xref: /spdk/lib/nvmf/ctrlr.c (revision e0d7428b482257aa6999b8b4cc44159dcc292df9)
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  *   Copyright (c) 2024 Samsung Electronics Co., Ltd. All rights reserved.
6  */
7 
8 #include "spdk/stdinc.h"
9 
10 #include "nvmf_internal.h"
11 #include "transport.h"
12 
13 #include "spdk/bdev.h"
14 #include "spdk/bdev_zone.h"
15 #include "spdk/bit_array.h"
16 #include "spdk/endian.h"
17 #include "spdk/thread.h"
18 #include "spdk/nvme_spec.h"
19 #include "spdk/nvmf_cmd.h"
20 #include "spdk/string.h"
21 #include "spdk/util.h"
22 #include "spdk/version.h"
23 #include "spdk/log.h"
24 #include "spdk_internal/usdt.h"
25 
26 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000
27 #define NVMF_DISC_KATO_IN_MS 120000
28 #define KAS_TIME_UNIT_IN_MS 100
29 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS)
30 
31 #define NVMF_CC_RESET_SHN_TIMEOUT_IN_MS	10000
32 
33 #define NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS	(NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + 5000)
34 
35 #define DUPLICATE_QID_RETRY_US 1000
36 
37 /*
38  * Report the SPDK version as the firmware revision.
39  * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts.
40  */
41 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING
42 
43 #define ANA_TRANSITION_TIME_IN_SEC 10
44 
45 #define NVMF_ABORT_COMMAND_LIMIT 3
46 
47 /*
48  * Support for custom admin command handlers
49  */
50 struct spdk_nvmf_custom_admin_cmd {
51 	spdk_nvmf_custom_cmd_hdlr hdlr;
52 	uint32_t nsid; /* nsid to forward */
53 };
54 
55 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1];
56 
57 static void _nvmf_request_complete(void *ctx);
58 int nvmf_passthru_admin_cmd_for_ctrlr(struct spdk_nvmf_request *req, struct spdk_nvmf_ctrlr *ctrlr);
59 static int nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req);
60 
61 static inline void
62 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp,
63 			      uint8_t iattr, uint16_t ipo)
64 {
65 	rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
66 	rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
67 	rsp->status_code_specific.invalid.iattr = iattr;
68 	rsp->status_code_specific.invalid.ipo = ipo;
69 }
70 
71 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field)	\
72 	nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field))
73 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field)	\
74 	nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field))
75 
76 
77 static void
78 nvmf_ctrlr_stop_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
79 {
80 	if (!ctrlr) {
81 		SPDK_ERRLOG("Controller is NULL\n");
82 		return;
83 	}
84 
85 	if (ctrlr->keep_alive_poller == NULL) {
86 		return;
87 	}
88 
89 	SPDK_DEBUGLOG(nvmf, "Stop keep alive poller\n");
90 	spdk_poller_unregister(&ctrlr->keep_alive_poller);
91 }
92 
93 static void
94 nvmf_ctrlr_stop_association_timer(struct spdk_nvmf_ctrlr *ctrlr)
95 {
96 	if (!ctrlr) {
97 		SPDK_ERRLOG("Controller is NULL\n");
98 		assert(false);
99 		return;
100 	}
101 
102 	if (ctrlr->association_timer == NULL) {
103 		return;
104 	}
105 
106 	SPDK_DEBUGLOG(nvmf, "Stop association timer\n");
107 	spdk_poller_unregister(&ctrlr->association_timer);
108 }
109 
110 static void
111 nvmf_ctrlr_disconnect_qpairs_done(struct spdk_io_channel_iter *i, int status)
112 {
113 	if (status == 0) {
114 		SPDK_DEBUGLOG(nvmf, "ctrlr disconnect qpairs complete successfully\n");
115 	} else {
116 		SPDK_ERRLOG("Fail to disconnect ctrlr qpairs\n");
117 	}
118 }
119 
120 static int
121 _nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i, bool include_admin)
122 {
123 	int rc = 0;
124 	struct spdk_nvmf_ctrlr *ctrlr;
125 	struct spdk_nvmf_qpair *qpair, *temp_qpair;
126 	struct spdk_io_channel *ch;
127 	struct spdk_nvmf_poll_group *group;
128 
129 	ctrlr = spdk_io_channel_iter_get_ctx(i);
130 	ch = spdk_io_channel_iter_get_channel(i);
131 	group = spdk_io_channel_get_ctx(ch);
132 
133 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, temp_qpair) {
134 		if (qpair->ctrlr == ctrlr && (include_admin || !nvmf_qpair_is_admin_queue(qpair))) {
135 			rc = spdk_nvmf_qpair_disconnect(qpair);
136 			if (rc) {
137 				if (rc == -EINPROGRESS) {
138 					rc = 0;
139 				} else {
140 					SPDK_ERRLOG("Qpair disconnect failed\n");
141 					return rc;
142 				}
143 			}
144 		}
145 	}
146 
147 	return rc;
148 }
149 
150 static void
151 nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i)
152 {
153 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, true));
154 }
155 
156 static void
157 nvmf_ctrlr_disconnect_io_qpairs_on_pg(struct spdk_io_channel_iter *i)
158 {
159 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, false));
160 }
161 
162 static int
163 nvmf_ctrlr_keep_alive_poll(void *ctx)
164 {
165 	uint64_t keep_alive_timeout_tick;
166 	uint64_t now = spdk_get_ticks();
167 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
168 
169 	if (ctrlr->in_destruct) {
170 		nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
171 		return SPDK_POLLER_IDLE;
172 	}
173 
174 	SPDK_DEBUGLOG(nvmf, "Polling ctrlr keep alive timeout\n");
175 
176 	/* If the Keep alive feature is in use and the timer expires */
177 	keep_alive_timeout_tick = ctrlr->last_keep_alive_tick +
178 				  ctrlr->feat.keep_alive_timer.bits.kato * spdk_get_ticks_hz() / UINT64_C(1000);
179 	if (now > keep_alive_timeout_tick) {
180 		SPDK_NOTICELOG("Disconnecting host %s from subsystem %s due to keep alive timeout.\n",
181 			       ctrlr->hostnqn, ctrlr->subsys->subnqn);
182 		/* set the Controller Fatal Status bit to '1' */
183 		if (ctrlr->vcprop.csts.bits.cfs == 0) {
184 			nvmf_ctrlr_set_fatal_status(ctrlr);
185 
186 			/*
187 			 * disconnect qpairs, terminate Transport connection
188 			 * destroy ctrlr, break the host to controller association
189 			 * disconnect qpairs with qpair->ctrlr == ctrlr
190 			 */
191 			spdk_for_each_channel(ctrlr->subsys->tgt,
192 					      nvmf_ctrlr_disconnect_qpairs_on_pg,
193 					      ctrlr,
194 					      nvmf_ctrlr_disconnect_qpairs_done);
195 			return SPDK_POLLER_BUSY;
196 		}
197 	}
198 
199 	return SPDK_POLLER_IDLE;
200 }
201 
202 static void
203 nvmf_ctrlr_start_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
204 {
205 	if (!ctrlr) {
206 		SPDK_ERRLOG("Controller is NULL\n");
207 		return;
208 	}
209 
210 	/* if cleared to 0 then the Keep Alive Timer is disabled */
211 	if (ctrlr->feat.keep_alive_timer.bits.kato != 0) {
212 
213 		ctrlr->last_keep_alive_tick = spdk_get_ticks();
214 
215 		SPDK_DEBUGLOG(nvmf, "Ctrlr add keep alive poller\n");
216 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
217 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
218 	}
219 }
220 
221 static void
222 nvmf_qpair_set_ctrlr(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_ctrlr *ctrlr)
223 {
224 	if (qpair->ctrlr != NULL) {
225 		/* Admin queues will call this function twice. */
226 		assert(qpair->ctrlr == ctrlr);
227 		return;
228 	}
229 
230 	qpair->ctrlr = ctrlr;
231 	spdk_trace_owner_append_description(qpair->trace_id,
232 					    spdk_nvmf_subsystem_get_nqn(ctrlr->subsys));
233 }
234 
235 static int _retry_qid_check(void *ctx);
236 
237 static void
238 nvmf_ctrlr_send_connect_rsp(void *ctx)
239 {
240 	struct spdk_nvmf_request *req = ctx;
241 	struct spdk_nvmf_qpair *qpair = req->qpair;
242 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
243 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
244 	int rc;
245 
246 	/* The qpair might have been disconnected in the meantime */
247 	assert(qpair->state == SPDK_NVMF_QPAIR_CONNECTING ||
248 	       qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
249 	if (qpair->state == SPDK_NVMF_QPAIR_CONNECTING) {
250 		if (nvmf_subsystem_host_auth_required(ctrlr->subsys, ctrlr->hostnqn)) {
251 			rc = nvmf_qpair_auth_init(qpair);
252 			if (rc != 0) {
253 				rsp->status.sct = SPDK_NVME_SCT_GENERIC;
254 				rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
255 				spdk_nvmf_request_complete(req);
256 				spdk_nvmf_qpair_disconnect(qpair);
257 				return;
258 			}
259 			rsp->status_code_specific.success.authreq.atr = 1;
260 			nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_AUTHENTICATING);
261 		} else {
262 			nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ENABLED);
263 		}
264 	}
265 
266 	SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n", ctrlr->cntlid);
267 
268 	assert(spdk_get_thread() == qpair->group->thread);
269 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
270 	rsp->status_code_specific.success.cntlid = ctrlr->cntlid;
271 	spdk_nvmf_request_complete(req);
272 }
273 
274 static void
275 nvmf_ctrlr_add_qpair(struct spdk_nvmf_qpair *qpair,
276 		     struct spdk_nvmf_ctrlr *ctrlr,
277 		     struct spdk_nvmf_request *req)
278 {
279 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
280 
281 	if (!ctrlr->admin_qpair) {
282 		SPDK_ERRLOG("Inactive admin qpair\n");
283 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
284 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
285 		qpair->connect_req = NULL;
286 		qpair->ctrlr = NULL;
287 		spdk_nvmf_request_complete(req);
288 		return;
289 	}
290 
291 	assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
292 
293 	if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) {
294 		if (qpair->connect_req != NULL) {
295 			SPDK_ERRLOG("Got I/O connect with duplicate QID %u (cntlid:%u)\n",
296 				    qpair->qid, ctrlr->cntlid);
297 			rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
298 			rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
299 			qpair->connect_req = NULL;
300 			qpair->ctrlr = NULL;
301 			spdk_nvmf_request_complete(req);
302 		} else {
303 			SPDK_WARNLOG("Duplicate QID detected (cntlid:%u, qid:%u), re-check in %dus\n",
304 				     ctrlr->cntlid, qpair->qid, DUPLICATE_QID_RETRY_US);
305 			qpair->connect_req = req;
306 			/* Set qpair->ctrlr here so that we'll have it when the poller expires. */
307 			nvmf_qpair_set_ctrlr(qpair, ctrlr);
308 			req->poller = SPDK_POLLER_REGISTER(_retry_qid_check, qpair,
309 							   DUPLICATE_QID_RETRY_US);
310 		}
311 		return;
312 	}
313 
314 	SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn,
315 				 ctrlr->hostnqn);
316 	nvmf_qpair_set_ctrlr(qpair, ctrlr);
317 	spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid);
318 	SPDK_DEBUGLOG(nvmf, "qpair_mask set, qid %u\n", qpair->qid);
319 
320 	spdk_thread_send_msg(qpair->group->thread, nvmf_ctrlr_send_connect_rsp, req);
321 }
322 
323 static int
324 _retry_qid_check(void *ctx)
325 {
326 	struct spdk_nvmf_qpair *qpair = ctx;
327 	struct spdk_nvmf_request *req = qpair->connect_req;
328 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
329 
330 	spdk_poller_unregister(&req->poller);
331 	SPDK_WARNLOG("Retrying adding qpair, qid:%d\n", qpair->qid);
332 	nvmf_ctrlr_add_qpair(qpair, ctrlr, req);
333 	return SPDK_POLLER_BUSY;
334 }
335 
336 static void
337 _nvmf_ctrlr_add_admin_qpair(void *ctx)
338 {
339 	struct spdk_nvmf_request *req = ctx;
340 	struct spdk_nvmf_qpair *qpair = req->qpair;
341 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
342 
343 	ctrlr->admin_qpair = qpair;
344 	ctrlr->association_timeout = qpair->transport->opts.association_timeout;
345 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
346 	nvmf_ctrlr_add_qpair(qpair, ctrlr, req);
347 }
348 
349 static void
350 _nvmf_subsystem_add_ctrlr(void *ctx)
351 {
352 	struct spdk_nvmf_request *req = ctx;
353 	struct spdk_nvmf_qpair *qpair = req->qpair;
354 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
355 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
356 
357 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
358 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
359 		spdk_bit_array_free(&ctrlr->qpair_mask);
360 		free(ctrlr);
361 		qpair->ctrlr = NULL;
362 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
363 		spdk_nvmf_request_complete(req);
364 		return;
365 	}
366 
367 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
368 }
369 
370 static void
371 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
372 		      struct spdk_nvmf_ctrlr_data *cdata)
373 {
374 	cdata->aerl = SPDK_NVMF_MAX_ASYNC_EVENTS - 1;
375 	cdata->kas = KAS_DEFAULT_VALUE;
376 	cdata->vid = SPDK_PCI_VID_INTEL;
377 	cdata->ssvid = SPDK_PCI_VID_INTEL;
378 	/* INTEL OUI */
379 	cdata->ieee[0] = 0xe4;
380 	cdata->ieee[1] = 0xd2;
381 	cdata->ieee[2] = 0x5c;
382 	cdata->oncs.compare = 1;
383 	cdata->oncs.dsm = 1;
384 	cdata->oncs.write_zeroes = 1;
385 	cdata->oncs.reservations = 1;
386 	cdata->oncs.copy = 1;
387 	cdata->fuses.compare_and_write = 1;
388 	cdata->sgls.supported = 1;
389 	cdata->sgls.keyed_sgl = 1;
390 	cdata->sgls.sgl_offset = 1;
391 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
392 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
393 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
394 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
395 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
396 	cdata->nvmf_specific.msdbd = 1;
397 
398 	if (transport->ops->cdata_init) {
399 		transport->ops->cdata_init(transport, subsystem, cdata);
400 	}
401 }
402 
403 static bool
404 nvmf_subsystem_has_zns_iocs(struct spdk_nvmf_subsystem *subsystem)
405 {
406 	struct spdk_nvmf_ns *ns;
407 	uint32_t i;
408 
409 	for (i = 0; i < subsystem->max_nsid; i++) {
410 		ns = subsystem->ns[i];
411 		if (ns && ns->bdev && spdk_bdev_is_zoned(ns->bdev)) {
412 			return true;
413 		}
414 	}
415 	return false;
416 }
417 
418 static void
419 nvmf_ctrlr_init_visible_ns(struct spdk_nvmf_ctrlr *ctrlr)
420 {
421 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
422 	struct spdk_nvmf_ns *ns;
423 
424 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
425 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
426 		if (ns->always_visible || nvmf_ns_find_host(ns, ctrlr->hostnqn) != NULL) {
427 			nvmf_ctrlr_ns_set_visible(ctrlr, ns->nsid, true);
428 		}
429 	}
430 }
431 
432 static struct spdk_nvmf_ctrlr *
433 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
434 		  struct spdk_nvmf_request *req,
435 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
436 		  struct spdk_nvmf_fabric_connect_data *connect_data)
437 {
438 	struct spdk_nvmf_ctrlr *ctrlr;
439 	struct spdk_nvmf_transport *transport = req->qpair->transport;
440 	struct spdk_nvme_transport_id listen_trid = {};
441 	bool subsys_has_multi_iocs = false;
442 
443 	ctrlr = calloc(1, sizeof(*ctrlr));
444 	if (ctrlr == NULL) {
445 		SPDK_ERRLOG("Memory allocation failed\n");
446 		return NULL;
447 	}
448 
449 	if (spdk_nvme_trtype_is_fabrics(transport->ops->type)) {
450 		ctrlr->dynamic_ctrlr = true;
451 	} else {
452 		ctrlr->cntlid = connect_data->cntlid;
453 	}
454 
455 	SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_create, ctrlr, subsystem->subnqn,
456 				 spdk_thread_get_id(req->qpair->group->thread));
457 
458 	STAILQ_INIT(&ctrlr->async_events);
459 	TAILQ_INIT(&ctrlr->log_head);
460 	ctrlr->subsys = subsystem;
461 	ctrlr->thread = req->qpair->group->thread;
462 	ctrlr->disconnect_in_progress = false;
463 
464 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
465 	if (!ctrlr->qpair_mask) {
466 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
467 		goto err_qpair_mask;
468 	}
469 
470 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
471 
472 	/*
473 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
474 	 * If this field is cleared to 0h, then Keep Alive is not supported.
475 	 */
476 	if (ctrlr->cdata.kas) {
477 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
478 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
479 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
480 	}
481 
482 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
483 	if (ctrlr->subsys->flags.ana_reporting) {
484 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
485 	}
486 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
487 	/* Coalescing Disable */
488 	ctrlr->feat.interrupt_vector_configuration.bits.cd = 1;
489 
490 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
491 		/*
492 		 * If keep-alive timeout is not set, discovery controllers use some
493 		 * arbitrary high value in order to cleanup stale discovery sessions
494 		 *
495 		 * From the 1.0a nvme-of spec:
496 		 * "The Keep Alive command is reserved for
497 		 * Discovery controllers. A transport may specify a
498 		 * fixed Discovery controller activity timeout value
499 		 * (e.g., 2 minutes). If no commands are received
500 		 * by a Discovery controller within that time
501 		 * period, the controller may perform the
502 		 * actions for Keep Alive Timer expiration".
503 		 *
504 		 * From the 1.1 nvme-of spec:
505 		 * "A host requests an explicit persistent connection
506 		 * to a Discovery controller and Asynchronous Event Notifications from
507 		 * the Discovery controller on that persistent connection by specifying
508 		 * a non-zero Keep Alive Timer value in the Connect command."
509 		 *
510 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
511 		 * otherwise we disable it and use default discovery controller KATO.
512 		 * KATO is in millisecond.
513 		 */
514 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
515 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
516 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
517 		} else {
518 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
519 		}
520 	}
521 
522 	/* Subtract 1 for admin queue, 1 for 0's based */
523 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
524 			1;
525 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
526 			1;
527 
528 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
529 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, SPDK_NVMF_NQN_MAX_LEN);
530 
531 	ctrlr->visible_ns = spdk_bit_array_create(subsystem->max_nsid);
532 	if (!ctrlr->visible_ns) {
533 		SPDK_ERRLOG("Failed to allocate visible namespace array\n");
534 		goto err_visible_ns;
535 	}
536 	nvmf_ctrlr_init_visible_ns(ctrlr);
537 
538 	ctrlr->vcprop.cap.raw = 0;
539 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
540 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
541 				      1; /* max queue depth */
542 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
543 	/* ready timeout - 500 msec units */
544 	ctrlr->vcprop.cap.bits.to = NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS / 500;
545 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
546 	subsys_has_multi_iocs = nvmf_subsystem_has_zns_iocs(subsystem);
547 	if (subsys_has_multi_iocs) {
548 		ctrlr->vcprop.cap.bits.css =
549 			SPDK_NVME_CAP_CSS_IOCS; /* One or more I/O command sets supported */
550 	} else {
551 		ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
552 	}
553 
554 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
555 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
556 
557 	/* Version Supported: 1.3 */
558 	ctrlr->vcprop.vs.bits.mjr = 1;
559 	ctrlr->vcprop.vs.bits.mnr = 3;
560 	ctrlr->vcprop.vs.bits.ter = 0;
561 
562 	ctrlr->vcprop.cc.raw = 0;
563 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
564 	if (subsys_has_multi_iocs) {
565 		ctrlr->vcprop.cc.bits.css =
566 			SPDK_NVME_CC_CSS_IOCS; /* All supported I/O Command Sets */
567 	}
568 
569 	ctrlr->vcprop.csts.raw = 0;
570 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
571 
572 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
573 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
574 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
575 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
576 
577 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
578 
579 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
580 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
581 			SPDK_ERRLOG("Could not get listener transport ID\n");
582 			goto err_listener;
583 		}
584 
585 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid);
586 		if (!ctrlr->listener) {
587 			SPDK_ERRLOG("Listener was not found\n");
588 			goto err_listener;
589 		}
590 	}
591 
592 	nvmf_qpair_set_ctrlr(req->qpair, ctrlr);
593 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
594 
595 	return ctrlr;
596 err_listener:
597 	spdk_bit_array_free(&ctrlr->visible_ns);
598 err_visible_ns:
599 	spdk_bit_array_free(&ctrlr->qpair_mask);
600 err_qpair_mask:
601 	free(ctrlr);
602 	return NULL;
603 }
604 
605 static void
606 _nvmf_ctrlr_destruct(void *ctx)
607 {
608 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
609 	struct spdk_nvmf_reservation_log *log, *log_tmp;
610 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
611 
612 	SPDK_DTRACE_PROBE3_TICKS(nvmf_ctrlr_destruct, ctrlr, ctrlr->subsys->subnqn,
613 				 spdk_thread_get_id(ctrlr->thread));
614 
615 	assert(spdk_get_thread() == ctrlr->thread);
616 	assert(ctrlr->in_destruct);
617 
618 	SPDK_DEBUGLOG(nvmf, "Destroy ctrlr 0x%hx\n", ctrlr->cntlid);
619 	if (ctrlr->disconnect_in_progress) {
620 		SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n");
621 		spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
622 		return;
623 	}
624 
625 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
626 	nvmf_ctrlr_stop_association_timer(ctrlr);
627 	spdk_bit_array_free(&ctrlr->qpair_mask);
628 
629 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
630 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
631 		free(log);
632 	}
633 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
634 		STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link);
635 		free(event);
636 	}
637 	spdk_bit_array_free(&ctrlr->visible_ns);
638 	free(ctrlr);
639 }
640 
641 void
642 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
643 {
644 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
645 
646 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
647 }
648 
649 static void
650 nvmf_ctrlr_add_io_qpair(void *ctx)
651 {
652 	struct spdk_nvmf_request *req = ctx;
653 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
654 	struct spdk_nvmf_qpair *qpair = req->qpair;
655 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
656 	struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair;
657 	struct spdk_nvmf_poll_group *admin_qpair_group = NULL;
658 	enum spdk_nvmf_qpair_state admin_qpair_state = SPDK_NVMF_QPAIR_UNINITIALIZED;
659 	bool admin_qpair_active = false;
660 
661 	SPDK_DTRACE_PROBE4_TICKS(nvmf_ctrlr_add_io_qpair, ctrlr, req->qpair, req->qpair->qid,
662 				 spdk_thread_get_id(ctrlr->thread));
663 
664 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
665 	  * For error case, the value should be NULL. So set it to NULL at first.
666 	  */
667 	qpair->ctrlr = NULL;
668 
669 	/* Make sure the controller is not being destroyed. */
670 	if (ctrlr->in_destruct) {
671 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
672 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
673 		goto end;
674 	}
675 
676 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
677 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
678 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
679 		goto end;
680 	}
681 
682 	if (!ctrlr->vcprop.cc.bits.en) {
683 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
684 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
685 		goto end;
686 	}
687 
688 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
689 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
690 			    ctrlr->vcprop.cc.bits.iosqes);
691 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
692 		goto end;
693 	}
694 
695 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
696 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
697 			    ctrlr->vcprop.cc.bits.iocqes);
698 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
699 		goto end;
700 	}
701 
702 	/* There is a chance that admin qpair was destroyed. This is an issue that was observed only with ESX initiators */
703 	if (admin_qpair) {
704 		admin_qpair_active = spdk_nvmf_qpair_is_active(admin_qpair);
705 		admin_qpair_group = admin_qpair->group;
706 		admin_qpair_state = admin_qpair->state;
707 	}
708 
709 	if (!admin_qpair_active || admin_qpair_group == NULL) {
710 		/* There is a chance that admin qpair was destroyed or is being destroyed at this moment due to e.g.
711 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
712 		 * state to DEACTIVATING and removing it from poll group */
713 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair_state, admin_qpair_group);
714 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
715 		goto end;
716 	}
717 
718 	/* check if we would exceed ctrlr connection limit */
719 	if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) {
720 		SPDK_ERRLOG("Requested QID %u but Max QID is %u\n",
721 			    qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1);
722 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
723 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
724 		goto end;
725 	}
726 
727 	nvmf_ctrlr_add_qpair(qpair, ctrlr, req);
728 	return;
729 end:
730 	spdk_nvmf_request_complete(req);
731 }
732 
733 static void
734 _nvmf_ctrlr_add_io_qpair(void *ctx)
735 {
736 	struct spdk_nvmf_request *req = ctx;
737 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
738 	struct spdk_nvmf_fabric_connect_data *data;
739 	struct spdk_nvmf_ctrlr *ctrlr;
740 	struct spdk_nvmf_qpair *qpair = req->qpair;
741 	struct spdk_nvmf_qpair *admin_qpair;
742 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
743 	struct spdk_nvmf_subsystem *subsystem;
744 	struct spdk_nvme_transport_id listen_trid = {};
745 	const struct spdk_nvmf_subsystem_listener *listener;
746 	struct spdk_nvmf_poll_group *admin_qpair_group = NULL;
747 	enum spdk_nvmf_qpair_state admin_qpair_state = SPDK_NVMF_QPAIR_UNINITIALIZED;
748 	bool admin_qpair_active = false;
749 
750 	assert(req->iovcnt == 1);
751 
752 	data = req->iov[0].iov_base;
753 
754 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
755 
756 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
757 	/* We already checked this in spdk_nvmf_ctrlr_connect */
758 	assert(subsystem != NULL);
759 
760 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
761 	if (ctrlr == NULL) {
762 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
763 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
764 		spdk_nvmf_request_complete(req);
765 		return;
766 	}
767 
768 	/* fail before passing a message to the controller thread. */
769 	if (ctrlr->in_destruct) {
770 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
771 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
772 		spdk_nvmf_request_complete(req);
773 		return;
774 	}
775 
776 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
777 	if (subsystem->flags.ana_reporting) {
778 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
779 			SPDK_ERRLOG("Could not get listener transport ID\n");
780 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
781 			spdk_nvmf_request_complete(req);
782 			return;
783 		}
784 
785 		listener = nvmf_subsystem_find_listener(subsystem, &listen_trid);
786 		if (listener != ctrlr->listener) {
787 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
788 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
789 			spdk_nvmf_request_complete(req);
790 			return;
791 		}
792 	}
793 
794 	admin_qpair = ctrlr->admin_qpair;
795 
796 	/* There is a chance that admin qpair was destroyed. This is an issue that was observed only with ESX initiators */
797 	if (admin_qpair) {
798 		admin_qpair_active = spdk_nvmf_qpair_is_active(admin_qpair);
799 		admin_qpair_group = admin_qpair->group;
800 		admin_qpair_state = admin_qpair->state;
801 	}
802 
803 	if (!admin_qpair_active || admin_qpair_group == NULL) {
804 		/* There is a chance that admin qpair was destroyed or is being destroyed at this moment due to e.g.
805 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
806 		 * state to DEACTIVATING and removing it from poll group */
807 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair_state, admin_qpair_group);
808 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
809 		spdk_nvmf_request_complete(req);
810 		return;
811 	}
812 	qpair->ctrlr = ctrlr;
813 	spdk_thread_send_msg(admin_qpair_group->thread, nvmf_ctrlr_add_io_qpair, req);
814 }
815 
816 static bool
817 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
818 			  const char *hostnqn)
819 {
820 	struct spdk_nvme_transport_id listen_trid = {};
821 
822 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
823 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
824 		return false;
825 	}
826 
827 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
828 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
829 			    subsystem->subnqn);
830 		return false;
831 	}
832 
833 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
834 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
835 			    subsystem->subnqn, hostnqn);
836 		return false;
837 	}
838 
839 	return true;
840 }
841 
842 static int
843 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
844 {
845 	struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base;
846 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
847 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
848 	struct spdk_nvmf_qpair *qpair = req->qpair;
849 	struct spdk_nvmf_transport *transport = qpair->transport;
850 	struct spdk_nvmf_ctrlr *ctrlr;
851 	struct spdk_nvmf_subsystem *subsystem;
852 
853 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
854 		      cmd->recfmt, cmd->qid, cmd->sqsize);
855 
856 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
857 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
858 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
859 		      ntohl(*(uint32_t *)&data->hostid[0]),
860 		      ntohs(*(uint16_t *)&data->hostid[4]),
861 		      ntohs(*(uint16_t *)&data->hostid[6]),
862 		      data->hostid[8],
863 		      data->hostid[9],
864 		      ntohs(*(uint16_t *)&data->hostid[10]),
865 		      ntohl(*(uint32_t *)&data->hostid[12]));
866 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
867 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
868 
869 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
870 	if (!subsystem) {
871 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
872 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
873 	}
874 
875 	if (cmd->recfmt != 0) {
876 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
877 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
878 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
879 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
880 	}
881 
882 	/*
883 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
884 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
885 	 */
886 	if (cmd->sqsize == 0) {
887 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
888 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
889 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
890 	}
891 
892 	if (cmd->qid == 0) {
893 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
894 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
895 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
896 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
897 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
898 		}
899 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
900 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
901 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
902 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
903 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
904 	}
905 
906 	qpair->sq_head_max = cmd->sqsize;
907 	qpair->qid = cmd->qid;
908 	qpair->connect_received = true;
909 
910 	pthread_mutex_lock(&qpair->group->mutex);
911 	qpair->group->current_unassociated_qpairs--;
912 	pthread_mutex_unlock(&qpair->group->mutex);
913 
914 	if (0 == qpair->qid) {
915 		qpair->group->stat.admin_qpairs++;
916 		qpair->group->stat.current_admin_qpairs++;
917 	} else {
918 		qpair->group->stat.io_qpairs++;
919 		qpair->group->stat.current_io_qpairs++;
920 	}
921 
922 	if (cmd->qid == 0) {
923 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
924 
925 		if (spdk_nvme_trtype_is_fabrics(transport->ops->type) && data->cntlid != 0xFFFF) {
926 			/* This NVMf target only supports dynamic mode. */
927 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
928 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
929 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
930 		}
931 
932 		/* Establish a new ctrlr */
933 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
934 		if (!ctrlr) {
935 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
936 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
937 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
938 		} else {
939 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
940 		}
941 	} else {
942 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
943 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
944 	}
945 }
946 
947 static struct spdk_nvmf_subsystem_poll_group *
948 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
949 {
950 	struct spdk_nvmf_fabric_connect_data *data;
951 	struct spdk_nvmf_subsystem *subsystem;
952 	struct spdk_nvmf_tgt *tgt;
953 
954 	assert(nvmf_request_is_fabric_connect(req));
955 	assert(req->qpair->ctrlr == NULL);
956 	assert(req->iovcnt == 1);
957 
958 	data = req->iov[0].iov_base;
959 	tgt = req->qpair->transport->tgt;
960 
961 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
962 	if (subsystem == NULL) {
963 		return NULL;
964 	}
965 
966 	return &req->qpair->group->sgroups[subsystem->id];
967 }
968 
969 int
970 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
971 {
972 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
973 	struct spdk_nvmf_subsystem_poll_group *sgroup;
974 	struct spdk_nvmf_qpair *qpair = req->qpair;
975 	enum spdk_nvmf_request_exec_status status;
976 
977 	if (req->iovcnt > 1) {
978 		SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt);
979 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
980 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
981 		goto out;
982 	}
983 
984 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
985 	if (!sgroup) {
986 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
987 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
988 		goto out;
989 	}
990 
991 	sgroup->mgmt_io_outstanding++;
992 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
993 
994 	status = _nvmf_ctrlr_connect(req);
995 
996 out:
997 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
998 		_nvmf_request_complete(req);
999 	}
1000 
1001 	return status;
1002 }
1003 
1004 static int
1005 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
1006 {
1007 	struct spdk_nvmf_fabric_connect_data *data = req->iov[0].iov_base;
1008 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
1009 	struct spdk_nvmf_transport *transport = req->qpair->transport;
1010 	struct spdk_nvmf_subsystem *subsystem;
1011 
1012 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
1013 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
1014 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1015 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1016 	}
1017 
1018 	if (req->iovcnt > 1) {
1019 		SPDK_ERRLOG("Connect command invalid iovcnt: %d\n", req->iovcnt);
1020 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1021 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1022 	}
1023 
1024 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
1025 	if (!subsystem) {
1026 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
1027 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1028 	}
1029 
1030 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
1031 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
1032 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
1033 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
1034 		struct spdk_nvmf_subsystem_poll_group *sgroup;
1035 
1036 		/* Subsystem is not ready to handle a connect. Decrement
1037 		 * the mgmt_io_outstanding to avoid the subsystem waiting
1038 		 * for this command to complete before unpausing. Queued
1039 		 * requests get retried when subsystem resumes.
1040 		 */
1041 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
1042 		assert(sgroup != NULL);
1043 		sgroup->mgmt_io_outstanding--;
1044 		TAILQ_REMOVE(&req->qpair->outstanding, req, link);
1045 		TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
1046 		SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn);
1047 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
1048 	}
1049 
1050 	/* Ensure that hostnqn is null terminated */
1051 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
1052 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
1053 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
1054 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1055 	}
1056 
1057 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
1058 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1059 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
1060 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1061 	}
1062 
1063 	return _nvmf_ctrlr_connect(req);
1064 }
1065 
1066 static int
1067 nvmf_ctrlr_association_remove(void *ctx)
1068 {
1069 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1070 	int rc;
1071 
1072 	nvmf_ctrlr_stop_association_timer(ctrlr);
1073 
1074 	if (ctrlr->in_destruct) {
1075 		return SPDK_POLLER_IDLE;
1076 	}
1077 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
1078 		      ctrlr->subsys->subnqn);
1079 
1080 	if (ctrlr->admin_qpair) {
1081 		rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair);
1082 		if (rc < 0 && rc != -EINPROGRESS) {
1083 			SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
1084 			assert(false);
1085 		}
1086 	}
1087 
1088 	return SPDK_POLLER_BUSY;
1089 }
1090 
1091 static int
1092 _nvmf_ctrlr_cc_reset_shn_done(void *ctx)
1093 {
1094 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1095 	uint64_t now = spdk_get_ticks();
1096 	uint32_t count;
1097 
1098 	if (ctrlr->cc_timer) {
1099 		spdk_poller_unregister(&ctrlr->cc_timer);
1100 	}
1101 
1102 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1103 	SPDK_DEBUGLOG(nvmf, "ctrlr %p active queue count %u\n", ctrlr, count);
1104 
1105 	if (count > 1) {
1106 		if (now < ctrlr->cc_timeout_tsc) {
1107 			/* restart cc timer */
1108 			ctrlr->cc_timer = SPDK_POLLER_REGISTER(_nvmf_ctrlr_cc_reset_shn_done, ctrlr, 100 * 1000);
1109 			return SPDK_POLLER_IDLE;
1110 		} else {
1111 			/* controller fatal status */
1112 			SPDK_WARNLOG("IO timeout, ctrlr %p is in fatal status\n", ctrlr);
1113 			nvmf_ctrlr_set_fatal_status(ctrlr);
1114 		}
1115 	}
1116 
1117 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
1118 
1119 	if (ctrlr->disconnect_is_shn) {
1120 		ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
1121 		ctrlr->disconnect_is_shn = false;
1122 	} else {
1123 		/* Only a subset of the registers are cleared out on a reset */
1124 		ctrlr->vcprop.cc.raw = 0;
1125 		ctrlr->vcprop.csts.raw = 0;
1126 	}
1127 
1128 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
1129 	 * between the host and controller shall be preserved for at least 2 minutes */
1130 	if (ctrlr->association_timer) {
1131 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
1132 		nvmf_ctrlr_stop_association_timer(ctrlr);
1133 	}
1134 	if (ctrlr->association_timeout) {
1135 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
1136 					   ctrlr->association_timeout * 1000);
1137 	}
1138 	ctrlr->disconnect_in_progress = false;
1139 	return SPDK_POLLER_BUSY;
1140 }
1141 
1142 static void
1143 nvmf_ctrlr_cc_reset_shn_done(struct spdk_io_channel_iter *i, int status)
1144 {
1145 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
1146 
1147 	if (status < 0) {
1148 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
1149 		assert(false);
1150 	}
1151 
1152 	_nvmf_ctrlr_cc_reset_shn_done((void *)ctrlr);
1153 }
1154 
1155 static void
1156 nvmf_bdev_complete_reset(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
1157 {
1158 	SPDK_NOTICELOG("Resetting bdev done with %s\n", success ? "success" : "failure");
1159 
1160 	spdk_bdev_free_io(bdev_io);
1161 }
1162 
1163 
1164 static int
1165 nvmf_ctrlr_cc_timeout(void *ctx)
1166 {
1167 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1168 	struct spdk_nvmf_poll_group *group;
1169 	struct spdk_nvmf_ns *ns;
1170 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1171 
1172 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
1173 	SPDK_DEBUGLOG(nvmf, "Ctrlr %p reset or shutdown timeout\n", ctrlr);
1174 
1175 	if (!ctrlr->admin_qpair) {
1176 		SPDK_NOTICELOG("Ctrlr %p admin qpair disconnected\n", ctrlr);
1177 		return SPDK_POLLER_IDLE;
1178 	}
1179 
1180 	group = ctrlr->admin_qpair->group;
1181 	assert(group != NULL && group->sgroups != NULL);
1182 
1183 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1184 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1185 		if (ns->bdev == NULL) {
1186 			continue;
1187 		}
1188 		ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[ns->opts.nsid - 1];
1189 		SPDK_NOTICELOG("Ctrlr %p resetting NSID %u\n", ctrlr, ns->opts.nsid);
1190 		spdk_bdev_reset(ns->desc, ns_info->channel, nvmf_bdev_complete_reset, NULL);
1191 	}
1192 
1193 	return SPDK_POLLER_BUSY;
1194 }
1195 
1196 const struct spdk_nvmf_registers *
1197 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
1198 {
1199 	return &ctrlr->vcprop;
1200 }
1201 
1202 void
1203 nvmf_ctrlr_set_fatal_status(struct spdk_nvmf_ctrlr *ctrlr)
1204 {
1205 	ctrlr->vcprop.csts.bits.cfs = 1;
1206 }
1207 
1208 static uint64_t
1209 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
1210 {
1211 	return ctrlr->vcprop.cap.raw;
1212 }
1213 
1214 static uint64_t
1215 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
1216 {
1217 	return ctrlr->vcprop.vs.raw;
1218 }
1219 
1220 static uint64_t
1221 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
1222 {
1223 	return ctrlr->vcprop.cc.raw;
1224 }
1225 
1226 static bool
1227 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1228 {
1229 	union spdk_nvme_cc_register cc, diff;
1230 	uint32_t cc_timeout_ms;
1231 
1232 	cc.raw = value;
1233 
1234 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
1235 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
1236 
1237 	/*
1238 	 * Calculate which bits changed between the current and new CC.
1239 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
1240 	 */
1241 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
1242 
1243 	if (diff.bits.en) {
1244 		if (cc.bits.en) {
1245 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
1246 			nvmf_ctrlr_stop_association_timer(ctrlr);
1247 
1248 			ctrlr->vcprop.cc.bits.en = 1;
1249 			ctrlr->vcprop.csts.bits.rdy = 1;
1250 		} else {
1251 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
1252 			if (ctrlr->disconnect_in_progress) {
1253 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1254 				return true;
1255 			}
1256 
1257 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1258 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1259 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1260 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1261 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1262 
1263 			ctrlr->vcprop.cc.bits.en = 0;
1264 			ctrlr->disconnect_in_progress = true;
1265 			ctrlr->disconnect_is_shn = false;
1266 			spdk_for_each_channel(ctrlr->subsys->tgt,
1267 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1268 					      ctrlr,
1269 					      nvmf_ctrlr_cc_reset_shn_done);
1270 		}
1271 		diff.bits.en = 0;
1272 	}
1273 
1274 	if (diff.bits.shn) {
1275 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
1276 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
1277 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
1278 				      cc.bits.shn >> 1, cc.bits.shn & 1);
1279 			if (ctrlr->disconnect_in_progress) {
1280 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1281 				return true;
1282 			}
1283 
1284 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1285 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1286 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1287 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1288 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1289 
1290 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
1291 			ctrlr->disconnect_in_progress = true;
1292 			ctrlr->disconnect_is_shn = true;
1293 			spdk_for_each_channel(ctrlr->subsys->tgt,
1294 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1295 					      ctrlr,
1296 					      nvmf_ctrlr_cc_reset_shn_done);
1297 
1298 			/* From the time a shutdown is initiated the controller shall disable
1299 			 * Keep Alive timer */
1300 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
1301 		} else if (cc.bits.shn == 0) {
1302 			ctrlr->vcprop.cc.bits.shn = 0;
1303 		} else {
1304 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
1305 				    cc.bits.shn >> 1, cc.bits.shn & 1);
1306 			return false;
1307 		}
1308 		diff.bits.shn = 0;
1309 	}
1310 
1311 	if (diff.bits.iosqes) {
1312 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
1313 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
1314 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
1315 		diff.bits.iosqes = 0;
1316 	}
1317 
1318 	if (diff.bits.iocqes) {
1319 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
1320 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
1321 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
1322 		diff.bits.iocqes = 0;
1323 	}
1324 
1325 	if (diff.bits.ams) {
1326 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
1327 		return false;
1328 	}
1329 
1330 	if (diff.bits.mps) {
1331 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
1332 		return false;
1333 	}
1334 
1335 	if (diff.bits.css) {
1336 		if (cc.bits.css > SPDK_NVME_CC_CSS_IOCS) {
1337 			SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
1338 			return false;
1339 		}
1340 		diff.bits.css = 0;
1341 	}
1342 
1343 	if (diff.raw != 0) {
1344 		/* Print an error message, but don't fail the command in this case.
1345 		 * If we did want to fail in this case, we'd need to ensure we acted
1346 		 * on no other bits or the initiator gets confused. */
1347 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
1348 	}
1349 
1350 	return true;
1351 }
1352 
1353 static uint64_t
1354 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
1355 {
1356 	return ctrlr->vcprop.csts.raw;
1357 }
1358 
1359 static uint64_t
1360 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
1361 {
1362 	return ctrlr->vcprop.aqa.raw;
1363 }
1364 
1365 static bool
1366 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1367 {
1368 	union spdk_nvme_aqa_register aqa;
1369 
1370 	aqa.raw = value;
1371 
1372 	/*
1373 	 * We don't need to explicitly check for maximum size, as the fields are
1374 	 * limited to 12 bits (4096).
1375 	 */
1376 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1377 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1378 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1379 		return false;
1380 	}
1381 
1382 	ctrlr->vcprop.aqa.raw = value;
1383 
1384 	return true;
1385 }
1386 
1387 static uint64_t
1388 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1389 {
1390 	return ctrlr->vcprop.asq;
1391 }
1392 
1393 static bool
1394 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1395 {
1396 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1397 
1398 	return true;
1399 }
1400 
1401 static bool
1402 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1403 {
1404 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1405 
1406 	return true;
1407 }
1408 
1409 static uint64_t
1410 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1411 {
1412 	return ctrlr->vcprop.acq;
1413 }
1414 
1415 static bool
1416 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1417 {
1418 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1419 
1420 	return true;
1421 }
1422 
1423 static bool
1424 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1425 {
1426 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1427 
1428 	return true;
1429 }
1430 
1431 struct nvmf_prop {
1432 	uint32_t ofst;
1433 	uint8_t size;
1434 	char name[11];
1435 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1436 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1437 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1438 };
1439 
1440 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1441 	{ \
1442 		offsetof(struct spdk_nvme_registers, field), \
1443 		size, \
1444 		#field, \
1445 		get_cb, set_cb, set_upper_cb \
1446 	}
1447 
1448 static const struct nvmf_prop nvmf_props[] = {
1449 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1450 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1451 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1452 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1453 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1454 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1455 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1456 };
1457 
1458 static const struct nvmf_prop *
1459 find_prop(uint32_t ofst, uint8_t size)
1460 {
1461 	size_t i;
1462 
1463 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1464 		const struct nvmf_prop *prop = &nvmf_props[i];
1465 
1466 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1467 			return prop;
1468 		}
1469 	}
1470 
1471 	return NULL;
1472 }
1473 
1474 static int
1475 nvmf_property_get(struct spdk_nvmf_request *req)
1476 {
1477 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1478 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1479 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1480 	const struct nvmf_prop *prop;
1481 	uint8_t size;
1482 
1483 	response->status.sc = 0;
1484 	response->value.u64 = 0;
1485 
1486 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1487 		      cmd->attrib.size, cmd->ofst);
1488 
1489 	switch (cmd->attrib.size) {
1490 	case SPDK_NVMF_PROP_SIZE_4:
1491 		size = 4;
1492 		break;
1493 	case SPDK_NVMF_PROP_SIZE_8:
1494 		size = 8;
1495 		break;
1496 	default:
1497 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1498 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1499 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1500 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1501 	}
1502 
1503 	prop = find_prop(cmd->ofst, size);
1504 	if (prop == NULL || prop->get_cb == NULL) {
1505 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1506 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1507 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1508 	}
1509 
1510 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1511 
1512 	response->value.u64 = prop->get_cb(ctrlr);
1513 
1514 	if (size != prop->size) {
1515 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1516 		assert(size == 4);
1517 		assert(prop->size == 8);
1518 
1519 		if (cmd->ofst == prop->ofst) {
1520 			/* Keep bottom 4 bytes only */
1521 			response->value.u64 &= 0xFFFFFFFF;
1522 		} else {
1523 			/* Keep top 4 bytes only */
1524 			response->value.u64 >>= 32;
1525 		}
1526 	}
1527 
1528 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1529 
1530 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1531 }
1532 
1533 static int
1534 nvmf_property_set(struct spdk_nvmf_request *req)
1535 {
1536 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1537 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1538 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1539 	const struct nvmf_prop *prop;
1540 	uint64_t value;
1541 	uint8_t size;
1542 	bool ret;
1543 
1544 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1545 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1546 
1547 	switch (cmd->attrib.size) {
1548 	case SPDK_NVMF_PROP_SIZE_4:
1549 		size = 4;
1550 		break;
1551 	case SPDK_NVMF_PROP_SIZE_8:
1552 		size = 8;
1553 		break;
1554 	default:
1555 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1556 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1557 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1558 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1559 	}
1560 
1561 	prop = find_prop(cmd->ofst, size);
1562 	if (prop == NULL || prop->set_cb == NULL) {
1563 		SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst);
1564 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1565 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1566 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1567 	}
1568 
1569 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1570 
1571 	value = cmd->value.u64;
1572 
1573 	if (prop->size == 4) {
1574 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1575 	} else if (size != prop->size) {
1576 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1577 		assert(size == 4);
1578 		assert(prop->size == 8);
1579 
1580 		if (cmd->ofst == prop->ofst) {
1581 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1582 		} else {
1583 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1584 		}
1585 	} else {
1586 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1587 		if (ret) {
1588 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1589 		}
1590 	}
1591 
1592 	if (!ret) {
1593 		SPDK_ERRLOG("prop set_cb failed\n");
1594 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1595 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1596 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1597 	}
1598 
1599 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1600 }
1601 
1602 static int
1603 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1604 {
1605 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1606 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1607 
1608 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1609 
1610 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1611 	ctrlr->feat.arbitration.bits.reserved = 0;
1612 
1613 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1614 }
1615 
1616 static int
1617 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1618 {
1619 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1620 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1621 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1622 
1623 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1624 
1625 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1626 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1627 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1628 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1629 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1630 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1631 	}
1632 
1633 	ctrlr->feat.power_management.raw = cmd->cdw11;
1634 	ctrlr->feat.power_management.bits.reserved = 0;
1635 
1636 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1637 }
1638 
1639 static bool
1640 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1641 {
1642 	/*
1643 	 * Valid TMPSEL values:
1644 	 *  0000b - 1000b: temperature sensors
1645 	 *  1111b: set all implemented temperature sensors
1646 	 */
1647 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1648 		/* 1001b - 1110b: reserved */
1649 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1650 		return false;
1651 	}
1652 
1653 	/*
1654 	 * Valid THSEL values:
1655 	 *  00b: over temperature threshold
1656 	 *  01b: under temperature threshold
1657 	 */
1658 	if (opts->bits.thsel > 1) {
1659 		/* 10b - 11b: reserved */
1660 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1661 		return false;
1662 	}
1663 
1664 	return true;
1665 }
1666 
1667 static int
1668 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1669 {
1670 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1671 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1672 
1673 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1674 
1675 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1676 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1677 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1678 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1679 	}
1680 
1681 	/* TODO: no sensors implemented - ignore new values */
1682 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1683 }
1684 
1685 static int
1686 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1687 {
1688 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1689 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1690 
1691 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1692 
1693 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1694 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1695 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1696 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1697 	}
1698 
1699 	/* TODO: no sensors implemented - return 0 for all thresholds */
1700 	rsp->cdw0 = 0;
1701 
1702 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1703 }
1704 
1705 static int
1706 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req)
1707 {
1708 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1709 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1710 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1711 	union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {};
1712 
1713 	SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11);
1714 
1715 	iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv;
1716 	iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd;
1717 	rsp->cdw0 = iv_conf.raw;
1718 
1719 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1720 }
1721 
1722 static int
1723 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1724 {
1725 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1726 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1727 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1728 
1729 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1730 
1731 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1732 		/*
1733 		 * Host is not allowed to set this bit, since we don't advertise it in
1734 		 * Identify Namespace.
1735 		 */
1736 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1737 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1738 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1739 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1740 	}
1741 
1742 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1743 	ctrlr->feat.error_recovery.bits.reserved = 0;
1744 
1745 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1746 }
1747 
1748 static int
1749 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1750 {
1751 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1752 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1753 
1754 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1755 
1756 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1757 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1758 
1759 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1760 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1761 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1762 }
1763 
1764 static int
1765 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1766 {
1767 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1768 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1769 
1770 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1771 
1772 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1773 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1774 
1775 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1776 }
1777 
1778 static int
1779 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1780 {
1781 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1782 
1783 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1784 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1785 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1786 }
1787 
1788 static int
1789 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1790 {
1791 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1792 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1793 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1794 	struct spdk_iov_xfer ix;
1795 
1796 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1797 
1798 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1799 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1800 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1801 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1802 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1803 	}
1804 
1805 	if (req->iovcnt < 1 || req->length < sizeof(ctrlr->hostid)) {
1806 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1807 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1808 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1809 	}
1810 
1811 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
1812 	spdk_iov_xfer_from_buf(&ix, &ctrlr->hostid, sizeof(ctrlr->hostid));
1813 
1814 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1815 }
1816 
1817 static int
1818 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1819 {
1820 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1821 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1822 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1823 	struct spdk_nvmf_ns *ns;
1824 
1825 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notification Mask\n");
1826 
1827 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1828 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1829 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1830 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1831 	}
1832 
1833 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1834 	if (ns == NULL) {
1835 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1836 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1837 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1838 	}
1839 	rsp->cdw0 = ns->mask;
1840 
1841 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1842 }
1843 
1844 static int
1845 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1846 {
1847 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1848 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1849 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1850 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1851 	struct spdk_nvmf_ns *ns;
1852 
1853 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notification Mask\n");
1854 
1855 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1856 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1857 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1858 			ns->mask = cmd->cdw11;
1859 		}
1860 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1861 	}
1862 
1863 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1864 	if (ns == NULL) {
1865 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1866 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1867 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1868 	}
1869 	ns->mask = cmd->cdw11;
1870 
1871 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1872 }
1873 
1874 static int
1875 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1876 {
1877 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1878 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1879 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1880 	struct spdk_nvmf_ns *ns;
1881 
1882 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1883 
1884 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1885 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1886 	if (ns == NULL) {
1887 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1888 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1889 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1890 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1891 	}
1892 
1893 	response->cdw0 = ns->ptpl_activated;
1894 
1895 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1896 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1897 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1898 }
1899 
1900 static int
1901 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req)
1902 {
1903 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1904 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1905 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1906 	struct spdk_nvmf_ns *ns;
1907 	bool ptpl;
1908 
1909 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1910 
1911 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
1912 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1913 
1914 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && nvmf_ns_is_ptpl_capable(ns)) {
1915 		ns->ptpl_activated = ptpl;
1916 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1917 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns;
1918 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1919 			if (nvmf_ns_is_ptpl_capable(ns)) {
1920 				ns->ptpl_activated = ptpl;
1921 			}
1922 		}
1923 	} else {
1924 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1925 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1926 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1927 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1928 	}
1929 
1930 	/* TODO: Feature not changeable for now */
1931 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1932 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1933 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1934 }
1935 
1936 static int
1937 nvmf_ctrlr_get_features_host_behavior_support(struct spdk_nvmf_request *req)
1938 {
1939 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1940 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1941 	struct spdk_nvme_host_behavior host_behavior = {};
1942 	struct spdk_iov_xfer ix;
1943 
1944 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Behavior Support\n");
1945 
1946 	if (req->iovcnt < 1 || req->length < sizeof(struct spdk_nvme_host_behavior)) {
1947 		SPDK_ERRLOG("invalid data buffer for Host Behavior Support\n");
1948 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1949 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1950 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1951 	}
1952 
1953 	host_behavior.acre = ctrlr->acre_enabled;
1954 	host_behavior.lbafee = ctrlr->lbafee_enabled;
1955 
1956 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
1957 	spdk_iov_xfer_from_buf(&ix, &host_behavior, sizeof(host_behavior));
1958 
1959 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1960 }
1961 
1962 static int
1963 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req)
1964 {
1965 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1966 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1967 	struct spdk_nvme_host_behavior *host_behavior;
1968 
1969 	SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n");
1970 	if (req->iovcnt != 1) {
1971 		SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt);
1972 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1973 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1974 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1975 	}
1976 	if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) {
1977 		SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len);
1978 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1979 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1980 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1981 	}
1982 
1983 	host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base;
1984 	if (host_behavior->acre == 0) {
1985 		ctrlr->acre_enabled = false;
1986 	} else if (host_behavior->acre == 1) {
1987 		ctrlr->acre_enabled = true;
1988 	} else {
1989 		SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre);
1990 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1991 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1992 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1993 	}
1994 	if (host_behavior->lbafee == 0) {
1995 		ctrlr->lbafee_enabled = false;
1996 	} else if (host_behavior->lbafee == 1) {
1997 		ctrlr->lbafee_enabled = true;
1998 	} else {
1999 		SPDK_ERRLOG("Host Behavior Support invalid lbafee: 0x%02x\n", host_behavior->lbafee);
2000 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2001 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2002 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2003 	}
2004 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2005 }
2006 
2007 static int
2008 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
2009 {
2010 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2011 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2012 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2013 
2014 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
2015 
2016 	/*
2017 	 * if attempts to disable keep alive by setting kato to 0h
2018 	 * a status value of keep alive invalid shall be returned
2019 	 */
2020 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
2021 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
2022 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
2023 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
2024 	} else {
2025 		/* round up to milliseconds */
2026 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
2027 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
2028 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
2029 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
2030 	}
2031 
2032 	/*
2033 	 * if change the keep alive timeout value successfully
2034 	 * update the keep alive poller.
2035 	 */
2036 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
2037 		if (ctrlr->keep_alive_poller != NULL) {
2038 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
2039 		}
2040 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
2041 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
2042 	}
2043 
2044 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
2045 		      ctrlr->feat.keep_alive_timer.bits.kato);
2046 
2047 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2048 }
2049 
2050 static int
2051 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
2052 {
2053 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2054 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2055 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2056 	uint32_t count;
2057 
2058 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
2059 		      req->cmd->nvme_cmd.cdw11);
2060 
2061 	if (cmd->cdw11_bits.feat_num_of_queues.bits.ncqr == UINT16_MAX ||
2062 	    cmd->cdw11_bits.feat_num_of_queues.bits.nsqr == UINT16_MAX) {
2063 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2064 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2065 	}
2066 
2067 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
2068 	/* verify that the controller is ready to process commands */
2069 	if (count > 1) {
2070 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
2071 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2072 	} else {
2073 		/*
2074 		 * Ignore the value requested by the host -
2075 		 * always return the pre-configured value based on max_qpairs_allowed.
2076 		 */
2077 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
2078 	}
2079 
2080 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2081 }
2082 
2083 SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ctrlr) == 4928,
2084 		   "Please check migration fields that need to be added or not");
2085 
2086 static void
2087 nvmf_ctrlr_migr_data_copy(struct spdk_nvmf_ctrlr_migr_data *data,
2088 			  const struct spdk_nvmf_ctrlr_migr_data *data_src, size_t data_size)
2089 {
2090 	assert(data);
2091 	assert(data_src);
2092 	assert(data_size);
2093 
2094 	memcpy(&data->regs, &data_src->regs, spdk_min(data->regs_size, data_src->regs_size));
2095 	memcpy(&data->feat, &data_src->feat, spdk_min(data->feat_size, data_src->feat_size));
2096 
2097 #define SET_FIELD(field) \
2098     if (offsetof(struct spdk_nvmf_ctrlr_migr_data, field) + sizeof(data->field) <= data_size) { \
2099         data->field = data_src->field; \
2100     } \
2101 
2102 	SET_FIELD(cntlid);
2103 	SET_FIELD(acre);
2104 	SET_FIELD(num_aer_cids);
2105 	SET_FIELD(num_async_events);
2106 	SET_FIELD(notice_aen_mask);
2107 #undef SET_FIELD
2108 
2109 #define SET_ARRAY(arr) \
2110     if (offsetof(struct spdk_nvmf_ctrlr_migr_data, arr) + sizeof(data->arr) <= data_size) { \
2111         memcpy(&data->arr, &data_src->arr, sizeof(data->arr)); \
2112     } \
2113 
2114 	SET_ARRAY(async_events);
2115 	SET_ARRAY(aer_cids);
2116 #undef SET_ARRAY
2117 }
2118 
2119 int
2120 spdk_nvmf_ctrlr_save_migr_data(struct spdk_nvmf_ctrlr *ctrlr,
2121 			       struct spdk_nvmf_ctrlr_migr_data *data)
2122 {
2123 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
2124 	uint32_t i;
2125 	struct spdk_nvmf_ctrlr_migr_data data_local = {
2126 		.data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused),
2127 		.regs_size = sizeof(struct spdk_nvmf_registers),
2128 		.feat_size = sizeof(struct spdk_nvmf_ctrlr_feat)
2129 	};
2130 
2131 	assert(data->data_size <= sizeof(data_local));
2132 	assert(spdk_get_thread() == ctrlr->thread);
2133 
2134 	memcpy(&data_local.regs, &ctrlr->vcprop, sizeof(struct spdk_nvmf_registers));
2135 	memcpy(&data_local.feat, &ctrlr->feat, sizeof(struct spdk_nvmf_ctrlr_feat));
2136 
2137 	data_local.cntlid = ctrlr->cntlid;
2138 	data_local.acre = ctrlr->acre_enabled;
2139 	data_local.num_aer_cids = ctrlr->nr_aer_reqs;
2140 
2141 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
2142 		if (data_local.num_async_events + 1 > SPDK_NVMF_MIGR_MAX_PENDING_AERS) {
2143 			SPDK_ERRLOG("ctrlr %p has too many pending AERs\n", ctrlr);
2144 			break;
2145 		}
2146 
2147 		data_local.async_events[data_local.num_async_events++].raw = event->event.raw;
2148 	}
2149 
2150 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2151 		struct spdk_nvmf_request *req = ctrlr->aer_req[i];
2152 		data_local.aer_cids[i] = req->cmd->nvme_cmd.cid;
2153 	}
2154 	data_local.notice_aen_mask = ctrlr->notice_aen_mask;
2155 
2156 	nvmf_ctrlr_migr_data_copy(data, &data_local, spdk_min(data->data_size, data_local.data_size));
2157 	return 0;
2158 }
2159 
2160 int
2161 spdk_nvmf_ctrlr_restore_migr_data(struct spdk_nvmf_ctrlr *ctrlr,
2162 				  const struct spdk_nvmf_ctrlr_migr_data *data)
2163 {
2164 	uint32_t i;
2165 	struct spdk_nvmf_ctrlr_migr_data data_local = {
2166 		.data_size = offsetof(struct spdk_nvmf_ctrlr_migr_data, unused),
2167 		.regs_size = sizeof(struct spdk_nvmf_registers),
2168 		.feat_size = sizeof(struct spdk_nvmf_ctrlr_feat)
2169 	};
2170 
2171 	assert(data->data_size <= sizeof(data_local));
2172 	assert(spdk_get_thread() == ctrlr->thread);
2173 
2174 	/* local version of data should have defaults set before copy */
2175 	nvmf_ctrlr_migr_data_copy(&data_local, data, spdk_min(data->data_size, data_local.data_size));
2176 	memcpy(&ctrlr->vcprop, &data_local.regs, sizeof(struct spdk_nvmf_registers));
2177 	memcpy(&ctrlr->feat, &data_local.feat, sizeof(struct spdk_nvmf_ctrlr_feat));
2178 
2179 	ctrlr->cntlid = data_local.cntlid;
2180 	ctrlr->acre_enabled = data_local.acre;
2181 
2182 	for (i = 0; i < data_local.num_async_events; i++) {
2183 		struct spdk_nvmf_async_event_completion *event;
2184 
2185 		event = calloc(1, sizeof(*event));
2186 		if (!event) {
2187 			return -ENOMEM;
2188 		}
2189 
2190 		event->event.raw = data_local.async_events[i].raw;
2191 		STAILQ_INSERT_TAIL(&ctrlr->async_events, event, link);
2192 	}
2193 	ctrlr->notice_aen_mask = data_local.notice_aen_mask;
2194 
2195 	return 0;
2196 }
2197 
2198 static int
2199 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
2200 {
2201 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2202 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2203 
2204 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
2205 		      cmd->cdw11);
2206 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
2207 	ctrlr->feat.async_event_configuration.bits.reserved1 = 0;
2208 	ctrlr->feat.async_event_configuration.bits.reserved2 = 0;
2209 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2210 }
2211 
2212 static int
2213 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
2214 {
2215 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2216 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2217 	struct spdk_nvmf_async_event_completion *pending_event;
2218 
2219 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
2220 
2221 	/* Four asynchronous events are supported for now */
2222 	if (ctrlr->nr_aer_reqs >= SPDK_NVMF_MAX_ASYNC_EVENTS) {
2223 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
2224 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
2225 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
2226 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2227 	}
2228 
2229 	if (!STAILQ_EMPTY(&ctrlr->async_events)) {
2230 		pending_event = STAILQ_FIRST(&ctrlr->async_events);
2231 		rsp->cdw0 = pending_event->event.raw;
2232 		STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link);
2233 		free(pending_event);
2234 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2235 	}
2236 
2237 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
2238 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2239 }
2240 
2241 static void
2242 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length)
2243 {
2244 	struct spdk_nvme_firmware_page fw_page;
2245 	size_t copy_len;
2246 	struct spdk_iov_xfer ix;
2247 
2248 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2249 
2250 	memset(&fw_page, 0, sizeof(fw_page));
2251 	fw_page.afi.active_slot = 1;
2252 	fw_page.afi.next_reset_slot = 0;
2253 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
2254 
2255 	if (offset < sizeof(fw_page)) {
2256 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
2257 		if (copy_len > 0) {
2258 			spdk_iov_xfer_from_buf(&ix, (const char *)&fw_page + offset, copy_len);
2259 		}
2260 	}
2261 }
2262 
2263 /*
2264  * Asynchronous Event Mask Bit
2265  */
2266 enum spdk_nvme_async_event_mask_bit {
2267 	/* Mask Namespace Change Notification */
2268 	SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT		= 0,
2269 	/* Mask Asymmetric Namespace Access Change Notification */
2270 	SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT		= 1,
2271 	/* Mask Discovery Log Change Notification */
2272 	SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT	= 2,
2273 	/* Mask Reservation Log Page Available Notification */
2274 	SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT	= 3,
2275 	/* Mask Error Event */
2276 	SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT			= 4,
2277 	/* 4 - 63 Reserved */
2278 };
2279 
2280 static inline void
2281 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2282 		      enum spdk_nvme_async_event_mask_bit mask)
2283 {
2284 	ctrlr->notice_aen_mask &= ~(1 << mask);
2285 }
2286 
2287 static inline bool
2288 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2289 		    enum spdk_nvme_async_event_mask_bit mask)
2290 {
2291 	if (ctrlr->notice_aen_mask & (1 << mask)) {
2292 		return false;
2293 	} else {
2294 		ctrlr->notice_aen_mask |= (1 << mask);
2295 		return true;
2296 	}
2297 }
2298 
2299 /* we have to use the typedef in the function declaration to appease astyle. */
2300 typedef enum spdk_nvme_ana_state spdk_nvme_ana_state_t;
2301 
2302 static inline spdk_nvme_ana_state_t
2303 nvmf_ctrlr_get_ana_state(struct spdk_nvmf_ctrlr *ctrlr, uint32_t anagrpid)
2304 {
2305 	if (!ctrlr->subsys->flags.ana_reporting) {
2306 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2307 	}
2308 
2309 	if (spdk_unlikely(ctrlr->listener == NULL)) {
2310 		return SPDK_NVME_ANA_INACCESSIBLE_STATE;
2311 	}
2312 
2313 	assert(anagrpid - 1 < ctrlr->subsys->max_nsid);
2314 	return ctrlr->listener->ana_state[anagrpid - 1];
2315 }
2316 
2317 static spdk_nvme_ana_state_t
2318 nvmf_ctrlr_get_ana_state_from_nsid(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2319 {
2320 	struct spdk_nvmf_ns *ns;
2321 
2322 	/* We do not have NVM subsystem specific ANA state. Hence if NSID is either
2323 	 * SPDK_NVMF_GLOBAL_NS_TAG, invalid, or for inactive namespace, return
2324 	 * the optimized state.
2325 	 */
2326 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
2327 	if (ns == NULL) {
2328 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2329 	}
2330 
2331 	return nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2332 }
2333 
2334 static void
2335 nvmf_get_error_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2336 			uint64_t offset, uint32_t length, uint32_t rae)
2337 {
2338 	if (!rae) {
2339 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT);
2340 	}
2341 
2342 	/* TODO: actually fill out log page data */
2343 }
2344 
2345 static void
2346 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2347 		      uint64_t offset, uint32_t length, uint32_t rae, uint32_t rgo)
2348 {
2349 	struct spdk_nvme_ana_page ana_hdr;
2350 	struct spdk_nvme_ana_group_descriptor ana_desc;
2351 	size_t copy_len, copied_len;
2352 	uint32_t num_anagrp = 0, anagrpid;
2353 	struct spdk_nvmf_ns *ns;
2354 	struct spdk_iov_xfer ix;
2355 
2356 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2357 
2358 	if (length == 0) {
2359 		goto done;
2360 	}
2361 
2362 	if (offset >= sizeof(ana_hdr)) {
2363 		offset -= sizeof(ana_hdr);
2364 	} else {
2365 		for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2366 			if (ctrlr->subsys->ana_group[anagrpid - 1] > 0) {
2367 				num_anagrp++;
2368 			}
2369 		}
2370 
2371 		memset(&ana_hdr, 0, sizeof(ana_hdr));
2372 
2373 		ana_hdr.num_ana_group_desc = num_anagrp;
2374 		/* TODO: Support Change Count. */
2375 		ana_hdr.change_count = 0;
2376 
2377 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
2378 		copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_hdr + offset, copy_len);
2379 		assert(copied_len == copy_len);
2380 		length -= copied_len;
2381 		offset = 0;
2382 	}
2383 
2384 	if (length == 0) {
2385 		goto done;
2386 	}
2387 
2388 	for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2389 		if (ctrlr->subsys->ana_group[anagrpid - 1] == 0) {
2390 			continue;
2391 		}
2392 
2393 		if (offset >= sizeof(ana_desc)) {
2394 			offset -= sizeof(ana_desc);
2395 		} else {
2396 			memset(&ana_desc, 0, sizeof(ana_desc));
2397 
2398 			ana_desc.ana_group_id = anagrpid;
2399 			if (rgo) {
2400 				ana_desc.num_of_nsid = 0;
2401 			} else {
2402 				ana_desc.num_of_nsid = ctrlr->subsys->ana_group[anagrpid - 1];
2403 			}
2404 			ana_desc.ana_state = nvmf_ctrlr_get_ana_state(ctrlr, anagrpid);
2405 
2406 			copy_len = spdk_min(sizeof(ana_desc) - offset, length);
2407 			copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ana_desc + offset,
2408 							    copy_len);
2409 			assert(copied_len == copy_len);
2410 			length -= copied_len;
2411 			offset = 0;
2412 
2413 			if (length == 0) {
2414 				goto done;
2415 			}
2416 		}
2417 
2418 		if (rgo) {
2419 			continue;
2420 		}
2421 
2422 		/* TODO: Revisit here about O(n^2) cost if we have subsystem with
2423 		 * many namespaces in the future.
2424 		 */
2425 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
2426 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
2427 			if (ns->anagrpid != anagrpid) {
2428 				continue;
2429 			}
2430 
2431 			if (offset >= sizeof(uint32_t)) {
2432 				offset -= sizeof(uint32_t);
2433 				continue;
2434 			}
2435 
2436 			copy_len = spdk_min(sizeof(uint32_t) - offset, length);
2437 			copied_len = spdk_iov_xfer_from_buf(&ix, (const char *)&ns->nsid + offset,
2438 							    copy_len);
2439 			assert(copied_len == copy_len);
2440 			length -= copied_len;
2441 			offset = 0;
2442 
2443 			if (length == 0) {
2444 				goto done;
2445 			}
2446 		}
2447 	}
2448 
2449 done:
2450 	if (!rae) {
2451 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT);
2452 	}
2453 }
2454 
2455 void
2456 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2457 {
2458 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
2459 	uint16_t i;
2460 	bool found = false;
2461 
2462 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
2463 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
2464 			/* nsid is already in the list */
2465 			found = true;
2466 			break;
2467 		}
2468 	}
2469 
2470 	if (!found) {
2471 		if (ctrlr->changed_ns_list_count == max_changes) {
2472 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
2473 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
2474 			for (i = 1; i < max_changes; i++) {
2475 				ctrlr->changed_ns_list.ns_list[i] = 0;
2476 			}
2477 		} else {
2478 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
2479 		}
2480 	}
2481 }
2482 
2483 static void
2484 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2485 				  struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2486 {
2487 	size_t copy_length;
2488 	struct spdk_iov_xfer ix;
2489 
2490 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2491 
2492 	if (offset < sizeof(ctrlr->changed_ns_list)) {
2493 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
2494 		if (copy_length) {
2495 			spdk_iov_xfer_from_buf(&ix, (char *)&ctrlr->changed_ns_list + offset, copy_length);
2496 		}
2497 	}
2498 
2499 	/* Clear log page each time it is read */
2500 	ctrlr->changed_ns_list_count = 0;
2501 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
2502 
2503 	if (!rae) {
2504 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT);
2505 	}
2506 }
2507 
2508 /* The structure can be modified if we provide support for other commands in future */
2509 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
2510 	.admin_cmds_supported = {
2511 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
2512 		/* Get Log Page */
2513 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2514 		/* Identify */
2515 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
2516 		/* Abort */
2517 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
2518 		/* Set Features */
2519 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2520 		/* Get Features */
2521 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2522 		/* Async Event Request */
2523 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
2524 		/* Keep Alive */
2525 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2526 	},
2527 	.io_cmds_supported = {
2528 		/* FLUSH */
2529 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
2530 		/* WRITE */
2531 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
2532 		/* READ */
2533 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
2534 		/* WRITE ZEROES */
2535 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
2536 		/* DATASET MANAGEMENT */
2537 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
2538 		/* COMPARE */
2539 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
2540 		/* ZONE MANAGEMENT SEND */
2541 		[SPDK_NVME_OPC_ZONE_MGMT_SEND]		= {1, 1, 0, 0, 0, 0, 0, 0},
2542 		/* ZONE MANAGEMENT RECEIVE */
2543 		[SPDK_NVME_OPC_ZONE_MGMT_RECV]		= {1, 0, 0, 0, 0, 0, 0, 0},
2544 		/* ZONE APPEND */
2545 		[SPDK_NVME_OPC_ZONE_APPEND]		= {1, 1, 0, 0, 0, 0, 0, 0},
2546 		/* COPY */
2547 		[SPDK_NVME_OPC_COPY]			= {1, 1, 0, 0, 0, 0, 0, 0},
2548 	},
2549 };
2550 
2551 static void
2552 nvmf_get_cmds_and_effects_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2553 				   uint64_t offset, uint32_t length)
2554 {
2555 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
2556 	size_t copy_len = 0;
2557 	struct spdk_nvme_cmds_and_effect_log_page cmds_and_effect_log_page = g_cmds_and_effect_log_page;
2558 	struct spdk_nvme_cmds_and_effect_entry zero = {};
2559 	struct spdk_iov_xfer ix;
2560 
2561 	if (!ctrlr->cdata.oncs.write_zeroes || !nvmf_ctrlr_write_zeroes_supported(ctrlr)) {
2562 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_WRITE_ZEROES] = zero;
2563 	}
2564 	if (!ctrlr->cdata.oncs.dsm || !nvmf_ctrlr_dsm_supported(ctrlr)) {
2565 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_DATASET_MANAGEMENT] = zero;
2566 	}
2567 	if (!ctrlr->cdata.oncs.compare) {
2568 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_COMPARE] = zero;
2569 	}
2570 	if (!nvmf_subsystem_has_zns_iocs(ctrlr->subsys)) {
2571 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_ZONE_MGMT_SEND] = zero;
2572 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_ZONE_MGMT_RECV] = zero;
2573 	}
2574 	if (!nvmf_subsystem_zone_append_supported(ctrlr->subsys)) {
2575 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_ZONE_APPEND] = zero;
2576 	}
2577 	if (!ctrlr->cdata.oncs.copy) {
2578 		cmds_and_effect_log_page.io_cmds_supported[SPDK_NVME_OPC_COPY] = zero;
2579 	}
2580 
2581 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2582 	if (offset < page_size) {
2583 		copy_len = spdk_min(page_size - offset, length);
2584 		spdk_iov_xfer_from_buf(&ix, (char *)(&cmds_and_effect_log_page) + offset, copy_len);
2585 	}
2586 }
2587 
2588 static void
2589 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2590 		struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2591 {
2592 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
2593 	struct spdk_nvmf_reservation_log *log, *log_tmp;
2594 	struct spdk_iov_xfer ix;
2595 
2596 	spdk_iov_xfer_init(&ix, iovs, iovcnt);
2597 
2598 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
2599 	/* No available log, return zeroed log pages */
2600 	if (!ctrlr->num_avail_log_pages) {
2601 		return;
2602 	}
2603 
2604 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
2605 	if (offset >= avail_log_len) {
2606 		return;
2607 	}
2608 
2609 	next_pos = 0;
2610 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
2611 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
2612 		ctrlr->num_avail_log_pages--;
2613 
2614 		next_pos += unit_log_len;
2615 		if (next_pos > offset) {
2616 			copy_len = spdk_min(next_pos - offset, length);
2617 			spdk_iov_xfer_from_buf(&ix, &log->log, copy_len);
2618 			length -= copy_len;
2619 			offset += copy_len;
2620 		}
2621 		free(log);
2622 
2623 		if (length == 0) {
2624 			break;
2625 		}
2626 	}
2627 
2628 	if (!rae) {
2629 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT);
2630 	}
2631 	return;
2632 }
2633 
2634 static int
2635 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
2636 {
2637 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2638 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2639 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2640 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2641 	struct spdk_nvme_transport_id cmd_source_trid;
2642 	uint64_t offset, len;
2643 	uint32_t rae, numdl, numdu;
2644 	uint8_t lid;
2645 
2646 	if (req->iovcnt < 1) {
2647 		SPDK_DEBUGLOG(nvmf, "get log command with no buffer\n");
2648 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2649 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2650 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2651 	}
2652 
2653 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
2654 	if (offset & 3) {
2655 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
2656 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2657 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2658 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2659 	}
2660 
2661 	rae = cmd->cdw10_bits.get_log_page.rae;
2662 	numdl = cmd->cdw10_bits.get_log_page.numdl;
2663 	numdu = cmd->cdw11_bits.get_log_page.numdu;
2664 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
2665 	if (len > req->length) {
2666 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
2667 			    len, req->length);
2668 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2669 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2670 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2671 	}
2672 
2673 	lid = cmd->cdw10_bits.get_log_page.lid;
2674 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n",
2675 		      lid, offset, len, rae);
2676 
2677 	if (spdk_nvmf_subsystem_is_discovery(subsystem)) {
2678 		switch (lid) {
2679 		case SPDK_NVME_LOG_DISCOVERY:
2680 			if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &cmd_source_trid)) {
2681 				SPDK_ERRLOG("Failed to get LOG_DISCOVERY source trid\n");
2682 				response->status.sct = SPDK_NVME_SCT_GENERIC;
2683 				response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
2684 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2685 			}
2686 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt,
2687 						    offset, len, &cmd_source_trid);
2688 			if (!rae) {
2689 				nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT);
2690 			}
2691 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2692 		default:
2693 			goto invalid_log_page;
2694 		}
2695 	} else {
2696 		if (offset > len) {
2697 			SPDK_ERRLOG("Get log page: offset (%" PRIu64 ") > len (%" PRIu64 ")\n",
2698 				    offset, len);
2699 			response->status.sct = SPDK_NVME_SCT_GENERIC;
2700 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2701 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2702 		}
2703 
2704 		switch (lid) {
2705 		case SPDK_NVME_LOG_ERROR:
2706 			nvmf_get_error_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2707 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2708 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
2709 			/* TODO: actually fill out log page data */
2710 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2711 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
2712 			nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len);
2713 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2714 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
2715 			if (subsystem->flags.ana_reporting) {
2716 				uint32_t rgo = cmd->cdw10_bits.get_log_page.lsp & 1;
2717 				nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae, rgo);
2718 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2719 			} else {
2720 				goto invalid_log_page;
2721 			}
2722 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
2723 			nvmf_get_cmds_and_effects_log_page(ctrlr, req->iov, req->iovcnt, offset, len);
2724 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2725 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
2726 			nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2727 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2728 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
2729 			nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2730 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2731 		default:
2732 			goto invalid_log_page;
2733 		}
2734 	}
2735 
2736 invalid_log_page:
2737 	SPDK_INFOLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid);
2738 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2739 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2740 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2741 }
2742 
2743 static struct spdk_nvmf_ns *
2744 _nvmf_ctrlr_get_ns_safe(struct spdk_nvmf_ctrlr *ctrlr,
2745 			uint32_t nsid,
2746 			struct spdk_nvme_cpl *rsp)
2747 {
2748 	struct spdk_nvmf_ns *ns;
2749 	if (nsid == 0 || nsid > ctrlr->subsys->max_nsid) {
2750 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", nsid);
2751 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2752 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2753 		return NULL;
2754 	}
2755 
2756 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
2757 	if (ns == NULL || ns->bdev == NULL) {
2758 		/*
2759 		 * Inactive namespaces should return a zero filled data structure.
2760 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2761 		 * so we can just return early here.
2762 		 */
2763 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", nsid);
2764 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2765 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2766 		return NULL;
2767 	}
2768 	return ns;
2769 }
2770 
2771 static void
2772 nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2773 		       struct spdk_nvme_cmd *cmd,
2774 		       struct spdk_nvme_cpl *rsp,
2775 		       struct spdk_nvme_ns_data *nsdata,
2776 		       uint32_t nsid)
2777 {
2778 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2779 	struct spdk_nvmf_ns *ns;
2780 	uint32_t max_num_blocks, format_index;
2781 	enum spdk_nvme_ana_state ana_state;
2782 
2783 	ns = _nvmf_ctrlr_get_ns_safe(ctrlr, nsid, rsp);
2784 	if (ns == NULL) {
2785 		return;
2786 	}
2787 
2788 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2789 
2790 	assert(ctrlr->admin_qpair);
2791 
2792 	format_index = spdk_nvme_ns_get_format_index(nsdata);
2793 
2794 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2795 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2796 			 (1U << nsdata->lbaf[format_index].lbads);
2797 	if (nsdata->noiob > max_num_blocks) {
2798 		nsdata->noiob = max_num_blocks;
2799 	}
2800 
2801 	/* Set NOWS equal to Controller MDTS */
2802 	if (nsdata->nsfeat.optperf) {
2803 		nsdata->nows = max_num_blocks - 1;
2804 	}
2805 
2806 	if (subsystem->flags.ana_reporting) {
2807 		assert(ns->anagrpid - 1 < subsystem->max_nsid);
2808 		nsdata->anagrpid = ns->anagrpid;
2809 
2810 		ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2811 		if (ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2812 		    ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2813 			nsdata->nuse = 0;
2814 		}
2815 	}
2816 }
2817 
2818 int
2819 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2820 			    struct spdk_nvme_cmd *cmd,
2821 			    struct spdk_nvme_cpl *rsp,
2822 			    struct spdk_nvme_ns_data *nsdata)
2823 {
2824 	nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, nsdata, cmd->nsid);
2825 
2826 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2827 }
2828 
2829 static void
2830 identify_ns_passthru_cb(struct spdk_nvmf_request *req)
2831 {
2832 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
2833 	struct spdk_nvmf_ctrlr *ctrlr = spdk_nvmf_request_get_ctrlr(req);
2834 	struct spdk_nvme_cpl *rsp = spdk_nvmf_request_get_response(req);
2835 	struct spdk_nvme_ns_data nvmf_nsdata = {};
2836 	struct spdk_nvme_ns_data nvme_nsdata = {};
2837 	size_t datalen;
2838 
2839 	/* This is the identify data from the NVMe drive */
2840 	datalen = spdk_nvmf_request_copy_to_buf(req, &nvme_nsdata,
2841 						sizeof(nvme_nsdata));
2842 	nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, &nvmf_nsdata, req->orig_nsid);
2843 
2844 	/* Update fabric's namespace according to SSD's namespace */
2845 	if (nvme_nsdata.nsfeat.optperf) {
2846 		nvmf_nsdata.nsfeat.optperf = nvme_nsdata.nsfeat.optperf;
2847 		nvmf_nsdata.npwg = nvme_nsdata.npwg;
2848 		nvmf_nsdata.npwa = nvme_nsdata.npwa;
2849 		nvmf_nsdata.npdg = nvme_nsdata.npdg;
2850 		nvmf_nsdata.npda = nvme_nsdata.npda;
2851 		nvmf_nsdata.nows = nvme_nsdata.nows;
2852 	}
2853 
2854 	if (nvme_nsdata.nsfeat.ns_atomic_write_unit) {
2855 		nvmf_nsdata.nsfeat.ns_atomic_write_unit = nvme_nsdata.nsfeat.ns_atomic_write_unit;
2856 		nvmf_nsdata.nawun = nvme_nsdata.nawun;
2857 		nvmf_nsdata.nawupf = nvme_nsdata.nawupf;
2858 		nvmf_nsdata.nacwu = nvme_nsdata.nacwu;
2859 	}
2860 
2861 	nvmf_nsdata.nabsn = nvme_nsdata.nabsn;
2862 	nvmf_nsdata.nabo = nvme_nsdata.nabo;
2863 	nvmf_nsdata.nabspf = nvme_nsdata.nabspf;
2864 
2865 	spdk_nvmf_request_copy_from_buf(req, &nvmf_nsdata, datalen);
2866 }
2867 
2868 int
2869 spdk_nvmf_ctrlr_identify_ns_ext(struct spdk_nvmf_request *req)
2870 {
2871 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
2872 	struct spdk_nvmf_ctrlr *ctrlr = spdk_nvmf_request_get_ctrlr(req);
2873 	struct spdk_nvmf_ns *ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
2874 	struct spdk_nvme_cpl *rsp = spdk_nvmf_request_get_response(req);
2875 	struct spdk_bdev *bdev;
2876 	struct spdk_bdev_desc *desc;
2877 	struct spdk_io_channel *ch;
2878 	struct spdk_nvme_ns_data nsdata = {};
2879 	struct spdk_iov_xfer ix;
2880 	int rc;
2881 
2882 	nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, &nsdata, cmd->nsid);
2883 
2884 	rc = spdk_nvmf_request_get_bdev(cmd->nsid, req, &bdev, &desc, &ch);
2885 	if (rc) {
2886 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2887 	}
2888 
2889 	if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN)) {
2890 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
2891 		spdk_iov_xfer_from_buf(&ix, &nsdata, sizeof(nsdata));
2892 
2893 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2894 	}
2895 
2896 	assert(ns->passthru_nsid != 0);
2897 	req->orig_nsid = ns->nsid;
2898 	cmd->nsid = ns->passthru_nsid;
2899 
2900 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, identify_ns_passthru_cb);
2901 }
2902 
2903 static void
2904 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2905 			 struct spdk_nvme_ctrlr_data *cdata)
2906 {
2907 	cdata->oacs = ctrlr->cdata.oacs;
2908 
2909 	cdata->oacs.virtualization_management =
2910 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2911 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2912 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2913 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2914 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2915 	cdata->oacs.device_self_test =
2916 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2917 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2918 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2919 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2920 			       NULL
2921 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2922 	cdata->oacs.format =
2923 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2924 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2925 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2926 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2927 				     NULL;
2928 }
2929 
2930 int
2931 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2932 {
2933 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2934 	struct spdk_nvmf_transport *transport;
2935 
2936 	/*
2937 	 * Common fields for discovery and NVM subsystems
2938 	 */
2939 	assert(ctrlr->admin_qpair);
2940 	transport = ctrlr->admin_qpair->transport;
2941 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2942 	assert((transport->opts.max_io_size % 4096) == 0);
2943 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2944 	cdata->cntlid = ctrlr->cntlid;
2945 	cdata->ver = ctrlr->vcprop.vs;
2946 	cdata->aerl = ctrlr->cdata.aerl;
2947 	cdata->lpa.edlp = 1;
2948 	cdata->elpe = 127;
2949 	cdata->maxcmd = transport->opts.max_queue_depth;
2950 	cdata->sgls = ctrlr->cdata.sgls;
2951 	cdata->fuses = ctrlr->cdata.fuses;
2952 	cdata->acwu = 0; /* ACWU is 0-based. */
2953 	if (subsystem->flags.ana_reporting) {
2954 		cdata->mnan = subsystem->max_nsid;
2955 	}
2956 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2957 
2958 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2959 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2960 
2961 
2962 	if (spdk_nvmf_subsystem_is_discovery(subsystem)) {
2963 		/*
2964 		 * NVM Discovery subsystem fields
2965 		 */
2966 		cdata->oaes.discovery_log_change_notices = 1;
2967 		cdata->cntrltype = SPDK_NVME_CTRLR_DISCOVERY;
2968 	} else {
2969 		cdata->vid = ctrlr->cdata.vid;
2970 		cdata->ssvid = ctrlr->cdata.ssvid;
2971 		cdata->ieee[0] = ctrlr->cdata.ieee[0];
2972 		cdata->ieee[1] = ctrlr->cdata.ieee[1];
2973 		cdata->ieee[2] = ctrlr->cdata.ieee[2];
2974 
2975 		/*
2976 		 * NVM subsystem fields (reserved for discovery subsystems)
2977 		 */
2978 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2979 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2980 		cdata->kas = ctrlr->cdata.kas;
2981 
2982 		cdata->rab = 6;
2983 		cdata->cmic.multi_port = 1;
2984 		cdata->cmic.multi_ctrlr = 1;
2985 		cdata->oaes.ns_attribute_notices = 1;
2986 		cdata->ctratt.bits.host_id_exhid_supported = 1;
2987 		cdata->ctratt.bits.fdps = ctrlr->subsys->fdp_supported;
2988 		cdata->cntrltype = SPDK_NVME_CTRLR_IO;
2989 		/* We do not have any actual limitation to the number of abort commands.
2990 		 * We follow the recommendation by the NVMe specification.
2991 		 */
2992 		cdata->acl = NVMF_ABORT_COMMAND_LIMIT;
2993 		cdata->frmw.slot1_ro = 1;
2994 		cdata->frmw.num_slots = 1;
2995 
2996 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2997 
2998 		cdata->sqes.min = 6;
2999 		cdata->sqes.max = 6;
3000 		cdata->cqes.min = 4;
3001 		cdata->cqes.max = 4;
3002 		cdata->nn = subsystem->max_nsid;
3003 		cdata->vwc.present = 1;
3004 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
3005 
3006 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
3007 
3008 		cdata->oncs.compare = ctrlr->cdata.oncs.compare;
3009 		cdata->oncs.dsm = ctrlr->cdata.oncs.dsm && nvmf_ctrlr_dsm_supported(ctrlr);
3010 		cdata->oncs.write_zeroes = ctrlr->cdata.oncs.write_zeroes &&
3011 					   nvmf_ctrlr_write_zeroes_supported(ctrlr);
3012 		cdata->oncs.reservations = ctrlr->cdata.oncs.reservations;
3013 		cdata->oncs.copy = ctrlr->cdata.oncs.copy;
3014 		cdata->ocfs.copy_format0 = cdata->oncs.copy;
3015 		if (subsystem->flags.ana_reporting) {
3016 			/* Asymmetric Namespace Access Reporting is supported. */
3017 			cdata->cmic.ana_reporting = 1;
3018 			cdata->oaes.ana_change_notices = 1;
3019 
3020 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
3021 			/* ANA Change state is not used, and ANA Persistent Loss state
3022 			 * is not supported for now.
3023 			 */
3024 			cdata->anacap.ana_optimized_state = 1;
3025 			cdata->anacap.ana_non_optimized_state = 1;
3026 			cdata->anacap.ana_inaccessible_state = 1;
3027 			/* ANAGRPID does not change while namespace is attached to controller */
3028 			cdata->anacap.no_change_anagrpid = 1;
3029 			cdata->anagrpmax = subsystem->max_nsid;
3030 			cdata->nanagrpid = subsystem->max_nsid;
3031 		}
3032 
3033 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
3034 
3035 		assert(subsystem->tgt != NULL);
3036 		cdata->crdt[0] = subsystem->tgt->crdt[0];
3037 		cdata->crdt[1] = subsystem->tgt->crdt[1];
3038 		cdata->crdt[2] = subsystem->tgt->crdt[2];
3039 
3040 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
3041 			      cdata->nvmf_specific.ioccsz);
3042 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
3043 			      cdata->nvmf_specific.iorcsz);
3044 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
3045 			      cdata->nvmf_specific.icdoff);
3046 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
3047 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
3048 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
3049 			      cdata->nvmf_specific.msdbd);
3050 	}
3051 
3052 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3053 }
3054 
3055 static int
3056 nvmf_ns_identify_iocs_zns(struct spdk_nvmf_ns *ns,
3057 			  struct spdk_nvme_cmd *cmd,
3058 			  struct spdk_nvme_cpl *rsp,
3059 			  struct spdk_nvme_zns_ns_data *nsdata_zns)
3060 {
3061 	nsdata_zns->zoc.variable_zone_capacity = 0;
3062 	nsdata_zns->zoc.zone_active_excursions = 0;
3063 	nsdata_zns->ozcs.read_across_zone_boundaries = 1;
3064 	/* Underflowing the zero based mar and mor bdev helper results in the correct
3065 	   value of FFFFFFFFh. */
3066 	nsdata_zns->mar = spdk_bdev_get_max_active_zones(ns->bdev) - 1;
3067 	nsdata_zns->mor = spdk_bdev_get_max_open_zones(ns->bdev) - 1;
3068 	nsdata_zns->rrl = 0;
3069 	nsdata_zns->frl = 0;
3070 	nsdata_zns->lbafe[0].zsze = spdk_bdev_get_zone_size(ns->bdev);
3071 
3072 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3073 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3074 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3075 }
3076 
3077 static int
3078 nvmf_ns_identify_iocs_nvm(struct spdk_nvmf_ns *ns,
3079 			  struct spdk_nvme_cpl *rsp,
3080 			  struct spdk_nvme_nvm_ns_data *nsdata_nvm)
3081 {
3082 	nvmf_bdev_ctrlr_identify_iocs_nvm(ns, nsdata_nvm);
3083 
3084 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3085 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3086 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3087 }
3088 
3089 int
3090 spdk_nvmf_ns_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr,
3091 				    struct spdk_nvme_cmd *cmd,
3092 				    struct spdk_nvme_cpl *rsp,
3093 				    void *nsdata,
3094 				    size_t nsdata_size)
3095 {
3096 	uint8_t csi = cmd->cdw11_bits.identify.csi;
3097 	struct spdk_nvmf_ns *ns = _nvmf_ctrlr_get_ns_safe(ctrlr, cmd->nsid, rsp);
3098 
3099 	memset(nsdata, 0, nsdata_size);
3100 
3101 	if (ns == NULL) {
3102 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3103 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3104 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3105 	}
3106 
3107 	switch (csi) {
3108 	case SPDK_NVME_CSI_ZNS:
3109 		return nvmf_ns_identify_iocs_zns(ns, cmd, rsp, nsdata);
3110 	case SPDK_NVME_CSI_NVM:
3111 		if (!ctrlr->dif_insert_or_strip) {
3112 			return nvmf_ns_identify_iocs_nvm(ns, rsp, nsdata);
3113 		}
3114 		break;
3115 	default:
3116 		break;
3117 	}
3118 
3119 	SPDK_DEBUGLOG(nvmf,
3120 		      "Returning zero filled struct for the iocs specific ns "
3121 		      "identify command and CSI 0x%02x\n",
3122 		      csi);
3123 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3124 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3125 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3126 }
3127 
3128 static int
3129 nvmf_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ctrlr *ctrlr,
3130 			     struct spdk_nvme_cmd *cmd,
3131 			     struct spdk_nvme_cpl *rsp,
3132 			     struct spdk_nvme_nvm_ctrlr_data *cdata_nvm)
3133 {
3134 	/* The unit of max_write_zeroes_size_kib is KiB.
3135 	 * The unit of wzsl is the minimum memory page size(2 ^ (12 + CAP.MPSMIN) bytes)
3136 	 * and is reported as a power of two (2^n).
3137 	 */
3138 	cdata_nvm->wzsl = spdk_u64log2(ctrlr->subsys->max_write_zeroes_size_kib >>
3139 				       (2 + ctrlr->vcprop.cap.bits.mpsmin));
3140 
3141 	/* The unit of max_discard_size_kib is KiB.
3142 	 * The dmrsl indicates the maximum number of logical blocks for
3143 	 * dataset management command.
3144 	 */
3145 	cdata_nvm->dmrsl = ctrlr->subsys->max_discard_size_kib << 1;
3146 	cdata_nvm->dmrl = 1;
3147 
3148 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3149 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3150 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3151 }
3152 
3153 static int
3154 nvmf_ctrlr_identify_iocs_zns(struct spdk_nvmf_ctrlr *ctrlr,
3155 			     struct spdk_nvme_cmd *cmd,
3156 			     struct spdk_nvme_cpl *rsp,
3157 			     struct spdk_nvme_zns_ctrlr_data *cdata_zns)
3158 {
3159 	/* The unit of max_zone_append_size_kib is KiB.
3160 	The unit of zasl is the minimum memory page size
3161 	(2 ^ (12 + CAP.MPSMIN) KiB)
3162 	and is reported as a power of two (2^n). */
3163 	cdata_zns->zasl = spdk_u64log2(ctrlr->subsys->max_zone_append_size_kib >>
3164 				       (12 + ctrlr->vcprop.cap.bits.mpsmin));
3165 
3166 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3167 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3168 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3169 }
3170 
3171 int
3172 spdk_nvmf_ctrlr_identify_iocs_specific(struct spdk_nvmf_ctrlr *ctrlr,
3173 				       struct spdk_nvme_cmd *cmd,
3174 				       struct spdk_nvme_cpl *rsp,
3175 				       void *cdata,
3176 				       size_t cdata_size)
3177 {
3178 	uint8_t csi = cmd->cdw11_bits.identify.csi;
3179 
3180 	memset(cdata, 0, cdata_size);
3181 
3182 	switch (csi) {
3183 	case SPDK_NVME_CSI_NVM:
3184 		return nvmf_ctrlr_identify_iocs_nvm(ctrlr, cmd, rsp, cdata);
3185 	case SPDK_NVME_CSI_ZNS:
3186 		return nvmf_ctrlr_identify_iocs_zns(ctrlr, cmd, rsp, cdata);
3187 	default:
3188 		break;
3189 	}
3190 
3191 	SPDK_DEBUGLOG(nvmf,
3192 		      "Returning zero filled struct for the iocs specific ctrlr "
3193 		      "identify command and CSI 0x%02x\n",
3194 		      csi);
3195 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3196 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3197 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3198 }
3199 
3200 static int
3201 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_ctrlr *ctrlr,
3202 				   struct spdk_nvme_cmd *cmd,
3203 				   struct spdk_nvme_cpl *rsp,
3204 				   struct spdk_nvme_ns_list *ns_list)
3205 {
3206 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
3207 	struct spdk_nvmf_ns *ns;
3208 	uint32_t count = 0;
3209 
3210 	if (cmd->nsid >= 0xfffffffeUL) {
3211 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
3212 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3213 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3214 	}
3215 
3216 	memset(ns_list, 0, sizeof(*ns_list));
3217 
3218 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
3219 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
3220 		if (ns->opts.nsid <= cmd->nsid || !nvmf_ctrlr_ns_is_visible(ctrlr, ns->opts.nsid)) {
3221 			continue;
3222 		}
3223 
3224 		ns_list->ns_list[count++] = ns->opts.nsid;
3225 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
3226 			break;
3227 		}
3228 	}
3229 
3230 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3231 }
3232 
3233 static void
3234 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
3235 		enum spdk_nvme_nidt type,
3236 		const void *data, size_t data_size)
3237 {
3238 	struct spdk_nvme_ns_id_desc *desc;
3239 	size_t desc_size = sizeof(*desc) + data_size;
3240 
3241 	/*
3242 	 * These should never fail in practice, since all valid NS ID descriptors
3243 	 * should be defined so that they fit in the available 4096-byte buffer.
3244 	 */
3245 	assert(data_size > 0);
3246 	assert(data_size <= UINT8_MAX);
3247 	assert(desc_size < *buf_remain);
3248 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
3249 		return;
3250 	}
3251 
3252 	desc = *buf_ptr;
3253 	desc->nidt = type;
3254 	desc->nidl = data_size;
3255 	memcpy(desc->nid, data, data_size);
3256 
3257 	*buf_ptr += desc_size;
3258 	*buf_remain -= desc_size;
3259 }
3260 
3261 static int
3262 nvmf_ctrlr_identify_ns_id_descriptor_list(
3263 	struct spdk_nvmf_ctrlr *ctrlr,
3264 	struct spdk_nvme_cmd *cmd,
3265 	struct spdk_nvme_cpl *rsp,
3266 	void *id_desc_list, size_t id_desc_list_size)
3267 {
3268 	struct spdk_nvmf_ns *ns;
3269 	size_t buf_remain = id_desc_list_size;
3270 	void *buf_ptr = id_desc_list;
3271 
3272 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
3273 	if (ns == NULL || ns->bdev == NULL) {
3274 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3275 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3276 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3277 	}
3278 
3279 #define ADD_ID_DESC(type, data, size) \
3280 	do { \
3281 		if (!spdk_mem_all_zero(data, size)) { \
3282 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
3283 		} \
3284 	} while (0)
3285 
3286 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
3287 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
3288 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
3289 	ADD_ID_DESC(SPDK_NVME_NIDT_CSI, &ns->csi, sizeof(uint8_t));
3290 
3291 	/*
3292 	 * The list is automatically 0-terminated, both in the temporary buffer
3293 	 * used by nvmf_ctrlr_identify(), and the eventual iov destination -
3294 	 * controller to host buffers in admin commands always get zeroed in
3295 	 * nvmf_ctrlr_process_admin_cmd().
3296 	 */
3297 
3298 #undef ADD_ID_DESC
3299 
3300 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3301 }
3302 
3303 static int
3304 nvmf_ctrlr_identify_iocs(struct spdk_nvmf_ctrlr *ctrlr,
3305 			 struct spdk_nvme_cmd *cmd,
3306 			 struct spdk_nvme_cpl *rsp,
3307 			 void *cdata, size_t cdata_size)
3308 {
3309 	struct spdk_nvme_iocs_vector *vector;
3310 	struct spdk_nvmf_ns *ns;
3311 
3312 	if (cdata_size < sizeof(struct spdk_nvme_iocs_vector)) {
3313 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3314 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3315 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3316 	}
3317 
3318 	/* For now we only support this command sent to the current
3319 	 * controller.
3320 	 */
3321 	if (cmd->cdw10_bits.identify.cntid != 0xFFFF &&
3322 	    cmd->cdw10_bits.identify.cntid != ctrlr->cntlid) {
3323 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3324 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3325 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3326 	}
3327 	memset(cdata, 0, cdata_size);
3328 
3329 	vector = cdata;
3330 	vector->nvm = 1;
3331 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
3332 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
3333 		if (ns->bdev == NULL) {
3334 			continue;
3335 		}
3336 		if (spdk_bdev_is_zoned(ns->bdev)) {
3337 			vector->zns = 1;
3338 		}
3339 	}
3340 
3341 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3342 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3343 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3344 }
3345 
3346 static int
3347 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
3348 {
3349 	uint8_t cns;
3350 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3351 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3352 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3353 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
3354 	int ret = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3355 	char tmpbuf[SPDK_NVME_IDENTIFY_BUFLEN] = "";
3356 	struct spdk_iov_xfer ix;
3357 
3358 	if (req->iovcnt < 1 || req->length < SPDK_NVME_IDENTIFY_BUFLEN) {
3359 		SPDK_DEBUGLOG(nvmf, "identify command with invalid buffer\n");
3360 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3361 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3362 		return ret;
3363 	}
3364 
3365 	cns = cmd->cdw10_bits.identify.cns;
3366 
3367 	if (spdk_nvmf_subsystem_is_discovery(subsystem) &&
3368 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
3369 		/* Discovery controllers only support Identify Controller */
3370 		goto invalid_cns;
3371 	}
3372 
3373 	/*
3374 	 * We must use a temporary buffer: it's entirely possible the out buffer
3375 	 * is split across more than one IOV.
3376 	 */
3377 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3378 
3379 	SPDK_DEBUGLOG(nvmf, "Received identify command with CNS 0x%02x\n", cns);
3380 
3381 	switch (cns) {
3382 	case SPDK_NVME_IDENTIFY_NS:
3383 		/* Function below can be asynchronous & we always need to have the data in request's buffer
3384 		 * So just return here */
3385 		return spdk_nvmf_ctrlr_identify_ns_ext(req);
3386 	case SPDK_NVME_IDENTIFY_CTRLR:
3387 		ret = spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, (void *)&tmpbuf);
3388 		break;
3389 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
3390 		ret = nvmf_ctrlr_identify_active_ns_list(ctrlr, cmd, rsp, (void *)&tmpbuf);
3391 		break;
3392 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
3393 		ret = nvmf_ctrlr_identify_ns_id_descriptor_list(ctrlr, cmd, rsp,
3394 				tmpbuf, req->length);
3395 		break;
3396 	case SPDK_NVME_IDENTIFY_NS_IOCS:
3397 		ret = spdk_nvmf_ns_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length);
3398 		break;
3399 	case SPDK_NVME_IDENTIFY_CTRLR_IOCS:
3400 		ret = spdk_nvmf_ctrlr_identify_iocs_specific(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length);
3401 		break;
3402 	case SPDK_NVME_IDENTIFY_IOCS:
3403 		ret = nvmf_ctrlr_identify_iocs(ctrlr, cmd, rsp, (void *)&tmpbuf, req->length);
3404 		break;
3405 	default:
3406 		goto invalid_cns;
3407 	}
3408 
3409 	if (ret == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3410 		spdk_iov_xfer_from_buf(&ix, tmpbuf, sizeof(tmpbuf));
3411 	}
3412 
3413 	return ret;
3414 
3415 invalid_cns:
3416 	SPDK_DEBUGLOG(nvmf, "Identify command with unsupported CNS 0x%02x\n", cns);
3417 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3418 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3419 	return ret;
3420 }
3421 
3422 static bool
3423 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
3424 {
3425 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3426 	struct spdk_nvmf_request *req;
3427 	int i;
3428 
3429 	if (!nvmf_qpair_is_admin_queue(qpair)) {
3430 		return false;
3431 	}
3432 
3433 	assert(spdk_get_thread() == ctrlr->thread);
3434 
3435 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3436 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
3437 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
3438 			req = ctrlr->aer_req[i];
3439 			ctrlr->aer_req[i] = NULL;
3440 			ctrlr->nr_aer_reqs--;
3441 
3442 			/* Move the last req to the aborting position for making aer_reqs
3443 			 * in continuous
3444 			 */
3445 			if (i < ctrlr->nr_aer_reqs) {
3446 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
3447 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3448 			}
3449 
3450 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3451 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3452 			_nvmf_request_complete(req);
3453 			return true;
3454 		}
3455 	}
3456 
3457 	return false;
3458 }
3459 
3460 void
3461 nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair)
3462 {
3463 	struct spdk_nvmf_request *req, *tmp;
3464 
3465 	TAILQ_FOREACH_SAFE(req, &qpair->outstanding, link, tmp) {
3466 		if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) {
3467 			/* Zero-copy requests are kept on the outstanding queue from the moment
3468 			 * zcopy_start is sent until a zcopy_end callback is received.  Therefore,
3469 			 * we can't remove them from the outstanding queue here, but need to rely on
3470 			 * the transport to do a zcopy_end to release their buffers and, in turn,
3471 			 * remove them from the queue.
3472 			 */
3473 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3474 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3475 			nvmf_transport_req_free(req);
3476 		}
3477 	}
3478 }
3479 
3480 static void
3481 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
3482 {
3483 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
3484 
3485 	if (nvmf_qpair_abort_aer(qpair, cid)) {
3486 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
3487 			      qpair->ctrlr, qpair->qid, cid);
3488 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
3489 
3490 		spdk_nvmf_request_complete(req);
3491 		return;
3492 	}
3493 
3494 	nvmf_transport_qpair_abort_request(qpair, req);
3495 }
3496 
3497 static void
3498 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
3499 {
3500 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
3501 
3502 	if (status == 0) {
3503 		/* There was no qpair whose ID matches SQID of the abort command.
3504 		 * Hence call _nvmf_request_complete() here.
3505 		 */
3506 		_nvmf_request_complete(req);
3507 	}
3508 }
3509 
3510 static void
3511 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
3512 {
3513 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
3514 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
3515 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
3516 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
3517 	struct spdk_nvmf_qpair *qpair;
3518 
3519 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
3520 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
3521 			/* Found the qpair */
3522 
3523 			nvmf_qpair_abort_request(qpair, req);
3524 
3525 			/* Return -1 for the status so the iteration across threads stops. */
3526 			spdk_for_each_channel_continue(i, -1);
3527 			return;
3528 		}
3529 	}
3530 
3531 	spdk_for_each_channel_continue(i, 0);
3532 }
3533 
3534 static int
3535 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
3536 {
3537 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3538 
3539 	rsp->cdw0 = 1U; /* Command not aborted */
3540 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3541 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3542 
3543 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
3544 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
3545 			      nvmf_ctrlr_abort_on_pg,
3546 			      req,
3547 			      nvmf_ctrlr_abort_done
3548 			     );
3549 
3550 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3551 }
3552 
3553 int
3554 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
3555 {
3556 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
3557 	struct spdk_bdev *bdev;
3558 	struct spdk_bdev_desc *desc;
3559 	struct spdk_io_channel *ch;
3560 	int rc;
3561 
3562 	assert(req_to_abort != NULL);
3563 
3564 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
3565 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
3566 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
3567 	}
3568 
3569 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
3570 					&bdev, &desc, &ch);
3571 	if (rc != 0) {
3572 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3573 	}
3574 
3575 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
3576 }
3577 
3578 static int
3579 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
3580 {
3581 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3582 
3583 	rsp->cdw0 = cdw0;
3584 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3585 }
3586 
3587 /* we have to use the typedef in the function declaration to appease astyle. */
3588 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
3589 
3590 static spdk_nvme_path_status_code_t
3591 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
3592 {
3593 	switch (ana_state) {
3594 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3595 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
3596 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3597 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3598 	case SPDK_NVME_ANA_CHANGE_STATE:
3599 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
3600 	default:
3601 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
3602 	}
3603 }
3604 
3605 static int
3606 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
3607 {
3608 	uint8_t feature;
3609 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3610 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3611 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3612 	enum spdk_nvme_ana_state ana_state;
3613 
3614 	feature = cmd->cdw10_bits.get_features.fid;
3615 
3616 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
3617 		/*
3618 		 * Features supported by Discovery controller
3619 		 */
3620 		switch (feature) {
3621 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3622 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3623 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3624 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3625 		default:
3626 			SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3627 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3628 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3629 		}
3630 	}
3631 	/*
3632 	 * Process Get Features command for non-discovery controller
3633 	 */
3634 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3635 	switch (ana_state) {
3636 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3637 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3638 	case SPDK_NVME_ANA_CHANGE_STATE:
3639 		switch (feature) {
3640 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
3641 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3642 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3643 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3644 			response->status.sct = SPDK_NVME_SCT_PATH;
3645 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3646 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3647 		default:
3648 			break;
3649 		}
3650 		break;
3651 	default:
3652 		break;
3653 	}
3654 
3655 	switch (feature) {
3656 	case SPDK_NVME_FEAT_ARBITRATION:
3657 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
3658 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3659 		return get_features_generic(req, ctrlr->feat.power_management.raw);
3660 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3661 		return nvmf_ctrlr_get_features_temperature_threshold(req);
3662 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3663 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
3664 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3665 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
3666 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3667 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
3668 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3669 		return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw);
3670 	case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION:
3671 		return nvmf_ctrlr_get_features_interrupt_vector_configuration(req);
3672 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3673 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
3674 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3675 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3676 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3677 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3678 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3679 		return nvmf_ctrlr_get_features_host_identifier(req);
3680 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3681 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
3682 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3683 		return nvmf_ctrlr_get_features_reservation_persistence(req);
3684 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3685 		return nvmf_ctrlr_get_features_host_behavior_support(req);
3686 	default:
3687 		SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3688 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3689 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3690 	}
3691 }
3692 
3693 static int
3694 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
3695 {
3696 	uint8_t feature, save;
3697 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3698 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3699 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3700 	enum spdk_nvme_ana_state ana_state;
3701 	/*
3702 	 * Features are not saveable by the controller as indicated by
3703 	 * ONCS field of the Identify Controller data.
3704 	 * */
3705 	save = cmd->cdw10_bits.set_features.sv;
3706 	if (save) {
3707 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
3708 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3709 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3710 	}
3711 
3712 	feature = cmd->cdw10_bits.set_features.fid;
3713 
3714 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
3715 		/*
3716 		 * Features supported by Discovery controller
3717 		 */
3718 		switch (feature) {
3719 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3720 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
3721 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3722 			return nvmf_ctrlr_set_features_async_event_configuration(req);
3723 		default:
3724 			SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3725 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3726 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3727 		}
3728 	}
3729 	/*
3730 	 * Process Set Features command for non-discovery controller
3731 	 */
3732 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3733 	switch (ana_state) {
3734 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3735 	case SPDK_NVME_ANA_CHANGE_STATE:
3736 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
3737 			response->status.sct = SPDK_NVME_SCT_PATH;
3738 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3739 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3740 		} else {
3741 			switch (feature) {
3742 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
3743 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3744 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3745 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3746 				response->status.sct = SPDK_NVME_SCT_PATH;
3747 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3748 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3749 			default:
3750 				break;
3751 			}
3752 		}
3753 		break;
3754 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3755 		response->status.sct = SPDK_NVME_SCT_PATH;
3756 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3757 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3758 	default:
3759 		break;
3760 	}
3761 
3762 	switch (feature) {
3763 	case SPDK_NVME_FEAT_ARBITRATION:
3764 		return nvmf_ctrlr_set_features_arbitration(req);
3765 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3766 		return nvmf_ctrlr_set_features_power_management(req);
3767 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3768 		return nvmf_ctrlr_set_features_temperature_threshold(req);
3769 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3770 		return nvmf_ctrlr_set_features_error_recovery(req);
3771 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3772 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
3773 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3774 		return nvmf_ctrlr_set_features_number_of_queues(req);
3775 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3776 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3777 		response->status.sc = SPDK_NVME_SC_FEATURE_NOT_CHANGEABLE;
3778 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3779 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3780 		return nvmf_ctrlr_set_features_write_atomicity(req);
3781 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3782 		return nvmf_ctrlr_set_features_async_event_configuration(req);
3783 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3784 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
3785 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3786 		return nvmf_ctrlr_set_features_host_identifier(req);
3787 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3788 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
3789 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3790 		return nvmf_ctrlr_set_features_reservation_persistence(req);
3791 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3792 		return nvmf_ctrlr_set_features_host_behavior_support(req);
3793 	default:
3794 		SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3795 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3796 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3797 	}
3798 }
3799 
3800 static int
3801 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
3802 {
3803 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3804 
3805 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
3806 	/*
3807 	 * To handle keep alive just clear or reset the
3808 	 * ctrlr based keep alive duration counter.
3809 	 * When added, a separate timer based process
3810 	 * will monitor if the time since last recorded
3811 	 * keep alive has exceeded the max duration and
3812 	 * take appropriate action.
3813 	 */
3814 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
3815 
3816 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3817 }
3818 
3819 int
3820 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
3821 {
3822 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3823 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3824 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3825 	struct spdk_nvmf_subsystem_poll_group *sgroup;
3826 	int rc;
3827 
3828 	assert(ctrlr != NULL);
3829 	if (cmd->opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST) {
3830 		/* We do not want to treat AERs as outstanding commands,
3831 		 * so decrement mgmt_io_outstanding here to offset
3832 		 * the increment that happened prior to this call.
3833 		 */
3834 		sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
3835 		assert(sgroup != NULL);
3836 		sgroup->mgmt_io_outstanding--;
3837 	}
3838 
3839 	assert(spdk_get_thread() == ctrlr->thread);
3840 
3841 	if (cmd->fuse != 0) {
3842 		/* Fused admin commands are not supported. */
3843 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3844 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3845 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3846 	}
3847 
3848 	if (ctrlr->vcprop.cc.bits.en != 1) {
3849 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
3850 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3851 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3852 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3853 	}
3854 
3855 	if (req->iovcnt && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
3856 		spdk_iov_memset(req->iov, req->iovcnt, 0);
3857 	}
3858 
3859 	if (spdk_nvmf_subsystem_is_discovery(ctrlr->subsys)) {
3860 		/* Discovery controllers only support these admin OPS. */
3861 		switch (cmd->opc) {
3862 		case SPDK_NVME_OPC_IDENTIFY:
3863 		case SPDK_NVME_OPC_GET_LOG_PAGE:
3864 		case SPDK_NVME_OPC_KEEP_ALIVE:
3865 		case SPDK_NVME_OPC_SET_FEATURES:
3866 		case SPDK_NVME_OPC_GET_FEATURES:
3867 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3868 			break;
3869 		default:
3870 			goto invalid_opcode;
3871 		}
3872 	}
3873 
3874 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
3875 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
3876 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
3877 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3878 			/* The handler took care of this command */
3879 			return rc;
3880 		}
3881 	}
3882 
3883 	/* We only want to send passthrough admin commands to namespaces.
3884 	 * However, we don't want to passthrough a command with intended for all namespaces.
3885 	 */
3886 	if (ctrlr->subsys->passthrough && cmd->nsid && cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG) {
3887 		return nvmf_passthru_admin_cmd(req);
3888 	}
3889 
3890 	switch (cmd->opc) {
3891 	case SPDK_NVME_OPC_GET_LOG_PAGE:
3892 		return nvmf_ctrlr_get_log_page(req);
3893 	case SPDK_NVME_OPC_IDENTIFY:
3894 		return nvmf_ctrlr_identify(req);
3895 	case SPDK_NVME_OPC_ABORT:
3896 		return nvmf_ctrlr_abort(req);
3897 	case SPDK_NVME_OPC_GET_FEATURES:
3898 		return nvmf_ctrlr_get_features(req);
3899 	case SPDK_NVME_OPC_SET_FEATURES:
3900 		return nvmf_ctrlr_set_features(req);
3901 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3902 		return nvmf_ctrlr_async_event_request(req);
3903 	case SPDK_NVME_OPC_KEEP_ALIVE:
3904 		return nvmf_ctrlr_keep_alive(req);
3905 
3906 	case SPDK_NVME_OPC_CREATE_IO_SQ:
3907 	case SPDK_NVME_OPC_CREATE_IO_CQ:
3908 	case SPDK_NVME_OPC_DELETE_IO_SQ:
3909 	case SPDK_NVME_OPC_DELETE_IO_CQ:
3910 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
3911 		goto invalid_opcode;
3912 
3913 	default:
3914 		goto invalid_opcode;
3915 	}
3916 
3917 invalid_opcode:
3918 	SPDK_INFOLOG(nvmf, "Unsupported admin opcode 0x%x\n", cmd->opc);
3919 	response->status.sct = SPDK_NVME_SCT_GENERIC;
3920 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3921 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3922 }
3923 
3924 static int
3925 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
3926 {
3927 	struct spdk_nvmf_qpair *qpair = req->qpair;
3928 	struct spdk_nvmf_capsule_cmd *cap_hdr;
3929 
3930 	cap_hdr = &req->cmd->nvmf_cmd;
3931 
3932 	if (qpair->ctrlr == NULL) {
3933 		/* No ctrlr established yet; the only valid command is Connect */
3934 		assert(cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT);
3935 		return nvmf_ctrlr_cmd_connect(req);
3936 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
3937 		/*
3938 		 * Controller session is established, and this is an admin queue.
3939 		 * Disallow Connect and allow other fabrics commands.
3940 		 */
3941 		switch (cap_hdr->fctype) {
3942 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
3943 			return nvmf_property_set(req);
3944 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
3945 			return nvmf_property_get(req);
3946 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_SEND:
3947 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_RECV:
3948 			return nvmf_auth_request_exec(req);
3949 		default:
3950 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
3951 				      cap_hdr->fctype);
3952 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3953 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3954 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3955 		}
3956 	} else {
3957 		/*
3958 		 * Controller session is established, and this is an I/O queue.
3959 		 * Disallow everything besides authentication commands.
3960 		 */
3961 		switch (cap_hdr->fctype) {
3962 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_SEND:
3963 		case SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_RECV:
3964 			return nvmf_auth_request_exec(req);
3965 		default:
3966 			SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
3967 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3968 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3969 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3970 		}
3971 	}
3972 }
3973 
3974 static inline void
3975 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr,
3976 				     union spdk_nvme_async_event_completion *event)
3977 {
3978 	struct spdk_nvmf_async_event_completion *nvmf_event;
3979 
3980 	nvmf_event = calloc(1, sizeof(*nvmf_event));
3981 	if (!nvmf_event) {
3982 		SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n");
3983 		return;
3984 	}
3985 	nvmf_event->event.raw = event->raw;
3986 	STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link);
3987 }
3988 
3989 static inline int
3990 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
3991 				    union spdk_nvme_async_event_completion *event)
3992 {
3993 	struct spdk_nvmf_request *req;
3994 	struct spdk_nvme_cpl *rsp;
3995 
3996 	assert(spdk_get_thread() == ctrlr->thread);
3997 
3998 	/* If there is no outstanding AER request, queue the event.  Then
3999 	 * if an AER is later submitted, this event can be sent as a
4000 	 * response.
4001 	 */
4002 	if (ctrlr->nr_aer_reqs == 0) {
4003 		nvmf_ctrlr_queue_pending_async_event(ctrlr, event);
4004 		return 0;
4005 	}
4006 
4007 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
4008 	rsp = &req->rsp->nvme_cpl;
4009 
4010 	rsp->cdw0 = event->raw;
4011 
4012 	_nvmf_request_complete(req);
4013 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
4014 
4015 	return 0;
4016 }
4017 
4018 int
4019 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
4020 {
4021 	union spdk_nvme_async_event_completion event = {0};
4022 
4023 	/* Users may disable the event notification */
4024 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
4025 		return 0;
4026 	}
4027 
4028 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) {
4029 		return 0;
4030 	}
4031 
4032 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
4033 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
4034 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
4035 
4036 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
4037 }
4038 
4039 int
4040 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
4041 {
4042 	union spdk_nvme_async_event_completion event = {0};
4043 
4044 	/* Users may disable the event notification */
4045 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
4046 		return 0;
4047 	}
4048 
4049 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) {
4050 		return 0;
4051 	}
4052 
4053 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
4054 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
4055 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
4056 
4057 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
4058 }
4059 
4060 void
4061 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
4062 {
4063 	union spdk_nvme_async_event_completion event = {0};
4064 
4065 	if (!ctrlr->num_avail_log_pages) {
4066 		return;
4067 	}
4068 
4069 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) {
4070 		return;
4071 	}
4072 
4073 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
4074 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
4075 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
4076 
4077 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
4078 }
4079 
4080 void
4081 nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx)
4082 {
4083 	union spdk_nvme_async_event_completion event = {0};
4084 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
4085 
4086 	/* Users may disable the event notification manually or
4087 	 * it may not be enabled due to keep alive timeout
4088 	 * not being set in connect command to discovery controller.
4089 	 */
4090 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
4091 		return;
4092 	}
4093 
4094 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) {
4095 		return;
4096 	}
4097 
4098 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
4099 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
4100 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
4101 
4102 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
4103 }
4104 
4105 int
4106 spdk_nvmf_ctrlr_async_event_error_event(struct spdk_nvmf_ctrlr *ctrlr,
4107 					enum spdk_nvme_async_event_info_error info)
4108 {
4109 	union spdk_nvme_async_event_completion event;
4110 
4111 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT)) {
4112 		return 0;
4113 	}
4114 
4115 	if (info > SPDK_NVME_ASYNC_EVENT_FW_IMAGE_LOAD) {
4116 		return 0;
4117 	}
4118 
4119 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_ERROR;
4120 	event.bits.log_page_identifier = SPDK_NVME_LOG_ERROR;
4121 	event.bits.async_event_info = info;
4122 
4123 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
4124 }
4125 
4126 void
4127 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
4128 {
4129 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4130 	int i;
4131 
4132 	if (ctrlr == NULL || !nvmf_qpair_is_admin_queue(qpair)) {
4133 		return;
4134 	}
4135 
4136 	assert(spdk_get_thread() == ctrlr->thread);
4137 
4138 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
4139 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
4140 		ctrlr->aer_req[i] = NULL;
4141 	}
4142 
4143 	ctrlr->nr_aer_reqs = 0;
4144 }
4145 
4146 void
4147 spdk_nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
4148 {
4149 	struct spdk_nvmf_request *req;
4150 	int i;
4151 
4152 	assert(spdk_get_thread() == ctrlr->thread);
4153 
4154 	if (!ctrlr->nr_aer_reqs) {
4155 		return;
4156 	}
4157 
4158 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
4159 		req = ctrlr->aer_req[i];
4160 
4161 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4162 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
4163 		_nvmf_request_complete(req);
4164 
4165 		ctrlr->aer_req[i] = NULL;
4166 	}
4167 
4168 	ctrlr->nr_aer_reqs = 0;
4169 }
4170 
4171 static void
4172 _nvmf_ctrlr_add_reservation_log(void *ctx)
4173 {
4174 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
4175 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
4176 
4177 	ctrlr->log_page_count++;
4178 
4179 	/* Maximum number of queued log pages is 255 */
4180 	if (ctrlr->num_avail_log_pages == 0xff) {
4181 		struct spdk_nvmf_reservation_log *entry;
4182 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
4183 		entry->log.log_page_count = ctrlr->log_page_count;
4184 		free(log);
4185 		return;
4186 	}
4187 
4188 	log->log.log_page_count = ctrlr->log_page_count;
4189 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
4190 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
4191 
4192 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
4193 }
4194 
4195 void
4196 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
4197 				  struct spdk_nvmf_ns *ns,
4198 				  enum spdk_nvme_reservation_notification_log_page_type type)
4199 {
4200 	struct spdk_nvmf_reservation_log *log;
4201 
4202 	switch (type) {
4203 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
4204 		return;
4205 	case SPDK_NVME_REGISTRATION_PREEMPTED:
4206 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
4207 			return;
4208 		}
4209 		break;
4210 	case SPDK_NVME_RESERVATION_RELEASED:
4211 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
4212 			return;
4213 		}
4214 		break;
4215 	case SPDK_NVME_RESERVATION_PREEMPTED:
4216 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
4217 			return;
4218 		}
4219 		break;
4220 	default:
4221 		return;
4222 	}
4223 
4224 	log = calloc(1, sizeof(*log));
4225 	if (!log) {
4226 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
4227 		return;
4228 	}
4229 	log->ctrlr = ctrlr;
4230 	log->log.type = type;
4231 	log->log.nsid = ns->nsid;
4232 
4233 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
4234 }
4235 
4236 /* Check from subsystem poll group's namespace information data structure */
4237 static bool
4238 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
4239 				 struct spdk_nvmf_ctrlr *ctrlr)
4240 {
4241 	uint32_t i;
4242 
4243 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
4244 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
4245 			return true;
4246 		}
4247 	}
4248 
4249 	return false;
4250 }
4251 
4252 /*
4253  * Check the NVMe command is permitted or not for current controller(Host).
4254  */
4255 static int
4256 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
4257 				  struct spdk_nvmf_ctrlr *ctrlr,
4258 				  struct spdk_nvmf_request *req)
4259 {
4260 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
4261 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
4262 	uint8_t status = SPDK_NVME_SC_SUCCESS;
4263 	uint8_t racqa;
4264 	bool is_registrant;
4265 
4266 	/* No valid reservation */
4267 	if (!rtype) {
4268 		return 0;
4269 	}
4270 
4271 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
4272 	/* All registrants type and current ctrlr is a valid registrant */
4273 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
4274 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
4275 		return 0;
4276 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
4277 		return 0;
4278 	}
4279 
4280 	/* Non-holder for current controller */
4281 	switch (cmd->opc) {
4282 	case SPDK_NVME_OPC_READ:
4283 	case SPDK_NVME_OPC_COMPARE:
4284 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
4285 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4286 			goto exit;
4287 		}
4288 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
4289 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
4290 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4291 		}
4292 		break;
4293 	case SPDK_NVME_OPC_FLUSH:
4294 	case SPDK_NVME_OPC_WRITE:
4295 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
4296 	case SPDK_NVME_OPC_WRITE_ZEROES:
4297 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
4298 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
4299 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
4300 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4301 			goto exit;
4302 		}
4303 		if (!is_registrant) {
4304 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4305 		}
4306 		break;
4307 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
4308 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
4309 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
4310 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4311 			goto exit;
4312 		}
4313 		if (!is_registrant) {
4314 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4315 		}
4316 		break;
4317 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
4318 		if (!is_registrant) {
4319 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
4320 		}
4321 		break;
4322 	default:
4323 		break;
4324 	}
4325 
4326 exit:
4327 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4328 	req->rsp->nvme_cpl.status.sc = status;
4329 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
4330 		return -EPERM;
4331 	}
4332 
4333 	return 0;
4334 }
4335 
4336 static int
4337 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
4338 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
4339 {
4340 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
4341 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
4342 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
4343 	int rc;
4344 
4345 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
4346 		/* first fused operation (should be compare) */
4347 		if (first_fused_req != NULL) {
4348 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
4349 
4350 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
4351 
4352 			/* abort req->qpair->first_fused_request and continue with new fused command */
4353 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4354 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4355 			_nvmf_request_complete(first_fused_req);
4356 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
4357 			SPDK_ERRLOG("Wrong op code of fused operations\n");
4358 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4359 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
4360 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4361 		}
4362 
4363 		req->qpair->first_fused_req = req;
4364 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
4365 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
4366 		/* second fused operation (should be write) */
4367 		if (first_fused_req == NULL) {
4368 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
4369 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4370 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4371 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4372 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
4373 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
4374 
4375 			SPDK_ERRLOG("Wrong op code of fused operations\n");
4376 
4377 			/* abort req->qpair->first_fused_request and fail current command */
4378 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4379 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4380 			_nvmf_request_complete(first_fused_req);
4381 
4382 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4383 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
4384 			req->qpair->first_fused_req = NULL;
4385 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4386 		}
4387 
4388 		/* save request of first command to generate response later */
4389 		req->first_fused_req = first_fused_req;
4390 		req->first_fused = true;
4391 		req->qpair->first_fused_req = NULL;
4392 	} else {
4393 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
4394 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4395 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
4396 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4397 	}
4398 
4399 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
4400 
4401 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4402 		if (spdk_nvme_cpl_is_error(rsp)) {
4403 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
4404 
4405 			fused_response->status = rsp->status;
4406 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
4407 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
4408 			/* Complete first of fused commands. Second will be completed by upper layer */
4409 			_nvmf_request_complete(first_fused_req);
4410 			req->first_fused_req = NULL;
4411 			req->first_fused = false;
4412 		}
4413 	}
4414 
4415 	return rc;
4416 }
4417 
4418 bool
4419 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req)
4420 {
4421 	struct spdk_nvmf_transport *transport = req->qpair->transport;
4422 	struct spdk_nvmf_ns *ns;
4423 
4424 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE);
4425 
4426 	if (!transport->opts.zcopy) {
4427 		return false;
4428 	}
4429 
4430 	if (nvmf_qpair_is_admin_queue(req->qpair)) {
4431 		/* Admin queue */
4432 		return false;
4433 	}
4434 
4435 	if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) &&
4436 	    (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) {
4437 		/* Not a READ or WRITE command */
4438 		return false;
4439 	}
4440 
4441 	if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) {
4442 		/* Fused commands dont use zcopy buffers */
4443 		return false;
4444 	}
4445 
4446 	ns = nvmf_ctrlr_get_ns(req->qpair->ctrlr, req->cmd->nvme_cmd.nsid);
4447 	if (ns == NULL || ns->bdev == NULL || !ns->zcopy) {
4448 		return false;
4449 	}
4450 
4451 	req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT;
4452 	return true;
4453 }
4454 
4455 void
4456 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req)
4457 {
4458 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
4459 
4460 	/* Set iovcnt to be the maximum number of iovs that the ZCOPY can use */
4461 	req->iovcnt = NVMF_REQ_MAX_BUFFERS;
4462 
4463 	spdk_nvmf_request_exec(req);
4464 }
4465 
4466 void
4467 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req, bool commit)
4468 {
4469 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE);
4470 	req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING;
4471 
4472 	nvmf_bdev_ctrlr_zcopy_end(req, commit);
4473 }
4474 
4475 int
4476 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
4477 {
4478 	uint32_t nsid;
4479 	struct spdk_nvmf_ns *ns;
4480 	struct spdk_bdev *bdev;
4481 	struct spdk_bdev_desc *desc;
4482 	struct spdk_io_channel *ch;
4483 	struct spdk_nvmf_qpair *qpair = req->qpair;
4484 	struct spdk_nvmf_poll_group *group = qpair->group;
4485 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4486 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
4487 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
4488 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4489 	enum spdk_nvme_ana_state ana_state;
4490 
4491 	/* pre-set response details for this command */
4492 	response->status.sc = SPDK_NVME_SC_SUCCESS;
4493 	nsid = cmd->nsid;
4494 
4495 	assert(ctrlr != NULL);
4496 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
4497 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
4498 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4499 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
4500 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4501 	}
4502 
4503 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
4504 	if (spdk_unlikely(ns == NULL || ns->bdev == NULL)) {
4505 		SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid);
4506 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4507 		response->status.dnr = 1;
4508 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4509 	}
4510 
4511 	ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
4512 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
4513 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
4514 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
4515 			      ana_state);
4516 		response->status.sct = SPDK_NVME_SCT_PATH;
4517 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
4518 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4519 	}
4520 
4521 	if (spdk_likely(ctrlr->listener != NULL)) {
4522 		SPDK_DTRACE_PROBE3_TICKS(nvmf_request_io_exec_path, req,
4523 					 ctrlr->listener->trid->traddr,
4524 					 ctrlr->listener->trid->trsvcid);
4525 	}
4526 
4527 	/* scan-build falsely reporting dereference of null pointer */
4528 	assert(group != NULL && group->sgroups != NULL);
4529 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
4530 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
4531 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
4532 			      cmd->nsid, cmd->opc);
4533 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4534 	}
4535 
4536 	bdev = ns->bdev;
4537 	desc = ns->desc;
4538 	ch = ns_info->channel;
4539 
4540 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
4541 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
4542 	} else if (spdk_unlikely(qpair->first_fused_req != NULL)) {
4543 		struct spdk_nvme_cpl *fused_response = &qpair->first_fused_req->rsp->nvme_cpl;
4544 
4545 		SPDK_ERRLOG("Second fused cmd expected - failing first one (cntlid:%u, qid:%u, opcode:0x%x)\n",
4546 			    ctrlr->cntlid, qpair->qid,
4547 			    req->qpair->first_fused_req->cmd->nvmf_cmd.opcode);
4548 
4549 		/* abort qpair->first_fused_request and continue with new command */
4550 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4551 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4552 		_nvmf_request_complete(qpair->first_fused_req);
4553 		qpair->first_fused_req = NULL;
4554 	}
4555 
4556 	if (ctrlr->subsys->passthrough) {
4557 		assert(ns->passthru_nsid > 0);
4558 		req->cmd->nvme_cmd.nsid = ns->passthru_nsid;
4559 
4560 		return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
4561 	}
4562 
4563 	if (spdk_nvmf_request_using_zcopy(req)) {
4564 		assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
4565 		return nvmf_bdev_ctrlr_zcopy_start(bdev, desc, ch, req);
4566 	} else {
4567 		switch (cmd->opc) {
4568 		case SPDK_NVME_OPC_READ:
4569 			return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
4570 		case SPDK_NVME_OPC_WRITE:
4571 			return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
4572 		case SPDK_NVME_OPC_FLUSH:
4573 			return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
4574 		case SPDK_NVME_OPC_COMPARE:
4575 			if (spdk_unlikely(!ctrlr->cdata.oncs.compare)) {
4576 				goto invalid_opcode;
4577 			}
4578 			return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
4579 		case SPDK_NVME_OPC_WRITE_ZEROES:
4580 			if (spdk_unlikely(!ctrlr->cdata.oncs.write_zeroes)) {
4581 				goto invalid_opcode;
4582 			}
4583 			return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
4584 		case SPDK_NVME_OPC_DATASET_MANAGEMENT:
4585 			if (spdk_unlikely(!ctrlr->cdata.oncs.dsm)) {
4586 				goto invalid_opcode;
4587 			}
4588 			return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
4589 		case SPDK_NVME_OPC_RESERVATION_REGISTER:
4590 		case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
4591 		case SPDK_NVME_OPC_RESERVATION_RELEASE:
4592 		case SPDK_NVME_OPC_RESERVATION_REPORT:
4593 			if (spdk_unlikely(!ctrlr->cdata.oncs.reservations)) {
4594 				goto invalid_opcode;
4595 			}
4596 			spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
4597 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
4598 		case SPDK_NVME_OPC_COPY:
4599 			if (spdk_unlikely(!ctrlr->cdata.oncs.copy)) {
4600 				goto invalid_opcode;
4601 			}
4602 			return nvmf_bdev_ctrlr_copy_cmd(bdev, desc, ch, req);
4603 		default:
4604 			if (spdk_unlikely(qpair->transport->opts.disable_command_passthru)) {
4605 				goto invalid_opcode;
4606 			}
4607 			if (ns->passthru_nsid) {
4608 				req->cmd->nvme_cmd.nsid = ns->passthru_nsid;
4609 			}
4610 			return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
4611 		}
4612 	}
4613 invalid_opcode:
4614 	SPDK_INFOLOG(nvmf, "Unsupported IO opcode 0x%x\n", cmd->opc);
4615 	response->status.sct = SPDK_NVME_SCT_GENERIC;
4616 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
4617 	response->status.dnr = 1;
4618 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4619 }
4620 
4621 static void
4622 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
4623 {
4624 	if (spdk_unlikely(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING)) {
4625 		assert(qpair->state_cb != NULL);
4626 
4627 		if (TAILQ_EMPTY(&qpair->outstanding)) {
4628 			qpair->state_cb(qpair->state_cb_arg, 0);
4629 		}
4630 	}
4631 }
4632 
4633 int
4634 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
4635 {
4636 	struct spdk_nvmf_qpair *qpair = req->qpair;
4637 
4638 	TAILQ_REMOVE(&qpair->outstanding, req, link);
4639 	if (spdk_unlikely(nvmf_transport_req_free(req))) {
4640 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
4641 	}
4642 
4643 	nvmf_qpair_request_cleanup(qpair);
4644 
4645 	return 0;
4646 }
4647 
4648 static void
4649 _nvmf_request_complete(void *ctx)
4650 {
4651 	struct spdk_nvmf_request *req = ctx;
4652 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
4653 	struct spdk_nvmf_qpair *qpair;
4654 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4655 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4656 	bool is_aer = false;
4657 	uint32_t nsid;
4658 	bool paused;
4659 	uint8_t opcode;
4660 
4661 	rsp->sqid = 0;
4662 	rsp->status.p = 0;
4663 	rsp->cid = req->cmd->nvme_cmd.cid;
4664 	nsid = req->cmd->nvme_cmd.nsid;
4665 	opcode = req->cmd->nvmf_cmd.opcode;
4666 
4667 	qpair = req->qpair;
4668 	if (spdk_likely(qpair->ctrlr)) {
4669 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4670 		assert(sgroup != NULL);
4671 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
4672 		if (spdk_likely(qpair->qid != 0)) {
4673 			qpair->group->stat.completed_nvme_io++;
4674 		}
4675 
4676 		/*
4677 		 * Set the crd value.
4678 		 * If the the IO has any error, and dnr (DoNotRetry) is not 1,
4679 		 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT.
4680 		 */
4681 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp) &&
4682 				  rsp->status.dnr == 0 &&
4683 				  qpair->ctrlr->acre_enabled)) {
4684 			rsp->status.crd = 1;
4685 		}
4686 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4687 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4688 	}
4689 
4690 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4691 		spdk_nvme_print_completion(qpair->qid, rsp);
4692 	}
4693 
4694 	switch (req->zcopy_phase) {
4695 	case NVMF_ZCOPY_PHASE_NONE:
4696 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4697 		break;
4698 	case NVMF_ZCOPY_PHASE_INIT:
4699 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) {
4700 			req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED;
4701 			TAILQ_REMOVE(&qpair->outstanding, req, link);
4702 		} else {
4703 			req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE;
4704 		}
4705 		break;
4706 	case NVMF_ZCOPY_PHASE_EXECUTE:
4707 		break;
4708 	case NVMF_ZCOPY_PHASE_END_PENDING:
4709 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4710 		req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE;
4711 		break;
4712 	default:
4713 		SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase);
4714 		break;
4715 	}
4716 
4717 	if (spdk_unlikely(nvmf_transport_req_complete(req))) {
4718 		SPDK_ERRLOG("Transport request completion error!\n");
4719 	}
4720 
4721 	/* AER cmd is an exception */
4722 	if (spdk_likely(sgroup && !is_aer)) {
4723 		if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC ||
4724 				  nvmf_qpair_is_admin_queue(qpair))) {
4725 			assert(sgroup->mgmt_io_outstanding > 0);
4726 			sgroup->mgmt_io_outstanding--;
4727 		} else {
4728 			if (req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE ||
4729 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE ||
4730 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT_FAILED) {
4731 				/* End of request */
4732 
4733 				/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4734 				if (spdk_likely(nsid - 1 < sgroup->num_ns)) {
4735 					sgroup->ns_info[nsid - 1].io_outstanding--;
4736 				}
4737 			}
4738 		}
4739 
4740 		if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4741 				  sgroup->mgmt_io_outstanding == 0)) {
4742 			paused = true;
4743 			for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
4744 				ns_info = &sgroup->ns_info[nsid];
4745 
4746 				if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4747 				    ns_info->io_outstanding > 0) {
4748 					paused = false;
4749 					break;
4750 				}
4751 			}
4752 
4753 			if (paused) {
4754 				sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
4755 				sgroup->cb_fn(sgroup->cb_arg, 0);
4756 				sgroup->cb_fn = NULL;
4757 				sgroup->cb_arg = NULL;
4758 			}
4759 		}
4760 
4761 	}
4762 
4763 	nvmf_qpair_request_cleanup(qpair);
4764 }
4765 
4766 int
4767 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
4768 {
4769 	struct spdk_nvmf_qpair *qpair = req->qpair;
4770 
4771 	spdk_thread_exec_msg(qpair->group->thread, _nvmf_request_complete, req);
4772 
4773 	return 0;
4774 }
4775 
4776 static bool
4777 nvmf_check_subsystem_active(struct spdk_nvmf_request *req)
4778 {
4779 	struct spdk_nvmf_qpair *qpair = req->qpair;
4780 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4781 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4782 	uint32_t nsid;
4783 
4784 	if (spdk_likely(qpair->ctrlr)) {
4785 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4786 		assert(sgroup != NULL);
4787 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4788 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4789 	}
4790 
4791 	/* Check if the subsystem is paused (if there is a subsystem) */
4792 	if (spdk_unlikely(sgroup == NULL)) {
4793 		return true;
4794 	}
4795 
4796 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC ||
4797 			  nvmf_qpair_is_admin_queue(qpair))) {
4798 		if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
4799 			/* The subsystem is not currently active. Queue this request. */
4800 			TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4801 			return false;
4802 		}
4803 		sgroup->mgmt_io_outstanding++;
4804 	} else {
4805 		nsid = req->cmd->nvme_cmd.nsid;
4806 
4807 		/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4808 		if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) {
4809 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4810 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4811 			req->rsp->nvme_cpl.status.dnr = 1;
4812 			TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4813 			_nvmf_request_complete(req);
4814 			return false;
4815 		}
4816 
4817 		ns_info = &sgroup->ns_info[nsid - 1];
4818 		if (spdk_unlikely(ns_info->channel == NULL)) {
4819 			/* This can can happen if host sends I/O to a namespace that is
4820 			 * in the process of being added, but before the full addition
4821 			 * process is complete.  Report invalid namespace in that case.
4822 			 */
4823 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4824 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4825 			req->rsp->nvme_cpl.status.dnr = 1;
4826 			TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4827 			ns_info->io_outstanding++;
4828 			_nvmf_request_complete(req);
4829 			return false;
4830 		}
4831 
4832 		if (spdk_unlikely(ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE)) {
4833 			/* The namespace is not currently active. Queue this request. */
4834 			TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4835 			return false;
4836 		}
4837 
4838 		ns_info->io_outstanding++;
4839 	}
4840 
4841 	return true;
4842 }
4843 
4844 static bool
4845 nvmf_check_qpair_active(struct spdk_nvmf_request *req)
4846 {
4847 	struct spdk_nvmf_qpair *qpair = req->qpair;
4848 	int sc, sct;
4849 
4850 	if (spdk_likely(qpair->state == SPDK_NVMF_QPAIR_ENABLED)) {
4851 		return true;
4852 	}
4853 
4854 	sct = SPDK_NVME_SCT_GENERIC;
4855 	sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
4856 
4857 	switch (qpair->state) {
4858 	case SPDK_NVMF_QPAIR_CONNECTING:
4859 		if (req->cmd->nvmf_cmd.opcode != SPDK_NVME_OPC_FABRIC) {
4860 			SPDK_ERRLOG("Received command 0x%x on qid %u before CONNECT\n",
4861 				    req->cmd->nvmf_cmd.opcode, qpair->qid);
4862 			break;
4863 		}
4864 		if (req->cmd->nvmf_cmd.fctype != SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
4865 			SPDK_ERRLOG("Received fctype 0x%x on qid %u before CONNECT\n",
4866 				    req->cmd->nvmf_cmd.fctype, qpair->qid);
4867 			break;
4868 		}
4869 		return true;
4870 	case SPDK_NVMF_QPAIR_AUTHENTICATING:
4871 		sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
4872 		sc = SPDK_NVMF_FABRIC_SC_AUTH_REQUIRED;
4873 		if (req->cmd->nvmf_cmd.opcode != SPDK_NVME_OPC_FABRIC) {
4874 			SPDK_ERRLOG("Received command 0x%x on qid %u before authentication\n",
4875 				    req->cmd->nvmf_cmd.opcode, qpair->qid);
4876 			break;
4877 		}
4878 		if (req->cmd->nvmf_cmd.fctype != SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_SEND &&
4879 		    req->cmd->nvmf_cmd.fctype != SPDK_NVMF_FABRIC_COMMAND_AUTHENTICATION_RECV) {
4880 			SPDK_ERRLOG("Received fctype 0x%x on qid %u before authentication\n",
4881 				    req->cmd->nvmf_cmd.fctype, qpair->qid);
4882 			break;
4883 		}
4884 		return true;
4885 	default:
4886 		SPDK_ERRLOG("Received command 0x%x on qid %u in state %d\n",
4887 			    req->cmd->nvmf_cmd.opcode, qpair->qid, qpair->state);
4888 		break;
4889 	}
4890 
4891 	req->rsp->nvme_cpl.status.sct = sct;
4892 	req->rsp->nvme_cpl.status.sc = sc;
4893 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4894 	_nvmf_request_complete(req);
4895 
4896 	return false;
4897 }
4898 
4899 void
4900 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
4901 {
4902 	struct spdk_nvmf_qpair *qpair = req->qpair;
4903 	enum spdk_nvmf_request_exec_status status;
4904 
4905 	if (spdk_unlikely(!nvmf_check_subsystem_active(req))) {
4906 		return;
4907 	}
4908 	if (spdk_unlikely(!nvmf_check_qpair_active(req))) {
4909 		return;
4910 	}
4911 
4912 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4913 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
4914 	}
4915 
4916 	/* Place the request on the outstanding list so we can keep track of it */
4917 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4918 
4919 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4920 		status = nvmf_ctrlr_process_fabrics_cmd(req);
4921 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4922 		status = nvmf_ctrlr_process_admin_cmd(req);
4923 	} else {
4924 		status = nvmf_ctrlr_process_io_cmd(req);
4925 	}
4926 
4927 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4928 		_nvmf_request_complete(req);
4929 	}
4930 }
4931 
4932 static bool
4933 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
4934 		       struct spdk_dif_ctx *dif_ctx)
4935 {
4936 	struct spdk_nvmf_ns *ns;
4937 	struct spdk_bdev *bdev;
4938 
4939 	if (ctrlr == NULL || cmd == NULL) {
4940 		return false;
4941 	}
4942 
4943 	ns = nvmf_ctrlr_get_ns(ctrlr, cmd->nsid);
4944 	if (ns == NULL || ns->bdev == NULL) {
4945 		return false;
4946 	}
4947 
4948 	bdev = ns->bdev;
4949 
4950 	switch (cmd->opc) {
4951 	case SPDK_NVME_OPC_READ:
4952 	case SPDK_NVME_OPC_WRITE:
4953 	case SPDK_NVME_OPC_COMPARE:
4954 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
4955 	default:
4956 		break;
4957 	}
4958 
4959 	return false;
4960 }
4961 
4962 bool
4963 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
4964 {
4965 	struct spdk_nvmf_qpair *qpair = req->qpair;
4966 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4967 
4968 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
4969 		return false;
4970 	}
4971 
4972 	if (spdk_unlikely(!spdk_nvmf_qpair_is_active(qpair))) {
4973 		return false;
4974 	}
4975 
4976 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4977 		return false;
4978 	}
4979 
4980 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4981 		return false;
4982 	}
4983 
4984 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
4985 }
4986 
4987 void
4988 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
4989 {
4990 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
4991 }
4992 
4993 static int
4994 nvmf_passthru_admin_cmd_for_bdev_nsid(struct spdk_nvmf_request *req, uint32_t bdev_nsid)
4995 {
4996 	struct spdk_bdev *bdev;
4997 	struct spdk_bdev_desc *desc;
4998 	struct spdk_io_channel *ch;
4999 	struct spdk_nvmf_ns *ns;
5000 	struct spdk_nvmf_ctrlr *ctrlr;
5001 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
5002 	int rc;
5003 
5004 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
5005 	if (rc) {
5006 		response->status.sct = SPDK_NVME_SCT_GENERIC;
5007 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
5008 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
5009 	}
5010 
5011 	ctrlr = req->qpair->ctrlr;
5012 	ns = nvmf_ctrlr_get_ns(ctrlr, bdev_nsid);
5013 
5014 	if (ns->passthru_nsid) {
5015 		req->cmd->nvme_cmd.nsid = ns->passthru_nsid;
5016 	}
5017 
5018 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
5019 }
5020 
5021 static int
5022 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
5023 {
5024 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
5025 	uint32_t bdev_nsid;
5026 
5027 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid != 0) {
5028 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
5029 	} else {
5030 		bdev_nsid = cmd->nsid;
5031 	}
5032 
5033 	return nvmf_passthru_admin_cmd_for_bdev_nsid(req, bdev_nsid);
5034 }
5035 
5036 int
5037 nvmf_passthru_admin_cmd_for_ctrlr(struct spdk_nvmf_request *req, struct spdk_nvmf_ctrlr *ctrlr)
5038 {
5039 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
5040 	struct spdk_nvmf_ns *ns;
5041 
5042 	ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys);
5043 	if (ns == NULL) {
5044 		/* Is there a better sc to use here? */
5045 		response->status.sct = SPDK_NVME_SCT_GENERIC;
5046 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
5047 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
5048 	}
5049 
5050 	return nvmf_passthru_admin_cmd_for_bdev_nsid(req, ns->nsid);
5051 }
5052 
5053 void
5054 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
5055 {
5056 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
5057 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
5058 }
5059 
5060 int
5061 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
5062 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
5063 {
5064 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
5065 	struct spdk_nvmf_ns *ns;
5066 	struct spdk_nvmf_poll_group *group = req->qpair->group;
5067 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
5068 
5069 	*bdev = NULL;
5070 	*desc = NULL;
5071 	*ch = NULL;
5072 
5073 	ns = nvmf_ctrlr_get_ns(ctrlr, nsid);
5074 	if (ns == NULL || ns->bdev == NULL) {
5075 		return -EINVAL;
5076 	}
5077 
5078 	assert(group != NULL && group->sgroups != NULL);
5079 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
5080 	*bdev = ns->bdev;
5081 	*desc = ns->desc;
5082 	*ch = ns_info->channel;
5083 
5084 	return 0;
5085 }
5086 
5087 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
5088 {
5089 	return req->qpair->ctrlr;
5090 }
5091 
5092 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
5093 {
5094 	return &req->cmd->nvme_cmd;
5095 }
5096 
5097 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
5098 {
5099 	return &req->rsp->nvme_cpl;
5100 }
5101 
5102 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
5103 {
5104 	return req->qpair->ctrlr->subsys;
5105 }
5106 
5107 size_t
5108 spdk_nvmf_request_copy_from_buf(struct spdk_nvmf_request *req,
5109 				void *buf, size_t buflen)
5110 {
5111 	struct spdk_iov_xfer ix;
5112 
5113 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
5114 	return spdk_iov_xfer_from_buf(&ix, buf, buflen);
5115 }
5116 
5117 size_t
5118 spdk_nvmf_request_copy_to_buf(struct spdk_nvmf_request *req,
5119 			      void *buf, size_t buflen)
5120 {
5121 	struct spdk_iov_xfer ix;
5122 
5123 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
5124 	return spdk_iov_xfer_to_buf(&ix, buf, buflen);
5125 }
5126 
5127 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
5128 {
5129 	return ctrlr->subsys;
5130 }
5131 
5132 uint16_t
5133 spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
5134 {
5135 	return ctrlr->cntlid;
5136 }
5137 
5138 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
5139 {
5140 	return req->req_to_abort;
5141 }
5142