xref: /spdk/lib/nvmf/ctrlr.c (revision e1d06d9954b871531c9b376069d620d2c6cee854)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "nvmf_internal.h"
37 #include "transport.h"
38 
39 #include "spdk/bit_array.h"
40 #include "spdk/endian.h"
41 #include "spdk/thread.h"
42 #include "spdk/nvme_spec.h"
43 #include "spdk/nvmf_cmd.h"
44 #include "spdk/string.h"
45 #include "spdk/util.h"
46 #include "spdk/version.h"
47 #include "spdk/log.h"
48 #include "spdk_internal/usdt.h"
49 
50 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000
51 #define NVMF_DISC_KATO_IN_MS 120000
52 #define KAS_TIME_UNIT_IN_MS 100
53 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS)
54 
55 /*
56  * Report the SPDK version as the firmware revision.
57  * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts.
58  */
59 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING
60 
61 #define ANA_TRANSITION_TIME_IN_SEC 10
62 
63 /*
64  * Support for custom admin command handlers
65  */
66 struct spdk_nvmf_custom_admin_cmd {
67 	spdk_nvmf_custom_cmd_hdlr hdlr;
68 	uint32_t nsid; /* nsid to forward */
69 };
70 
71 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1];
72 
73 static void _nvmf_request_complete(void *ctx);
74 
75 static inline void
76 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp,
77 			      uint8_t iattr, uint16_t ipo)
78 {
79 	rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
80 	rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
81 	rsp->status_code_specific.invalid.iattr = iattr;
82 	rsp->status_code_specific.invalid.ipo = ipo;
83 }
84 
85 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field)	\
86 	nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field))
87 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field)	\
88 	nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field))
89 
90 
91 static void
92 nvmf_ctrlr_stop_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
93 {
94 	if (!ctrlr) {
95 		SPDK_ERRLOG("Controller is NULL\n");
96 		return;
97 	}
98 
99 	if (ctrlr->keep_alive_poller == NULL) {
100 		return;
101 	}
102 
103 	SPDK_DEBUGLOG(nvmf, "Stop keep alive poller\n");
104 	spdk_poller_unregister(&ctrlr->keep_alive_poller);
105 }
106 
107 static void
108 nvmf_ctrlr_stop_association_timer(struct spdk_nvmf_ctrlr *ctrlr)
109 {
110 	if (!ctrlr) {
111 		SPDK_ERRLOG("Controller is NULL\n");
112 		assert(false);
113 		return;
114 	}
115 
116 	if (ctrlr->association_timer == NULL) {
117 		return;
118 	}
119 
120 	SPDK_DEBUGLOG(nvmf, "Stop association timer\n");
121 	spdk_poller_unregister(&ctrlr->association_timer);
122 }
123 
124 static void
125 nvmf_ctrlr_disconnect_qpairs_done(struct spdk_io_channel_iter *i, int status)
126 {
127 	if (status == 0) {
128 		SPDK_DEBUGLOG(nvmf, "ctrlr disconnect qpairs complete successfully\n");
129 	} else {
130 		SPDK_ERRLOG("Fail to disconnect ctrlr qpairs\n");
131 	}
132 }
133 
134 static int
135 _nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i, bool include_admin)
136 {
137 	int rc = 0;
138 	struct spdk_nvmf_ctrlr *ctrlr;
139 	struct spdk_nvmf_qpair *qpair, *temp_qpair;
140 	struct spdk_io_channel *ch;
141 	struct spdk_nvmf_poll_group *group;
142 
143 	ctrlr = spdk_io_channel_iter_get_ctx(i);
144 	ch = spdk_io_channel_iter_get_channel(i);
145 	group = spdk_io_channel_get_ctx(ch);
146 
147 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, temp_qpair) {
148 		if (qpair->ctrlr == ctrlr && (include_admin || !nvmf_qpair_is_admin_queue(qpair))) {
149 			rc = spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
150 			if (rc) {
151 				SPDK_ERRLOG("Qpair disconnect failed\n");
152 				return rc;
153 			}
154 		}
155 	}
156 
157 	return rc;
158 }
159 
160 static void
161 nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i)
162 {
163 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, true));
164 }
165 
166 static void
167 nvmf_ctrlr_disconnect_io_qpairs_on_pg(struct spdk_io_channel_iter *i)
168 {
169 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, false));
170 }
171 
172 static int
173 nvmf_ctrlr_keep_alive_poll(void *ctx)
174 {
175 	uint64_t keep_alive_timeout_tick;
176 	uint64_t now = spdk_get_ticks();
177 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
178 
179 	SPDK_DEBUGLOG(nvmf, "Polling ctrlr keep alive timeout\n");
180 
181 	/* If the Keep alive feature is in use and the timer expires */
182 	keep_alive_timeout_tick = ctrlr->last_keep_alive_tick +
183 				  ctrlr->feat.keep_alive_timer.bits.kato * spdk_get_ticks_hz() / UINT64_C(1000);
184 	if (now > keep_alive_timeout_tick) {
185 		SPDK_NOTICELOG("Disconnecting host %s from subsystem %s due to keep alive timeout.\n",
186 			       ctrlr->hostnqn, ctrlr->subsys->subnqn);
187 		/* set the Controller Fatal Status bit to '1' */
188 		if (ctrlr->vcprop.csts.bits.cfs == 0) {
189 			ctrlr->vcprop.csts.bits.cfs = 1;
190 
191 			/*
192 			 * disconnect qpairs, terminate Transport connection
193 			 * destroy ctrlr, break the host to controller association
194 			 * disconnect qpairs with qpair->ctrlr == ctrlr
195 			 */
196 			spdk_for_each_channel(ctrlr->subsys->tgt,
197 					      nvmf_ctrlr_disconnect_qpairs_on_pg,
198 					      ctrlr,
199 					      nvmf_ctrlr_disconnect_qpairs_done);
200 		}
201 	}
202 
203 	return SPDK_POLLER_BUSY;
204 }
205 
206 static void
207 nvmf_ctrlr_start_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
208 {
209 	if (!ctrlr) {
210 		SPDK_ERRLOG("Controller is NULL\n");
211 		return;
212 	}
213 
214 	/* if cleared to 0 then the Keep Alive Timer is disabled */
215 	if (ctrlr->feat.keep_alive_timer.bits.kato != 0) {
216 
217 		ctrlr->last_keep_alive_tick = spdk_get_ticks();
218 
219 		SPDK_DEBUGLOG(nvmf, "Ctrlr add keep alive poller\n");
220 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
221 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
222 	}
223 }
224 
225 static void
226 ctrlr_add_qpair_and_update_rsp(struct spdk_nvmf_qpair *qpair,
227 			       struct spdk_nvmf_ctrlr *ctrlr,
228 			       struct spdk_nvmf_fabric_connect_rsp *rsp)
229 {
230 	assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
231 
232 	/* check if we would exceed ctrlr connection limit */
233 	if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) {
234 		SPDK_ERRLOG("Requested QID %u but Max QID is %u\n",
235 			    qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1);
236 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
237 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
238 		return;
239 	}
240 
241 	if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) {
242 		SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid);
243 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
244 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
245 		return;
246 	}
247 
248 	qpair->ctrlr = ctrlr;
249 	spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid);
250 
251 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
252 	rsp->status_code_specific.success.cntlid = ctrlr->cntlid;
253 	SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n",
254 		      rsp->status_code_specific.success.cntlid);
255 
256 	SPDK_DTRACE_PROBE4(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn,
257 			   ctrlr->hostnqn);
258 }
259 
260 static void
261 _nvmf_ctrlr_add_admin_qpair(void *ctx)
262 {
263 	struct spdk_nvmf_request *req = ctx;
264 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
265 	struct spdk_nvmf_qpair *qpair = req->qpair;
266 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
267 
268 	ctrlr->admin_qpair = qpair;
269 	ctrlr->association_timeout = qpair->transport->opts.association_timeout;
270 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
271 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
272 	_nvmf_request_complete(req);
273 }
274 
275 static void
276 _nvmf_subsystem_add_ctrlr(void *ctx)
277 {
278 	struct spdk_nvmf_request *req = ctx;
279 	struct spdk_nvmf_qpair *qpair = req->qpair;
280 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
281 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
282 
283 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
284 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
285 		spdk_bit_array_free(&ctrlr->qpair_mask);
286 		free(ctrlr);
287 		qpair->ctrlr = NULL;
288 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
289 		spdk_nvmf_request_complete(req);
290 		return;
291 	}
292 
293 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
294 }
295 
296 static void
297 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
298 		      struct spdk_nvmf_ctrlr_data *cdata)
299 {
300 	cdata->kas = KAS_DEFAULT_VALUE;
301 	cdata->sgls.supported = 1;
302 	cdata->sgls.keyed_sgl = 1;
303 	cdata->sgls.sgl_offset = 1;
304 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
305 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
306 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
307 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
308 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
309 	cdata->nvmf_specific.msdbd = 1;
310 
311 	if (transport->ops->cdata_init) {
312 		transport->ops->cdata_init(transport, subsystem, cdata);
313 	}
314 }
315 
316 static struct spdk_nvmf_ctrlr *
317 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
318 		  struct spdk_nvmf_request *req,
319 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
320 		  struct spdk_nvmf_fabric_connect_data *connect_data)
321 {
322 	struct spdk_nvmf_ctrlr	*ctrlr;
323 	struct spdk_nvmf_transport *transport;
324 	struct spdk_nvme_transport_id listen_trid = {};
325 
326 	ctrlr = calloc(1, sizeof(*ctrlr));
327 	if (ctrlr == NULL) {
328 		SPDK_ERRLOG("Memory allocation failed\n");
329 		return NULL;
330 	}
331 
332 	STAILQ_INIT(&ctrlr->async_events);
333 	TAILQ_INIT(&ctrlr->log_head);
334 	ctrlr->subsys = subsystem;
335 	ctrlr->thread = req->qpair->group->thread;
336 	ctrlr->disconnect_in_progress = false;
337 
338 	transport = req->qpair->transport;
339 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
340 	if (!ctrlr->qpair_mask) {
341 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
342 		goto err_qpair_mask;
343 	}
344 
345 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
346 
347 	/*
348 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
349 	 * If this field is cleared to 0h, then Keep Alive is not supported.
350 	 */
351 	if (ctrlr->cdata.kas) {
352 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
353 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
354 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
355 	}
356 
357 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
358 	if (ctrlr->subsys->flags.ana_reporting) {
359 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
360 	}
361 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
362 
363 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
364 		/*
365 		 * If keep-alive timeout is not set, discovery controllers use some
366 		 * arbitrary high value in order to cleanup stale discovery sessions
367 		 *
368 		 * From the 1.0a nvme-of spec:
369 		 * "The Keep Alive command is reserved for
370 		 * Discovery controllers. A transport may specify a
371 		 * fixed Discovery controller activity timeout value
372 		 * (e.g., 2 minutes). If no commands are received
373 		 * by a Discovery controller within that time
374 		 * period, the controller may perform the
375 		 * actions for Keep Alive Timer expiration".
376 		 *
377 		 * From the 1.1 nvme-of spec:
378 		 * "A host requests an explicit persistent connection
379 		 * to a Discovery controller and Asynchronous Event Notifications from
380 		 * the Discovery controller on that persistent connection by specifying
381 		 * a non-zero Keep Alive Timer value in the Connect command."
382 		 *
383 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
384 		 * otherwise we disable it and use default discovery controller KATO.
385 		 * KATO is in millisecond.
386 		 */
387 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
388 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
389 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
390 		} else {
391 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
392 		}
393 	}
394 
395 	/* Subtract 1 for admin queue, 1 for 0's based */
396 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
397 			1;
398 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
399 			1;
400 
401 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
402 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn));
403 
404 	ctrlr->vcprop.cap.raw = 0;
405 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
406 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
407 				      1; /* max queue depth */
408 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
409 	ctrlr->vcprop.cap.bits.to = 1; /* ready timeout - 500 msec units */
410 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
411 	ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
412 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
413 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
414 
415 	/* Version Supported: 1.3 */
416 	ctrlr->vcprop.vs.bits.mjr = 1;
417 	ctrlr->vcprop.vs.bits.mnr = 3;
418 	ctrlr->vcprop.vs.bits.ter = 0;
419 
420 	ctrlr->vcprop.cc.raw = 0;
421 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
422 
423 	ctrlr->vcprop.csts.raw = 0;
424 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
425 
426 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
427 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
428 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
429 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
430 
431 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
432 
433 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
434 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
435 			SPDK_ERRLOG("Could not get listener transport ID\n");
436 			goto err_listener;
437 		}
438 
439 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid);
440 		if (!ctrlr->listener) {
441 			SPDK_ERRLOG("Listener was not found\n");
442 			goto err_listener;
443 		}
444 	}
445 
446 	req->qpair->ctrlr = ctrlr;
447 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
448 
449 	return ctrlr;
450 err_listener:
451 	spdk_bit_array_free(&ctrlr->qpair_mask);
452 err_qpair_mask:
453 	free(ctrlr);
454 	return NULL;
455 }
456 
457 static void
458 _nvmf_ctrlr_destruct(void *ctx)
459 {
460 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
461 	struct spdk_nvmf_reservation_log *log, *log_tmp;
462 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
463 
464 	if (ctrlr->disconnect_in_progress) {
465 		SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n");
466 		spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
467 		return;
468 	}
469 
470 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
471 	nvmf_ctrlr_stop_association_timer(ctrlr);
472 	spdk_bit_array_free(&ctrlr->qpair_mask);
473 
474 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
475 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
476 		free(log);
477 	}
478 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
479 		STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link);
480 		free(event);
481 	}
482 	free(ctrlr);
483 }
484 
485 void
486 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
487 {
488 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
489 
490 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
491 }
492 
493 static void
494 nvmf_ctrlr_add_io_qpair(void *ctx)
495 {
496 	struct spdk_nvmf_request *req = ctx;
497 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
498 	struct spdk_nvmf_qpair *qpair = req->qpair;
499 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
500 	struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair;
501 
502 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
503 	  * For error case, the value should be NULL. So set it to NULL at first.
504 	  */
505 	qpair->ctrlr = NULL;
506 
507 	/* Make sure the controller is not being destroyed. */
508 	if (ctrlr->in_destruct) {
509 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
510 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
511 		goto end;
512 	}
513 
514 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
515 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
516 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
517 		goto end;
518 	}
519 
520 	if (!ctrlr->vcprop.cc.bits.en) {
521 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
522 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
523 		goto end;
524 	}
525 
526 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
527 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
528 			    ctrlr->vcprop.cc.bits.iosqes);
529 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
530 		goto end;
531 	}
532 
533 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
534 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
535 			    ctrlr->vcprop.cc.bits.iocqes);
536 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
537 		goto end;
538 	}
539 
540 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
541 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
542 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
543 		 * state to DEACTIVATING and removing it from poll group */
544 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
545 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
546 		goto end;
547 	}
548 
549 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
550 end:
551 	spdk_nvmf_request_complete(req);
552 }
553 
554 static void
555 _nvmf_ctrlr_add_io_qpair(void *ctx)
556 {
557 	struct spdk_nvmf_request *req = ctx;
558 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
559 	struct spdk_nvmf_fabric_connect_data *data = req->data;
560 	struct spdk_nvmf_ctrlr *ctrlr;
561 	struct spdk_nvmf_qpair *qpair = req->qpair;
562 	struct spdk_nvmf_qpair *admin_qpair;
563 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
564 	struct spdk_nvmf_subsystem *subsystem;
565 	struct spdk_nvme_transport_id listen_trid = {};
566 	const struct spdk_nvmf_subsystem_listener *listener;
567 
568 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
569 
570 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
571 	/* We already checked this in spdk_nvmf_ctrlr_connect */
572 	assert(subsystem != NULL);
573 
574 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
575 	if (ctrlr == NULL) {
576 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
577 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
578 		spdk_nvmf_request_complete(req);
579 		return;
580 	}
581 
582 	/* fail before passing a message to the controller thread. */
583 	if (ctrlr->in_destruct) {
584 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
585 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
586 		spdk_nvmf_request_complete(req);
587 		return;
588 	}
589 
590 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
591 	if (subsystem->flags.ana_reporting) {
592 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
593 			SPDK_ERRLOG("Could not get listener transport ID\n");
594 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
595 			spdk_nvmf_request_complete(req);
596 			return;
597 		}
598 
599 		listener = nvmf_subsystem_find_listener(subsystem, &listen_trid);
600 		if (listener != ctrlr->listener) {
601 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
602 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
603 			spdk_nvmf_request_complete(req);
604 			return;
605 		}
606 	}
607 
608 	admin_qpair = ctrlr->admin_qpair;
609 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
610 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
611 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
612 		 * state to DEACTIVATING and removing it from poll group */
613 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
614 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
615 		spdk_nvmf_request_complete(req);
616 		return;
617 	}
618 	qpair->ctrlr = ctrlr;
619 	spdk_thread_send_msg(admin_qpair->group->thread, nvmf_ctrlr_add_io_qpair, req);
620 }
621 
622 static bool
623 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
624 			  const char *hostnqn)
625 {
626 	struct spdk_nvme_transport_id listen_trid = {};
627 
628 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
629 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
630 		return false;
631 	}
632 
633 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
634 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
635 			    subsystem->subnqn);
636 		return false;
637 	}
638 
639 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
640 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
641 			    subsystem->subnqn, hostnqn);
642 		return false;
643 	}
644 
645 	return true;
646 }
647 
648 static int
649 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
650 {
651 	struct spdk_nvmf_fabric_connect_data *data = req->data;
652 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
653 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
654 	struct spdk_nvmf_qpair *qpair = req->qpair;
655 	struct spdk_nvmf_transport *transport = qpair->transport;
656 	struct spdk_nvmf_ctrlr *ctrlr;
657 	struct spdk_nvmf_subsystem *subsystem;
658 
659 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
660 		      cmd->recfmt, cmd->qid, cmd->sqsize);
661 
662 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
663 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
664 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
665 		      ntohl(*(uint32_t *)&data->hostid[0]),
666 		      ntohs(*(uint16_t *)&data->hostid[4]),
667 		      ntohs(*(uint16_t *)&data->hostid[6]),
668 		      data->hostid[8],
669 		      data->hostid[9],
670 		      ntohs(*(uint16_t *)&data->hostid[10]),
671 		      ntohl(*(uint32_t *)&data->hostid[12]));
672 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
673 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
674 
675 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
676 	if (!subsystem) {
677 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
678 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
679 	}
680 
681 	if (cmd->recfmt != 0) {
682 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
683 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
684 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
685 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
686 	}
687 
688 	/*
689 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
690 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
691 	 */
692 	if (cmd->sqsize == 0) {
693 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
694 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
695 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
696 	}
697 
698 	if (cmd->qid == 0) {
699 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
700 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
701 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
702 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
703 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
704 		}
705 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
706 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
707 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
708 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
709 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
710 	}
711 
712 	qpair->sq_head_max = cmd->sqsize;
713 	qpair->qid = cmd->qid;
714 
715 	if (0 == qpair->qid) {
716 		qpair->group->stat.admin_qpairs++;
717 		qpair->group->stat.current_admin_qpairs++;
718 	} else {
719 		qpair->group->stat.io_qpairs++;
720 		qpair->group->stat.current_io_qpairs++;
721 	}
722 
723 	if (cmd->qid == 0) {
724 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
725 
726 		if (data->cntlid != 0xFFFF) {
727 			/* This NVMf target only supports dynamic mode. */
728 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
729 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
730 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
731 		}
732 
733 		/* Establish a new ctrlr */
734 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
735 		if (!ctrlr) {
736 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
737 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
738 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
739 		} else {
740 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
741 		}
742 	} else {
743 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
744 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
745 	}
746 }
747 
748 static inline bool
749 nvmf_request_is_fabric_connect(struct spdk_nvmf_request *req)
750 {
751 	return req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC &&
752 	       req->cmd->nvmf_cmd.fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT;
753 }
754 
755 static struct spdk_nvmf_subsystem_poll_group *
756 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
757 {
758 	struct spdk_nvmf_fabric_connect_data *data;
759 	struct spdk_nvmf_subsystem *subsystem;
760 	struct spdk_nvmf_tgt *tgt;
761 
762 	assert(nvmf_request_is_fabric_connect(req));
763 	assert(req->qpair->ctrlr == NULL);
764 
765 	data = req->data;
766 	tgt = req->qpair->transport->tgt;
767 
768 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
769 	if (subsystem == NULL) {
770 		return NULL;
771 	}
772 
773 	return &req->qpair->group->sgroups[subsystem->id];
774 }
775 
776 int
777 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
778 {
779 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
780 	struct spdk_nvmf_qpair *qpair = req->qpair;
781 	struct spdk_nvmf_subsystem_poll_group *sgroup;
782 	enum spdk_nvmf_request_exec_status status;
783 
784 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
785 	if (!sgroup) {
786 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
787 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
788 		goto out;
789 	}
790 
791 	sgroup->mgmt_io_outstanding++;
792 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
793 
794 	status = _nvmf_ctrlr_connect(req);
795 
796 out:
797 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
798 		_nvmf_request_complete(req);
799 	}
800 
801 	return status;
802 }
803 
804 static int nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req);
805 
806 static int
807 retry_connect(void *arg)
808 {
809 	struct spdk_nvmf_request *req = arg;
810 	struct spdk_nvmf_subsystem_poll_group *sgroup;
811 	int rc;
812 
813 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
814 	sgroup->mgmt_io_outstanding++;
815 	spdk_poller_unregister(&req->poller);
816 	rc = nvmf_ctrlr_cmd_connect(req);
817 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
818 		_nvmf_request_complete(req);
819 	}
820 	return SPDK_POLLER_BUSY;
821 }
822 
823 static int
824 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
825 {
826 	struct spdk_nvmf_fabric_connect_data *data = req->data;
827 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
828 	struct spdk_nvmf_transport *transport = req->qpair->transport;
829 	struct spdk_nvmf_subsystem *subsystem;
830 
831 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
832 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
833 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
834 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
835 	}
836 
837 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
838 	if (!subsystem) {
839 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
840 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
841 	}
842 
843 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
844 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
845 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
846 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
847 		struct spdk_nvmf_subsystem_poll_group *sgroup;
848 
849 		if (req->timeout_tsc == 0) {
850 			/* We will only retry the request up to 1 second. */
851 			req->timeout_tsc = spdk_get_ticks() + spdk_get_ticks_hz();
852 		} else if (spdk_get_ticks() > req->timeout_tsc) {
853 			SPDK_ERRLOG("Subsystem '%s' was not ready for 1 second\n", subsystem->subnqn);
854 			rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
855 			rsp->status.sc = SPDK_NVMF_FABRIC_SC_CONTROLLER_BUSY;
856 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
857 		}
858 
859 		/* Subsystem is not ready to handle a connect. Use a poller to retry it
860 		 * again later. Decrement the mgmt_io_outstanding to avoid the
861 		 * subsystem waiting for this command to complete before unpausing.
862 		 */
863 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
864 		sgroup->mgmt_io_outstanding--;
865 		SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn);
866 		req->poller = SPDK_POLLER_REGISTER(retry_connect, req, 100);
867 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
868 	}
869 
870 	/* Ensure that hostnqn is null terminated */
871 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
872 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
873 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
874 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
875 	}
876 
877 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
878 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
879 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
880 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
881 	}
882 
883 	return _nvmf_ctrlr_connect(req);
884 }
885 
886 static int
887 nvmf_ctrlr_association_remove(void *ctx)
888 {
889 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
890 	int rc;
891 
892 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
893 		      ctrlr->subsys->subnqn);
894 
895 	rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL);
896 	if (rc < 0) {
897 		SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
898 		assert(false);
899 	}
900 
901 	nvmf_ctrlr_stop_association_timer(ctrlr);
902 	return 1;
903 }
904 
905 static void
906 nvmf_ctrlr_cc_shn_done(struct spdk_io_channel_iter *i, int status)
907 {
908 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
909 
910 	if (status < 0) {
911 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
912 		assert(false);
913 	}
914 
915 	ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
916 
917 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
918 	 * between the host and controller shall be preserved for at least 2 minutes */
919 	if (ctrlr->association_timer) {
920 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
921 		nvmf_ctrlr_stop_association_timer(ctrlr);
922 	}
923 	if (ctrlr->association_timeout) {
924 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
925 					   ctrlr->association_timeout * 1000);
926 	}
927 	ctrlr->disconnect_in_progress = false;
928 }
929 
930 static void
931 nvmf_ctrlr_cc_reset_done(struct spdk_io_channel_iter *i, int status)
932 {
933 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
934 
935 	if (status < 0) {
936 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
937 		assert(false);
938 	}
939 
940 	/* Only a subset of the registers are cleared out on a reset */
941 	ctrlr->vcprop.cc.raw = 0;
942 	ctrlr->vcprop.csts.raw = 0;
943 
944 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
945 	 * between the host and controller shall be preserved for at least 2 minutes */
946 	if (ctrlr->association_timer) {
947 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
948 		nvmf_ctrlr_stop_association_timer(ctrlr);
949 	}
950 	if (ctrlr->association_timeout) {
951 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
952 					   ctrlr->association_timeout * 1000);
953 	}
954 	ctrlr->disconnect_in_progress = false;
955 }
956 
957 const struct spdk_nvmf_registers *
958 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
959 {
960 	return &ctrlr->vcprop;
961 }
962 
963 static uint64_t
964 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
965 {
966 	return ctrlr->vcprop.cap.raw;
967 }
968 
969 static uint64_t
970 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
971 {
972 	return ctrlr->vcprop.vs.raw;
973 }
974 
975 static uint64_t
976 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
977 {
978 	return ctrlr->vcprop.cc.raw;
979 }
980 
981 static bool
982 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
983 {
984 	union spdk_nvme_cc_register cc, diff;
985 
986 	cc.raw = value;
987 
988 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
989 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
990 
991 	/*
992 	 * Calculate which bits changed between the current and new CC.
993 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
994 	 */
995 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
996 
997 	if (diff.bits.en) {
998 		if (cc.bits.en) {
999 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
1000 			nvmf_ctrlr_stop_association_timer(ctrlr);
1001 
1002 			ctrlr->vcprop.cc.bits.en = 1;
1003 			ctrlr->vcprop.csts.bits.rdy = 1;
1004 		} else {
1005 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
1006 			ctrlr->vcprop.cc.bits.en = 0;
1007 			ctrlr->disconnect_in_progress = true;
1008 			spdk_for_each_channel(ctrlr->subsys->tgt,
1009 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1010 					      ctrlr,
1011 					      nvmf_ctrlr_cc_reset_done);
1012 		}
1013 		diff.bits.en = 0;
1014 	}
1015 
1016 	if (diff.bits.shn) {
1017 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
1018 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
1019 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
1020 				      cc.bits.shn >> 1, cc.bits.shn & 1);
1021 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
1022 			ctrlr->disconnect_in_progress = true;
1023 			spdk_for_each_channel(ctrlr->subsys->tgt,
1024 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1025 					      ctrlr,
1026 					      nvmf_ctrlr_cc_shn_done);
1027 
1028 			/* From the time a shutdown is initiated the controller shall disable
1029 			 * Keep Alive timer */
1030 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
1031 		} else if (cc.bits.shn == 0) {
1032 			ctrlr->vcprop.cc.bits.shn = 0;
1033 		} else {
1034 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
1035 				    cc.bits.shn >> 1, cc.bits.shn & 1);
1036 			return false;
1037 		}
1038 		diff.bits.shn = 0;
1039 	}
1040 
1041 	if (diff.bits.iosqes) {
1042 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
1043 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
1044 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
1045 		diff.bits.iosqes = 0;
1046 	}
1047 
1048 	if (diff.bits.iocqes) {
1049 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
1050 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
1051 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
1052 		diff.bits.iocqes = 0;
1053 	}
1054 
1055 	if (diff.bits.ams) {
1056 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
1057 		return false;
1058 	}
1059 
1060 	if (diff.bits.mps) {
1061 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
1062 		return false;
1063 	}
1064 
1065 	if (diff.bits.css) {
1066 		SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
1067 		return false;
1068 	}
1069 
1070 	if (diff.raw != 0) {
1071 		/* Print an error message, but don't fail the command in this case.
1072 		 * If we did want to fail in this case, we'd need to ensure we acted
1073 		 * on no other bits or the initiator gets confused. */
1074 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
1075 	}
1076 
1077 	return true;
1078 }
1079 
1080 static uint64_t
1081 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
1082 {
1083 	return ctrlr->vcprop.csts.raw;
1084 }
1085 
1086 static uint64_t
1087 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
1088 {
1089 	return ctrlr->vcprop.aqa.raw;
1090 }
1091 
1092 static bool
1093 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1094 {
1095 	union spdk_nvme_aqa_register aqa;
1096 
1097 	aqa.raw = value;
1098 
1099 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1100 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1101 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1102 		return false;
1103 	}
1104 
1105 	ctrlr->vcprop.aqa.raw = value;
1106 
1107 	return true;
1108 }
1109 
1110 static uint64_t
1111 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1112 {
1113 	return ctrlr->vcprop.asq;
1114 }
1115 
1116 static bool
1117 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1118 {
1119 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1120 
1121 	return true;
1122 }
1123 
1124 static bool
1125 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1126 {
1127 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1128 
1129 	return true;
1130 }
1131 
1132 static uint64_t
1133 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1134 {
1135 	return ctrlr->vcprop.acq;
1136 }
1137 
1138 static bool
1139 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1140 {
1141 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1142 
1143 	return true;
1144 }
1145 
1146 static bool
1147 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1148 {
1149 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1150 
1151 	return true;
1152 }
1153 
1154 struct nvmf_prop {
1155 	uint32_t ofst;
1156 	uint8_t size;
1157 	char name[11];
1158 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1159 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1160 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1161 };
1162 
1163 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1164 	{ \
1165 		offsetof(struct spdk_nvme_registers, field), \
1166 		size, \
1167 		#field, \
1168 		get_cb, set_cb, set_upper_cb \
1169 	}
1170 
1171 static const struct nvmf_prop nvmf_props[] = {
1172 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1173 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1174 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1175 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1176 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1177 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1178 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1179 };
1180 
1181 static const struct nvmf_prop *
1182 find_prop(uint32_t ofst, uint8_t size)
1183 {
1184 	size_t i;
1185 
1186 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1187 		const struct nvmf_prop *prop = &nvmf_props[i];
1188 
1189 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1190 			return prop;
1191 		}
1192 	}
1193 
1194 	return NULL;
1195 }
1196 
1197 static int
1198 nvmf_property_get(struct spdk_nvmf_request *req)
1199 {
1200 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1201 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1202 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1203 	const struct nvmf_prop *prop;
1204 	uint8_t size;
1205 
1206 	response->status.sc = 0;
1207 	response->value.u64 = 0;
1208 
1209 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1210 		      cmd->attrib.size, cmd->ofst);
1211 
1212 	switch (cmd->attrib.size) {
1213 	case SPDK_NVMF_PROP_SIZE_4:
1214 		size = 4;
1215 		break;
1216 	case SPDK_NVMF_PROP_SIZE_8:
1217 		size = 8;
1218 		break;
1219 	default:
1220 		SPDK_ERRLOG("Invalid size value %d\n", cmd->attrib.size);
1221 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1222 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1223 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1224 	}
1225 
1226 	prop = find_prop(cmd->ofst, size);
1227 	if (prop == NULL || prop->get_cb == NULL) {
1228 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1229 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1230 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1231 	}
1232 
1233 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1234 
1235 	response->value.u64 = prop->get_cb(ctrlr);
1236 
1237 	if (size != prop->size) {
1238 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1239 		assert(size == 4);
1240 		assert(prop->size == 8);
1241 
1242 		if (cmd->ofst == prop->ofst) {
1243 			/* Keep bottom 4 bytes only */
1244 			response->value.u64 &= 0xFFFFFFFF;
1245 		} else {
1246 			/* Keep top 4 bytes only */
1247 			response->value.u64 >>= 32;
1248 		}
1249 	}
1250 
1251 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1252 
1253 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1254 }
1255 
1256 static int
1257 nvmf_property_set(struct spdk_nvmf_request *req)
1258 {
1259 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1260 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1261 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1262 	const struct nvmf_prop *prop;
1263 	uint64_t value;
1264 	uint8_t size;
1265 	bool ret;
1266 
1267 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1268 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1269 
1270 	switch (cmd->attrib.size) {
1271 	case SPDK_NVMF_PROP_SIZE_4:
1272 		size = 4;
1273 		break;
1274 	case SPDK_NVMF_PROP_SIZE_8:
1275 		size = 8;
1276 		break;
1277 	default:
1278 		SPDK_ERRLOG("Invalid size value %d\n", cmd->attrib.size);
1279 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1280 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1281 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1282 	}
1283 
1284 	prop = find_prop(cmd->ofst, size);
1285 	if (prop == NULL || prop->set_cb == NULL) {
1286 		SPDK_ERRLOG("Invalid offset 0x%x\n", cmd->ofst);
1287 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1288 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1289 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1290 	}
1291 
1292 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1293 
1294 	value = cmd->value.u64;
1295 
1296 	if (prop->size == 4) {
1297 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1298 	} else if (size != prop->size) {
1299 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1300 		assert(size == 4);
1301 		assert(prop->size == 8);
1302 
1303 		if (cmd->ofst == prop->ofst) {
1304 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1305 		} else {
1306 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1307 		}
1308 	} else {
1309 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1310 		if (ret) {
1311 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1312 		}
1313 	}
1314 
1315 	if (!ret) {
1316 		SPDK_ERRLOG("prop set_cb failed\n");
1317 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1318 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1319 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1320 	}
1321 
1322 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1323 }
1324 
1325 static int
1326 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1327 {
1328 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1329 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1330 
1331 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1332 
1333 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1334 	ctrlr->feat.arbitration.bits.reserved = 0;
1335 
1336 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1337 }
1338 
1339 static int
1340 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1341 {
1342 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1343 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1344 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1345 
1346 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1347 
1348 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1349 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1350 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1351 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1352 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1353 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1354 	}
1355 
1356 	ctrlr->feat.power_management.raw = cmd->cdw11;
1357 	ctrlr->feat.power_management.bits.reserved = 0;
1358 
1359 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1360 }
1361 
1362 static bool
1363 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1364 {
1365 	/*
1366 	 * Valid TMPSEL values:
1367 	 *  0000b - 1000b: temperature sensors
1368 	 *  1111b: set all implemented temperature sensors
1369 	 */
1370 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1371 		/* 1001b - 1110b: reserved */
1372 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1373 		return false;
1374 	}
1375 
1376 	/*
1377 	 * Valid THSEL values:
1378 	 *  00b: over temperature threshold
1379 	 *  01b: under temperature threshold
1380 	 */
1381 	if (opts->bits.thsel > 1) {
1382 		/* 10b - 11b: reserved */
1383 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1384 		return false;
1385 	}
1386 
1387 	return true;
1388 }
1389 
1390 static int
1391 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1392 {
1393 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1394 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1395 
1396 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1397 
1398 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1399 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1400 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1401 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1402 	}
1403 
1404 	/* TODO: no sensors implemented - ignore new values */
1405 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1406 }
1407 
1408 static int
1409 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1410 {
1411 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1412 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1413 
1414 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1415 
1416 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1417 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1418 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1419 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1420 	}
1421 
1422 	/* TODO: no sensors implemented - return 0 for all thresholds */
1423 	rsp->cdw0 = 0;
1424 
1425 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1426 }
1427 
1428 static int
1429 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1430 {
1431 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1432 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1433 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1434 
1435 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1436 
1437 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1438 		/*
1439 		 * Host is not allowed to set this bit, since we don't advertise it in
1440 		 * Identify Namespace.
1441 		 */
1442 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1443 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1444 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1445 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1446 	}
1447 
1448 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1449 	ctrlr->feat.error_recovery.bits.reserved = 0;
1450 
1451 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1452 }
1453 
1454 static int
1455 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1456 {
1457 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1458 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1459 
1460 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1461 
1462 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1463 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1464 
1465 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1466 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1467 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1468 }
1469 
1470 static int
1471 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1472 {
1473 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1474 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1475 
1476 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1477 
1478 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1479 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1480 
1481 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1482 }
1483 
1484 static int
1485 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1486 {
1487 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1488 
1489 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1490 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1491 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1492 }
1493 
1494 static int
1495 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1496 {
1497 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1498 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1499 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1500 
1501 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1502 
1503 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1504 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1505 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1506 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1507 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1508 	}
1509 
1510 	if (req->data == NULL || req->length < sizeof(ctrlr->hostid)) {
1511 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1512 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1513 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1514 	}
1515 
1516 	spdk_uuid_copy((struct spdk_uuid *)req->data, &ctrlr->hostid);
1517 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1518 }
1519 
1520 static int
1521 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1522 {
1523 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1524 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1525 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1526 	struct spdk_nvmf_ns *ns;
1527 
1528 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notificaton Mask\n");
1529 
1530 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1531 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1532 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1533 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1534 	}
1535 
1536 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1537 	if (ns == NULL) {
1538 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1539 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1540 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1541 	}
1542 	rsp->cdw0 = ns->mask;
1543 
1544 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1545 }
1546 
1547 static int
1548 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1549 {
1550 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1551 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1552 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1553 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1554 	struct spdk_nvmf_ns *ns;
1555 
1556 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notificaton Mask\n");
1557 
1558 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1559 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1560 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1561 			ns->mask = cmd->cdw11;
1562 		}
1563 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1564 	}
1565 
1566 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1567 	if (ns == NULL) {
1568 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1569 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1570 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1571 	}
1572 	ns->mask = cmd->cdw11;
1573 
1574 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1575 }
1576 
1577 static int
1578 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1579 {
1580 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1581 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1582 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1583 	struct spdk_nvmf_ns *ns;
1584 
1585 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1586 
1587 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1588 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1589 	if (ns == NULL) {
1590 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1591 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1592 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1593 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1594 	}
1595 
1596 	response->cdw0 = ns->ptpl_activated;
1597 
1598 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1599 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1600 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1601 }
1602 
1603 static int
1604 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req)
1605 {
1606 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1607 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1608 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1609 	struct spdk_nvmf_ns *ns;
1610 	bool ptpl;
1611 
1612 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1613 
1614 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1615 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1616 
1617 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && ns->ptpl_file) {
1618 		ns->ptpl_activated = ptpl;
1619 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1620 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns && ns->ptpl_file;
1621 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1622 			ns->ptpl_activated = ptpl;
1623 		}
1624 	} else {
1625 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1626 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1627 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1628 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1629 	}
1630 
1631 	/* TODO: Feature not changeable for now */
1632 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1633 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1634 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1635 }
1636 
1637 static int
1638 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req)
1639 {
1640 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1641 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1642 	struct spdk_nvme_host_behavior *host_behavior;
1643 
1644 	SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n");
1645 	if (req->iovcnt != 1) {
1646 		SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt);
1647 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1648 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1649 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1650 	}
1651 	if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) {
1652 		SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len);
1653 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1654 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1655 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1656 	}
1657 
1658 	host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base;
1659 	if (host_behavior->acre == 0) {
1660 		ctrlr->acre_enabled = false;
1661 	} else if (host_behavior->acre == 1) {
1662 		ctrlr->acre_enabled = true;
1663 	} else {
1664 		SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre);
1665 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1666 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1667 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1668 	}
1669 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1670 }
1671 
1672 static int
1673 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
1674 {
1675 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1676 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1677 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1678 
1679 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
1680 
1681 	/*
1682 	 * if attempts to disable keep alive by setting kato to 0h
1683 	 * a status value of keep alive invalid shall be returned
1684 	 */
1685 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
1686 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
1687 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
1688 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
1689 	} else {
1690 		/* round up to milliseconds */
1691 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
1692 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
1693 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
1694 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
1695 	}
1696 
1697 	/*
1698 	 * if change the keep alive timeout value successfully
1699 	 * update the keep alive poller.
1700 	 */
1701 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
1702 		if (ctrlr->keep_alive_poller != NULL) {
1703 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
1704 		}
1705 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
1706 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
1707 	}
1708 
1709 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
1710 		      ctrlr->feat.keep_alive_timer.bits.kato);
1711 
1712 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1713 }
1714 
1715 static int
1716 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
1717 {
1718 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1719 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1720 	uint32_t count;
1721 
1722 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
1723 		      req->cmd->nvme_cmd.cdw11);
1724 
1725 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1726 	/* verify that the controller is ready to process commands */
1727 	if (count > 1) {
1728 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
1729 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1730 	} else {
1731 		/*
1732 		 * Ignore the value requested by the host -
1733 		 * always return the pre-configured value based on max_qpairs_allowed.
1734 		 */
1735 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
1736 	}
1737 
1738 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1739 }
1740 
1741 static int
1742 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
1743 {
1744 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1745 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1746 
1747 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
1748 		      cmd->cdw11);
1749 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
1750 	ctrlr->feat.async_event_configuration.bits.reserved1 = 0;
1751 	ctrlr->feat.async_event_configuration.bits.reserved2 = 0;
1752 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1753 }
1754 
1755 static int
1756 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
1757 {
1758 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1759 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1760 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1761 	struct spdk_nvmf_async_event_completion *pending_event;
1762 
1763 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
1764 
1765 	/* AER cmd is an exception */
1766 	sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
1767 	assert(sgroup != NULL);
1768 	sgroup->mgmt_io_outstanding--;
1769 
1770 	/* Four asynchronous events are supported for now */
1771 	if (ctrlr->nr_aer_reqs >= NVMF_MAX_ASYNC_EVENTS) {
1772 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
1773 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1774 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
1775 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1776 	}
1777 
1778 	if (!STAILQ_EMPTY(&ctrlr->async_events)) {
1779 		pending_event = STAILQ_FIRST(&ctrlr->async_events);
1780 		rsp->cdw0 = pending_event->event.raw;
1781 		STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link);
1782 		free(pending_event);
1783 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1784 	}
1785 
1786 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
1787 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
1788 }
1789 
1790 struct copy_iovs_ctx {
1791 	struct iovec *iovs;
1792 	int iovcnt;
1793 	int cur_iov_idx;
1794 	size_t cur_iov_offset;
1795 };
1796 
1797 static void
1798 _init_copy_iovs_ctx(struct copy_iovs_ctx *copy_ctx, struct iovec *iovs, int iovcnt)
1799 {
1800 	int iov_idx = 0;
1801 	struct iovec *iov;
1802 
1803 	copy_ctx->iovs = iovs;
1804 	copy_ctx->iovcnt = iovcnt;
1805 	copy_ctx->cur_iov_idx = 0;
1806 	copy_ctx->cur_iov_offset = 0;
1807 
1808 	while (iov_idx < copy_ctx->iovcnt) {
1809 		iov = &copy_ctx->iovs[iov_idx];
1810 		memset(iov->iov_base, 0, iov->iov_len);
1811 		iov_idx++;
1812 	}
1813 }
1814 
1815 static size_t
1816 _copy_buf_to_iovs(struct copy_iovs_ctx *copy_ctx, const void *buf, size_t buf_len)
1817 {
1818 	size_t len, iov_remain_len, copied_len = 0;
1819 	struct iovec *iov;
1820 
1821 	if (buf_len == 0) {
1822 		return 0;
1823 	}
1824 
1825 	while (copy_ctx->cur_iov_idx < copy_ctx->iovcnt) {
1826 		iov = &copy_ctx->iovs[copy_ctx->cur_iov_idx];
1827 		iov_remain_len = iov->iov_len - copy_ctx->cur_iov_offset;
1828 		if (iov_remain_len == 0) {
1829 			copy_ctx->cur_iov_idx++;
1830 			copy_ctx->cur_iov_offset = 0;
1831 			continue;
1832 		}
1833 
1834 		len = spdk_min(iov_remain_len, buf_len - copied_len);
1835 		memcpy((char *)iov->iov_base + copy_ctx->cur_iov_offset,
1836 		       (const char *)buf + copied_len,
1837 		       len);
1838 		copied_len += len;
1839 		copy_ctx->cur_iov_offset += len;
1840 
1841 		if (buf_len == copied_len) {
1842 			return copied_len;
1843 		}
1844 	}
1845 
1846 	return copied_len;
1847 }
1848 
1849 static void
1850 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length)
1851 {
1852 	struct spdk_nvme_firmware_page fw_page;
1853 	size_t copy_len;
1854 	struct copy_iovs_ctx copy_ctx;
1855 
1856 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
1857 
1858 	memset(&fw_page, 0, sizeof(fw_page));
1859 	fw_page.afi.active_slot = 1;
1860 	fw_page.afi.next_reset_slot = 0;
1861 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
1862 
1863 	if (offset < sizeof(fw_page)) {
1864 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
1865 		if (copy_len > 0) {
1866 			_copy_buf_to_iovs(&copy_ctx, (const char *)&fw_page + offset, copy_len);
1867 		}
1868 	}
1869 }
1870 
1871 /*
1872  * Asynchronous Event Mask Bit
1873  */
1874 enum spdk_nvme_async_event_mask_bit {
1875 	/* Mask Namespace Change Notificaton */
1876 	SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT		= 0,
1877 	/* Mask Asymmetric Namespace Access Change Notification */
1878 	SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT		= 1,
1879 	/* Mask Discovery Log Change Notification */
1880 	SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT	= 2,
1881 	/* Mask Reservation Log Page Available Notification */
1882 	SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT	= 3,
1883 
1884 	/* 4 - 63 Reserved */
1885 };
1886 
1887 static inline void
1888 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr,
1889 		      enum spdk_nvme_async_event_mask_bit mask)
1890 {
1891 	ctrlr->notice_aen_mask &= ~(1 << mask);
1892 }
1893 
1894 static inline bool
1895 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr,
1896 		    enum spdk_nvme_async_event_mask_bit mask)
1897 {
1898 	if (ctrlr->notice_aen_mask & (1 << mask)) {
1899 		return false;
1900 	} else {
1901 		ctrlr->notice_aen_mask |= (1 << mask);
1902 		return true;
1903 	}
1904 }
1905 
1906 #define SPDK_NVMF_ANA_DESC_SIZE	(sizeof(struct spdk_nvme_ana_group_descriptor) +	\
1907 				 sizeof(uint32_t))
1908 static void
1909 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
1910 		      uint64_t offset, uint32_t length, uint32_t rae)
1911 {
1912 	struct spdk_nvme_ana_page ana_hdr;
1913 	char _ana_desc[SPDK_NVMF_ANA_DESC_SIZE];
1914 	struct spdk_nvme_ana_group_descriptor *ana_desc;
1915 	size_t copy_len, copied_len;
1916 	uint32_t num_ns = 0;
1917 	struct spdk_nvmf_ns *ns;
1918 	struct copy_iovs_ctx copy_ctx;
1919 
1920 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
1921 
1922 	if (length == 0) {
1923 		return;
1924 	}
1925 
1926 	if (offset >= sizeof(ana_hdr)) {
1927 		offset -= sizeof(ana_hdr);
1928 	} else {
1929 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1930 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1931 			num_ns++;
1932 		}
1933 
1934 		memset(&ana_hdr, 0, sizeof(ana_hdr));
1935 
1936 		ana_hdr.num_ana_group_desc = num_ns;
1937 		/* TODO: Support Change Count. */
1938 		ana_hdr.change_count = 0;
1939 
1940 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
1941 		copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ana_hdr + offset, copy_len);
1942 		assert(copied_len == copy_len);
1943 		length -= copied_len;
1944 		offset = 0;
1945 	}
1946 
1947 	if (length == 0) {
1948 		return;
1949 	}
1950 
1951 	ana_desc = (void *)_ana_desc;
1952 
1953 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1954 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1955 		if (offset >= SPDK_NVMF_ANA_DESC_SIZE) {
1956 			offset -= SPDK_NVMF_ANA_DESC_SIZE;
1957 			continue;
1958 		}
1959 
1960 		memset(ana_desc, 0, SPDK_NVMF_ANA_DESC_SIZE);
1961 
1962 		ana_desc->ana_group_id = ns->nsid;
1963 		ana_desc->num_of_nsid = 1;
1964 		ana_desc->ana_state = ctrlr->listener->ana_state;
1965 		ana_desc->nsid[0] = ns->nsid;
1966 		/* TODO: Support Change Count. */
1967 		ana_desc->change_count = 0;
1968 
1969 		copy_len = spdk_min(SPDK_NVMF_ANA_DESC_SIZE - offset, length);
1970 		copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)ana_desc + offset, copy_len);
1971 		assert(copied_len == copy_len);
1972 		length -= copied_len;
1973 		offset = 0;
1974 
1975 		if (length == 0) {
1976 			goto done;
1977 		}
1978 	}
1979 
1980 done:
1981 	if (!rae) {
1982 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT);
1983 	}
1984 }
1985 
1986 void
1987 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
1988 {
1989 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
1990 	uint16_t i;
1991 	bool found = false;
1992 
1993 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
1994 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
1995 			/* nsid is already in the list */
1996 			found = true;
1997 			break;
1998 		}
1999 	}
2000 
2001 	if (!found) {
2002 		if (ctrlr->changed_ns_list_count == max_changes) {
2003 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
2004 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
2005 			for (i = 1; i < max_changes; i++) {
2006 				ctrlr->changed_ns_list.ns_list[i] = 0;
2007 			}
2008 		} else {
2009 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
2010 		}
2011 	}
2012 }
2013 
2014 static void
2015 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2016 				  struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2017 {
2018 	size_t copy_length;
2019 	struct copy_iovs_ctx copy_ctx;
2020 
2021 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2022 
2023 	if (offset < sizeof(ctrlr->changed_ns_list)) {
2024 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
2025 		if (copy_length) {
2026 			_copy_buf_to_iovs(&copy_ctx, (char *)&ctrlr->changed_ns_list + offset, copy_length);
2027 		}
2028 	}
2029 
2030 	/* Clear log page each time it is read */
2031 	ctrlr->changed_ns_list_count = 0;
2032 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
2033 
2034 	if (!rae) {
2035 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT);
2036 	}
2037 }
2038 
2039 /* The structure can be modified if we provide support for other commands in future */
2040 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
2041 	.admin_cmds_supported = {
2042 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
2043 		/* Get Log Page */
2044 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2045 		/* Identify */
2046 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
2047 		/* Abort */
2048 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
2049 		/* Set Features */
2050 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2051 		/* Get Features */
2052 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2053 		/* Async Event Request */
2054 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
2055 		/* Keep Alive */
2056 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2057 	},
2058 	.io_cmds_supported = {
2059 		/* FLUSH */
2060 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
2061 		/* WRITE */
2062 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
2063 		/* READ */
2064 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
2065 		/* WRITE ZEROES */
2066 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
2067 		/* DATASET MANAGEMENT */
2068 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
2069 		/* COMPARE */
2070 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
2071 	},
2072 };
2073 
2074 static void
2075 nvmf_get_cmds_and_effects_log_page(struct iovec *iovs, int iovcnt,
2076 				   uint64_t offset, uint32_t length)
2077 {
2078 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
2079 	size_t copy_len = 0;
2080 	struct copy_iovs_ctx copy_ctx;
2081 
2082 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2083 
2084 	if (offset < page_size) {
2085 		copy_len = spdk_min(page_size - offset, length);
2086 		_copy_buf_to_iovs(&copy_ctx, (char *)(&g_cmds_and_effect_log_page) + offset, copy_len);
2087 	}
2088 }
2089 
2090 static void
2091 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2092 		struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2093 {
2094 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
2095 	struct spdk_nvmf_reservation_log *log, *log_tmp;
2096 	struct copy_iovs_ctx copy_ctx;
2097 
2098 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2099 
2100 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
2101 	/* No available log, return zeroed log pages */
2102 	if (!ctrlr->num_avail_log_pages) {
2103 		return;
2104 	}
2105 
2106 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
2107 	if (offset >= avail_log_len) {
2108 		return;
2109 	}
2110 
2111 	next_pos = 0;
2112 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
2113 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
2114 		ctrlr->num_avail_log_pages--;
2115 
2116 		next_pos += unit_log_len;
2117 		if (next_pos > offset) {
2118 			copy_len = spdk_min(next_pos - offset, length);
2119 			_copy_buf_to_iovs(&copy_ctx, &log->log, copy_len);
2120 			length -= copy_len;
2121 			offset += copy_len;
2122 		}
2123 		free(log);
2124 
2125 		if (length == 0) {
2126 			break;
2127 		}
2128 	}
2129 
2130 	if (!rae) {
2131 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT);
2132 	}
2133 	return;
2134 }
2135 
2136 static int
2137 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
2138 {
2139 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2140 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2141 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2142 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2143 	uint64_t offset, len;
2144 	uint32_t rae, numdl, numdu;
2145 	uint8_t lid;
2146 
2147 	if (req->data == NULL) {
2148 		SPDK_ERRLOG("get log command with no buffer\n");
2149 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2150 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2151 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2152 	}
2153 
2154 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
2155 	if (offset & 3) {
2156 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
2157 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2158 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2159 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2160 	}
2161 
2162 	rae = cmd->cdw10_bits.get_log_page.rae;
2163 	numdl = cmd->cdw10_bits.get_log_page.numdl;
2164 	numdu = cmd->cdw11_bits.get_log_page.numdu;
2165 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
2166 	if (len > req->length) {
2167 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
2168 			    len, req->length);
2169 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2170 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2171 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2172 	}
2173 
2174 	lid = cmd->cdw10_bits.get_log_page.lid;
2175 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n",
2176 		      lid, offset, len, rae);
2177 
2178 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2179 		switch (lid) {
2180 		case SPDK_NVME_LOG_DISCOVERY:
2181 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt, offset,
2182 						    len);
2183 			if (!rae) {
2184 				nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT);
2185 			}
2186 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2187 		default:
2188 			goto invalid_log_page;
2189 		}
2190 	} else {
2191 		switch (lid) {
2192 		case SPDK_NVME_LOG_ERROR:
2193 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
2194 			/* TODO: actually fill out log page data */
2195 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2196 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
2197 			nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len);
2198 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2199 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
2200 			if (subsystem->flags.ana_reporting) {
2201 				nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2202 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2203 			} else {
2204 				goto invalid_log_page;
2205 			}
2206 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
2207 			nvmf_get_cmds_and_effects_log_page(req->iov, req->iovcnt, offset, len);
2208 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2209 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
2210 			nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2211 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2212 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
2213 			nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2214 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2215 		default:
2216 			goto invalid_log_page;
2217 		}
2218 	}
2219 
2220 invalid_log_page:
2221 	SPDK_DEBUGLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid);
2222 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2223 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2224 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2225 }
2226 
2227 int
2228 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2229 			    struct spdk_nvme_cmd *cmd,
2230 			    struct spdk_nvme_cpl *rsp,
2231 			    struct spdk_nvme_ns_data *nsdata)
2232 {
2233 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2234 	struct spdk_nvmf_ns *ns;
2235 	uint32_t max_num_blocks;
2236 
2237 	if (cmd->nsid == 0 || cmd->nsid > subsystem->max_nsid) {
2238 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", cmd->nsid);
2239 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2240 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2241 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2242 	}
2243 
2244 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2245 	if (ns == NULL || ns->bdev == NULL) {
2246 		/*
2247 		 * Inactive namespaces should return a zero filled data structure.
2248 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2249 		 * so we can just return early here.
2250 		 */
2251 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", cmd->nsid);
2252 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2253 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2254 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2255 	}
2256 
2257 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2258 
2259 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2260 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2261 			 (1U << nsdata->lbaf[nsdata->flbas.format].lbads);
2262 	if (nsdata->noiob > max_num_blocks) {
2263 		nsdata->noiob = max_num_blocks;
2264 	}
2265 
2266 	if (subsystem->flags.ana_reporting) {
2267 		/* ANA group ID matches NSID. */
2268 		nsdata->anagrpid = ns->nsid;
2269 
2270 		if (ctrlr->listener->ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2271 		    ctrlr->listener->ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2272 			nsdata->nuse = 0;
2273 		}
2274 	}
2275 
2276 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2277 }
2278 
2279 static void
2280 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2281 			 struct spdk_nvme_ctrlr_data *cdata)
2282 {
2283 	cdata->oacs.virtualization_management =
2284 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2285 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2286 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2287 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2288 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2289 	cdata->oacs.device_self_test =
2290 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2291 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2292 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2293 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2294 			       NULL
2295 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2296 	cdata->oacs.format =
2297 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2298 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2299 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2300 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2301 				     NULL;
2302 }
2303 
2304 int
2305 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2306 {
2307 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2308 	struct spdk_nvmf_transport *transport = ctrlr->admin_qpair->transport;
2309 
2310 	/*
2311 	 * Common fields for discovery and NVM subsystems
2312 	 */
2313 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2314 	assert((transport->opts.max_io_size % 4096) == 0);
2315 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2316 	cdata->cntlid = ctrlr->cntlid;
2317 	cdata->ver = ctrlr->vcprop.vs;
2318 	cdata->aerl = NVMF_MAX_ASYNC_EVENTS - 1;
2319 	cdata->lpa.edlp = 1;
2320 	cdata->elpe = 127;
2321 	cdata->maxcmd = transport->opts.max_queue_depth;
2322 	cdata->sgls = ctrlr->cdata.sgls;
2323 	cdata->fuses.compare_and_write = 1;
2324 	cdata->acwu = 1;
2325 	if (subsystem->flags.ana_reporting) {
2326 		cdata->mnan = subsystem->max_nsid;
2327 	}
2328 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2329 
2330 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2331 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2332 
2333 
2334 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2335 		/*
2336 		 * NVM Discovery subsystem fields
2337 		 */
2338 		cdata->oaes.discovery_log_change_notices = 1;
2339 	} else {
2340 		/*
2341 		 * NVM subsystem fields (reserved for discovery subsystems)
2342 		 */
2343 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2344 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2345 		cdata->kas = ctrlr->cdata.kas;
2346 
2347 		cdata->rab = 6;
2348 		cdata->cmic.multi_port = 1;
2349 		cdata->cmic.multi_ctrlr = 1;
2350 		if (subsystem->flags.ana_reporting) {
2351 			/* Asymmetric Namespace Access Reporting is supported. */
2352 			cdata->cmic.ana_reporting = 1;
2353 		}
2354 		cdata->oaes.ns_attribute_notices = 1;
2355 		if (subsystem->flags.ana_reporting) {
2356 			cdata->oaes.ana_change_notices = 1;
2357 		}
2358 		cdata->ctratt.host_id_exhid_supported = 1;
2359 		/* TODO: Concurrent execution of multiple abort commands. */
2360 		cdata->acl = 0;
2361 		cdata->frmw.slot1_ro = 1;
2362 		cdata->frmw.num_slots = 1;
2363 
2364 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2365 
2366 		cdata->sqes.min = 6;
2367 		cdata->sqes.max = 6;
2368 		cdata->cqes.min = 4;
2369 		cdata->cqes.max = 4;
2370 		cdata->nn = subsystem->max_nsid;
2371 		cdata->vwc.present = 1;
2372 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
2373 
2374 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
2375 
2376 		cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr);
2377 		cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr);
2378 		cdata->oncs.reservations = 1;
2379 		if (subsystem->flags.ana_reporting) {
2380 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
2381 			/* ANA Change state is not used, and ANA Persistent Loss state
2382 			 * is not supported for now.
2383 			 */
2384 			cdata->anacap.ana_optimized_state = 1;
2385 			cdata->anacap.ana_non_optimized_state = 1;
2386 			cdata->anacap.ana_inaccessible_state = 1;
2387 			/* ANAGRPID does not change while namespace is attached to controller */
2388 			cdata->anacap.no_change_anagrpid = 1;
2389 			cdata->anagrpmax = subsystem->max_nsid;
2390 			cdata->nanagrpid = subsystem->max_nsid;
2391 		}
2392 
2393 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
2394 
2395 		assert(subsystem->tgt != NULL);
2396 		cdata->crdt[0] = subsystem->tgt->crdt[0];
2397 		cdata->crdt[1] = subsystem->tgt->crdt[1];
2398 		cdata->crdt[2] = subsystem->tgt->crdt[2];
2399 
2400 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
2401 			      cdata->nvmf_specific.ioccsz);
2402 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
2403 			      cdata->nvmf_specific.iorcsz);
2404 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
2405 			      cdata->nvmf_specific.icdoff);
2406 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
2407 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
2408 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
2409 			      cdata->nvmf_specific.msdbd);
2410 	}
2411 
2412 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2413 }
2414 
2415 static int
2416 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_subsystem *subsystem,
2417 				   struct spdk_nvme_cmd *cmd,
2418 				   struct spdk_nvme_cpl *rsp,
2419 				   struct spdk_nvme_ns_list *ns_list)
2420 {
2421 	struct spdk_nvmf_ns *ns;
2422 	uint32_t count = 0;
2423 
2424 	if (cmd->nsid >= 0xfffffffeUL) {
2425 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
2426 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2427 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2428 	}
2429 
2430 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
2431 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2432 		if (ns->opts.nsid <= cmd->nsid) {
2433 			continue;
2434 		}
2435 
2436 		ns_list->ns_list[count++] = ns->opts.nsid;
2437 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
2438 			break;
2439 		}
2440 	}
2441 
2442 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2443 }
2444 
2445 static void
2446 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
2447 		enum spdk_nvme_nidt type,
2448 		const void *data, size_t data_size)
2449 {
2450 	struct spdk_nvme_ns_id_desc *desc;
2451 	size_t desc_size = sizeof(*desc) + data_size;
2452 
2453 	/*
2454 	 * These should never fail in practice, since all valid NS ID descriptors
2455 	 * should be defined so that they fit in the available 4096-byte buffer.
2456 	 */
2457 	assert(data_size > 0);
2458 	assert(data_size <= UINT8_MAX);
2459 	assert(desc_size < *buf_remain);
2460 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
2461 		return;
2462 	}
2463 
2464 	desc = *buf_ptr;
2465 	desc->nidt = type;
2466 	desc->nidl = data_size;
2467 	memcpy(desc->nid, data, data_size);
2468 
2469 	*buf_ptr += desc_size;
2470 	*buf_remain -= desc_size;
2471 }
2472 
2473 static int
2474 nvmf_ctrlr_identify_ns_id_descriptor_list(
2475 	struct spdk_nvmf_subsystem *subsystem,
2476 	struct spdk_nvme_cmd *cmd,
2477 	struct spdk_nvme_cpl *rsp,
2478 	void *id_desc_list, size_t id_desc_list_size)
2479 {
2480 	struct spdk_nvmf_ns *ns;
2481 	size_t buf_remain = id_desc_list_size;
2482 	void *buf_ptr = id_desc_list;
2483 
2484 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2485 	if (ns == NULL || ns->bdev == NULL) {
2486 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2487 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2488 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2489 	}
2490 
2491 #define ADD_ID_DESC(type, data, size) \
2492 	do { \
2493 		if (!spdk_mem_all_zero(data, size)) { \
2494 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
2495 		} \
2496 	} while (0)
2497 
2498 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
2499 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
2500 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
2501 
2502 	/*
2503 	 * The list is automatically 0-terminated because controller to host buffers in
2504 	 * admin commands always get zeroed in nvmf_ctrlr_process_admin_cmd().
2505 	 */
2506 
2507 #undef ADD_ID_DESC
2508 
2509 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2510 }
2511 
2512 static int
2513 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
2514 {
2515 	uint8_t cns;
2516 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2517 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2518 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2519 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2520 
2521 	if (req->data == NULL || req->length < 4096) {
2522 		SPDK_ERRLOG("identify command with invalid buffer\n");
2523 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2524 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2525 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2526 	}
2527 
2528 	cns = cmd->cdw10_bits.identify.cns;
2529 
2530 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY &&
2531 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
2532 		/* Discovery controllers only support Identify Controller */
2533 		goto invalid_cns;
2534 	}
2535 
2536 	switch (cns) {
2537 	case SPDK_NVME_IDENTIFY_NS:
2538 		return spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, req->data);
2539 	case SPDK_NVME_IDENTIFY_CTRLR:
2540 		return spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, req->data);
2541 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
2542 		return nvmf_ctrlr_identify_active_ns_list(subsystem, cmd, rsp, req->data);
2543 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
2544 		return nvmf_ctrlr_identify_ns_id_descriptor_list(subsystem, cmd, rsp, req->data, req->length);
2545 	default:
2546 		goto invalid_cns;
2547 	}
2548 
2549 invalid_cns:
2550 	SPDK_ERRLOG("Identify command with unsupported CNS 0x%02x\n", cns);
2551 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2552 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2553 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2554 }
2555 
2556 static bool
2557 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
2558 {
2559 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
2560 	struct spdk_nvmf_request *req;
2561 	int i;
2562 
2563 	if (!nvmf_qpair_is_admin_queue(qpair)) {
2564 		return false;
2565 	}
2566 
2567 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2568 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
2569 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
2570 			req = ctrlr->aer_req[i];
2571 			ctrlr->aer_req[i] = NULL;
2572 			ctrlr->nr_aer_reqs--;
2573 
2574 			/* Move the last req to the aborting position for making aer_reqs
2575 			 * in continuous
2576 			 */
2577 			if (i < ctrlr->nr_aer_reqs) {
2578 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
2579 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
2580 			}
2581 
2582 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2583 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2584 			_nvmf_request_complete(req);
2585 			return true;
2586 		}
2587 	}
2588 
2589 	return false;
2590 }
2591 
2592 static void
2593 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
2594 {
2595 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
2596 
2597 	if (nvmf_qpair_abort_aer(qpair, cid)) {
2598 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
2599 			      qpair->ctrlr, qpair->qid, cid);
2600 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
2601 
2602 		spdk_nvmf_request_complete(req);
2603 		return;
2604 	}
2605 
2606 	nvmf_transport_qpair_abort_request(qpair, req);
2607 }
2608 
2609 static void
2610 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
2611 {
2612 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2613 
2614 	if (status == 0) {
2615 		/* There was no qpair whose ID matches SQID of the abort command.
2616 		 * Hence call _nvmf_request_complete() here.
2617 		 */
2618 		_nvmf_request_complete(req);
2619 	}
2620 }
2621 
2622 static void
2623 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
2624 {
2625 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2626 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
2627 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
2628 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
2629 	struct spdk_nvmf_qpair *qpair;
2630 
2631 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
2632 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
2633 			/* Found the qpair */
2634 
2635 			nvmf_qpair_abort_request(qpair, req);
2636 
2637 			/* Return -1 for the status so the iteration across threads stops. */
2638 			spdk_for_each_channel_continue(i, -1);
2639 			return;
2640 		}
2641 	}
2642 
2643 	spdk_for_each_channel_continue(i, 0);
2644 }
2645 
2646 static int
2647 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
2648 {
2649 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2650 
2651 	rsp->cdw0 = 1U; /* Command not aborted */
2652 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2653 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2654 
2655 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
2656 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
2657 			      nvmf_ctrlr_abort_on_pg,
2658 			      req,
2659 			      nvmf_ctrlr_abort_done
2660 			     );
2661 
2662 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2663 }
2664 
2665 int
2666 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
2667 {
2668 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
2669 	struct spdk_bdev *bdev;
2670 	struct spdk_bdev_desc *desc;
2671 	struct spdk_io_channel *ch;
2672 	int rc;
2673 
2674 	assert(req_to_abort != NULL);
2675 
2676 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
2677 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
2678 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
2679 	}
2680 
2681 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
2682 					&bdev, &desc, &ch);
2683 	if (rc != 0) {
2684 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2685 	}
2686 
2687 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
2688 }
2689 
2690 static int
2691 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
2692 {
2693 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2694 
2695 	rsp->cdw0 = cdw0;
2696 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2697 }
2698 
2699 /* we have to use the typedef in the function declaration to appease astyle. */
2700 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
2701 
2702 static spdk_nvme_path_status_code_t
2703 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
2704 {
2705 	switch (ana_state) {
2706 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2707 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
2708 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2709 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
2710 	case SPDK_NVME_ANA_CHANGE_STATE:
2711 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
2712 	default:
2713 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
2714 	}
2715 }
2716 
2717 static int
2718 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
2719 {
2720 	uint8_t feature;
2721 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2722 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2723 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2724 	enum spdk_nvme_ana_state ana_state;
2725 
2726 	feature = cmd->cdw10_bits.get_features.fid;
2727 
2728 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2729 		/*
2730 		 * Features supported by Discovery controller
2731 		 */
2732 		switch (feature) {
2733 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2734 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
2735 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2736 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
2737 		default:
2738 			SPDK_DEBUGLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
2739 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2740 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2741 		}
2742 	}
2743 	/*
2744 	 * Process Get Features command for non-discovery controller
2745 	 */
2746 	ana_state = ctrlr->listener->ana_state;
2747 	switch (ana_state) {
2748 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2749 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2750 	case SPDK_NVME_ANA_CHANGE_STATE:
2751 		switch (feature) {
2752 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
2753 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2754 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2755 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2756 			response->status.sct = SPDK_NVME_SCT_PATH;
2757 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2758 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2759 		default:
2760 			break;
2761 		}
2762 		break;
2763 	default:
2764 		break;
2765 	}
2766 
2767 	switch (feature) {
2768 	case SPDK_NVME_FEAT_ARBITRATION:
2769 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
2770 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
2771 		return get_features_generic(req, ctrlr->feat.power_management.raw);
2772 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
2773 		return nvmf_ctrlr_get_features_temperature_threshold(req);
2774 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
2775 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
2776 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
2777 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
2778 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
2779 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
2780 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2781 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
2782 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2783 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
2784 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2785 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
2786 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2787 		return nvmf_ctrlr_get_features_host_identifier(req);
2788 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2789 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
2790 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2791 		return nvmf_ctrlr_get_features_reservation_persistence(req);
2792 	default:
2793 		SPDK_ERRLOG("Get Features command with unsupported feature ID 0x%02x\n", feature);
2794 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2795 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2796 	}
2797 }
2798 
2799 static int
2800 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
2801 {
2802 	uint8_t feature, save;
2803 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2804 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2805 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2806 	enum spdk_nvme_ana_state ana_state;
2807 	/*
2808 	 * Features are not saveable by the controller as indicated by
2809 	 * ONCS field of the Identify Controller data.
2810 	 * */
2811 	save = cmd->cdw10_bits.set_features.sv;
2812 	if (save) {
2813 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
2814 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
2815 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2816 	}
2817 
2818 	feature = cmd->cdw10_bits.set_features.fid;
2819 
2820 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2821 		/*
2822 		 * Features supported by Discovery controller
2823 		 */
2824 		switch (feature) {
2825 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2826 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
2827 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2828 			return nvmf_ctrlr_set_features_async_event_configuration(req);
2829 		default:
2830 			SPDK_ERRLOG("Set Features command with unsupported feature ID 0x%02x\n", feature);
2831 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2832 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2833 		}
2834 	}
2835 	/*
2836 	 * Process Set Features command for non-discovery controller
2837 	 */
2838 	ana_state = ctrlr->listener->ana_state;
2839 	switch (ana_state) {
2840 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2841 	case SPDK_NVME_ANA_CHANGE_STATE:
2842 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
2843 			response->status.sct = SPDK_NVME_SCT_PATH;
2844 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2845 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2846 		} else {
2847 			switch (feature) {
2848 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
2849 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2850 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2851 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2852 				response->status.sct = SPDK_NVME_SCT_PATH;
2853 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2854 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2855 			default:
2856 				break;
2857 			}
2858 		}
2859 		break;
2860 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2861 		response->status.sct = SPDK_NVME_SCT_PATH;
2862 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
2863 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2864 	default:
2865 		break;
2866 	}
2867 
2868 	switch (feature) {
2869 	case SPDK_NVME_FEAT_ARBITRATION:
2870 		return nvmf_ctrlr_set_features_arbitration(req);
2871 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
2872 		return nvmf_ctrlr_set_features_power_management(req);
2873 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
2874 		return nvmf_ctrlr_set_features_temperature_threshold(req);
2875 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
2876 		return nvmf_ctrlr_set_features_error_recovery(req);
2877 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
2878 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
2879 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
2880 		return nvmf_ctrlr_set_features_number_of_queues(req);
2881 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2882 		return nvmf_ctrlr_set_features_write_atomicity(req);
2883 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2884 		return nvmf_ctrlr_set_features_async_event_configuration(req);
2885 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2886 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
2887 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2888 		return nvmf_ctrlr_set_features_host_identifier(req);
2889 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2890 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
2891 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2892 		return nvmf_ctrlr_set_features_reservation_persistence(req);
2893 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
2894 		return nvmf_ctrlr_set_features_host_behavior_support(req);
2895 	default:
2896 		SPDK_ERRLOG("Set Features command with unsupported feature ID 0x%02x\n", feature);
2897 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2898 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2899 	}
2900 }
2901 
2902 static int
2903 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
2904 {
2905 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2906 
2907 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
2908 	/*
2909 	 * To handle keep alive just clear or reset the
2910 	 * ctrlr based keep alive duration counter.
2911 	 * When added, a separate timer based process
2912 	 * will monitor if the time since last recorded
2913 	 * keep alive has exceeded the max duration and
2914 	 * take appropriate action.
2915 	 */
2916 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
2917 
2918 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2919 }
2920 
2921 int
2922 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
2923 {
2924 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2925 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2926 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2927 	int rc;
2928 
2929 	if (ctrlr == NULL) {
2930 		SPDK_ERRLOG("Admin command sent before CONNECT\n");
2931 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2932 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2933 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2934 	}
2935 
2936 	if (ctrlr->vcprop.cc.bits.en != 1) {
2937 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
2938 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2939 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2940 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2941 	}
2942 
2943 	if (req->data && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2944 		memset(req->data, 0, req->length);
2945 	}
2946 
2947 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2948 		/* Discovery controllers only support these admin OPS. */
2949 		switch (cmd->opc) {
2950 		case SPDK_NVME_OPC_IDENTIFY:
2951 		case SPDK_NVME_OPC_GET_LOG_PAGE:
2952 		case SPDK_NVME_OPC_KEEP_ALIVE:
2953 		case SPDK_NVME_OPC_SET_FEATURES:
2954 		case SPDK_NVME_OPC_GET_FEATURES:
2955 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
2956 			break;
2957 		default:
2958 			goto invalid_opcode;
2959 		}
2960 	}
2961 
2962 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
2963 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
2964 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
2965 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
2966 			/* The handler took care of this commmand */
2967 			return rc;
2968 		}
2969 	}
2970 
2971 	switch (cmd->opc) {
2972 	case SPDK_NVME_OPC_GET_LOG_PAGE:
2973 		return nvmf_ctrlr_get_log_page(req);
2974 	case SPDK_NVME_OPC_IDENTIFY:
2975 		return nvmf_ctrlr_identify(req);
2976 	case SPDK_NVME_OPC_ABORT:
2977 		return nvmf_ctrlr_abort(req);
2978 	case SPDK_NVME_OPC_GET_FEATURES:
2979 		return nvmf_ctrlr_get_features(req);
2980 	case SPDK_NVME_OPC_SET_FEATURES:
2981 		return nvmf_ctrlr_set_features(req);
2982 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
2983 		return nvmf_ctrlr_async_event_request(req);
2984 	case SPDK_NVME_OPC_KEEP_ALIVE:
2985 		return nvmf_ctrlr_keep_alive(req);
2986 
2987 	case SPDK_NVME_OPC_CREATE_IO_SQ:
2988 	case SPDK_NVME_OPC_CREATE_IO_CQ:
2989 	case SPDK_NVME_OPC_DELETE_IO_SQ:
2990 	case SPDK_NVME_OPC_DELETE_IO_CQ:
2991 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
2992 		goto invalid_opcode;
2993 
2994 	default:
2995 		goto invalid_opcode;
2996 	}
2997 
2998 invalid_opcode:
2999 	SPDK_ERRLOG("Unsupported admin opcode 0x%x\n", cmd->opc);
3000 	response->status.sct = SPDK_NVME_SCT_GENERIC;
3001 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3002 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3003 }
3004 
3005 static int
3006 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
3007 {
3008 	struct spdk_nvmf_qpair *qpair = req->qpair;
3009 	struct spdk_nvmf_capsule_cmd *cap_hdr;
3010 
3011 	cap_hdr = &req->cmd->nvmf_cmd;
3012 
3013 	if (qpair->ctrlr == NULL) {
3014 		/* No ctrlr established yet; the only valid command is Connect */
3015 		if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
3016 			return nvmf_ctrlr_cmd_connect(req);
3017 		} else {
3018 			SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n",
3019 				      cap_hdr->fctype);
3020 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3021 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3022 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3023 		}
3024 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
3025 		/*
3026 		 * Controller session is established, and this is an admin queue.
3027 		 * Disallow Connect and allow other fabrics commands.
3028 		 */
3029 		switch (cap_hdr->fctype) {
3030 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
3031 			return nvmf_property_set(req);
3032 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
3033 			return nvmf_property_get(req);
3034 		default:
3035 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
3036 				      cap_hdr->fctype);
3037 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3038 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3039 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3040 		}
3041 	} else {
3042 		/* Controller session is established, and this is an I/O queue */
3043 		/* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */
3044 		SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
3045 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3046 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3047 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3048 	}
3049 }
3050 
3051 static inline int
3052 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
3053 				    union spdk_nvme_async_event_completion *event)
3054 {
3055 	struct spdk_nvmf_request *req;
3056 	struct spdk_nvme_cpl *rsp;
3057 
3058 	assert(ctrlr->nr_aer_reqs > 0);
3059 
3060 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
3061 	rsp = &req->rsp->nvme_cpl;
3062 
3063 	rsp->cdw0 = event->raw;
3064 
3065 	_nvmf_request_complete(req);
3066 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3067 
3068 	return 0;
3069 }
3070 
3071 static inline void
3072 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr,
3073 				     union spdk_nvme_async_event_completion *event)
3074 {
3075 	struct spdk_nvmf_async_event_completion *nvmf_event;
3076 
3077 	nvmf_event = calloc(1, sizeof(*nvmf_event));
3078 	if (!nvmf_event) {
3079 		SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n");
3080 		return;
3081 	}
3082 	nvmf_event->event.raw = event->raw;
3083 	STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link);
3084 }
3085 
3086 int
3087 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
3088 {
3089 	union spdk_nvme_async_event_completion event = {0};
3090 
3091 	/* Users may disable the event notification */
3092 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
3093 		return 0;
3094 	}
3095 
3096 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) {
3097 		return 0;
3098 	}
3099 
3100 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3101 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
3102 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
3103 
3104 	/* If there is no outstanding AER request, queue the event.  Then
3105 	 * if an AER is later submitted, this event can be sent as a
3106 	 * response.
3107 	 */
3108 	if (ctrlr->nr_aer_reqs == 0) {
3109 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3110 		return 0;
3111 	}
3112 
3113 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3114 }
3115 
3116 int
3117 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3118 {
3119 	union spdk_nvme_async_event_completion event = {0};
3120 
3121 	/* Users may disable the event notification */
3122 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
3123 		return 0;
3124 	}
3125 
3126 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) {
3127 		return 0;
3128 	}
3129 
3130 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3131 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
3132 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
3133 
3134 	/* If there is no outstanding AER request, queue the event.  Then
3135 	 * if an AER is later submitted, this event can be sent as a
3136 	 * response.
3137 	 */
3138 	if (ctrlr->nr_aer_reqs == 0) {
3139 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3140 		return 0;
3141 	}
3142 
3143 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3144 }
3145 
3146 void
3147 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
3148 {
3149 	union spdk_nvme_async_event_completion event = {0};
3150 
3151 	if (!ctrlr->num_avail_log_pages) {
3152 		return;
3153 	}
3154 
3155 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) {
3156 		return;
3157 	}
3158 
3159 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
3160 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
3161 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
3162 
3163 	/* If there is no outstanding AER request, queue the event.  Then
3164 	 * if an AER is later submitted, this event can be sent as a
3165 	 * response.
3166 	 */
3167 	if (ctrlr->nr_aer_reqs == 0) {
3168 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3169 		return;
3170 	}
3171 
3172 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3173 }
3174 
3175 int
3176 nvmf_ctrlr_async_event_discovery_log_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3177 {
3178 	union spdk_nvme_async_event_completion event = {0};
3179 
3180 	/* Users may disable the event notification manually or
3181 	 * it may not be enabled due to keep alive timeout
3182 	 * not being set in connect command to discovery controller.
3183 	 */
3184 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
3185 		return 0;
3186 	}
3187 
3188 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) {
3189 		return 0;
3190 	}
3191 
3192 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3193 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
3194 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
3195 
3196 	/* If there is no outstanding AER request, queue the event.  Then
3197 	 * if an AER is later submitted, this event can be sent as a
3198 	 * response.
3199 	 */
3200 	if (ctrlr->nr_aer_reqs == 0) {
3201 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3202 		return 0;
3203 	}
3204 
3205 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3206 }
3207 
3208 void
3209 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
3210 {
3211 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3212 	int i;
3213 
3214 	if (!nvmf_qpair_is_admin_queue(qpair)) {
3215 		return;
3216 	}
3217 
3218 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3219 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
3220 		ctrlr->aer_req[i] = NULL;
3221 	}
3222 
3223 	ctrlr->nr_aer_reqs = 0;
3224 }
3225 
3226 void
3227 nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
3228 {
3229 	struct spdk_nvmf_request *req;
3230 	int i;
3231 
3232 	if (!ctrlr->nr_aer_reqs) {
3233 		return;
3234 	}
3235 
3236 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3237 		req = ctrlr->aer_req[i];
3238 
3239 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3240 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3241 		_nvmf_request_complete(req);
3242 
3243 		ctrlr->aer_req[i] = NULL;
3244 	}
3245 
3246 	ctrlr->nr_aer_reqs = 0;
3247 }
3248 
3249 static void
3250 _nvmf_ctrlr_add_reservation_log(void *ctx)
3251 {
3252 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
3253 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
3254 
3255 	ctrlr->log_page_count++;
3256 
3257 	/* Maximum number of queued log pages is 255 */
3258 	if (ctrlr->num_avail_log_pages == 0xff) {
3259 		struct spdk_nvmf_reservation_log *entry;
3260 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
3261 		entry->log.log_page_count = ctrlr->log_page_count;
3262 		free(log);
3263 		return;
3264 	}
3265 
3266 	log->log.log_page_count = ctrlr->log_page_count;
3267 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
3268 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
3269 
3270 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
3271 }
3272 
3273 void
3274 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
3275 				  struct spdk_nvmf_ns *ns,
3276 				  enum spdk_nvme_reservation_notification_log_page_type type)
3277 {
3278 	struct spdk_nvmf_reservation_log *log;
3279 
3280 	switch (type) {
3281 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
3282 		return;
3283 	case SPDK_NVME_REGISTRATION_PREEMPTED:
3284 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
3285 			return;
3286 		}
3287 		break;
3288 	case SPDK_NVME_RESERVATION_RELEASED:
3289 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
3290 			return;
3291 		}
3292 		break;
3293 	case SPDK_NVME_RESERVATION_PREEMPTED:
3294 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
3295 			return;
3296 		}
3297 		break;
3298 	default:
3299 		return;
3300 	}
3301 
3302 	log = calloc(1, sizeof(*log));
3303 	if (!log) {
3304 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
3305 		return;
3306 	}
3307 	log->ctrlr = ctrlr;
3308 	log->log.type = type;
3309 	log->log.nsid = ns->nsid;
3310 
3311 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
3312 }
3313 
3314 /* Check from subsystem poll group's namespace information data structure */
3315 static bool
3316 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3317 				 struct spdk_nvmf_ctrlr *ctrlr)
3318 {
3319 	uint32_t i;
3320 
3321 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
3322 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
3323 			return true;
3324 		}
3325 	}
3326 
3327 	return false;
3328 }
3329 
3330 /*
3331  * Check the NVMe command is permitted or not for current controller(Host).
3332  */
3333 static int
3334 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3335 				  struct spdk_nvmf_ctrlr *ctrlr,
3336 				  struct spdk_nvmf_request *req)
3337 {
3338 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3339 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
3340 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3341 	uint8_t racqa;
3342 	bool is_registrant;
3343 
3344 	/* No valid reservation */
3345 	if (!rtype) {
3346 		return 0;
3347 	}
3348 
3349 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
3350 	/* All registrants type and current ctrlr is a valid registrant */
3351 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3352 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
3353 		return 0;
3354 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
3355 		return 0;
3356 	}
3357 
3358 	/* Non-holder for current controller */
3359 	switch (cmd->opc) {
3360 	case SPDK_NVME_OPC_READ:
3361 	case SPDK_NVME_OPC_COMPARE:
3362 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3363 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3364 			goto exit;
3365 		}
3366 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
3367 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
3368 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3369 		}
3370 		break;
3371 	case SPDK_NVME_OPC_FLUSH:
3372 	case SPDK_NVME_OPC_WRITE:
3373 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
3374 	case SPDK_NVME_OPC_WRITE_ZEROES:
3375 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3376 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
3377 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3378 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3379 			goto exit;
3380 		}
3381 		if (!is_registrant) {
3382 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3383 		}
3384 		break;
3385 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3386 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
3387 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
3388 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3389 			goto exit;
3390 		}
3391 		if (!is_registrant) {
3392 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3393 		}
3394 		break;
3395 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3396 		if (!is_registrant) {
3397 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3398 		}
3399 		break;
3400 	default:
3401 		break;
3402 	}
3403 
3404 exit:
3405 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3406 	req->rsp->nvme_cpl.status.sc = status;
3407 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
3408 		return -EPERM;
3409 	}
3410 
3411 	return 0;
3412 }
3413 
3414 static int
3415 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
3416 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
3417 {
3418 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3419 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3420 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
3421 	int rc;
3422 
3423 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
3424 		/* first fused operation (should be compare) */
3425 		if (first_fused_req != NULL) {
3426 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3427 
3428 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3429 
3430 			/* abort req->qpair->first_fused_request and continue with new fused command */
3431 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3432 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3433 			_nvmf_request_complete(first_fused_req);
3434 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
3435 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3436 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3437 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3438 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3439 		}
3440 
3441 		req->qpair->first_fused_req = req;
3442 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3443 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
3444 		/* second fused operation (should be write) */
3445 		if (first_fused_req == NULL) {
3446 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3447 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3448 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3449 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3450 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
3451 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3452 
3453 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3454 
3455 			/* abort req->qpair->first_fused_request and fail current command */
3456 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3457 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3458 			_nvmf_request_complete(first_fused_req);
3459 
3460 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3461 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3462 			req->qpair->first_fused_req = NULL;
3463 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3464 		}
3465 
3466 		/* save request of first command to generate response later */
3467 		req->first_fused_req = first_fused_req;
3468 		req->qpair->first_fused_req = NULL;
3469 	} else {
3470 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
3471 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3472 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3473 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3474 	}
3475 
3476 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
3477 
3478 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3479 		if (spdk_nvme_cpl_is_error(rsp)) {
3480 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3481 
3482 			fused_response->status = rsp->status;
3483 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3484 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
3485 			/* Complete first of fused commands. Second will be completed by upper layer */
3486 			_nvmf_request_complete(first_fused_req);
3487 			req->first_fused_req = NULL;
3488 		}
3489 	}
3490 
3491 	return rc;
3492 }
3493 
3494 int
3495 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
3496 {
3497 	uint32_t nsid;
3498 	struct spdk_nvmf_ns *ns;
3499 	struct spdk_bdev *bdev;
3500 	struct spdk_bdev_desc *desc;
3501 	struct spdk_io_channel *ch;
3502 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3503 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3504 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3505 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3506 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3507 	enum spdk_nvme_ana_state ana_state;
3508 
3509 	/* pre-set response details for this command */
3510 	response->status.sc = SPDK_NVME_SC_SUCCESS;
3511 	nsid = cmd->nsid;
3512 
3513 	if (spdk_unlikely(ctrlr == NULL)) {
3514 		SPDK_ERRLOG("I/O command sent before CONNECT\n");
3515 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3516 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3517 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3518 	}
3519 
3520 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
3521 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
3522 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3523 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3524 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3525 	}
3526 
3527 	/* It will be lower overhead to check if ANA state is optimized or
3528 	 * non-optimized.
3529 	 */
3530 	ana_state = ctrlr->listener->ana_state;
3531 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
3532 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
3533 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
3534 			      ana_state);
3535 		response->status.sct = SPDK_NVME_SCT_PATH;
3536 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3537 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3538 	}
3539 
3540 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3541 	if (ns == NULL || ns->bdev == NULL) {
3542 		SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid);
3543 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3544 		response->status.dnr = 1;
3545 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3546 	}
3547 
3548 	/* scan-build falsely reporting dereference of null pointer */
3549 	assert(group != NULL && group->sgroups != NULL);
3550 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3551 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
3552 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
3553 			      cmd->nsid, cmd->opc);
3554 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3555 	}
3556 
3557 	bdev = ns->bdev;
3558 	desc = ns->desc;
3559 	ch = ns_info->channel;
3560 
3561 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
3562 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
3563 	} else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) {
3564 		struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl;
3565 
3566 		SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n");
3567 
3568 		/* abort req->qpair->first_fused_request and continue with new command */
3569 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3570 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3571 		_nvmf_request_complete(req->qpair->first_fused_req);
3572 		req->qpair->first_fused_req = NULL;
3573 	}
3574 
3575 	switch (cmd->opc) {
3576 	case SPDK_NVME_OPC_READ:
3577 		return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
3578 	case SPDK_NVME_OPC_WRITE:
3579 		return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
3580 	case SPDK_NVME_OPC_COMPARE:
3581 		return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
3582 	case SPDK_NVME_OPC_WRITE_ZEROES:
3583 		return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
3584 	case SPDK_NVME_OPC_FLUSH:
3585 		return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
3586 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3587 		return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
3588 	case SPDK_NVME_OPC_RESERVATION_REGISTER:
3589 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3590 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3591 	case SPDK_NVME_OPC_RESERVATION_REPORT:
3592 		spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
3593 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3594 	default:
3595 		return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
3596 	}
3597 }
3598 
3599 static void
3600 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
3601 {
3602 	if (qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING) {
3603 		assert(qpair->state_cb != NULL);
3604 
3605 		if (TAILQ_EMPTY(&qpair->outstanding)) {
3606 			qpair->state_cb(qpair->state_cb_arg, 0);
3607 		}
3608 	}
3609 }
3610 
3611 int
3612 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
3613 {
3614 	struct spdk_nvmf_qpair *qpair = req->qpair;
3615 
3616 	TAILQ_REMOVE(&qpair->outstanding, req, link);
3617 	if (nvmf_transport_req_free(req)) {
3618 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
3619 	}
3620 
3621 	nvmf_qpair_request_cleanup(qpair);
3622 
3623 	return 0;
3624 }
3625 
3626 static void
3627 _nvmf_request_complete(void *ctx)
3628 {
3629 	struct spdk_nvmf_request *req = ctx;
3630 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3631 	struct spdk_nvmf_qpair *qpair;
3632 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3633 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3634 	bool is_aer = false;
3635 	uint32_t nsid;
3636 	bool paused;
3637 	uint8_t opcode;
3638 
3639 	rsp->sqid = 0;
3640 	rsp->status.p = 0;
3641 	rsp->cid = req->cmd->nvme_cmd.cid;
3642 	nsid = req->cmd->nvme_cmd.nsid;
3643 	opcode = req->cmd->nvmf_cmd.opcode;
3644 
3645 	qpair = req->qpair;
3646 	if (qpair->ctrlr) {
3647 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3648 		assert(sgroup != NULL);
3649 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
3650 
3651 		/*
3652 		 * Set the crd value.
3653 		 * If the the IO has any error, and dnr (DoNotRetry) is not 1,
3654 		 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT.
3655 		 */
3656 		if (spdk_nvme_cpl_is_error(rsp) &&
3657 		    rsp->status.dnr == 0 &&
3658 		    qpair->ctrlr->acre_enabled) {
3659 			rsp->status.crd = 1;
3660 		}
3661 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3662 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3663 	}
3664 
3665 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
3666 		spdk_nvme_print_completion(qpair->qid, rsp);
3667 	}
3668 
3669 	TAILQ_REMOVE(&qpair->outstanding, req, link);
3670 	if (nvmf_transport_req_complete(req)) {
3671 		SPDK_ERRLOG("Transport request completion error!\n");
3672 	}
3673 
3674 	/* AER cmd is an exception */
3675 	if (sgroup && !is_aer) {
3676 		if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC ||
3677 				  nvmf_qpair_is_admin_queue(qpair))) {
3678 			assert(sgroup->mgmt_io_outstanding > 0);
3679 			sgroup->mgmt_io_outstanding--;
3680 		} else {
3681 			/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
3682 			if (spdk_likely(nsid - 1 < sgroup->num_ns)) {
3683 				sgroup->ns_info[nsid - 1].io_outstanding--;
3684 			}
3685 		}
3686 
3687 		if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
3688 				  sgroup->mgmt_io_outstanding == 0)) {
3689 			paused = true;
3690 			for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
3691 				ns_info = &sgroup->ns_info[nsid];
3692 
3693 				if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
3694 				    ns_info->io_outstanding > 0) {
3695 					paused = false;
3696 					break;
3697 				}
3698 			}
3699 
3700 			if (paused) {
3701 				sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
3702 				sgroup->cb_fn(sgroup->cb_arg, 0);
3703 				sgroup->cb_fn = NULL;
3704 				sgroup->cb_arg = NULL;
3705 			}
3706 		}
3707 
3708 	}
3709 
3710 	nvmf_qpair_request_cleanup(qpair);
3711 }
3712 
3713 int
3714 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
3715 {
3716 	struct spdk_nvmf_qpair *qpair = req->qpair;
3717 
3718 	if (spdk_likely(qpair->group->thread == spdk_get_thread())) {
3719 		_nvmf_request_complete(req);
3720 	} else {
3721 		spdk_thread_send_msg(qpair->group->thread,
3722 				     _nvmf_request_complete, req);
3723 	}
3724 
3725 	return 0;
3726 }
3727 
3728 void
3729 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req)
3730 {
3731 	struct spdk_nvmf_qpair *qpair = req->qpair;
3732 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3733 	enum spdk_nvmf_request_exec_status status;
3734 
3735 	if (qpair->ctrlr) {
3736 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3737 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3738 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3739 	}
3740 
3741 	assert(sgroup != NULL);
3742 	sgroup->mgmt_io_outstanding++;
3743 
3744 	/* Place the request on the outstanding list so we can keep track of it */
3745 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3746 
3747 	assert(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC);
3748 	status = nvmf_ctrlr_process_fabrics_cmd(req);
3749 
3750 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3751 		_nvmf_request_complete(req);
3752 	}
3753 }
3754 
3755 void
3756 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
3757 {
3758 	struct spdk_nvmf_qpair *qpair = req->qpair;
3759 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3760 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3761 	enum spdk_nvmf_request_exec_status status;
3762 	uint32_t nsid;
3763 
3764 	if (qpair->ctrlr) {
3765 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3766 		assert(sgroup != NULL);
3767 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3768 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3769 	}
3770 
3771 	/* Check if the subsystem is paused (if there is a subsystem) */
3772 	if (sgroup != NULL) {
3773 		if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC ||
3774 				  nvmf_qpair_is_admin_queue(qpair))) {
3775 			if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
3776 				/* The subsystem is not currently active. Queue this request. */
3777 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
3778 				return;
3779 			}
3780 			sgroup->mgmt_io_outstanding++;
3781 		} else {
3782 			nsid = req->cmd->nvme_cmd.nsid;
3783 
3784 			/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
3785 			if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) {
3786 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3787 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3788 				req->rsp->nvme_cpl.status.dnr = 1;
3789 				TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3790 				_nvmf_request_complete(req);
3791 				return;
3792 			}
3793 
3794 			ns_info = &sgroup->ns_info[nsid - 1];
3795 			if (ns_info->channel == NULL) {
3796 				/* This can can happen if host sends I/O to a namespace that is
3797 				 * in the process of being added, but before the full addition
3798 				 * process is complete.  Report invalid namespace in that case.
3799 				 */
3800 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3801 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3802 				req->rsp->nvme_cpl.status.dnr = 1;
3803 				TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3804 				ns_info->io_outstanding++;
3805 				_nvmf_request_complete(req);
3806 				return;
3807 			}
3808 
3809 			if (ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
3810 				/* The namespace is not currently active. Queue this request. */
3811 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
3812 				return;
3813 			}
3814 			ns_info->io_outstanding++;
3815 		}
3816 	}
3817 
3818 	if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
3819 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3820 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3821 		TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3822 		_nvmf_request_complete(req);
3823 		return;
3824 	}
3825 
3826 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
3827 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
3828 	}
3829 
3830 	/* Place the request on the outstanding list so we can keep track of it */
3831 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3832 
3833 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
3834 		status = nvmf_ctrlr_process_fabrics_cmd(req);
3835 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
3836 		status = nvmf_ctrlr_process_admin_cmd(req);
3837 	} else {
3838 		status = nvmf_ctrlr_process_io_cmd(req);
3839 	}
3840 
3841 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3842 		_nvmf_request_complete(req);
3843 	}
3844 }
3845 
3846 static bool
3847 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
3848 		       struct spdk_dif_ctx *dif_ctx)
3849 {
3850 	struct spdk_nvmf_ns *ns;
3851 	struct spdk_bdev *bdev;
3852 
3853 	if (ctrlr == NULL || cmd == NULL) {
3854 		return false;
3855 	}
3856 
3857 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
3858 	if (ns == NULL || ns->bdev == NULL) {
3859 		return false;
3860 	}
3861 
3862 	bdev = ns->bdev;
3863 
3864 	switch (cmd->opc) {
3865 	case SPDK_NVME_OPC_READ:
3866 	case SPDK_NVME_OPC_WRITE:
3867 	case SPDK_NVME_OPC_COMPARE:
3868 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
3869 	default:
3870 		break;
3871 	}
3872 
3873 	return false;
3874 }
3875 
3876 bool
3877 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
3878 {
3879 	struct spdk_nvmf_qpair *qpair = req->qpair;
3880 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3881 
3882 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
3883 		return false;
3884 	}
3885 
3886 	if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) {
3887 		return false;
3888 	}
3889 
3890 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
3891 		return false;
3892 	}
3893 
3894 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
3895 		return false;
3896 	}
3897 
3898 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
3899 }
3900 
3901 void
3902 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
3903 {
3904 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
3905 }
3906 
3907 static int
3908 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
3909 {
3910 	struct spdk_bdev *bdev;
3911 	struct spdk_bdev_desc *desc;
3912 	struct spdk_io_channel *ch;
3913 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
3914 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
3915 	uint32_t bdev_nsid;
3916 	int rc;
3917 
3918 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid == 0) {
3919 		bdev_nsid = cmd->nsid;
3920 	} else {
3921 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
3922 	}
3923 
3924 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
3925 	if (rc) {
3926 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3927 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3928 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3929 	}
3930 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
3931 }
3932 
3933 void
3934 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
3935 {
3936 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
3937 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
3938 }
3939 
3940 int
3941 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
3942 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
3943 {
3944 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3945 	struct spdk_nvmf_ns *ns;
3946 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3947 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3948 
3949 	*bdev = NULL;
3950 	*desc = NULL;
3951 	*ch = NULL;
3952 
3953 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3954 	if (ns == NULL || ns->bdev == NULL) {
3955 		return -EINVAL;
3956 	}
3957 
3958 	assert(group != NULL && group->sgroups != NULL);
3959 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3960 	*bdev = ns->bdev;
3961 	*desc = ns->desc;
3962 	*ch = ns_info->channel;
3963 
3964 	return 0;
3965 }
3966 
3967 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
3968 {
3969 	return req->qpair->ctrlr;
3970 }
3971 
3972 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
3973 {
3974 	return &req->cmd->nvme_cmd;
3975 }
3976 
3977 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
3978 {
3979 	return &req->rsp->nvme_cpl;
3980 }
3981 
3982 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
3983 {
3984 	return req->qpair->ctrlr->subsys;
3985 }
3986 
3987 void spdk_nvmf_request_get_data(struct spdk_nvmf_request *req, void **data, uint32_t *length)
3988 {
3989 	*data = req->data;
3990 	*length = req->length;
3991 }
3992 
3993 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
3994 {
3995 	return ctrlr->subsys;
3996 }
3997 
3998 uint16_t spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
3999 {
4000 	return ctrlr->cntlid;
4001 }
4002 
4003 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
4004 {
4005 	return req->req_to_abort;
4006 }
4007