539 				csupp_and_lbcc_effect_entry;
2540 		}
2541 		copy_len = spdk_min(page_size - offset, length);
2542 		spdk_iov_xfer_from_buf(&ix, (char *)(&cmds_and_effect_log_page) + offset, copy_len);
2543 	}
2544 }
2545 
2546 static void
2547 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2548 		struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2549 {
2550 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
2551 	struct spdk_nvmf_reservation_log *log, *log_tmp;
2552 	struct spdk_iov_xfer ix;
2553 
2554 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2555 
2556 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
2557 	/* No available log, return zeroed log pages */
2558 	if (!ctrlr->num_avail_log_pages) {
2559 		return;
2560 	}
2561 
2562 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
2563 	if (offset >= avail_log_len) {
2564 		return;
2565 	}
2566 
2567 	next_pos = 0;
2568 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
2569 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
2570 		ctrlr->num_avail_log_pages--;
2571 
2572 		next_pos += unit_log_len;
2573 		if (next_pos > offset) {
2574 			copy_len = spdk_min(next_pos - offset, length);
2575 			spdk_iov_xfer_from_buf(&ix, &log->log, copy_len);
2576 			length -= copy_len;
2577 			offset += copy_len;
2578 		}
2579 		free(log);
2580 
2581 		if (length == 0) {
2582 			break;
2583 		}
2584 	}
2585 
2586 	if (!rae) {
2587 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT);
2588 	}
2589 	return;
2590 }
2591 
2592 static int
2593 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
2594 {
2595 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2596 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2597 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2598 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2599 	struct spdk_nvme_transport_id cmd_source_trid;
2600 	uint64_t offset, len;
2601 	uint32_t rae, numdl, numdu;
2602 	uint8_t lid;
2603 
2604 	if (req->iovcnt < 1) {
2605 		SPDK_DEBUGLOG(nvmf, "get log command with no buffer\n");
2606 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2607 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2608 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2609 	}
2610 
2611 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
2612 	if (offset & 3) {
2613 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
2614 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2615 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2616 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2617 	}
2618 
2619 	rae = cmd->cdw10_bits.get_log_page.rae;
2620 	numdl = cmd->cdw10_bits.get_log_page.numdl;
2621 	numdu = cmd->cdw11_bits.get_log_page.numdu;
2622 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
2623 	if (len > req->length) {
2624 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
2625 			    len, req->length);
2626 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2627 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2628 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2629 	}
2630 
2631 	lid = cmd->cdw10_bits.get_log_page.lid;
2632 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n",
2633 		      lid, offset, len, rae);
2634 
2635 	if (spdk_nvmf_subsystem_is_discovery(subsystem)) {
2636 		switch (lid) {
2637 		case SPDK_NVME_LOG_DISCOVERY:
2638 			if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &cmd_source_trid)) {
2639 				SPDK_ERRLOG("Failed to get LOG_DISCOVERY source trid\n");
2640 				response->status.sct = SPDK_NVME_SCT_GENERIC;
2641 				response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
2642 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2643 			}
2644 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt,
2645 						    offset, len, &cmd_source_trid);
2646 			if (!rae) {
2647 				nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT);
2648 			}
2649 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2650 		default:
2651 			goto invalid_log_page;
2652 		}
2653 	} else {
2654 		if (offset > len) {
2655 			SPDK_ERRLOG("Get log page: offset (%" PRIu64 ") > len (%" PRIu64 ")\n",
2656 				    offset, len);
2657 			response->status.sct = SPDK_NVME_SCT_GENERIC;
2658 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2659 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2660 		}
2661 
2662 		switch (lid) {
2663 		case SPDK_NVME_LOG_ERROR:
2664 			nvmf_get_error_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2665 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2666 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
2667 			/* TODO: actually fill out log page data */
2668 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2669 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
2670 			nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len);
2671 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2672 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
2673 			if (subsystem->flags.ana_reporting) {
2674 				nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2675 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2676 			} else {
2677 				goto invalid_log_page;
2678 			}
2679 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
2680 			nvmf_get_cmds_and_effects_log_page(ctrlr, req->iov, req->iovcnt, offset, len);
2681 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2682 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
2683 			nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2684 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2685 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
2686 			nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2687 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2688 		default:
2689 			goto invalid_log_page;
2690 		}
2691 	}
2692 
2693 invalid_log_page:
2694 	SPDK_INFOLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid);
2695 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2696 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2697 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2698 }
2699 
2700 static struct spdk_nvmf_ns *
2701 _nvmf_ctrlr_get_ns_safe(struct spdk_nvmf_ctrlr *ctrlr,
2702 			uint32_t nsid,
2703 			struct spdk_nvme_cpl *rsp)
2704 {
2705 	struct spdk_nvmf_ns *ns;
2706 	if (nsid == 0 || nsid > ctrlr->subsys->max_nsid) {
2707 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", nsid);
2708 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2709 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2710 		return NULL;
2711 	}
2712 
2713 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
2714 	if (ns == NULL || ns->bdev == NULL) {
2715 		/*
2716 		 * Inactive namespaces should return a zero filled data structure.
2717 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2718 		 * so we can just return early here.
2719 		 */
2720 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", nsid);
2721 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2722 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2723 		return NULL;
2724 	}
2725 	return ns;
2726 }
2727 
2728 int
2729 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2730 			    struct spdk_nvme_cmd *cmd,
2731 			    struct spdk_nvme_cpl *rsp,
2732 			    struct spdk_nvme_ns_data *nsdata)
2733 {
2734 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2735 	struct spdk_nvmf_ns *ns;
2736 	uint32_t max_num_blocks, format_index;
2737 	enum spdk_nvme_ana_state ana_state;
2738 
2739 	ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp);
2740 	if (ns == NULL) {
2741 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2742 	}
2743 
2744 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2745 
2746 	assert(ctrlr->admin_qpair);
2747 
2748 	format_index = spdk_nvme_ns_get_format_index(nsdata);
2749 
2750 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2751 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2752 			 (1U << nsdata->lbaf[format_index].lbads);
2753 	if (nsdata->noiob > max_num_blocks) {
2754 		nsdata->noiob = max_num_blocks;
2755 	}
2756 
2757 	/* Set NOWS equal to Controller MDTS */
2758 	if (nsdata->nsfeat.optperf) {
2759 		nsdata->nows = max_num_blocks - 1;
2760 	}
2761 
2762 	if (subsystem->flags.ana_reporting) {
2763 		assert(ns->anagrpid - 1 < subsystem->max_nsid);
2764 		nsdata->anagrpid = ns->anagrpid;
2765 
2766 		ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2767 		if (ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2768 		    ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2769 			nsdata->nuse = 0;
2770 		}
2771 	}
2772 
2773 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2774 }
2775 
2776 static void
2777 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2778 			 struct spdk_nvme_ctrlr_data *cdata)
2779 {
2780 	cdata->oacs = ctrlr->cdata.oacs;
2781 
2782 	cdata->oacs.virtualization_management =
2783 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2784 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2785 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2786 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2787 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2788 	cdata->oacs.device_self_test =
2789 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2790 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2791 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2792 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2793 			       NULL
2794 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2795 	cdata->oacs.format =
2796 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2797 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2798 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2799 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2800 				     NULL;
2801 }
2802 
2803 int
2804 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2805 {
2806 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2807 	struct spdk_nvmf_transport *transport;
2808 
2809 	/*
2810 	 * Common fields for discovery and NVM subsystems
2811 	 */
2812 	assert(ctrlr->admin_qpair);
2813 	transport = ctrlr->admin_qpair->transport;
2814 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2815 	assert((transport->opts.max_io_size % 4096) == 0);
2816 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2817 	cdata->cntlid = ctrlr->cntlid;
2818 	cdata->ver = ctrlr->vcprop.vs;
2819 	cdata->aerl = ctrlr->cdata.aerl;
2820 	cdata->lpa.edlp = 1;
2821 	cdata->elpe = 127;
2822 	cdata->maxcmd = transport->opts.max_queue_depth;
2823 	cdata->sgls = ctrlr->cdata.sgls;
2824 	cdata->fuses = ctrlr->cdata.fuses;
2825 	cdata->acwu = 0; /* ACWU is 0-based. */
2826 	if (subsystem->flags.ana_reporting) {
2827 		cdata->mnan = subsystem->max_nsid;
2828 	}
2829 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2830 
2831 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2832 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2833 
2834 
2835 	if (spdk_nvmf_subsystem_is_discovery(subsystem)) {
2836 		/*
2837 		 * NVM Discovery subsystem fields
2838 		 */
2839 		cdata->oaes.discovery_log_change_notices = 1;
2840 	} else {
2841 		cdata->vid = ctrlr->cdata.vid;
2842 		cdata->ssvid = ctrlr->cdata.ssvid;
2843 		cdata->ieee[0] = ctrlr->cdata.ieee[0];
2844 		cdata->ieee[1] = ctrlr->cdata.ieee[1];
2845 		cdata->ieee[2] = ctrlr->cdata.ieee[2];
2846 
2847 		/*
2848 		 * NVM subsystem fields (reserved for discovery subsystems)
2849 		 */
2850 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2851 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2852 		cdata->kas = ctrlr->cdata.kas;
2853 
2854 		cdata->rab = 6;
2855 		cdata->cmic.multi_port = 1;
2856 		cdata->cmic.multi_ctrlr = 1;
2857 		cdata->oaes.ns_attribute_notices = 1;
2858 		cdata->ctratt.bits.host_id_exhid_supported = 1;
2859 		cdata->ctratt.bits.fdps = ctrlr->subsys->fdp_supported;
2860 		/* We do not have any actual limitation to the number of abort commands.
2861 		 * We follow the recommendation by the NVMe specification.
2862 		 */
2863 		cdata->acl = NVMF_ABORT_COMMAND_LIMIT;
2864 		cdata->frmw.slot1_ro = 1;
2865 		cdata->frmw.num_slots = 1;
2866 
2867 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2868 
2869 		cdata->sqes.min = 6;
2870 		cdata->sqes.max = 6;
2871 		cdata->cqes.min = 4;
2872 		cdata->cqes.max = 4;
2873 		cdata->nn = subsystem->max_nsid;
2874 		cdata->vwc.present = 1;
2875 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
2876 
2877 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
2878 
2879 		cdata->oncs.compare = ctrlr->cdata.oncs.compare;
2880 		cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr);
2881 		cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr);
2882 		cdata->oncs.reservations = ctrlr->cdata.oncs.reservations;
2883 		cdata->oncs.copy = ctrlr->cdata.oncs.copy;
2884 		cdata->ocfs.copy_format0 = cdata->oncs.copy;
2885 		if (subsystem->flags.ana_reporting) {
2886 			/* Asymmetric Namespace Access Reporting is supported. */
2887 			cdata->cmic.ana_reporting = 1;
2888 			cdata->oaes.ana_change_notices = 1;
2889 
2890 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
2891 			/* ANA Change state is not used, and ANA Persistent Loss state
2892 			 * is not supported for now.
2893 			 */
2894 			cdata->anacap.ana_optimized_state = 1;
2895 			cdata->anacap.ana_non_optimized_state = 1;
2896 			cdata->anacap.ana_inaccessible_state = 1;
2897 			/* ANAGRPID does not change while namespace is attached to controller */
2898 			cdata->anacap.no_change_anagrpid = 1;
2899 			cdata->anagrpmax = subsystem->max_nsid;
2900 			cdata->nanagrpid = subsystem->max_nsid;
2901 		}
2902 
2903 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
2904 
2905 		assert(subsystem->tgt != NULL);
2906 		cdata->crdt[0] = subsystem->tgt->crdt[0];
2907 		cdata->crdt[1] = subsystem->tgt->crdt[1];
2908 		cdata->crdt[2] = subsystem->tgt->crdt[2];
2909 
2910 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
2911 			      cdata->nvmf_specific.ioccsz);
2912 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
2913 			      cdata->nvmf_specific.iorcsz);
2914 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
2915 			      cdata->nvmf_specific.icdoff);
2916 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
2917 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
2918 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
2919 			      cdata->nvmf_specific.msdbd);
2920 	}
2921 
2922 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2923 }
2924 
2925 static int
2926 nvmf_ns_identify_iocs_zns(struct spdk_nvmf_ns *ns,
2927 			  struct spdk_nvme_cmd *cmd,
2928 			  struct spdk_nvme_cpl *rsp,
2929 			  struct spdk_nvme_zns_ns_data *nsdata_zns)
2930 {
2931 	nsdata_zns->zoc.variable_zone_capacity = 0;
2932 	nsdata_zns->zoc.zone_active_excursions = 0;
2933 	nsdata_zns->ozcs.read_across_zone_boundaries = 1;
2934 	/* Underflowing the zero based mar and mor bdev helper results in the correct
2935 	   value of FFFFFFFFh. */
2936 	nsdata_zns->mar = spdk_bdev_get_max_active_zones(ns->bdev) - 1;
2937 	nsdata_zns->mor = spdk_bdev_get_max_open_zones(ns->bdev) - 1;
2938 	nsdata_zns->rrl = 0;
2939 	nsdata_zns->frl = 0;
2940 	nsdata_zns->lbafe[0].zsze = spdk_bdev_get_zone_size(ns->bdev);
2941 
2942 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2943 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2944 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2945 }
2946 
2947 int
2948 spdk_nvmf_ns_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr,
2949 				    struct spdk_nvme_cmd *cmd,
2950 				    struct spdk_nvme_cpl *rsp,
2951 				    void *nsdata,
2952 				    size_t nsdata_size)
2953 {
2954 	uint8_t csi = cmd->cdw11_bits.identify.csi;
2955 	struct spdk_nvmf_ns *ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp);
2956 
2957 	memset(nsdata, 0, nsdata_size);
2958 
2959 	if (ns == NULL) {
2960 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2961 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2962 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2963 	}
2964 
2965 	switch (csi) {
2966 	case SPDK_NVME_CSI_ZNS:
2967 		return nvmf_ns_identify_iocs_zns(ns, cmd, rsp, nsdata);
2968 	default:
2969 		break;
2970 	}
2971 
2972 	SPDK_DEBUGLOG(nvmf,
2973 		      "Returning zero filled struct for the iocs specific ns "
2974 		      "identify command and CSI 0x%02x\n",
2975 		      csi);
2976 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2977 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2978 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2979 }
2980 
2981 static int
2982 nvmf_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ctrlr *ctrlr,
2983 			     struct spdk_nvme_cmd *cmd,
2984 			     struct spdk_nvme_cpl *rsp,
2985 			     struct spdk_nvme_nvm_ctrlr_data *cdata_nvm)
2986 {
2987 	/* The unit of max_write_zeroes_size_kib is KiB.
2988 	 * The unit of wzsl is the minimum memory page size(2 ^ (12 + CAP.MPSMIN) bytes)
2989 	 * and is reported as a power of two (2^n).
2990 	 */
2991 	cdata_nvm->wzsl = spdk_u64log2(ctrlr->subsys->max_write_zeroes_size_kib >>
2992 				       (2 + ctrlr->vcprop.cap.bits.mpsmin));
2993 
2994 	/* The unit of max_discard_size_kib is KiB.
2995 	 * The dmrsl indicates the maximum number of logical blocks for
2996 	 * dataset management command.
2997 	 */
2998 	cdata_nvm->dmrsl = ctrlr->subsys->max_discard_size_kib << 1;
2999 	cdata_nvm->dmrl = 1;
3000 
3001 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3002 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3003 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3004 }
3005 
3006 static int
3007 nvmf_ctrlr_identify_iocs_zns(struct spdk_nvmf_ctrlr *ctrlr,
3008 			     struct spdk_nvme_cmd *cmd,
3009 			     struct spdk_nvme_cpl *rsp,
3010 			     struct spdk_nvme_zns_ctrlr_data *cdata_zns)
3011 {
3012 	/* The unit of max_zone_append_size_kib is KiB.
3013 	The unit of zasl is the minimum memory page size
3014 	(2 ^ (12 + CAP.MPSMIN) KiB)
3015 	and is reported as a power of two (2^n). */
3016 	cdata_zns->zasl = spdk_u64log2(ctrlr->subsys->max_zone_append_size_kib >>
3017 				       (12 + ctrlr->vcprop.cap.bits.mpsmin));
3018 
3019 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3020 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3021 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3022 }
3023 
3024 int
3025 spdk_nvmf_ctrlr_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr,
3026 				       struct spdk_nvme_cmd *cmd,
3027 				       struct spdk_nvme_cpl *rsp,
3028 				       void *cdata,
3029 				       size_t cdata_size)
3030 {
3031 	uint8_t csi = cmd->cdw11_bits.identify.csi;
3032 
3033 	memset(cdata, 0, cdata_size);
3034 
3035 	switch (csi) {
3036 	case SPDK_NVME_CSI_NVM:
3037 		return nvmf_ctrlr_identify_iocs_nvm(ctrlr, cmd, rsp, cdata);
3038 	case SPDK_NVME_CSI_ZNS:
3039 		return nvmf_ctrlr_identify_iocs_zns(ctrlr, cmd, rsp, cdata);
3040 	default:
3041 		break;
3042 	}
3043 
3044 	SPDK_DEBUGLOG(nvmf,
3045 		      "Returning zero filled struct for the iocs specific ctrlr "
3046 		      "identify command and CSI 0x%02x\n",
3047 		      csi);
3048 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3049 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3050 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3051 }
3052 
3053 static int
3054 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_ctrlr *ctrlr,
3055 				   struct spdk_nvme_cmd *cmd,
3056 				   struct spdk_nvme_cpl *rsp,
3057 				   struct spdk_nvme_ns_list *ns_list)
3058 {
3059 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
3060 	struct spdk_nvmf_ns *ns;
3061 	uint32_t count = 0;
3062 
3063 	if (cmd->nsid >= 0xfffffffeUL) {
3064 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
3065 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3066 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3067 	}
3068 
3069 	memset(ns_list, 0, sizeof(*ns_list));
3070 
3071 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
3072 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
3073 		if (ns->opts.nsid <= cmd->nsid || !nvmf_ctrlr_ns_is_visible(ctrlr, ns->opts.nsid)) {
3074 			continue;
3075 		}
3076 
3077 		ns_list->ns_list[count++] = ns->opts.nsid;
3078 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
3079 			break;
3080 		}
3081 	}
3082 
3083 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3084 }
3085 
3086 static void
3087 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
3088 		enum spdk_nvme_nidt type,
3089 		const void *data, size_t data_size)
3090 {
3091 	struct spdk_nvme_ns_id_desc *desc;
3092 	size_t desc_size = sizeof(*desc) + data_size;
3093 
3094 	/*
3095 	 * These should never fail in practice, since all valid NS ID descriptors
3096 	 * should be defined so that they fit in the available 4096-byte buffer.
3097 	 */
3098 	assert(data_size > 0);
3099 	assert(data_size <= UINT8_MAX);
3100 	assert(desc_size < *buf_remain);
3101 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
3102 		return;
3103 	}
3104 
3105 	desc = *buf_ptr;
3106 	desc->nidt = type;
3107 	desc->nidl = data_size;
3108 	memcpy(desc->nid, data, data_size);
3109 
3110 	*buf_ptr += desc_size;
3111 	*buf_remain -= desc_size;
3112 }
3113 
3114 static int
3115 nvmf_ctrlr_identify_ns_id_descriptor_list(
3116 	struct spdk_nvmf_ctrlr *ctrlr,
3117 	struct spdk_nvme_cmd *cmd,
3118 	struct spdk_nvme_cpl *rsp,
3119 	void *id_desc_list, size_t id_desc_list_size)
3120 {
3121 	struct spdk_nvmf_ns *ns;
3122 	size_t buf_remain = id_desc_list_size;
3123 	void *buf_ptr = id_desc_list;
3124 
3125 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
3126 	if (ns == NULL || ns->bdev == NULL) {
3127 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3128 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3129 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3130 	}
3131 
3132 #define ADD_ID_DESC(type, data, size) \
3133 	do { \
3134 		if (!spdk_mem_all_zero(data, size)) { \
3135 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
3136 		} \
3137 	} while (0)
3138 
3139 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
3140 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
3141 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
3142 	ADD_ID_DESC(SPDK_NVME_NIDT_CSI, &ns->csi, sizeof(uint8_t));
3143 
3144 	/*
3145 	 * The list is automatically 0-terminated, both in the temporary buffer
3146 	 * used by nvmf_ctrlr_identify(), and the eventual iov destination -
3147 	 * controller to host buffers in admin commands always get zeroed in
3148 	 * nvmf_ctrlr_process_admin_cmd().
3149 	 */
3150 
3151 #undef ADD_ID_DESC
3152 
3153 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3154 }
3155 
3156 static int
3157 nvmf_ctrlr_identify_iocs(struct spdk_nvmf_ctrlr *ctrlr,
3158 			 struct spdk_nvme_cmd *cmd,
3159 			 struct spdk_nvme_cpl *rsp,
3160 			 void *cdata, size_t cdata_size)
3161 {
3162 	struct spdk_nvme_iocs_vector *vector;
3163 	struct spdk_nvmf_ns *ns;
3164 
3165 	if (cdata_size < sizeof(struct spdk_nvme_iocs_vector)) {
3166 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3167 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3168 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3169 	}
3170 
3171 	/* For now we only support this command sent to the current
3172 	 * controller.
3173 	 */
3174 	if (cmd->cdw10_bits.identify.cntid != 0xFFFF &&
3175 	    cmd->cdw10_bits.identify.cntid != ctrlr->cntlid) {
3176 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3177 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3178 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3179 	}
3180 	memset(cdata, 0, cdata_size);
3181 
3182 	vector = cdata;
3183 	vector->nvm = 1;
3184 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
3185 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
3186 		if (ns->bdev == NULL) {
3187 			continue;
3188 		}
3189 		if (spdk_bdev_is_zoned(ns->bdev)) {
3190 			vector->zns = 1;
3191 		}
3192 	}
3193 
3194 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3195 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3196 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3197 }
3198 
3199 static int
3200 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
3201 {
3202 	uint8_t cns;
3203 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3204 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3205 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3206 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
3207 	int ret = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3208 	char tmpbuf[SPDK_NVME_IDENTIFY_BUFLEN] = "";
3209 	struct spdk_iov_xfer ix;
3210 
3211 	if (req->iovcnt < 1 || req->length < SPDK_NVME_IDENTIFY_BUFLEN) {
3212 		SPDK_DEBUGLOG(nvmf, "identify command with invalid buffer\n");
3213 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3214 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3215 		return ret;
3216 	}
3217 
3218 	cns = cmd->cdw10_bits.identify.cns;
3219 
3220 	if (spdk_nvmf_subsystem_is_discovery(subsystem) &&
3221 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
3222 		/* Discovery controllers only support Identify Controller */
3223 		goto invalid_cns;
3224 	}
3225 
3226 	/*
3227 	 * We must use a temporary buffer: it's entirely possible the out buffer
3228 	 * is split across more than one IOV.
3229 	 */
3230 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3231 
3232 	SPDK_DEBUGLOG(nvmf, "Received identify command with CNS 0x%02x\n", cns);
3233 
3234 	switch (cns) {
3235 	case SPDK_NVME_IDENTIFY_NS:
3236 		ret = spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, (void *)&tmpbuf);
3237 		break;
3238 	case SPDK_NVME_IDENTIFY_CTRLR:
3239 		ret = spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, (void *)&tmpbuf);
3240 		break;
3241 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
3242 		ret = nvmf_ctrlr_identify_active_ns_list(ctrlr, cmd, rsp, (void *)&tmpbuf);
3243 		break;
3244 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
3245 		ret = nvmf_ctrlr_identify_ns_id_descriptor_list(ctrlr, cmd, rsp,
3246 				tmpbuf, req->length);
3247 		break;
3248 	case SPDK_NVME_IDENTIFY_NS_IOCS:
3249 		ret = spdk_nvmf_ns_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length);
3250 		break;
3251 	case SPDK_NVME_IDENTIFY_CTRLR_IOCS:
3252 		ret = spdk_nvmf_ctrlr_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length);
3253 		break;
3254 	case SPDK_NVME_IDENTIFY_IOCS:
3255 		ret = nvmf_ctrlr_identify_iocs(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length);
3256 		break;
3257 	default:
3258 		goto invalid_cns;
3259 	}
3260 
3261 	if (ret == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3262 		spdk_iov_xfer_from_buf(&ix, tmpbuf, sizeof(tmpbuf));
3263 	}
3264 
3265 	return ret;
3266 
3267 invalid_cns:
3268 	SPDK_DEBUGLOG(nvmf, "Identify command with unsupported CNS 0x%02x\n", cns);
3269 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3270 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3271 	return ret;
3272 }
3273 
3274 static bool
3275 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
3276 {
3277 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3278 	struct spdk_nvmf_request *req;
3279 	int i;
3280 
3281 	if (!nvmf_qpair_is_admin_queue(qpair)) {
3282 		return false;
3283 	}
3284 
3285 	assert(spdk_get_thread() == ctrlr->thread);
3286 
3287 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3288 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
3289 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
3290 			req = ctrlr->aer_req[i];
3291 			ctrlr->aer_req[i] = NULL;
3292 			ctrlr->nr_aer_reqs--;
3293 
3294 			/* Move the last req to the aborting position for making aer_reqs
3295 			 * in continuous
3296 			 */
3297 			if (i < ctrlr->nr_aer_reqs) {
3298 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
3299 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3300 			}
3301 
3302 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3303 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3304 			_nvmf_request_complete(req);
3305 			return true;
3306 		}
3307 	}
3308 
3309 	return false;
3310 }
3311 
3312 void
3313 nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair)
3314 {
3315 	struct spdk_nvmf_request *req, *tmp;
3316 
3317 	TAILQ_FOREACH_SAFE(req, &qpair->outstanding, link, tmp) {
3318 		if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) {
3319 			/* Zero-copy requests are kept on the outstanding queue from the moment
3320 			 * zcopy_start is sent until a zcopy_end callback is received.  Therefore,
3321 			 * we can't remove them from the outstanding queue here, but need to rely on
3322 			 * the transport to do a zcopy_end to release their buffers and, in turn,
3323 			 * remove them from the queue.
3324 			 */
3325 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3326 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3327 			nvmf_transport_req_free(req);
3328 		}
3329 	}
3330 }
3331 
3332 static void
3333 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
3334 {
3335 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
3336 
3337 	if (nvmf_qpair_abort_aer(qpair, cid)) {
3338 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
3339 			      qpair->ctrlr, qpair->qid, cid);
3340 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
3341 
3342 		spdk_nvmf_request_complete(req);
3343 		return;
3344 	}
3345 
3346 	nvmf_transport_qpair_abort_request(qpair, req);
3347 }
3348 
3349 static void
3350 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
3351 {
3352 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
3353 
3354 	if (status == 0) {
3355 		/* There was no qpair whose ID matches SQID of the abort command.
3356 		 * Hence call _nvmf_request_complete() here.
3357 		 */
3358 		_nvmf_request_complete(req);
3359 	}
3360 }
3361 
3362 static void
3363 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
3364 {
3365 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
3366 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
3367 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
3368 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
3369 	struct spdk_nvmf_qpair *qpair;
3370 
3371 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
3372 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
3373 			/* Found the qpair */
3374 
3375 			nvmf_qpair_abort_request(qpair, req);
3376 
3377 			/* Return -1 for the status so the iteration across threads stops. */
3378 			spdk_for_each_channel_continue(i, -1);
3379 			return;
3380 		}
3381 	}
3382 
3383 	spdk_for_each_channel_continue(i, 0);
3384 }
3385 
3386 static int
3387 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
3388 {
3389 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3390 
3391 	rsp->cdw0 = 1U; /* Command not aborted */
3392 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3393 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3394 
3395 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
3396 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
3397 			      nvmf_ctrlr_abort_on_pg,
3398 			      req,
3399 			      nvmf_ctrlr_abort_done
3400 			     );
3401 
3402 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3403 }
3404 
3405 int
3406 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
3407 {
3408 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
3409 	struct spdk_bdev *bdev;
3410 	struct spdk_bdev_desc *desc;
3411 	struct spdk_io_channel *ch;
3412 	int rc;
3413 
3414 	assert(req_to_abort != NULL);
3415 
3416 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
3417 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
3418 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
3419 	}
3420 
3421 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
3422 					&bdev, &desc, &ch);
3423 	if (rc != 0) {
3424 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3425 	}
3426 
3427 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
3428 }
3429 
3430 static int
3431 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
3432 {
3433 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3434 
3435 	rsp->cdw0 = cdw0;
3436 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3437 }
3438 
3439 /* we have to use the typedef in the function declaration to appease astyle. */
3440 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
3441 
3442 static spdk_nvme_path_status_code_t
3443 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
3444 {
3445 	switch (ana_state) {
3446 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3447 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
3448 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3449 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3450 	case SPDK_NVME_ANA_CHANGE_STATE:
3451 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
3452 	default:
3453 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
3454 	}
3455 }
3456 
3457 static int
3458 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
3459 {
3460 	uint8_t feature;
3461 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3462 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3463 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3464 	enum spdk_nvme_ana_state ana_state;
3465 
3466 	feature = cmd->cdw10_bits.get_features.fid;
3467 
3468 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
3469 		/*
3470 		 * Features supported by Discovery controller
3471 		 */
3472 		switch (feature) {
3473 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3474 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3475 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3476 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3477 		default:
3478 			SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3479 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3480 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3481 		}
3482 	}
3483 	/*
3484 	 * Process Get Features command for non-discovery controller
3485 	 */
3486 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3487 	switch (ana_state) {
3488 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3489 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3490 	case SPDK_NVME_ANA_CHANGE_STATE:
3491 		switch (feature) {
3492 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
3493 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3494 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3495 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3496 			response->status.sct = SPDK_NVME_SCT_PATH;
3497 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3498 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3499 		default:
3500 			break;
3501 		}
3502 		break;
3503 	default:
3504 		break;
3505 	}
3506 
3507 	switch (feature) {
3508 	case SPDK_NVME_FEAT_ARBITRATION:
3509 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
3510 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3511 		return get_features_generic(req, ctrlr->feat.power_management.raw);
3512 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3513 		return nvmf_ctrlr_get_features_temperature_threshold(req);
3514 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3515 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
3516 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3517 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
3518 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3519 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
3520 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3521 		return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw);
3522 	case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION:
3523 		return nvmf_ctrlr_get_features_interrupt_vector_configuration(req);
3524 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3525 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
3526 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3527 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3528 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3529 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3530 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3531 		return nvmf_ctrlr_get_features_host_identifier(req);
3532 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3533 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
3534 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3535 		return nvmf_ctrlr_get_features_reservation_persistence(req);
3536 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3537 		return nvmf_ctrlr_get_features_host_behavior_support(req);
3538 	default:
3539 		SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3540 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3541 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3542 	}
3543 }
3544 
3545 static int
3546 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
3547 {
3548 	uint8_t feature, save;
3549 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3550 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3551 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3552 	enum spdk_nvme_ana_state ana_state;
3553 	/*
3554 	 * Features are not saveable by the controller as indicated by
3555 	 * ONCS field of the Identify Controller data.
3556 	 * */
3557 	save = cmd->cdw10_bits.set_features.sv;
3558 	if (save) {
3559 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
3560 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3561 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3562 	}
3563 
3564 	feature = cmd->cdw10_bits.set_features.fid;
3565 
3566 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
3567 		/*
3568 		 * Features supported by Discovery controller
3569 		 */
3570 		switch (feature) {
3571 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3572 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
3573 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3574 			return nvmf_ctrlr_set_features_async_event_configuration(req);
3575 		default:
3576 			SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3577 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3578 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3579 		}
3580 	}
3581 	/*
3582 	 * Process Set Features command for non-discovery controller
3583 	 */
3584 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3585 	switch (ana_state) {
3586 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3587 	case SPDK_NVME_ANA_CHANGE_STATE:
3588 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
3589 			response->status.sct = SPDK_NVME_SCT_PATH;
3590 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3591 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3592 		} else {
3593 			switch (feature) {
3594 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
3595 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3596 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3597 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3598 				response->status.sct = SPDK_NVME_SCT_PATH;
3599 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3600 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3601 			default:
3602 				break;
3603 			}
3604 		}
3605 		break;
3606 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3607 		response->status.sct = SPDK_NVME_SCT_PATH;
3608 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3609 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3610 	default:
3611 		break;
3612 	}
3613 
3614 	switch (feature) {
3615 	case SPDK_NVME_FEAT_ARBITRATION:
3616 		return nvmf_ctrlr_set_features_arbitration(req);
3617 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3618 		return nvmf_ctrlr_set_features_power_management(req);
3619 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3620 		return nvmf_ctrlr_set_features_temperature_threshold(req);
3621 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3622 		return nvmf_ctrlr_set_features_error_recovery(req);
3623 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3624 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
3625 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3626 		return nvmf_ctrlr_set_features_number_of_queues(req);
3627 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3628 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3629 		response->status.sc = SPDK_NVME_SC_FEATURE_NOT_CHANGEABLE;
3630 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3631 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3632 		return nvmf_ctrlr_set_features_write_atomicity(req);
3633 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3634 		return nvmf_ctrlr_set_features_async_event_configuration(req);
3635 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3636 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
3637 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3638 		return nvmf_ctrlr_set_features_host_identifier(req);
3639 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3640 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
3641 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3642 		return nvmf_ctrlr_set_features_reservation_persistence(req);
3643 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3644 		return nvmf_ctrlr_set_features_host_behavior_support(req);
3645 	default:
3646 		SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3647 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3648 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3649 	}
3650 }
3651 
3652 static int
3653 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
3654 {
3655 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3656 
3657 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
3658 	/*
3659 	 * To handle keep alive just clear or reset the
3660 	 * ctrlr based keep alive duration counter.
3661 	 * When added, a separate timer based process
3662 	 * will monitor if the time since last recorded
3663 	 * keep alive has exceeded the max duration and
3664 	 * take appropriate action.
3665 	 */
3666 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
3667 
3668 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3669 }
3670 
3671 int
3672 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
3673 {
3674 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3675 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3676 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3677 	struct spdk_nvmf_subsystem_poll_group *sgroup;
3678 	int rc;
3679 
3680 	assert(ctrlr != NULL);
3681 	if (cmd->opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST) {
3682 		/* We do not want to treat AERs as outstanding commands,
3683 		 * so decrement mgmt_io_outstanding here to offset
3684 		 * the increment that happened prior to this call.
3685 		 */
3686 		sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
3687 		assert(sgroup != NULL);
3688 		sgroup->mgmt_io_outstanding--;
3689 	}
3690 
3691 	assert(spdk_get_thread() == ctrlr->thread);
3692 
3693 	if (cmd->fuse != 0) {
3694 		/* Fused admin commands are not supported. */
3695 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3696 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3697 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3698 	}
3699 
3700 	if (ctrlr->vcprop.cc.bits.en != 1) {
3701 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
3702 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3703 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3704 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3705 	}
3706 
3707 	if (req->iovcnt && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
3708 		spdk_iov_memset(req->iov, req->iovcnt, 0);
3709 	}
3710 
3711 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
3712 		/* Discovery controllers only support these admin OPS. */
3713 		switch (cmd->opc) {
3714 		case SPDK_NVME_OPC_IDENTIFY:
3715 		case SPDK_NVME_OPC_GET_LOG_PAGE:
3716 		case SPDK_NVME_OPC_KEEP_ALIVE:
3717 		case SPDK_NVME_OPC_SET_FEATURES:
3718 		case SPDK_NVME_OPC_GET_FEATURES:
3719 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3720 			break;
3721 		default:
3722 			goto invalid_opcode;
3723 		}
3724 	}
3725 
3726 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
3727 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
3728 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
3729 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3730 			/* The handler took care of this command */
3731 			return rc;
3732 		}
3733 	}
3734 
3735 	switch (cmd->opc) {
3736 	case SPDK_NVME_OPC_GET_LOG_PAGE:
3737 		return nvmf_ctrlr_get_log_page(req);
3738 	case SPDK_NVME_OPC_IDENTIFY:
3739 		return nvmf_ctrlr_identify(req);
3740 	case SPDK_NVME_OPC_ABORT:
3741 		return nvmf_ctrlr_abort(req);
3742 	case SPDK_NVME_OPC_GET_FEATURES:
3743 		return nvmf_ctrlr_get_features(req);
3744 	case SPDK_NVME_OPC_SET_FEATURES:
3745 		return nvmf_ctrlr_set_features(req);
3746 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3747 		return nvmf_ctrlr_async_event_request(req);
3748 	case SPDK_NVME_OPC_KEEP_ALIVE:
3749 		return nvmf_ctrlr_keep_alive(req);
3750 
3751 	case SPDK_NVME_OPC_CREATE_IO_SQ:
3752 	case SPDK_NVME_OPC_CREATE_IO_CQ:
3753 	case SPDK_NVME_OPC_DELETE_IO_SQ:
3754 	case SPDK_NVME_OPC_DELETE_IO_CQ:
3755 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
3756 		goto invalid_opcode;
3757 
3758 	default:
3759 		goto invalid_opcode;
3760 	}
3761 
3762 invalid_opcode:
3763 	SPDK_INFOLOG(nvmf, "Unsupported admin opcode 0x%x\n", cmd->opc);
3764 	response->status.sct = SPDK_NVME_SCT_GENERIC;
3765 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3766 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3767 }
3768 
3769 static int
3770 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
3771 {
3772 	struct spdk_nvmf_qpair *qpair = req->qpair;
3773 	struct spdk_nvmf_capsule_cmd *cap_hdr;
3774 
3775 	cap_hdr = &req->cmd->nvmf_cmd;
3776 
3777 	if (qpair->ctrlr == NULL) {
3778 		/* No ctrlr established yet; the only valid command is Connect */
3779 		assert(cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT);
3780 		return nvmf_ctrlr_cmd_connect(req);
3781 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
3782 		/*
3783 		 * Controller session is established, and this is an admin queue.
3784 		 * Disallow Connect and allow other fabrics commands.
3785 		 */
3786 		switch (cap_hdr->fctype) {
3787 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
3788 			return nvmf_property_set(req);
3789 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
3790 			return nvmf_property_get(req);
3791 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_SEND:
3792 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_RECV:
3793 			return nvmf_auth_request_exec(req);
3794 		default:
3795 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
3796 				      cap_hdr->fctype);
3797 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3798 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3799 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3800 		}
3801 	} else {
3802 		/*
3803 		 * Controller session is established, and this is an I/O queue.
3804 		 * Disallow everything besides authentication commands.
3805 		 */
3806 		switch (cap_hdr->fctype) {
3807 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_SEND:
3808 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_RECV:
3809 			return nvmf_auth_request_exec(req);
3810 		default:
3811 			SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
3812 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3813 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3814 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3815 		}
3816 	}
3817 }
3818 
3819 static inline void
3820 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr,
3821 				     union spdk_nvme_async_event_completion *event)
3822 {
3823 	struct spdk_nvmf_async_event_completion *nvmf_event;
3824 
3825 	nvmf_event = calloc(1, sizeof(*nvmf_event));
3826 	if (!nvmf_event) {
3827 		SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n");
3828 		return;
3829 	}
3830 	nvmf_event->event.raw = event->raw;
3831 	STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link);
3832 }
3833 
3834 static inline int
3835 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
3836 				    union spdk_nvme_async_event_completion *event)
3837 {
3838 	struct spdk_nvmf_request *req;
3839 	struct spdk_nvme_cpl *rsp;
3840 
3841 	assert(spdk_get_thread() == ctrlr->thread);
3842 
3843 	/* If there is no outstanding AER request, queue the event.  Then
3844 	 * if an AER is later submitted, this event can be sent as a
3845 	 * response.
3846 	 */
3847 	if (ctrlr->nr_aer_reqs == 0) {
3848 		nvmf_ctrlr_queue_pending_async_event(ctrlr, event);
3849 		return 0;
3850 	}
3851 
3852 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
3853 	rsp = &req->rsp->nvme_cpl;
3854 
3855 	rsp->cdw0 = event->raw;
3856 
3857 	_nvmf_request_complete(req);
3858 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3859 
3860 	return 0;
3861 }
3862 
3863 int
3864 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
3865 {
3866 	union spdk_nvme_async_event_completion event = {0};
3867 
3868 	/* Users may disable the event notification */
3869 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
3870 		return 0;
3871 	}
3872 
3873 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) {
3874 		return 0;
3875 	}
3876 
3877 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3878 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
3879 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
3880 
3881 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3882 }
3883 
3884 int
3885 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3886 {
3887 	union spdk_nvme_async_event_completion event = {0};
3888 
3889 	/* Users may disable the event notification */
3890 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
3891 		return 0;
3892 	}
3893 
3894 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) {
3895 		return 0;
3896 	}
3897 
3898 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3899 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
3900 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
3901 
3902 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3903 }
3904 
3905 void
3906 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
3907 {
3908 	union spdk_nvme_async_event_completion event = {0};
3909 
3910 	if (!ctrlr->num_avail_log_pages) {
3911 		return;
3912 	}
3913 
3914 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) {
3915 		return;
3916 	}
3917 
3918 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
3919 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
3920 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
3921 
3922 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3923 }
3924 
3925 void
3926 nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx)
3927 {
3928 	union spdk_nvme_async_event_completion event = {0};
3929 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
3930 
3931 	/* Users may disable the event notification manually or
3932 	 * it may not be enabled due to keep alive timeout
3933 	 * not being set in connect command to discovery controller.
3934 	 */
3935 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
3936 		return;
3937 	}
3938 
3939 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) {
3940 		return;
3941 	}
3942 
3943 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3944 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
3945 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
3946 
3947 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3948 }
3949 
3950 int
3951 spdk_nvmf_ctrlr_async_event_error_event(struct spdk_nvmf_ctrlr *ctrlr,
3952 					enum spdk_nvme_async_event_info_error info)
3953 {
3954 	union spdk_nvme_async_event_completion event;
3955 
3956 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT)) {
3957 		return 0;
3958 	}
3959 
3960 	if (info > SPDK_NVME_ASYNC_EVENT_FW_IMAGE_LOAD) {
3961 		return 0;
3962 	}
3963 
3964 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_ERROR;
3965 	event.bits.log_page_identifier = SPDK_NVME_LOG_ERROR;
3966 	event.bits.async_event_info = info;
3967 
3968 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3969 }
3970 
3971 void
3972 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
3973 {
3974 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3975 	int i;
3976 
3977 	if (ctrlr == NULL || !nvmf_qpair_is_admin_queue(qpair)) {
3978 		return;
3979 	}
3980 
3981 	assert(spdk_get_thread() == ctrlr->thread);
3982 
3983 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3984 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
3985 		ctrlr->aer_req[i] = NULL;
3986 	}
3987 
3988 	ctrlr->nr_aer_reqs = 0;
3989 }
3990 
3991 void
3992 spdk_nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
3993 {
3994 	struct spdk_nvmf_request *req;
3995 	int i;
3996 
3997 	assert(spdk_get_thread() == ctrlr->thread);
3998 
3999 	if (!ctrlr->nr_aer_reqs) {
4000 		return;
4001 	}
4002 
4003 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
4004 		req = ctrlr->aer_req[i];
4005 
4006 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4007 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
4008 		_nvmf_request_complete(req);
4009 
4010 		ctrlr->aer_req[i] = NULL;
4011 	}
4012 
4013 	ctrlr->nr_aer_reqs = 0;
4014 }
4015 
4016 static void
4017 _nvmf_ctrlr_add_reservation_log(void *ctx)
4018 {
4019 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
4020 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
4021 
4022 	ctrlr->log_page_count++;
4023 
4024 	/* Maximum number of queued log pages is 255 */
4025 	if (ctrlr->num_avail_log_pages == 0xff) {
4026 		struct spdk_nvmf_reservation_log *entry;
4027 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
4028 		entry->log.log_page_count = ctrlr->log_page_count;
4029 		free(log);
4030 		return;
4031 	}
4032 
4033 	log->log.log_page_count = ctrlr->log_page_count;
4034 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
4035 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
4036 
4037 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
4038 }
4039 
4040 void
4041 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
4042 				  struct spdk_nvmf_ns *ns,
4043 				  enum spdk_nvme_reservation_notification_log_page_type type)
4044 {
4045 	struct spdk_nvmf_reservation_log *log;
4046 
4047 	switch (type) {
4048 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
4049 		return;
4050 	case SPDK_NVME_REGISTRATION_PREEMPTED:
4051 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
4052 			return;
4053 		}
4054 		break;
4055 	case SPDK_NVME_RESERVATION_RELEASED:
4056 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
4057 			return;
4058 		}
4059 		break;
4060 	case SPDK_NVME_RESERVATION_PREEMPTED:
4061 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
4062 			return;
4063 		}
4064 		break;
4065 	default:
4066 		return;
4067 	}
4068 
4069 	log = calloc(1, sizeof(*log));
4070 	if (!log) {
4071 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
4072 		return;
4073 	}
4074 	log->ctrlr = ctrlr;
4075 	log->log.type = type;
4076 	log->log.nsid = ns->nsid;
4077 
4078 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
4079 }
4080 
4081 /* Check from subsystem poll group's namespace information data structure */
4082 static bool
4083 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
4084 				 struct spdk_nvmf_ctrlr *ctrlr)
4085 {
4086 	uint32_t i;
4087 
4088 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
4089 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
4090 			return true;
4091 		}
4092 	}
4093 
4094 	return false;
4095 }
4096 
4097 /*
4098  * Check the NVMe command is permitted or not for current controller(Host).
4099  */
4100 static int
4101 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
4102 				  struct spdk_nvmf_ctrlr *ctrlr,
4103 				  struct spdk_nvmf_request *req)
4104 {
4105 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
4106 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
4107 	uint8_t status = SPDK_NVME_SC_SUCCESS;
4108 	uint8_t racqa;
4109 	bool is_registrant;
4110 
4111 	/* No valid reservation */
4112 	if (!rtype) {
4113 		return 0;
4114 	}
4115 
4116 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
4117 	/* All registrants type and current ctrlr is a valid registrant */
4118 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
4119 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
4120 		return 0;
4121 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
4122 		return 0;
4123 	}
4124 
4125 	/* Non-holder for current controller */
4126 	switch (cmd->opc) {
4127 	case SPDK_NVME_OPC_READ:
4128 	case SPDK_NVME_OPC_COMPARE:
4129 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
4130 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4131 			goto exit;
4132 		}
4133 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
4134 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
4135 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4136 		}
4137 		break;
4138 	case SPDK_NVME_OPC_FLUSH:
4139 	case SPDK_NVME_OPC_WRITE:
4140 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
4141 	case SPDK_NVME_OPC_WRITE_ZEROES:
4142 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
4143 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
4144 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
4145 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4146 			goto exit;
4147 		}
4148 		if (!is_registrant) {
4149 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4150 		}
4151 		break;
4152 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
4153 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
4154 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
4155 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4156 			goto exit;
4157 		}
4158 		if (!is_registrant) {
4159 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4160 		}
4161 		break;
4162 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
4163 		if (!is_registrant) {
4164 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4165 		}
4166 		break;
4167 	default:
4168 		break;
4169 	}
4170 
4171 exit:
4172 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4173 	req->rsp->nvme_cpl.status.sc = status;
4174 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
4175 		return -EPERM;
4176 	}
4177 
4178 	return 0;
4179 }
4180 
4181 static int
4182 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
4183 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
4184 {
4185 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
4186 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
4187 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
4188 	int rc;
4189 
4190 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
4191 		/* first fused operation (should be compare) */
4192 		if (first_fused_req != NULL) {
4193 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
4194 
4195 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
4196 
4197 			/* abort req->qpair->first_fused_request and continue with new fused command */
4198 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4199 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4200 			_nvmf_request_complete(first_fused_req);
4201 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
4202 			SPDK_ERRLOG("Wrong op code of fused operations\n");
4203 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4204 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
4205 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4206 		}
4207 
4208 		req->qpair->first_fused_req = req;
4209 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
4210 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
4211 		/* second fused operation (should be write) */
4212 		if (first_fused_req == NULL) {
4213 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
4214 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4215 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4216 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4217 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
4218 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
4219 
4220 			SPDK_ERRLOG("Wrong op code of fused operations\n");
4221 
4222 			/* abort req->qpair->first_fused_request and fail current command */
4223 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4224 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4225 			_nvmf_request_complete(first_fused_req);
4226 
4227 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4228 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
4229 			req->qpair->first_fused_req = NULL;
4230 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4231 		}
4232 
4233 		/* save request of first command to generate response later */
4234 		req->first_fused_req = first_fused_req;
4235 		req->first_fused = true;
4236 		req->qpair->first_fused_req = NULL;
4237 	} else {
4238 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
4239 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4240 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
4241 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4242 	}
4243 
4244 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
4245 
4246 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4247 		if (spdk_nvme_cpl_is_error(rsp)) {
4248 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
4249 
4250 			fused_response->status = rsp->status;
4251 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4252 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
4253 			/* Complete first of fused commands. Second will be completed by upper layer */
4254 			_nvmf_request_complete(first_fused_req);
4255 			req->first_fused_req = NULL;
4256 			req->first_fused = false;
4257 		}
4258 	}
4259 
4260 	return rc;
4261 }
4262 
4263 bool
4264 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req)
4265 {
4266 	struct spdk_nvmf_transport *transport = req->qpair->transport;
4267 	struct spdk_nvmf_ns *ns;
4268 
4269 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE);
4270 
4271 	if (!transport->opts.zcopy) {
4272 		return false;
4273 	}
4274 
4275 	if (nvmf_qpair_is_admin_queue(req->qpair)) {
4276 		/* Admin queue */
4277 		return false;
4278 	}
4279 
4280 	if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) &&
4281 	    (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) {
4282 		/* Not a READ or WRITE command */
4283 		return false;
4284 	}
4285 
4286 	if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) {
4287 		/* Fused commands dont use zcopy buffers */
4288 		return false;
4289 	}
4290 
4291 	ns = nvmf_ctrlr_get_ns(req->qpair->ctrlr, req->cmd->nvme_cmd.nsid);
4292 	if (ns == NULL || ns->bdev == NULL || !ns->zcopy) {
4293 		return false;
4294 	}
4295 
4296 	req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT;
4297 	return true;
4298 }
4299 
4300 void
4301 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req)
4302 {
4303 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
4304 
4305 	/* Set iovcnt to be the maximum number of iovs that the ZCOPY can use */
4306 	req->iovcnt = NVMF_REQ_MAX_BUFFERS;
4307 
4308 	spdk_nvmf_request_exec(req);
4309 }
4310 
4311 void
4312 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req, bool commit)
4313 {
4314 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE);
4315 	req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING;
4316 
4317 	nvmf_bdev_ctrlr_zcopy_end(req, commit);
4318 }
4319 
4320 int
4321 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
4322 {
4323 	uint32_t nsid;
4324 	struct spdk_nvmf_ns *ns;
4325 	struct spdk_bdev *bdev;
4326 	struct spdk_bdev_desc *desc;
4327 	struct spdk_io_channel *ch;
4328 	struct spdk_nvmf_poll_group *group = req->qpair->group;
4329 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
4330 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
4331 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
4332 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4333 	enum spdk_nvme_ana_state ana_state;
4334 
4335 	/* pre-set response details for this command */
4336 	response->status.sc = SPDK_NVME_SC_SUCCESS;
4337 	nsid = cmd->nsid;
4338 
4339 	assert(ctrlr != NULL);
4340 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
4341 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
4342 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4343 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
4344 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4345 	}
4346 
4347 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
4348 	if (spdk_unlikely(ns == NULL || ns->bdev == NULL)) {
4349 		SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid);
4350 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4351 		response->status.dnr = 1;
4352 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4353 	}
4354 
4355 	ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
4356 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
4357 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
4358 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
4359 			      ana_state);
4360 		response->status.sct = SPDK_NVME_SCT_PATH;
4361 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
4362 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4363 	}
4364 
4365 	if (spdk_likely(ctrlr->listener != NULL)) {
4366 		SPDK_DTRACE_PROBE3_TICKS(nvmf_request_io_exec_path, req,
4367 					 ctrlr->listener->trid->traddr,
4368 					 ctrlr->listener->trid->trsvcid);
4369 	}
4370 
4371 	/* scan-build falsely reporting dereference of null pointer */
4372 	assert(group != NULL && group->sgroups != NULL);
4373 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
4374 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
4375 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
4376 			      cmd->nsid, cmd->opc);
4377 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4378 	}
4379 
4380 	bdev = ns->bdev;
4381 	desc = ns->desc;
4382 	ch = ns_info->channel;
4383 
4384 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
4385 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
4386 	} else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) {
4387 		struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl;
4388 
4389 		SPDK_ERRLOG("Second fused cmd expected - failing first one (opcode:0x%x)\n",
4390 			    req->qpair->first_fused_req->cmd->nvmf_cmd.opcode);
4391 
4392 		/* abort req->qpair->first_fused_request and continue with new command */
4393 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4394 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4395 		_nvmf_request_complete(req->qpair->first_fused_req);
4396 		req->qpair->first_fused_req = NULL;
4397 	}
4398 
4399 	if (spdk_nvmf_request_using_zcopy(req)) {
4400 		assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
4401 		return nvmf_bdev_ctrlr_zcopy_start(bdev, desc, ch, req);
4402 	} else {
4403 		switch (cmd->opc) {
4404 		case SPDK_NVME_OPC_READ:
4405 			return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
4406 		case SPDK_NVME_OPC_WRITE:
4407 			return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
4408 		case SPDK_NVME_OPC_COMPARE:
4409 			return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
4410 		case SPDK_NVME_OPC_WRITE_ZEROES:
4411 			return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
4412 		case SPDK_NVME_OPC_FLUSH:
4413 			return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
4414 		case SPDK_NVME_OPC_DATASET_MANAGEMENT:
4415 			return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
4416 		case SPDK_NVME_OPC_RESERVATION_REGISTER:
4417 		case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
4418 		case SPDK_NVME_OPC_RESERVATION_RELEASE:
4419 		case SPDK_NVME_OPC_RESERVATION_REPORT:
4420 			spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
4421 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
4422 		case SPDK_NVME_OPC_COPY:
4423 			return nvmf_bdev_ctrlr_copy_cmd(bdev, desc, ch, req);
4424 		default:
4425 			return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
4426 		}
4427 	}
4428 }
4429 
4430 static void
4431 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
4432 {
4433 	if (spdk_unlikely(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING)) {
4434 		assert(qpair->state_cb != NULL);
4435 
4436 		if (TAILQ_EMPTY(&qpair->outstanding)) {
4437 			qpair->state_cb(qpair->state_cb_arg, 0);
4438 		}
4439 	}
4440 }
4441 
4442 int
4443 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
4444 {
4445 	struct spdk_nvmf_qpair *qpair = req->qpair;
4446 
4447 	TAILQ_REMOVE(&qpair->outstanding, req, link);
4448 	if (spdk_unlikely(nvmf_transport_req_free(req))) {
4449 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
4450 	}
4451 
4452 	nvmf_qpair_request_cleanup(qpair);
4453 
4454 	return 0;
4455 }
4456 
4457 static void
4458 _nvmf_request_complete(void *ctx)
4459 {
4460 	struct spdk_nvmf_request *req = ctx;
4461 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
4462 	struct spdk_nvmf_qpair *qpair;
4463 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4464 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4465 	bool is_aer = false;
4466 	uint32_t nsid;
4467 	bool paused;
4468 	uint8_t opcode;
4469 
4470 	rsp->sqid = 0;
4471 	rsp->status.p = 0;
4472 	rsp->cid = req->cmd->nvme_cmd.cid;
4473 	nsid = req->cmd->nvme_cmd.nsid;
4474 	opcode = req->cmd->nvmf_cmd.opcode;
4475 
4476 	qpair = req->qpair;
4477 	if (spdk_likely(qpair->ctrlr)) {
4478 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4479 		assert(sgroup != NULL);
4480 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
4481 		if (spdk_likely(qpair->qid != 0)) {
4482 			qpair->group->stat.completed_nvme_io++;
4483 		}
4484 
4485 		/*
4486 		 * Set the crd value.
4487 		 * If the the IO has any error, and dnr (DoNotRetry) is not 1,
4488 		 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT.
4489 		 */
4490 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp) &&
4491 				  rsp->status.dnr == 0 &&
4492 				  qpair->ctrlr->acre_enabled)) {
4493 			rsp->status.crd = 1;
4494 		}
4495 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4496 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4497 	}
4498 
4499 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4500 		spdk_nvme_print_completion(qpair->qid, rsp);
4501 	}
4502 
4503 	switch (req->zcopy_phase) {
4504 	case NVMF_ZCOPY_PHASE_NONE:
4505 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4506 		break;
4507 	case NVMF_ZCOPY_PHASE_INIT:
4508 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) {
4509 			req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED;
4510 			TAILQ_REMOVE(&qpair->outstanding, req, link);
4511 		} else {
4512 			req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE;
4513 		}
4514 		break;
4515 	case NVMF_ZCOPY_PHASE_EXECUTE:
4516 		break;
4517 	case NVMF_ZCOPY_PHASE_END_PENDING:
4518 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4519 		req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE;
4520 		break;
4521 	default:
4522 		SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase);
4523 		break;
4524 	}
4525 
4526 	if (spdk_unlikely(nvmf_transport_req_complete(req))) {
4527 		SPDK_ERRLOG("Transport request completion error!\n");
4528 	}
4529 
4530 	/* AER cmd is an exception */
4531 	if (spdk_likely(sgroup && !is_aer)) {
4532 		if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC ||
4533 				  nvmf_qpair_is_admin_queue(qpair))) {
4534 			assert(sgroup->mgmt_io_outstanding > 0);
4535 			sgroup->mgmt_io_outstanding--;
4536 		} else {
4537 			if (req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE ||
4538 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE ||
4539 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT_FAILED) {
4540 				/* End of request */
4541 
4542 				/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4543 				if (spdk_likely(nsid - 1 < sgroup->num_ns)) {
4544 					sgroup->ns_info[nsid - 1].io_outstanding--;
4545 				}
4546 			}
4547 		}
4548 
4549 		if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4550 				  sgroup->mgmt_io_outstanding == 0)) {
4551 			paused = true;
4552 			for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
4553 				ns_info = &sgroup->ns_info[nsid];
4554 
4555 				if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4556 				    ns_info->io_outstanding > 0) {
4557 					paused = false;
4558 					break;
4559 				}
4560 			}
4561 
4562 			if (paused) {
4563 				sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
4564 				sgroup->cb_fn(sgroup->cb_arg, 0);
4565 				sgroup->cb_fn = NULL;
4566 				sgroup->cb_arg = NULL;
4567 			}
4568 		}
4569 
4570 	}
4571 
4572 	nvmf_qpair_request_cleanup(qpair);
4573 }
4574 
4575 int
4576 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
4577 {
4578 	struct spdk_nvmf_qpair *qpair = req->qpair;
4579 
4580 	spdk_thread_exec_msg(qpair->group->thread, _nvmf_request_complete, req);
4581 
4582 	return 0;
4583 }
4584 
4585 SPDK_LOG_DEPRECATION_REGISTER(nvmf_request_exec_fabrics, "spdk_nvmf_request_exec_fabrics()",
4586 			      "v24.09", 1);
4587 void
4588 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req)
4589 {
4590 	SPDK_LOG_DEPRECATED(nvmf_request_exec_fabrics);
4591 
4592 	return spdk_nvmf_request_exec(req);
4593 }
4594 
4595 static bool
4596 nvmf_check_subsystem_active(struct spdk_nvmf_request *req)
4597 {
4598 	struct spdk_nvmf_qpair *qpair = req->qpair;
4599 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4600 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4601 	uint32_t nsid;
4602 
4603 	if (spdk_likely(qpair->ctrlr)) {
4604 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4605 		assert(sgroup != NULL);
4606 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4607 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4608 	}
4609 
4610 	/* Check if the subsystem is paused (if there is a subsystem) */
4611 	if (spdk_unlikely(sgroup == NULL)) {
4612 		return true;
4613 	}
4614 
4615 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC ||
4616 			  nvmf_qpair_is_admin_queue(qpair))) {
4617 		if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
4618 			/* The subsystem is not currently active. Queue this request. */
4619 			TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4620 			return false;
4621 		}
4622 		sgroup->mgmt_io_outstanding++;
4623 	} else {
4624 		nsid = req->cmd->nvme_cmd.nsid;
4625 
4626 		/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4627 		if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) {
4628 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4629 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4630 			req->rsp->nvme_cpl.status.dnr = 1;
4631 			TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4632 			_nvmf_request_complete(req);
4633 			return false;
4634 		}
4635 
4636 		ns_info = &sgroup->ns_info[nsid - 1];
4637 		if (spdk_unlikely(ns_info->channel == NULL)) {
4638 			/* This can can happen if host sends I/O to a namespace that is
4639 			 * in the process of being added, but before the full addition
4640 			 * process is complete.  Report invalid namespace in that case.
4641 			 */
4642 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4643 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4644 			req->rsp->nvme_cpl.status.dnr = 1;
4645 			TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4646 			ns_info->io_outstanding++;
4647 			_nvmf_request_complete(req);
4648 			return false;
4649 		}
4650 
4651 		if (spdk_unlikely(ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE)) {
4652 			/* The namespace is not currently active. Queue this request. */
4653 			TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4654 			return false;
4655 		}
4656 
4657 		ns_info->io_outstanding++;
4658 	}
4659 
4660 	return true;
4661 }
4662 
4663 static bool
4664 nvmf_check_qpair_active(struct spdk_nvmf_request *req)
4665 {
4666 	struct spdk_nvmf_qpair *qpair = req->qpair;
4667 	int sc, sct;
4668 
4669 	if (spdk_likely(qpair->state == SPDK_NVMF_QPAIR_ENABLED)) {
4670 		return true;
4671 	}
4672 
4673 	sct = SPDK_NVME_SCT_GENERIC;
4674 	sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
4675 
4676 	switch (qpair->state) {
4677 	case SPDK_NVMF_QPAIR_CONNECTING:
4678 		if (req->cmd->nvmf_cmd.opcode != SPDK_NVME_OPC_FABRIC) {
4679 			SPDK_ERRLOG("Received command 0x%x on qid %u before CONNECT\n",
4680 				    req->cmd->nvmf_cmd.opcode, qpair->qid);
4681 			break;
4682 		}
4683 		if (req->cmd->nvmf_cmd.fctype != SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
4684 			SPDK_ERRLOG("Received fctype 0x%x on qid %u before CONNECT\n",
4685 				    req->cmd->nvmf_cmd.fctype, qpair->qid);
4686 			break;
4687 		}
4688 		return true;
4689 	case SPDK_NVMF_QPAIR_AUTHENTICATING:
4690 		sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
4691 		sc = SPDK_NVMF_FABRIC_SC_AUTH_REQUIRED;
4692 		if (req->cmd->nvmf_cmd.opcode != SPDK_NVME_OPC_FABRIC) {
4693 			SPDK_ERRLOG("Received command 0x%x on qid %u before authentication\n",
4694 				    req->cmd->nvmf_cmd.opcode, qpair->qid);
4695 			break;
4696 		}
4697 		if (req->cmd->nvmf_cmd.fctype != SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_SEND &&
4698 		    req->cmd->nvmf_cmd.fctype != SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_RECV) {
4699 			SPDK_ERRLOG("Received fctype 0x%x on qid %u before authentication\n",
4700 				    req->cmd->nvmf_cmd.fctype, qpair->qid);
4701 			break;
4702 		}
4703 		return true;
4704 	default:
4705 		SPDK_ERRLOG("Received command 0x%x on qid %u in state %d\n",
4706 			    req->cmd->nvmf_cmd.opcode, qpair->qid, qpair->state);
4707 		break;
4708 	}
4709 
4710 	req->rsp->nvme_cpl.status.sct = sct;
4711 	req->rsp->nvme_cpl.status.sc = sc;
4712 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4713 	_nvmf_request_complete(req);
4714 
4715 	return false;
4716 }
4717 
4718 void
4719 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
4720 {
4721 	struct spdk_nvmf_qpair *qpair = req->qpair;
4722 	enum spdk_nvmf_request_exec_status status;
4723 
4724 	if (spdk_unlikely(!nvmf_check_subsystem_active(req))) {
4725 		return;
4726 	}
4727 	if (spdk_unlikely(!nvmf_check_qpair_active(req))) {
4728 		return;
4729 	}
4730 
4731 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4732 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
4733 	}
4734 
4735 	/* Place the request on the outstanding list so we can keep track of it */
4736 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4737 
4738 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4739 		status = nvmf_ctrlr_process_fabrics_cmd(req);
4740 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4741 		status = nvmf_ctrlr_process_admin_cmd(req);
4742 	} else {
4743 		status = nvmf_ctrlr_process_io_cmd(req);
4744 	}
4745 
4746 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4747 		_nvmf_request_complete(req);
4748 	}
4749 }
4750 
4751 static bool
4752 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
4753 		       struct spdk_dif_ctx *dif_ctx)
4754 {
4755 	struct spdk_nvmf_ns *ns;
4756 	struct spdk_bdev *bdev;
4757 
4758 	if (ctrlr == NULL || cmd == NULL) {
4759 		return false;
4760 	}
4761 
4762 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
4763 	if (ns == NULL || ns->bdev == NULL) {
4764 		return false;
4765 	}
4766 
4767 	bdev = ns->bdev;
4768 
4769 	switch (cmd->opc) {
4770 	case SPDK_NVME_OPC_READ:
4771 	case SPDK_NVME_OPC_WRITE:
4772 	case SPDK_NVME_OPC_COMPARE:
4773 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
4774 	default:
4775 		break;
4776 	}
4777 
4778 	return false;
4779 }
4780 
4781 bool
4782 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
4783 {
4784 	struct spdk_nvmf_qpair *qpair = req->qpair;
4785 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4786 
4787 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
4788 		return false;
4789 	}
4790 
4791 	if (spdk_unlikely(!spdk_nvmf_qpair_is_active(qpair))) {
4792 		return false;
4793 	}
4794 
4795 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4796 		return false;
4797 	}
4798 
4799 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4800 		return false;
4801 	}
4802 
4803 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
4804 }
4805 
4806 void
4807 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
4808 {
4809 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
4810 }
4811 
4812 static int
4813 nvmf_passthru_admin_cmd_for_bdev_nsid(struct spdk_nvmf_request *req, uint32_t bdev_nsid)
4814 {
4815 	struct spdk_bdev *bdev;
4816 	struct spdk_bdev_desc *desc;
4817 	struct spdk_io_channel *ch;
4818 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
4819 	int rc;
4820 
4821 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
4822 	if (rc) {
4823 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4824 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4825 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4826 	}
4827 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
4828 }
4829 
4830 static int
4831 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
4832 {
4833 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
4834 	uint32_t bdev_nsid;
4835 
4836 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid != 0) {
4837 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
4838 	} else {
4839 		bdev_nsid = cmd->nsid;
4840 	}
4841 
4842 	return nvmf_passthru_admin_cmd_for_bdev_nsid(req, bdev_nsid);
4843 }
4844 
4845 int
4846 nvmf_passthru_admin_cmd_for_ctrlr(struct spdk_nvmf_request *req, struct spdk_nvmf_ctrlr *ctrlr)
4847 {
4848 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
4849 	struct spdk_nvmf_ns *ns;
4850 
4851 	ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys);
4852 	if (ns == NULL) {
4853 		/* Is there a better sc to use here? */
4854 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4855 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4856 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4857 	}
4858 
4859 	return nvmf_passthru_admin_cmd_for_bdev_nsid(req, ns->nsid);
4860 }
4861 
4862 void
4863 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
4864 {
4865 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
4866 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
4867 }
4868 
4869 int
4870 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
4871 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
4872 {
4873 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
4874 	struct spdk_nvmf_ns *ns;
4875 	struct spdk_nvmf_poll_group *group = req->qpair->group;
4876 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4877 
4878 	*bdev = NULL;
4879 	*desc = NULL;
4880 	*ch = NULL;
4881 
4882 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
4883 	if (ns == NULL || ns->bdev == NULL) {
4884 		return -EINVAL;
4885 	}
4886 
4887 	assert(group != NULL && group->sgroups != NULL);
4888 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
4889 	*bdev = ns->bdev;
4890 	*desc = ns->desc;
4891 	*ch = ns_info->channel;
4892 
4893 	return 0;
4894 }
4895 
4896 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
4897 {
4898 	return req->qpair->ctrlr;
4899 }
4900 
4901 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
4902 {
4903 	return &req->cmd->nvme_cmd;
4904 }
4905 
4906 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
4907 {
4908 	return &req->rsp->nvme_cpl;
4909 }
4910 
4911 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
4912 {
4913 	return req->qpair->ctrlr->subsys;
4914 }
4915 
4916 size_t
4917 spdk_nvmf_request_copy_from_buf(struct spdk_nvmf_request *req,
4918 				void *buf, size_t buflen)
4919 {
4920 	struct spdk_iov_xfer ix;
4921 
4922 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
4923 	return spdk_iov_xfer_from_buf(&ix, buf, buflen);
4924 }
4925 
4926 size_t
4927 spdk_nvmf_request_copy_to_buf(struct spdk_nvmf_request *req,
4928 			      void *buf, size_t buflen)
4929 {
4930 	struct spdk_iov_xfer ix;
4931 
4932 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
4933 	return spdk_iov_xfer_to_buf(&ix, buf, buflen);
4934 }
4935 
4936 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
4937 {
4938 	return ctrlr->subsys;
4939 }
4940 
4941 uint16_t
4942 spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
4943 {
4944 	return ctrlr->cntlid;
4945 }
4946 
4947 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
4948 {
4949 	return req->req_to_abort;
4950 }
4951