xref: /spdk/lib/nvmf/ctrlr.c (revision c39647df83e4be9bcc49025132c48bf2414ef8b1)
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  *   Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "spdk/stdinc.h"
36 
37 #include "nvmf_internal.h"
38 #include "transport.h"
39 
40 #include "spdk/bit_array.h"
41 #include "spdk/endian.h"
42 #include "spdk/thread.h"
43 #include "spdk/nvme_spec.h"
44 #include "spdk/nvmf_cmd.h"
45 #include "spdk/string.h"
46 #include "spdk/util.h"
47 #include "spdk/version.h"
48 #include "spdk/log.h"
49 #include "spdk_internal/usdt.h"
50 
51 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000
52 #define NVMF_DISC_KATO_IN_MS 120000
53 #define KAS_TIME_UNIT_IN_MS 100
54 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS)
55 
56 #define NVMF_CC_RESET_SHN_TIMEOUT_IN_MS	10000
57 
58 #define NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS	(NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + 5000)
59 
60 /*
61  * Report the SPDK version as the firmware revision.
62  * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts.
63  */
64 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING
65 
66 #define ANA_TRANSITION_TIME_IN_SEC 10
67 
68 #define NVMF_ABORT_COMMAND_LIMIT 3
69 
70 /*
71  * Support for custom admin command handlers
72  */
73 struct spdk_nvmf_custom_admin_cmd {
74 	spdk_nvmf_custom_cmd_hdlr hdlr;
75 	uint32_t nsid; /* nsid to forward */
76 };
77 
78 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1];
79 
80 static void _nvmf_request_complete(void *ctx);
81 
82 static inline void
83 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp,
84 			      uint8_t iattr, uint16_t ipo)
85 {
86 	rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
87 	rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
88 	rsp->status_code_specific.invalid.iattr = iattr;
89 	rsp->status_code_specific.invalid.ipo = ipo;
90 }
91 
92 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field)	\
93 	nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field))
94 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field)	\
95 	nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field))
96 
97 
98 static void
99 nvmf_ctrlr_stop_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
100 {
101 	if (!ctrlr) {
102 		SPDK_ERRLOG("Controller is NULL\n");
103 		return;
104 	}
105 
106 	if (ctrlr->keep_alive_poller == NULL) {
107 		return;
108 	}
109 
110 	SPDK_DEBUGLOG(nvmf, "Stop keep alive poller\n");
111 	spdk_poller_unregister(&ctrlr->keep_alive_poller);
112 }
113 
114 static void
115 nvmf_ctrlr_stop_association_timer(struct spdk_nvmf_ctrlr *ctrlr)
116 {
117 	if (!ctrlr) {
118 		SPDK_ERRLOG("Controller is NULL\n");
119 		assert(false);
120 		return;
121 	}
122 
123 	if (ctrlr->association_timer == NULL) {
124 		return;
125 	}
126 
127 	SPDK_DEBUGLOG(nvmf, "Stop association timer\n");
128 	spdk_poller_unregister(&ctrlr->association_timer);
129 }
130 
131 static void
132 nvmf_ctrlr_disconnect_qpairs_done(struct spdk_io_channel_iter *i, int status)
133 {
134 	if (status == 0) {
135 		SPDK_DEBUGLOG(nvmf, "ctrlr disconnect qpairs complete successfully\n");
136 	} else {
137 		SPDK_ERRLOG("Fail to disconnect ctrlr qpairs\n");
138 	}
139 }
140 
141 static int
142 _nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i, bool include_admin)
143 {
144 	int rc = 0;
145 	struct spdk_nvmf_ctrlr *ctrlr;
146 	struct spdk_nvmf_qpair *qpair, *temp_qpair;
147 	struct spdk_io_channel *ch;
148 	struct spdk_nvmf_poll_group *group;
149 
150 	ctrlr = spdk_io_channel_iter_get_ctx(i);
151 	ch = spdk_io_channel_iter_get_channel(i);
152 	group = spdk_io_channel_get_ctx(ch);
153 
154 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, temp_qpair) {
155 		if (qpair->ctrlr == ctrlr && (include_admin || !nvmf_qpair_is_admin_queue(qpair))) {
156 			rc = spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
157 			if (rc) {
158 				SPDK_ERRLOG("Qpair disconnect failed\n");
159 				return rc;
160 			}
161 		}
162 	}
163 
164 	return rc;
165 }
166 
167 static void
168 nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i)
169 {
170 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, true));
171 }
172 
173 static void
174 nvmf_ctrlr_disconnect_io_qpairs_on_pg(struct spdk_io_channel_iter *i)
175 {
176 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, false));
177 }
178 
179 static int
180 nvmf_ctrlr_keep_alive_poll(void *ctx)
181 {
182 	uint64_t keep_alive_timeout_tick;
183 	uint64_t now = spdk_get_ticks();
184 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
185 
186 	if (ctrlr->in_destruct) {
187 		nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
188 		return SPDK_POLLER_IDLE;
189 	}
190 
191 	SPDK_DEBUGLOG(nvmf, "Polling ctrlr keep alive timeout\n");
192 
193 	/* If the Keep alive feature is in use and the timer expires */
194 	keep_alive_timeout_tick = ctrlr->last_keep_alive_tick +
195 				  ctrlr->feat.keep_alive_timer.bits.kato * spdk_get_ticks_hz() / UINT64_C(1000);
196 	if (now > keep_alive_timeout_tick) {
197 		SPDK_NOTICELOG("Disconnecting host %s from subsystem %s due to keep alive timeout.\n",
198 			       ctrlr->hostnqn, ctrlr->subsys->subnqn);
199 		/* set the Controller Fatal Status bit to '1' */
200 		if (ctrlr->vcprop.csts.bits.cfs == 0) {
201 			ctrlr->vcprop.csts.bits.cfs = 1;
202 
203 			/*
204 			 * disconnect qpairs, terminate Transport connection
205 			 * destroy ctrlr, break the host to controller association
206 			 * disconnect qpairs with qpair->ctrlr == ctrlr
207 			 */
208 			spdk_for_each_channel(ctrlr->subsys->tgt,
209 					      nvmf_ctrlr_disconnect_qpairs_on_pg,
210 					      ctrlr,
211 					      nvmf_ctrlr_disconnect_qpairs_done);
212 			return SPDK_POLLER_BUSY;
213 		}
214 	}
215 
216 	return SPDK_POLLER_IDLE;
217 }
218 
219 static void
220 nvmf_ctrlr_start_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
221 {
222 	if (!ctrlr) {
223 		SPDK_ERRLOG("Controller is NULL\n");
224 		return;
225 	}
226 
227 	/* if cleared to 0 then the Keep Alive Timer is disabled */
228 	if (ctrlr->feat.keep_alive_timer.bits.kato != 0) {
229 
230 		ctrlr->last_keep_alive_tick = spdk_get_ticks();
231 
232 		SPDK_DEBUGLOG(nvmf, "Ctrlr add keep alive poller\n");
233 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
234 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
235 	}
236 }
237 
238 static void
239 ctrlr_add_qpair_and_update_rsp(struct spdk_nvmf_qpair *qpair,
240 			       struct spdk_nvmf_ctrlr *ctrlr,
241 			       struct spdk_nvmf_fabric_connect_rsp *rsp)
242 {
243 	assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
244 
245 	/* check if we would exceed ctrlr connection limit */
246 	if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) {
247 		SPDK_ERRLOG("Requested QID %u but Max QID is %u\n",
248 			    qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1);
249 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
250 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
251 		return;
252 	}
253 
254 	if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) {
255 		SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid);
256 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
257 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
258 		return;
259 	}
260 
261 	qpair->ctrlr = ctrlr;
262 	spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid);
263 
264 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
265 	rsp->status_code_specific.success.cntlid = ctrlr->cntlid;
266 	SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n",
267 		      rsp->status_code_specific.success.cntlid);
268 
269 	SPDK_DTRACE_PROBE4(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn,
270 			   ctrlr->hostnqn);
271 }
272 
273 static void
274 _nvmf_ctrlr_add_admin_qpair(void *ctx)
275 {
276 	struct spdk_nvmf_request *req = ctx;
277 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
278 	struct spdk_nvmf_qpair *qpair = req->qpair;
279 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
280 
281 	ctrlr->admin_qpair = qpair;
282 	ctrlr->association_timeout = qpair->transport->opts.association_timeout;
283 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
284 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
285 	_nvmf_request_complete(req);
286 }
287 
288 static void
289 _nvmf_subsystem_add_ctrlr(void *ctx)
290 {
291 	struct spdk_nvmf_request *req = ctx;
292 	struct spdk_nvmf_qpair *qpair = req->qpair;
293 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
294 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
295 
296 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
297 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
298 		spdk_bit_array_free(&ctrlr->qpair_mask);
299 		free(ctrlr);
300 		qpair->ctrlr = NULL;
301 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
302 		spdk_nvmf_request_complete(req);
303 		return;
304 	}
305 
306 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
307 }
308 
309 static void
310 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
311 		      struct spdk_nvmf_ctrlr_data *cdata)
312 {
313 	cdata->aerl = NVMF_MAX_ASYNC_EVENTS - 1;
314 	cdata->kas = KAS_DEFAULT_VALUE;
315 	cdata->vid = SPDK_PCI_VID_INTEL;
316 	cdata->ssvid = SPDK_PCI_VID_INTEL;
317 	/* INTEL OUI */
318 	cdata->ieee[0] = 0xe4;
319 	cdata->ieee[1] = 0xd2;
320 	cdata->ieee[2] = 0x5c;
321 	cdata->oncs.reservations = 1;
322 	cdata->sgls.supported = 1;
323 	cdata->sgls.keyed_sgl = 1;
324 	cdata->sgls.sgl_offset = 1;
325 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
326 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
327 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
328 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
329 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
330 	cdata->nvmf_specific.msdbd = 1;
331 
332 	if (transport->ops->cdata_init) {
333 		transport->ops->cdata_init(transport, subsystem, cdata);
334 	}
335 }
336 
337 static struct spdk_nvmf_ctrlr *
338 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
339 		  struct spdk_nvmf_request *req,
340 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
341 		  struct spdk_nvmf_fabric_connect_data *connect_data)
342 {
343 	struct spdk_nvmf_ctrlr *ctrlr;
344 	struct spdk_nvmf_transport *transport = req->qpair->transport;
345 	struct spdk_nvme_transport_id listen_trid = {};
346 
347 	ctrlr = calloc(1, sizeof(*ctrlr));
348 	if (ctrlr == NULL) {
349 		SPDK_ERRLOG("Memory allocation failed\n");
350 		return NULL;
351 	}
352 
353 	if (spdk_nvme_trtype_is_fabrics(transport->ops->type)) {
354 		ctrlr->dynamic_ctrlr = true;
355 	} else {
356 		ctrlr->cntlid = connect_data->cntlid;
357 	}
358 
359 	SPDK_DTRACE_PROBE3(nvmf_ctrlr_create, ctrlr, subsystem->subnqn,
360 			   spdk_thread_get_id(req->qpair->group->thread));
361 
362 	STAILQ_INIT(&ctrlr->async_events);
363 	TAILQ_INIT(&ctrlr->log_head);
364 	ctrlr->subsys = subsystem;
365 	ctrlr->thread = req->qpair->group->thread;
366 	ctrlr->disconnect_in_progress = false;
367 
368 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
369 	if (!ctrlr->qpair_mask) {
370 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
371 		goto err_qpair_mask;
372 	}
373 
374 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
375 
376 	/*
377 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
378 	 * If this field is cleared to 0h, then Keep Alive is not supported.
379 	 */
380 	if (ctrlr->cdata.kas) {
381 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
382 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
383 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
384 	}
385 
386 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
387 	if (ctrlr->subsys->flags.ana_reporting) {
388 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
389 	}
390 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
391 	/* Coalescing Disable */
392 	ctrlr->feat.interrupt_vector_configuration.bits.cd = 1;
393 
394 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
395 		/*
396 		 * If keep-alive timeout is not set, discovery controllers use some
397 		 * arbitrary high value in order to cleanup stale discovery sessions
398 		 *
399 		 * From the 1.0a nvme-of spec:
400 		 * "The Keep Alive command is reserved for
401 		 * Discovery controllers. A transport may specify a
402 		 * fixed Discovery controller activity timeout value
403 		 * (e.g., 2 minutes). If no commands are received
404 		 * by a Discovery controller within that time
405 		 * period, the controller may perform the
406 		 * actions for Keep Alive Timer expiration".
407 		 *
408 		 * From the 1.1 nvme-of spec:
409 		 * "A host requests an explicit persistent connection
410 		 * to a Discovery controller and Asynchronous Event Notifications from
411 		 * the Discovery controller on that persistent connection by specifying
412 		 * a non-zero Keep Alive Timer value in the Connect command."
413 		 *
414 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
415 		 * otherwise we disable it and use default discovery controller KATO.
416 		 * KATO is in millisecond.
417 		 */
418 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
419 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
420 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
421 		} else {
422 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
423 		}
424 	}
425 
426 	/* Subtract 1 for admin queue, 1 for 0's based */
427 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
428 			1;
429 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
430 			1;
431 
432 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
433 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn));
434 
435 	ctrlr->vcprop.cap.raw = 0;
436 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
437 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
438 				      1; /* max queue depth */
439 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
440 	/* ready timeout - 500 msec units */
441 	ctrlr->vcprop.cap.bits.to = NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS / 500;
442 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
443 	ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
444 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
445 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
446 
447 	/* Version Supported: 1.3 */
448 	ctrlr->vcprop.vs.bits.mjr = 1;
449 	ctrlr->vcprop.vs.bits.mnr = 3;
450 	ctrlr->vcprop.vs.bits.ter = 0;
451 
452 	ctrlr->vcprop.cc.raw = 0;
453 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
454 
455 	ctrlr->vcprop.csts.raw = 0;
456 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
457 
458 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
459 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
460 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
461 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
462 
463 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
464 
465 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
466 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
467 			SPDK_ERRLOG("Could not get listener transport ID\n");
468 			goto err_listener;
469 		}
470 
471 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid);
472 		if (!ctrlr->listener) {
473 			SPDK_ERRLOG("Listener was not found\n");
474 			goto err_listener;
475 		}
476 	}
477 
478 	req->qpair->ctrlr = ctrlr;
479 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
480 
481 	return ctrlr;
482 err_listener:
483 	spdk_bit_array_free(&ctrlr->qpair_mask);
484 err_qpair_mask:
485 	free(ctrlr);
486 	return NULL;
487 }
488 
489 static void
490 _nvmf_ctrlr_destruct(void *ctx)
491 {
492 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
493 	struct spdk_nvmf_reservation_log *log, *log_tmp;
494 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
495 
496 	SPDK_DTRACE_PROBE3(nvmf_ctrlr_destruct, ctrlr, ctrlr->subsys->subnqn,
497 			   spdk_thread_get_id(ctrlr->thread));
498 
499 	assert(spdk_get_thread() == ctrlr->thread);
500 	assert(ctrlr->in_destruct);
501 
502 	SPDK_DEBUGLOG(nvmf, "Destroy ctrlr 0x%hx\n", ctrlr->cntlid);
503 	if (ctrlr->disconnect_in_progress) {
504 		SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n");
505 		spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
506 		return;
507 	}
508 
509 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
510 	nvmf_ctrlr_stop_association_timer(ctrlr);
511 	spdk_bit_array_free(&ctrlr->qpair_mask);
512 
513 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
514 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
515 		free(log);
516 	}
517 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
518 		STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link);
519 		free(event);
520 	}
521 	free(ctrlr);
522 }
523 
524 void
525 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
526 {
527 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
528 
529 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
530 }
531 
532 static void
533 nvmf_ctrlr_add_io_qpair(void *ctx)
534 {
535 	struct spdk_nvmf_request *req = ctx;
536 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
537 	struct spdk_nvmf_qpair *qpair = req->qpair;
538 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
539 	struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair;
540 
541 	SPDK_DTRACE_PROBE4(nvmf_ctrlr_add_io_qpair, ctrlr, req->qpair, req->qpair->qid,
542 			   spdk_thread_get_id(ctrlr->thread));
543 
544 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
545 	  * For error case, the value should be NULL. So set it to NULL at first.
546 	  */
547 	qpair->ctrlr = NULL;
548 
549 	/* Make sure the controller is not being destroyed. */
550 	if (ctrlr->in_destruct) {
551 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
552 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
553 		goto end;
554 	}
555 
556 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
557 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
558 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
559 		goto end;
560 	}
561 
562 	if (!ctrlr->vcprop.cc.bits.en) {
563 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
564 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
565 		goto end;
566 	}
567 
568 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
569 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
570 			    ctrlr->vcprop.cc.bits.iosqes);
571 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
572 		goto end;
573 	}
574 
575 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
576 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
577 			    ctrlr->vcprop.cc.bits.iocqes);
578 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
579 		goto end;
580 	}
581 
582 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
583 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
584 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
585 		 * state to DEACTIVATING and removing it from poll group */
586 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
587 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
588 		goto end;
589 	}
590 
591 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
592 end:
593 	spdk_nvmf_request_complete(req);
594 }
595 
596 static void
597 _nvmf_ctrlr_add_io_qpair(void *ctx)
598 {
599 	struct spdk_nvmf_request *req = ctx;
600 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
601 	struct spdk_nvmf_fabric_connect_data *data = req->data;
602 	struct spdk_nvmf_ctrlr *ctrlr;
603 	struct spdk_nvmf_qpair *qpair = req->qpair;
604 	struct spdk_nvmf_qpair *admin_qpair;
605 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
606 	struct spdk_nvmf_subsystem *subsystem;
607 	struct spdk_nvme_transport_id listen_trid = {};
608 	const struct spdk_nvmf_subsystem_listener *listener;
609 
610 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
611 
612 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
613 	/* We already checked this in spdk_nvmf_ctrlr_connect */
614 	assert(subsystem != NULL);
615 
616 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
617 	if (ctrlr == NULL) {
618 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
619 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
620 		spdk_nvmf_request_complete(req);
621 		return;
622 	}
623 
624 	/* fail before passing a message to the controller thread. */
625 	if (ctrlr->in_destruct) {
626 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
627 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
628 		spdk_nvmf_request_complete(req);
629 		return;
630 	}
631 
632 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
633 	if (subsystem->flags.ana_reporting) {
634 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
635 			SPDK_ERRLOG("Could not get listener transport ID\n");
636 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
637 			spdk_nvmf_request_complete(req);
638 			return;
639 		}
640 
641 		listener = nvmf_subsystem_find_listener(subsystem, &listen_trid);
642 		if (listener != ctrlr->listener) {
643 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
644 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
645 			spdk_nvmf_request_complete(req);
646 			return;
647 		}
648 	}
649 
650 	admin_qpair = ctrlr->admin_qpair;
651 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
652 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
653 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
654 		 * state to DEACTIVATING and removing it from poll group */
655 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
656 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
657 		spdk_nvmf_request_complete(req);
658 		return;
659 	}
660 	qpair->ctrlr = ctrlr;
661 	spdk_thread_send_msg(admin_qpair->group->thread, nvmf_ctrlr_add_io_qpair, req);
662 }
663 
664 static bool
665 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
666 			  const char *hostnqn)
667 {
668 	struct spdk_nvme_transport_id listen_trid = {};
669 
670 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
671 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
672 		return false;
673 	}
674 
675 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
676 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
677 			    subsystem->subnqn);
678 		return false;
679 	}
680 
681 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
682 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
683 			    subsystem->subnqn, hostnqn);
684 		return false;
685 	}
686 
687 	return true;
688 }
689 
690 static int
691 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
692 {
693 	struct spdk_nvmf_fabric_connect_data *data = req->data;
694 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
695 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
696 	struct spdk_nvmf_qpair *qpair = req->qpair;
697 	struct spdk_nvmf_transport *transport = qpair->transport;
698 	struct spdk_nvmf_ctrlr *ctrlr;
699 	struct spdk_nvmf_subsystem *subsystem;
700 
701 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
702 		      cmd->recfmt, cmd->qid, cmd->sqsize);
703 
704 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
705 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
706 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
707 		      ntohl(*(uint32_t *)&data->hostid[0]),
708 		      ntohs(*(uint16_t *)&data->hostid[4]),
709 		      ntohs(*(uint16_t *)&data->hostid[6]),
710 		      data->hostid[8],
711 		      data->hostid[9],
712 		      ntohs(*(uint16_t *)&data->hostid[10]),
713 		      ntohl(*(uint32_t *)&data->hostid[12]));
714 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
715 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
716 
717 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
718 	if (!subsystem) {
719 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
720 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
721 	}
722 
723 	if (cmd->recfmt != 0) {
724 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
725 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
726 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
727 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
728 	}
729 
730 	/*
731 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
732 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
733 	 */
734 	if (cmd->sqsize == 0) {
735 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
736 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
737 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
738 	}
739 
740 	if (cmd->qid == 0) {
741 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
742 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
743 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
744 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
745 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
746 		}
747 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
748 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
749 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
750 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
751 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
752 	}
753 
754 	qpair->sq_head_max = cmd->sqsize;
755 	qpair->qid = cmd->qid;
756 
757 	if (0 == qpair->qid) {
758 		qpair->group->stat.admin_qpairs++;
759 		qpair->group->stat.current_admin_qpairs++;
760 	} else {
761 		qpair->group->stat.io_qpairs++;
762 		qpair->group->stat.current_io_qpairs++;
763 	}
764 
765 	if (cmd->qid == 0) {
766 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
767 
768 		if (spdk_nvme_trtype_is_fabrics(transport->ops->type) && data->cntlid != 0xFFFF) {
769 			/* This NVMf target only supports dynamic mode. */
770 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
771 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
772 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
773 		}
774 
775 		/* Establish a new ctrlr */
776 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
777 		if (!ctrlr) {
778 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
779 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
780 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
781 		} else {
782 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
783 		}
784 	} else {
785 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
786 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
787 	}
788 }
789 
790 static inline bool
791 nvmf_request_is_fabric_connect(struct spdk_nvmf_request *req)
792 {
793 	return req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC &&
794 	       req->cmd->nvmf_cmd.fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT;
795 }
796 
797 static struct spdk_nvmf_subsystem_poll_group *
798 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
799 {
800 	struct spdk_nvmf_fabric_connect_data *data;
801 	struct spdk_nvmf_subsystem *subsystem;
802 	struct spdk_nvmf_tgt *tgt;
803 
804 	assert(nvmf_request_is_fabric_connect(req));
805 	assert(req->qpair->ctrlr == NULL);
806 
807 	data = req->data;
808 	tgt = req->qpair->transport->tgt;
809 
810 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
811 	if (subsystem == NULL) {
812 		return NULL;
813 	}
814 
815 	return &req->qpair->group->sgroups[subsystem->id];
816 }
817 
818 int
819 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
820 {
821 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
822 	struct spdk_nvmf_subsystem_poll_group *sgroup;
823 	struct spdk_nvmf_qpair *qpair = req->qpair;
824 	enum spdk_nvmf_request_exec_status status;
825 
826 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
827 	if (!sgroup) {
828 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
829 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
830 		goto out;
831 	}
832 
833 	sgroup->mgmt_io_outstanding++;
834 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
835 
836 	status = _nvmf_ctrlr_connect(req);
837 
838 out:
839 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
840 		_nvmf_request_complete(req);
841 	}
842 
843 	return status;
844 }
845 
846 static int nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req);
847 
848 static int
849 retry_connect(void *arg)
850 {
851 	struct spdk_nvmf_request *req = arg;
852 	struct spdk_nvmf_subsystem_poll_group *sgroup;
853 	int rc;
854 
855 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
856 	assert(sgroup != NULL);
857 	sgroup->mgmt_io_outstanding++;
858 	spdk_poller_unregister(&req->poller);
859 	rc = nvmf_ctrlr_cmd_connect(req);
860 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
861 		_nvmf_request_complete(req);
862 	}
863 	return SPDK_POLLER_BUSY;
864 }
865 
866 static int
867 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
868 {
869 	struct spdk_nvmf_fabric_connect_data *data = req->data;
870 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
871 	struct spdk_nvmf_transport *transport = req->qpair->transport;
872 	struct spdk_nvmf_subsystem *subsystem;
873 
874 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
875 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
876 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
877 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
878 	}
879 
880 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
881 	if (!subsystem) {
882 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
883 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
884 	}
885 
886 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
887 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
888 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
889 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
890 		struct spdk_nvmf_subsystem_poll_group *sgroup;
891 
892 		if (req->timeout_tsc == 0) {
893 			/* We will only retry the request up to 1 second. */
894 			req->timeout_tsc = spdk_get_ticks() + spdk_get_ticks_hz();
895 		} else if (spdk_get_ticks() > req->timeout_tsc) {
896 			SPDK_ERRLOG("Subsystem '%s' was not ready for 1 second\n", subsystem->subnqn);
897 			rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
898 			rsp->status.sc = SPDK_NVMF_FABRIC_SC_CONTROLLER_BUSY;
899 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
900 		}
901 
902 		/* Subsystem is not ready to handle a connect. Use a poller to retry it
903 		 * again later. Decrement the mgmt_io_outstanding to avoid the
904 		 * subsystem waiting for this command to complete before unpausing.
905 		 */
906 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
907 		assert(sgroup != NULL);
908 		sgroup->mgmt_io_outstanding--;
909 		SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn);
910 		req->poller = SPDK_POLLER_REGISTER(retry_connect, req, 100);
911 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
912 	}
913 
914 	/* Ensure that hostnqn is null terminated */
915 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
916 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
917 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
918 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
919 	}
920 
921 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
922 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
923 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
924 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
925 	}
926 
927 	return _nvmf_ctrlr_connect(req);
928 }
929 
930 static int
931 nvmf_ctrlr_association_remove(void *ctx)
932 {
933 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
934 	int rc;
935 
936 	nvmf_ctrlr_stop_association_timer(ctrlr);
937 
938 	if (ctrlr->in_destruct) {
939 		return SPDK_POLLER_IDLE;
940 	}
941 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
942 		      ctrlr->subsys->subnqn);
943 
944 	if (ctrlr->admin_qpair) {
945 		rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL);
946 		if (rc < 0) {
947 			SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
948 			assert(false);
949 		}
950 	}
951 
952 	return SPDK_POLLER_BUSY;
953 }
954 
955 static int
956 _nvmf_ctrlr_cc_reset_shn_done(void *ctx)
957 {
958 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
959 	uint64_t now = spdk_get_ticks();
960 	uint32_t count;
961 
962 	if (ctrlr->cc_timer) {
963 		spdk_poller_unregister(&ctrlr->cc_timer);
964 	}
965 
966 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
967 	SPDK_DEBUGLOG(nvmf, "ctrlr %p active queue count %u\n", ctrlr, count);
968 
969 	if (count > 1) {
970 		if (now < ctrlr->cc_timeout_tsc) {
971 			/* restart cc timer */
972 			ctrlr->cc_timer = SPDK_POLLER_REGISTER(_nvmf_ctrlr_cc_reset_shn_done, ctrlr, 100 * 1000);
973 			return SPDK_POLLER_IDLE;
974 		} else {
975 			/* controller fatal status */
976 			SPDK_WARNLOG("IO timeout, ctrlr %p is in fatal status\n", ctrlr);
977 			ctrlr->vcprop.csts.bits.cfs = 1;
978 		}
979 	}
980 
981 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
982 
983 	if (ctrlr->disconnect_is_shn) {
984 		ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
985 		ctrlr->disconnect_is_shn = false;
986 	} else {
987 		/* Only a subset of the registers are cleared out on a reset */
988 		ctrlr->vcprop.cc.raw = 0;
989 		ctrlr->vcprop.csts.raw = 0;
990 	}
991 
992 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
993 	 * between the host and controller shall be preserved for at least 2 minutes */
994 	if (ctrlr->association_timer) {
995 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
996 		nvmf_ctrlr_stop_association_timer(ctrlr);
997 	}
998 	if (ctrlr->association_timeout) {
999 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
1000 					   ctrlr->association_timeout * 1000);
1001 	}
1002 	ctrlr->disconnect_in_progress = false;
1003 	return SPDK_POLLER_BUSY;
1004 }
1005 
1006 static void
1007 nvmf_ctrlr_cc_reset_shn_done(struct spdk_io_channel_iter *i, int status)
1008 {
1009 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
1010 
1011 	if (status < 0) {
1012 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
1013 		assert(false);
1014 	}
1015 
1016 	_nvmf_ctrlr_cc_reset_shn_done((void *)ctrlr);
1017 }
1018 
1019 static void
1020 nvmf_bdev_complete_reset(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
1021 {
1022 	SPDK_NOTICELOG("Resetting bdev done with %s\n", success ? "success" : "failure");
1023 
1024 	spdk_bdev_free_io(bdev_io);
1025 }
1026 
1027 
1028 static int
1029 nvmf_ctrlr_cc_timeout(void *ctx)
1030 {
1031 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1032 	struct spdk_nvmf_poll_group *group = ctrlr->admin_qpair->group;
1033 	struct spdk_nvmf_ns *ns;
1034 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1035 
1036 	assert(group != NULL && group->sgroups != NULL);
1037 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
1038 	SPDK_DEBUGLOG(nvmf, "Ctrlr %p reset or shutdown timeout\n", ctrlr);
1039 
1040 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1041 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1042 		if (ns->bdev == NULL) {
1043 			continue;
1044 		}
1045 		ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[ns->opts.nsid - 1];
1046 		SPDK_NOTICELOG("Ctrlr %p resetting NSID %u\n", ctrlr, ns->opts.nsid);
1047 		spdk_bdev_reset(ns->desc, ns_info->channel, nvmf_bdev_complete_reset, NULL);
1048 	}
1049 
1050 	return SPDK_POLLER_BUSY;
1051 }
1052 
1053 const struct spdk_nvmf_registers *
1054 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
1055 {
1056 	return &ctrlr->vcprop;
1057 }
1058 
1059 static uint64_t
1060 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
1061 {
1062 	return ctrlr->vcprop.cap.raw;
1063 }
1064 
1065 static uint64_t
1066 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
1067 {
1068 	return ctrlr->vcprop.vs.raw;
1069 }
1070 
1071 static uint64_t
1072 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
1073 {
1074 	return ctrlr->vcprop.cc.raw;
1075 }
1076 
1077 static bool
1078 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1079 {
1080 	union spdk_nvme_cc_register cc, diff;
1081 	uint32_t cc_timeout_ms;
1082 
1083 	cc.raw = value;
1084 
1085 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
1086 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
1087 
1088 	/*
1089 	 * Calculate which bits changed between the current and new CC.
1090 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
1091 	 */
1092 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
1093 
1094 	if (diff.bits.en) {
1095 		if (cc.bits.en) {
1096 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
1097 			nvmf_ctrlr_stop_association_timer(ctrlr);
1098 
1099 			ctrlr->vcprop.cc.bits.en = 1;
1100 			ctrlr->vcprop.csts.bits.rdy = 1;
1101 		} else {
1102 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
1103 			if (ctrlr->disconnect_in_progress) {
1104 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1105 				return true;
1106 			}
1107 
1108 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1109 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1110 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1111 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1112 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1113 
1114 			ctrlr->vcprop.cc.bits.en = 0;
1115 			ctrlr->disconnect_in_progress = true;
1116 			ctrlr->disconnect_is_shn = false;
1117 			spdk_for_each_channel(ctrlr->subsys->tgt,
1118 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1119 					      ctrlr,
1120 					      nvmf_ctrlr_cc_reset_shn_done);
1121 		}
1122 		diff.bits.en = 0;
1123 	}
1124 
1125 	if (diff.bits.shn) {
1126 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
1127 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
1128 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
1129 				      cc.bits.shn >> 1, cc.bits.shn & 1);
1130 			if (ctrlr->disconnect_in_progress) {
1131 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1132 				return true;
1133 			}
1134 
1135 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1136 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1137 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1138 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1139 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1140 
1141 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
1142 			ctrlr->disconnect_in_progress = true;
1143 			ctrlr->disconnect_is_shn = true;
1144 			spdk_for_each_channel(ctrlr->subsys->tgt,
1145 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1146 					      ctrlr,
1147 					      nvmf_ctrlr_cc_reset_shn_done);
1148 
1149 			/* From the time a shutdown is initiated the controller shall disable
1150 			 * Keep Alive timer */
1151 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
1152 		} else if (cc.bits.shn == 0) {
1153 			ctrlr->vcprop.cc.bits.shn = 0;
1154 		} else {
1155 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
1156 				    cc.bits.shn >> 1, cc.bits.shn & 1);
1157 			return false;
1158 		}
1159 		diff.bits.shn = 0;
1160 	}
1161 
1162 	if (diff.bits.iosqes) {
1163 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
1164 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
1165 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
1166 		diff.bits.iosqes = 0;
1167 	}
1168 
1169 	if (diff.bits.iocqes) {
1170 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
1171 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
1172 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
1173 		diff.bits.iocqes = 0;
1174 	}
1175 
1176 	if (diff.bits.ams) {
1177 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
1178 		return false;
1179 	}
1180 
1181 	if (diff.bits.mps) {
1182 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
1183 		return false;
1184 	}
1185 
1186 	if (diff.bits.css) {
1187 		SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
1188 		return false;
1189 	}
1190 
1191 	if (diff.raw != 0) {
1192 		/* Print an error message, but don't fail the command in this case.
1193 		 * If we did want to fail in this case, we'd need to ensure we acted
1194 		 * on no other bits or the initiator gets confused. */
1195 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
1196 	}
1197 
1198 	return true;
1199 }
1200 
1201 static uint64_t
1202 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
1203 {
1204 	return ctrlr->vcprop.csts.raw;
1205 }
1206 
1207 static uint64_t
1208 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
1209 {
1210 	return ctrlr->vcprop.aqa.raw;
1211 }
1212 
1213 static bool
1214 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1215 {
1216 	union spdk_nvme_aqa_register aqa;
1217 
1218 	aqa.raw = value;
1219 
1220 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1221 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1222 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1223 		return false;
1224 	}
1225 
1226 	ctrlr->vcprop.aqa.raw = value;
1227 
1228 	return true;
1229 }
1230 
1231 static uint64_t
1232 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1233 {
1234 	return ctrlr->vcprop.asq;
1235 }
1236 
1237 static bool
1238 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1239 {
1240 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1241 
1242 	return true;
1243 }
1244 
1245 static bool
1246 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1247 {
1248 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1249 
1250 	return true;
1251 }
1252 
1253 static uint64_t
1254 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1255 {
1256 	return ctrlr->vcprop.acq;
1257 }
1258 
1259 static bool
1260 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1261 {
1262 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1263 
1264 	return true;
1265 }
1266 
1267 static bool
1268 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1269 {
1270 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1271 
1272 	return true;
1273 }
1274 
1275 struct nvmf_prop {
1276 	uint32_t ofst;
1277 	uint8_t size;
1278 	char name[11];
1279 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1280 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1281 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1282 };
1283 
1284 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1285 	{ \
1286 		offsetof(struct spdk_nvme_registers, field), \
1287 		size, \
1288 		#field, \
1289 		get_cb, set_cb, set_upper_cb \
1290 	}
1291 
1292 static const struct nvmf_prop nvmf_props[] = {
1293 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1294 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1295 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1296 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1297 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1298 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1299 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1300 };
1301 
1302 static const struct nvmf_prop *
1303 find_prop(uint32_t ofst, uint8_t size)
1304 {
1305 	size_t i;
1306 
1307 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1308 		const struct nvmf_prop *prop = &nvmf_props[i];
1309 
1310 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1311 			return prop;
1312 		}
1313 	}
1314 
1315 	return NULL;
1316 }
1317 
1318 static int
1319 nvmf_property_get(struct spdk_nvmf_request *req)
1320 {
1321 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1322 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1323 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1324 	const struct nvmf_prop *prop;
1325 	uint8_t size;
1326 
1327 	response->status.sc = 0;
1328 	response->value.u64 = 0;
1329 
1330 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1331 		      cmd->attrib.size, cmd->ofst);
1332 
1333 	switch (cmd->attrib.size) {
1334 	case SPDK_NVMF_PROP_SIZE_4:
1335 		size = 4;
1336 		break;
1337 	case SPDK_NVMF_PROP_SIZE_8:
1338 		size = 8;
1339 		break;
1340 	default:
1341 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1342 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1343 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1344 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1345 	}
1346 
1347 	prop = find_prop(cmd->ofst, size);
1348 	if (prop == NULL || prop->get_cb == NULL) {
1349 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1350 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1351 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1352 	}
1353 
1354 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1355 
1356 	response->value.u64 = prop->get_cb(ctrlr);
1357 
1358 	if (size != prop->size) {
1359 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1360 		assert(size == 4);
1361 		assert(prop->size == 8);
1362 
1363 		if (cmd->ofst == prop->ofst) {
1364 			/* Keep bottom 4 bytes only */
1365 			response->value.u64 &= 0xFFFFFFFF;
1366 		} else {
1367 			/* Keep top 4 bytes only */
1368 			response->value.u64 >>= 32;
1369 		}
1370 	}
1371 
1372 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1373 
1374 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1375 }
1376 
1377 static int
1378 nvmf_property_set(struct spdk_nvmf_request *req)
1379 {
1380 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1381 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1382 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1383 	const struct nvmf_prop *prop;
1384 	uint64_t value;
1385 	uint8_t size;
1386 	bool ret;
1387 
1388 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1389 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1390 
1391 	switch (cmd->attrib.size) {
1392 	case SPDK_NVMF_PROP_SIZE_4:
1393 		size = 4;
1394 		break;
1395 	case SPDK_NVMF_PROP_SIZE_8:
1396 		size = 8;
1397 		break;
1398 	default:
1399 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1400 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1401 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1402 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1403 	}
1404 
1405 	prop = find_prop(cmd->ofst, size);
1406 	if (prop == NULL || prop->set_cb == NULL) {
1407 		SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst);
1408 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1409 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1410 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1411 	}
1412 
1413 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1414 
1415 	value = cmd->value.u64;
1416 
1417 	if (prop->size == 4) {
1418 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1419 	} else if (size != prop->size) {
1420 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1421 		assert(size == 4);
1422 		assert(prop->size == 8);
1423 
1424 		if (cmd->ofst == prop->ofst) {
1425 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1426 		} else {
1427 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1428 		}
1429 	} else {
1430 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1431 		if (ret) {
1432 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1433 		}
1434 	}
1435 
1436 	if (!ret) {
1437 		SPDK_ERRLOG("prop set_cb failed\n");
1438 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1439 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1440 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1441 	}
1442 
1443 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1444 }
1445 
1446 static int
1447 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1448 {
1449 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1450 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1451 
1452 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1453 
1454 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1455 	ctrlr->feat.arbitration.bits.reserved = 0;
1456 
1457 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1458 }
1459 
1460 static int
1461 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1462 {
1463 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1464 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1465 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1466 
1467 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1468 
1469 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1470 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1471 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1472 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1473 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1474 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1475 	}
1476 
1477 	ctrlr->feat.power_management.raw = cmd->cdw11;
1478 	ctrlr->feat.power_management.bits.reserved = 0;
1479 
1480 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1481 }
1482 
1483 static bool
1484 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1485 {
1486 	/*
1487 	 * Valid TMPSEL values:
1488 	 *  0000b - 1000b: temperature sensors
1489 	 *  1111b: set all implemented temperature sensors
1490 	 */
1491 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1492 		/* 1001b - 1110b: reserved */
1493 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1494 		return false;
1495 	}
1496 
1497 	/*
1498 	 * Valid THSEL values:
1499 	 *  00b: over temperature threshold
1500 	 *  01b: under temperature threshold
1501 	 */
1502 	if (opts->bits.thsel > 1) {
1503 		/* 10b - 11b: reserved */
1504 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1505 		return false;
1506 	}
1507 
1508 	return true;
1509 }
1510 
1511 static int
1512 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1513 {
1514 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1515 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1516 
1517 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1518 
1519 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1520 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1521 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1522 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1523 	}
1524 
1525 	/* TODO: no sensors implemented - ignore new values */
1526 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1527 }
1528 
1529 static int
1530 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1531 {
1532 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1533 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1534 
1535 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1536 
1537 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1538 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1539 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1540 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1541 	}
1542 
1543 	/* TODO: no sensors implemented - return 0 for all thresholds */
1544 	rsp->cdw0 = 0;
1545 
1546 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1547 }
1548 
1549 static int
1550 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req)
1551 {
1552 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1553 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1554 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1555 	union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {};
1556 
1557 	SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11);
1558 
1559 	iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv;
1560 	iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd;
1561 	rsp->cdw0 = iv_conf.raw;
1562 
1563 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1564 }
1565 
1566 static int
1567 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1568 {
1569 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1570 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1571 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1572 
1573 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1574 
1575 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1576 		/*
1577 		 * Host is not allowed to set this bit, since we don't advertise it in
1578 		 * Identify Namespace.
1579 		 */
1580 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1581 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1582 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1583 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1584 	}
1585 
1586 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1587 	ctrlr->feat.error_recovery.bits.reserved = 0;
1588 
1589 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1590 }
1591 
1592 static int
1593 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1594 {
1595 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1596 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1597 
1598 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1599 
1600 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1601 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1602 
1603 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1604 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1605 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1606 }
1607 
1608 static int
1609 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1610 {
1611 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1612 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1613 
1614 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1615 
1616 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1617 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1618 
1619 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1620 }
1621 
1622 static int
1623 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1624 {
1625 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1626 
1627 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1628 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1629 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1630 }
1631 
1632 static int
1633 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1634 {
1635 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1636 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1637 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1638 
1639 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1640 
1641 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1642 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1643 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1644 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1645 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1646 	}
1647 
1648 	if (req->data == NULL || req->length < sizeof(ctrlr->hostid)) {
1649 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1650 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1651 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1652 	}
1653 
1654 	spdk_uuid_copy((struct spdk_uuid *)req->data, &ctrlr->hostid);
1655 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1656 }
1657 
1658 static int
1659 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1660 {
1661 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1662 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1663 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1664 	struct spdk_nvmf_ns *ns;
1665 
1666 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notification Mask\n");
1667 
1668 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1669 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1670 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1671 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1672 	}
1673 
1674 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1675 	if (ns == NULL) {
1676 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1677 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1678 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1679 	}
1680 	rsp->cdw0 = ns->mask;
1681 
1682 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1683 }
1684 
1685 static int
1686 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1687 {
1688 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1689 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1690 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1691 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1692 	struct spdk_nvmf_ns *ns;
1693 
1694 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notification Mask\n");
1695 
1696 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1697 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1698 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1699 			ns->mask = cmd->cdw11;
1700 		}
1701 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1702 	}
1703 
1704 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1705 	if (ns == NULL) {
1706 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1707 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1708 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1709 	}
1710 	ns->mask = cmd->cdw11;
1711 
1712 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1713 }
1714 
1715 static int
1716 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1717 {
1718 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1719 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1720 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1721 	struct spdk_nvmf_ns *ns;
1722 
1723 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1724 
1725 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1726 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1727 	if (ns == NULL) {
1728 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1729 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1730 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1731 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1732 	}
1733 
1734 	response->cdw0 = ns->ptpl_activated;
1735 
1736 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1737 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1738 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1739 }
1740 
1741 static int
1742 nvmf_ctrlr_set_features_reservation_persistence(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 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1747 	struct spdk_nvmf_ns *ns;
1748 	bool ptpl;
1749 
1750 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1751 
1752 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1753 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1754 
1755 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && ns->ptpl_file) {
1756 		ns->ptpl_activated = ptpl;
1757 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1758 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns && ns->ptpl_file;
1759 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1760 			ns->ptpl_activated = ptpl;
1761 		}
1762 	} else {
1763 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1764 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1765 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1766 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1767 	}
1768 
1769 	/* TODO: Feature not changeable for now */
1770 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1771 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1772 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1773 }
1774 
1775 static int
1776 nvmf_ctrlr_get_features_host_behavior_support(struct spdk_nvmf_request *req)
1777 {
1778 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1779 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1780 	struct spdk_nvme_host_behavior host_behavior = {};
1781 
1782 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Behavior Support\n");
1783 
1784 	if (req->data == NULL || req->length < sizeof(struct spdk_nvme_host_behavior)) {
1785 		SPDK_ERRLOG("invalid data buffer for Host Behavior Support\n");
1786 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1787 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1788 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1789 	}
1790 
1791 	host_behavior.acre = ctrlr->acre_enabled;
1792 	memcpy(req->data, &host_behavior, sizeof(host_behavior));
1793 
1794 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1795 }
1796 
1797 static int
1798 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req)
1799 {
1800 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1801 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1802 	struct spdk_nvme_host_behavior *host_behavior;
1803 
1804 	SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n");
1805 	if (req->iovcnt != 1) {
1806 		SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt);
1807 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1808 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1809 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1810 	}
1811 	if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) {
1812 		SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len);
1813 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1814 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1815 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1816 	}
1817 
1818 	host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base;
1819 	if (host_behavior->acre == 0) {
1820 		ctrlr->acre_enabled = false;
1821 	} else if (host_behavior->acre == 1) {
1822 		ctrlr->acre_enabled = true;
1823 	} else {
1824 		SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre);
1825 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1826 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1827 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1828 	}
1829 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1830 }
1831 
1832 static int
1833 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
1834 {
1835 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1836 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1837 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1838 
1839 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
1840 
1841 	/*
1842 	 * if attempts to disable keep alive by setting kato to 0h
1843 	 * a status value of keep alive invalid shall be returned
1844 	 */
1845 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
1846 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
1847 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
1848 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
1849 	} else {
1850 		/* round up to milliseconds */
1851 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
1852 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
1853 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
1854 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
1855 	}
1856 
1857 	/*
1858 	 * if change the keep alive timeout value successfully
1859 	 * update the keep alive poller.
1860 	 */
1861 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
1862 		if (ctrlr->keep_alive_poller != NULL) {
1863 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
1864 		}
1865 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
1866 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
1867 	}
1868 
1869 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
1870 		      ctrlr->feat.keep_alive_timer.bits.kato);
1871 
1872 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1873 }
1874 
1875 static int
1876 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
1877 {
1878 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1879 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1880 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1881 	uint32_t count;
1882 
1883 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
1884 		      req->cmd->nvme_cmd.cdw11);
1885 
1886 	if (cmd->cdw11_bits.feat_num_of_queues.bits.ncqr == UINT16_MAX ||
1887 	    cmd->cdw11_bits.feat_num_of_queues.bits.nsqr == UINT16_MAX) {
1888 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1889 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1890 	}
1891 
1892 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1893 	/* verify that the controller is ready to process commands */
1894 	if (count > 1) {
1895 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
1896 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1897 	} else {
1898 		/*
1899 		 * Ignore the value requested by the host -
1900 		 * always return the pre-configured value based on max_qpairs_allowed.
1901 		 */
1902 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
1903 	}
1904 
1905 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1906 }
1907 
1908 int
1909 nvmf_ctrlr_save_aers(struct spdk_nvmf_ctrlr *ctrlr, uint16_t *aer_cids,
1910 		     uint16_t max_aers)
1911 {
1912 	struct spdk_nvmf_request *req;
1913 	uint16_t i;
1914 
1915 	if (!aer_cids || max_aers < ctrlr->nr_aer_reqs) {
1916 		return -EINVAL;
1917 	}
1918 
1919 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
1920 		req = ctrlr->aer_req[i];
1921 		aer_cids[i] = req->cmd->nvme_cmd.cid;
1922 	}
1923 
1924 	return ctrlr->nr_aer_reqs;
1925 }
1926 
1927 int
1928 nvmf_ctrlr_save_migr_data(struct spdk_nvmf_ctrlr *ctrlr, struct nvmf_ctrlr_migr_data *data)
1929 {
1930 	uint32_t num_async_events = 0;
1931 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
1932 
1933 	memcpy(&data->feat, &ctrlr->feat, sizeof(struct spdk_nvmf_ctrlr_feat));
1934 	data->cntlid = ctrlr->cntlid;
1935 	data->acre_enabled = ctrlr->acre_enabled;
1936 	data->notice_aen_mask = ctrlr->notice_aen_mask;
1937 
1938 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
1939 		data->async_events[num_async_events++].raw = event->event.raw;
1940 		if (num_async_events == NVMF_MIGR_MAX_PENDING_AERS) {
1941 			SPDK_ERRLOG("%p has too many pending AERs\n", ctrlr);
1942 			break;
1943 		}
1944 	}
1945 	data->num_async_events = num_async_events;
1946 
1947 	return 0;
1948 }
1949 
1950 int
1951 nvmf_ctrlr_restore_migr_data(struct spdk_nvmf_ctrlr *ctrlr, struct nvmf_ctrlr_migr_data *data)
1952 {
1953 	struct spdk_nvmf_async_event_completion *event;
1954 	uint32_t i;
1955 
1956 	memcpy(&ctrlr->feat, &data->feat, sizeof(struct spdk_nvmf_ctrlr_feat));
1957 	ctrlr->acre_enabled = data->acre_enabled;
1958 	ctrlr->notice_aen_mask = data->notice_aen_mask;
1959 
1960 	for (i = 0; i < data->num_async_events; i++) {
1961 		event = calloc(1, sizeof(struct spdk_nvmf_async_event_completion));
1962 		if (!event) {
1963 			return -ENOMEM;
1964 		}
1965 		event->event.raw = data->async_events[i].raw;
1966 		STAILQ_INSERT_TAIL(&ctrlr->async_events, event, link);
1967 	}
1968 
1969 	return 0;
1970 }
1971 
1972 static int
1973 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
1974 {
1975 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1976 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1977 
1978 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
1979 		      cmd->cdw11);
1980 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
1981 	ctrlr->feat.async_event_configuration.bits.reserved1 = 0;
1982 	ctrlr->feat.async_event_configuration.bits.reserved2 = 0;
1983 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1984 }
1985 
1986 static int
1987 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
1988 {
1989 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1990 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1991 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1992 	struct spdk_nvmf_async_event_completion *pending_event;
1993 
1994 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
1995 
1996 	/* AER cmd is an exception */
1997 	sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
1998 	assert(sgroup != NULL);
1999 	sgroup->mgmt_io_outstanding--;
2000 
2001 	/* Four asynchronous events are supported for now */
2002 	if (ctrlr->nr_aer_reqs >= NVMF_MAX_ASYNC_EVENTS) {
2003 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
2004 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
2005 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
2006 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2007 	}
2008 
2009 	if (!STAILQ_EMPTY(&ctrlr->async_events)) {
2010 		pending_event = STAILQ_FIRST(&ctrlr->async_events);
2011 		rsp->cdw0 = pending_event->event.raw;
2012 		STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link);
2013 		free(pending_event);
2014 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2015 	}
2016 
2017 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
2018 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2019 }
2020 
2021 struct copy_iovs_ctx {
2022 	struct iovec *iovs;
2023 	int iovcnt;
2024 	int cur_iov_idx;
2025 	size_t cur_iov_offset;
2026 };
2027 
2028 static void
2029 _init_copy_iovs_ctx(struct copy_iovs_ctx *copy_ctx, struct iovec *iovs, int iovcnt)
2030 {
2031 	int iov_idx = 0;
2032 	struct iovec *iov;
2033 
2034 	copy_ctx->iovs = iovs;
2035 	copy_ctx->iovcnt = iovcnt;
2036 	copy_ctx->cur_iov_idx = 0;
2037 	copy_ctx->cur_iov_offset = 0;
2038 
2039 	while (iov_idx < copy_ctx->iovcnt) {
2040 		iov = &copy_ctx->iovs[iov_idx];
2041 		memset(iov->iov_base, 0, iov->iov_len);
2042 		iov_idx++;
2043 	}
2044 }
2045 
2046 static size_t
2047 _copy_buf_to_iovs(struct copy_iovs_ctx *copy_ctx, const void *buf, size_t buf_len)
2048 {
2049 	size_t len, iov_remain_len, copied_len = 0;
2050 	struct iovec *iov;
2051 
2052 	if (buf_len == 0) {
2053 		return 0;
2054 	}
2055 
2056 	while (copy_ctx->cur_iov_idx < copy_ctx->iovcnt) {
2057 		iov = &copy_ctx->iovs[copy_ctx->cur_iov_idx];
2058 		iov_remain_len = iov->iov_len - copy_ctx->cur_iov_offset;
2059 		if (iov_remain_len == 0) {
2060 			copy_ctx->cur_iov_idx++;
2061 			copy_ctx->cur_iov_offset = 0;
2062 			continue;
2063 		}
2064 
2065 		len = spdk_min(iov_remain_len, buf_len - copied_len);
2066 		memcpy((char *)iov->iov_base + copy_ctx->cur_iov_offset,
2067 		       (const char *)buf + copied_len,
2068 		       len);
2069 		copied_len += len;
2070 		copy_ctx->cur_iov_offset += len;
2071 
2072 		if (buf_len == copied_len) {
2073 			return copied_len;
2074 		}
2075 	}
2076 
2077 	return copied_len;
2078 }
2079 
2080 static void
2081 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length)
2082 {
2083 	struct spdk_nvme_firmware_page fw_page;
2084 	size_t copy_len;
2085 	struct copy_iovs_ctx copy_ctx;
2086 
2087 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2088 
2089 	memset(&fw_page, 0, sizeof(fw_page));
2090 	fw_page.afi.active_slot = 1;
2091 	fw_page.afi.next_reset_slot = 0;
2092 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
2093 
2094 	if (offset < sizeof(fw_page)) {
2095 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
2096 		if (copy_len > 0) {
2097 			_copy_buf_to_iovs(&copy_ctx, (const char *)&fw_page + offset, copy_len);
2098 		}
2099 	}
2100 }
2101 
2102 /*
2103  * Asynchronous Event Mask Bit
2104  */
2105 enum spdk_nvme_async_event_mask_bit {
2106 	/* Mask Namespace Change Notification */
2107 	SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT		= 0,
2108 	/* Mask Asymmetric Namespace Access Change Notification */
2109 	SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT		= 1,
2110 	/* Mask Discovery Log Change Notification */
2111 	SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT	= 2,
2112 	/* Mask Reservation Log Page Available Notification */
2113 	SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT	= 3,
2114 	/* Mask Error Event */
2115 	SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT			= 4,
2116 	/* 4 - 63 Reserved */
2117 };
2118 
2119 static inline void
2120 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2121 		      enum spdk_nvme_async_event_mask_bit mask)
2122 {
2123 	ctrlr->notice_aen_mask &= ~(1 << mask);
2124 }
2125 
2126 static inline bool
2127 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2128 		    enum spdk_nvme_async_event_mask_bit mask)
2129 {
2130 	if (ctrlr->notice_aen_mask & (1 << mask)) {
2131 		return false;
2132 	} else {
2133 		ctrlr->notice_aen_mask |= (1 << mask);
2134 		return true;
2135 	}
2136 }
2137 
2138 /* we have to use the typedef in the function declaration to appease astyle. */
2139 typedef enum spdk_nvme_ana_state spdk_nvme_ana_state_t;
2140 
2141 static inline spdk_nvme_ana_state_t
2142 nvmf_ctrlr_get_ana_state(struct spdk_nvmf_ctrlr *ctrlr, uint32_t anagrpid)
2143 {
2144 	if (!ctrlr->subsys->flags.ana_reporting) {
2145 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2146 	}
2147 
2148 	if (spdk_unlikely(ctrlr->listener == NULL)) {
2149 		return SPDK_NVME_ANA_INACCESSIBLE_STATE;
2150 	}
2151 
2152 	assert(anagrpid - 1 < ctrlr->subsys->max_nsid);
2153 	return ctrlr->listener->ana_state[anagrpid - 1];
2154 }
2155 
2156 static spdk_nvme_ana_state_t
2157 nvmf_ctrlr_get_ana_state_from_nsid(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2158 {
2159 	struct spdk_nvmf_ns *ns;
2160 
2161 	/* We do not have NVM subsystem specific ANA state. Hence if NSID is either
2162 	 * SPDK_NVMF_GLOBAL_NS_TAG, invalid, or for inactive namespace, return
2163 	 * the optimized state.
2164 	 */
2165 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
2166 	if (ns == NULL) {
2167 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2168 	}
2169 
2170 	return nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2171 }
2172 
2173 static void
2174 nvmf_get_error_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2175 			uint64_t offset, uint32_t length, uint32_t rae)
2176 {
2177 	if (!rae) {
2178 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT);
2179 	}
2180 
2181 	/* TODO: actually fill out log page data */
2182 }
2183 
2184 static void
2185 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2186 		      uint64_t offset, uint32_t length, uint32_t rae)
2187 {
2188 	struct spdk_nvme_ana_page ana_hdr;
2189 	struct spdk_nvme_ana_group_descriptor ana_desc;
2190 	size_t copy_len, copied_len;
2191 	uint32_t num_anagrp = 0, anagrpid;
2192 	struct spdk_nvmf_ns *ns;
2193 	struct copy_iovs_ctx copy_ctx;
2194 
2195 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2196 
2197 	if (length == 0) {
2198 		goto done;
2199 	}
2200 
2201 	if (offset >= sizeof(ana_hdr)) {
2202 		offset -= sizeof(ana_hdr);
2203 	} else {
2204 		for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2205 			if (ctrlr->subsys->ana_group[anagrpid - 1] > 0) {
2206 				num_anagrp++;
2207 			}
2208 		}
2209 
2210 		memset(&ana_hdr, 0, sizeof(ana_hdr));
2211 
2212 		ana_hdr.num_ana_group_desc = num_anagrp;
2213 		/* TODO: Support Change Count. */
2214 		ana_hdr.change_count = 0;
2215 
2216 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
2217 		copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ana_hdr + offset, copy_len);
2218 		assert(copied_len == copy_len);
2219 		length -= copied_len;
2220 		offset = 0;
2221 	}
2222 
2223 	if (length == 0) {
2224 		goto done;
2225 	}
2226 
2227 	for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2228 		if (ctrlr->subsys->ana_group[anagrpid - 1] == 0) {
2229 			continue;
2230 		}
2231 
2232 		if (offset >= sizeof(ana_desc)) {
2233 			offset -= sizeof(ana_desc);
2234 		} else {
2235 			memset(&ana_desc, 0, sizeof(ana_desc));
2236 
2237 			ana_desc.ana_group_id = anagrpid;
2238 			ana_desc.num_of_nsid = ctrlr->subsys->ana_group[anagrpid - 1];
2239 			ana_desc.ana_state = nvmf_ctrlr_get_ana_state(ctrlr, anagrpid);
2240 
2241 			copy_len = spdk_min(sizeof(ana_desc) - offset, length);
2242 			copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ana_desc + offset,
2243 						       copy_len);
2244 			assert(copied_len == copy_len);
2245 			length -= copied_len;
2246 			offset = 0;
2247 
2248 			if (length == 0) {
2249 				goto done;
2250 			}
2251 		}
2252 
2253 		/* TODO: Revisit here about O(n^2) cost if we have subsystem with
2254 		 * many namespaces in the future.
2255 		 */
2256 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
2257 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
2258 			if (ns->anagrpid != anagrpid) {
2259 				continue;
2260 			}
2261 
2262 			if (offset >= sizeof(uint32_t)) {
2263 				offset -= sizeof(uint32_t);
2264 				continue;
2265 			}
2266 
2267 			copy_len = spdk_min(sizeof(uint32_t) - offset, length);
2268 			copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ns->nsid + offset,
2269 						       copy_len);
2270 			assert(copied_len == copy_len);
2271 			length -= copied_len;
2272 			offset = 0;
2273 
2274 			if (length == 0) {
2275 				goto done;
2276 			}
2277 		}
2278 	}
2279 
2280 done:
2281 	if (!rae) {
2282 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT);
2283 	}
2284 }
2285 
2286 void
2287 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2288 {
2289 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
2290 	uint16_t i;
2291 	bool found = false;
2292 
2293 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
2294 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
2295 			/* nsid is already in the list */
2296 			found = true;
2297 			break;
2298 		}
2299 	}
2300 
2301 	if (!found) {
2302 		if (ctrlr->changed_ns_list_count == max_changes) {
2303 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
2304 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
2305 			for (i = 1; i < max_changes; i++) {
2306 				ctrlr->changed_ns_list.ns_list[i] = 0;
2307 			}
2308 		} else {
2309 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
2310 		}
2311 	}
2312 }
2313 
2314 static void
2315 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2316 				  struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2317 {
2318 	size_t copy_length;
2319 	struct copy_iovs_ctx copy_ctx;
2320 
2321 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2322 
2323 	if (offset < sizeof(ctrlr->changed_ns_list)) {
2324 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
2325 		if (copy_length) {
2326 			_copy_buf_to_iovs(&copy_ctx, (char *)&ctrlr->changed_ns_list + offset, copy_length);
2327 		}
2328 	}
2329 
2330 	/* Clear log page each time it is read */
2331 	ctrlr->changed_ns_list_count = 0;
2332 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
2333 
2334 	if (!rae) {
2335 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT);
2336 	}
2337 }
2338 
2339 /* The structure can be modified if we provide support for other commands in future */
2340 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
2341 	.admin_cmds_supported = {
2342 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
2343 		/* Get Log Page */
2344 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2345 		/* Identify */
2346 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
2347 		/* Abort */
2348 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
2349 		/* Set Features */
2350 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2351 		/* Get Features */
2352 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2353 		/* Async Event Request */
2354 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
2355 		/* Keep Alive */
2356 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2357 	},
2358 	.io_cmds_supported = {
2359 		/* FLUSH */
2360 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
2361 		/* WRITE */
2362 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
2363 		/* READ */
2364 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
2365 		/* WRITE ZEROES */
2366 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
2367 		/* DATASET MANAGEMENT */
2368 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
2369 		/* COMPARE */
2370 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
2371 	},
2372 };
2373 
2374 static void
2375 nvmf_get_cmds_and_effects_log_page(struct iovec *iovs, int iovcnt,
2376 				   uint64_t offset, uint32_t length)
2377 {
2378 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
2379 	size_t copy_len = 0;
2380 	struct copy_iovs_ctx copy_ctx;
2381 
2382 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2383 
2384 	if (offset < page_size) {
2385 		copy_len = spdk_min(page_size - offset, length);
2386 		_copy_buf_to_iovs(&copy_ctx, (char *)(&g_cmds_and_effect_log_page) + offset, copy_len);
2387 	}
2388 }
2389 
2390 static void
2391 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2392 		struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2393 {
2394 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
2395 	struct spdk_nvmf_reservation_log *log, *log_tmp;
2396 	struct copy_iovs_ctx copy_ctx;
2397 
2398 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2399 
2400 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
2401 	/* No available log, return zeroed log pages */
2402 	if (!ctrlr->num_avail_log_pages) {
2403 		return;
2404 	}
2405 
2406 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
2407 	if (offset >= avail_log_len) {
2408 		return;
2409 	}
2410 
2411 	next_pos = 0;
2412 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
2413 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
2414 		ctrlr->num_avail_log_pages--;
2415 
2416 		next_pos += unit_log_len;
2417 		if (next_pos > offset) {
2418 			copy_len = spdk_min(next_pos - offset, length);
2419 			_copy_buf_to_iovs(&copy_ctx, &log->log, copy_len);
2420 			length -= copy_len;
2421 			offset += copy_len;
2422 		}
2423 		free(log);
2424 
2425 		if (length == 0) {
2426 			break;
2427 		}
2428 	}
2429 
2430 	if (!rae) {
2431 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT);
2432 	}
2433 	return;
2434 }
2435 
2436 static int
2437 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
2438 {
2439 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2440 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2441 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2442 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2443 	struct spdk_nvme_transport_id cmd_source_trid;
2444 	uint64_t offset, len;
2445 	uint32_t rae, numdl, numdu;
2446 	uint8_t lid;
2447 
2448 	if (req->data == NULL) {
2449 		SPDK_DEBUGLOG(nvmf, "get log command with no buffer\n");
2450 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2451 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2452 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2453 	}
2454 
2455 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
2456 	if (offset & 3) {
2457 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
2458 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2459 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2460 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2461 	}
2462 
2463 	rae = cmd->cdw10_bits.get_log_page.rae;
2464 	numdl = cmd->cdw10_bits.get_log_page.numdl;
2465 	numdu = cmd->cdw11_bits.get_log_page.numdu;
2466 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
2467 	if (len > req->length) {
2468 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
2469 			    len, req->length);
2470 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2471 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2472 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2473 	}
2474 
2475 	lid = cmd->cdw10_bits.get_log_page.lid;
2476 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n",
2477 		      lid, offset, len, rae);
2478 
2479 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2480 		switch (lid) {
2481 		case SPDK_NVME_LOG_DISCOVERY:
2482 			if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &cmd_source_trid)) {
2483 				SPDK_ERRLOG("Failed to get LOG_DISCOVERY source trid\n");
2484 				response->status.sct = SPDK_NVME_SCT_GENERIC;
2485 				response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
2486 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2487 			}
2488 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt,
2489 						    offset, len, &cmd_source_trid);
2490 			if (!rae) {
2491 				nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT);
2492 			}
2493 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2494 		default:
2495 			goto invalid_log_page;
2496 		}
2497 	} else {
2498 		if (offset > len) {
2499 			SPDK_ERRLOG("Get log page: offset (%" PRIu64 ") > len (%" PRIu64 ")\n",
2500 				    offset, len);
2501 			response->status.sct = SPDK_NVME_SCT_GENERIC;
2502 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2503 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2504 		}
2505 
2506 		switch (lid) {
2507 		case SPDK_NVME_LOG_ERROR:
2508 			nvmf_get_error_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2509 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2510 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
2511 			/* TODO: actually fill out log page data */
2512 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2513 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
2514 			nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len);
2515 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2516 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
2517 			if (subsystem->flags.ana_reporting) {
2518 				nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2519 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2520 			} else {
2521 				goto invalid_log_page;
2522 			}
2523 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
2524 			nvmf_get_cmds_and_effects_log_page(req->iov, req->iovcnt, offset, len);
2525 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2526 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
2527 			nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2528 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2529 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
2530 			nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2531 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2532 		default:
2533 			goto invalid_log_page;
2534 		}
2535 	}
2536 
2537 invalid_log_page:
2538 	SPDK_INFOLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid);
2539 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2540 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2541 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2542 }
2543 
2544 int
2545 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2546 			    struct spdk_nvme_cmd *cmd,
2547 			    struct spdk_nvme_cpl *rsp,
2548 			    struct spdk_nvme_ns_data *nsdata)
2549 {
2550 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2551 	struct spdk_nvmf_ns *ns;
2552 	uint32_t max_num_blocks;
2553 	enum spdk_nvme_ana_state ana_state;
2554 
2555 	if (cmd->nsid == 0 || cmd->nsid > subsystem->max_nsid) {
2556 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", cmd->nsid);
2557 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2558 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2559 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2560 	}
2561 
2562 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2563 	if (ns == NULL || ns->bdev == NULL) {
2564 		/*
2565 		 * Inactive namespaces should return a zero filled data structure.
2566 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2567 		 * so we can just return early here.
2568 		 */
2569 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", cmd->nsid);
2570 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2571 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2572 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2573 	}
2574 
2575 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2576 
2577 	assert(ctrlr->admin_qpair);
2578 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2579 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2580 			 (1U << nsdata->lbaf[nsdata->flbas.format].lbads);
2581 	if (nsdata->noiob > max_num_blocks) {
2582 		nsdata->noiob = max_num_blocks;
2583 	}
2584 
2585 	/* Set NOWS equal to Controller MDTS */
2586 	if (nsdata->nsfeat.optperf) {
2587 		nsdata->nows = max_num_blocks - 1;
2588 	}
2589 
2590 	if (subsystem->flags.ana_reporting) {
2591 		assert(ns->anagrpid - 1 < subsystem->max_nsid);
2592 		nsdata->anagrpid = ns->anagrpid;
2593 
2594 		ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2595 		if (ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2596 		    ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2597 			nsdata->nuse = 0;
2598 		}
2599 	}
2600 
2601 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2602 }
2603 
2604 static void
2605 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2606 			 struct spdk_nvme_ctrlr_data *cdata)
2607 {
2608 	cdata->oacs.virtualization_management =
2609 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2610 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2611 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2612 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2613 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2614 	cdata->oacs.device_self_test =
2615 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2616 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2617 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2618 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2619 			       NULL
2620 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2621 	cdata->oacs.format =
2622 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2623 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2624 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2625 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2626 				     NULL;
2627 }
2628 
2629 int
2630 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2631 {
2632 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2633 	struct spdk_nvmf_transport *transport;
2634 
2635 	/*
2636 	 * Common fields for discovery and NVM subsystems
2637 	 */
2638 	assert(ctrlr->admin_qpair);
2639 	transport = ctrlr->admin_qpair->transport;
2640 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2641 	assert((transport->opts.max_io_size % 4096) == 0);
2642 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2643 	cdata->cntlid = ctrlr->cntlid;
2644 	cdata->ver = ctrlr->vcprop.vs;
2645 	cdata->aerl = ctrlr->cdata.aerl;
2646 	cdata->lpa.edlp = 1;
2647 	cdata->elpe = 127;
2648 	cdata->maxcmd = transport->opts.max_queue_depth;
2649 	cdata->sgls = ctrlr->cdata.sgls;
2650 	cdata->fuses.compare_and_write = 1;
2651 	cdata->acwu = 1;
2652 	if (subsystem->flags.ana_reporting) {
2653 		cdata->mnan = subsystem->max_nsid;
2654 	}
2655 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2656 
2657 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2658 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2659 
2660 
2661 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2662 		/*
2663 		 * NVM Discovery subsystem fields
2664 		 */
2665 		cdata->oaes.discovery_log_change_notices = 1;
2666 	} else {
2667 		cdata->vid = ctrlr->cdata.vid;
2668 		cdata->ssvid = ctrlr->cdata.ssvid;
2669 		cdata->ieee[0] = ctrlr->cdata.ieee[0];
2670 		cdata->ieee[1] = ctrlr->cdata.ieee[1];
2671 		cdata->ieee[2] = ctrlr->cdata.ieee[2];
2672 
2673 		/*
2674 		 * NVM subsystem fields (reserved for discovery subsystems)
2675 		 */
2676 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2677 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2678 		cdata->kas = ctrlr->cdata.kas;
2679 
2680 		cdata->rab = 6;
2681 		cdata->cmic.multi_port = 1;
2682 		cdata->cmic.multi_ctrlr = 1;
2683 		if (subsystem->flags.ana_reporting) {
2684 			/* Asymmetric Namespace Access Reporting is supported. */
2685 			cdata->cmic.ana_reporting = 1;
2686 		}
2687 		cdata->oaes.ns_attribute_notices = 1;
2688 		if (subsystem->flags.ana_reporting) {
2689 			cdata->oaes.ana_change_notices = 1;
2690 		}
2691 		cdata->ctratt.host_id_exhid_supported = 1;
2692 		/* We do not have any actual limitation to the number of abort commands.
2693 		 * We follow the recommendation by the NVMe specification.
2694 		 */
2695 		cdata->acl = NVMF_ABORT_COMMAND_LIMIT;
2696 		cdata->frmw.slot1_ro = 1;
2697 		cdata->frmw.num_slots = 1;
2698 
2699 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2700 
2701 		cdata->sqes.min = 6;
2702 		cdata->sqes.max = 6;
2703 		cdata->cqes.min = 4;
2704 		cdata->cqes.max = 4;
2705 		cdata->nn = subsystem->max_nsid;
2706 		cdata->vwc.present = 1;
2707 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
2708 
2709 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
2710 
2711 		cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr);
2712 		cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr);
2713 		cdata->oncs.reservations = ctrlr->cdata.oncs.reservations;
2714 		if (subsystem->flags.ana_reporting) {
2715 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
2716 			/* ANA Change state is not used, and ANA Persistent Loss state
2717 			 * is not supported for now.
2718 			 */
2719 			cdata->anacap.ana_optimized_state = 1;
2720 			cdata->anacap.ana_non_optimized_state = 1;
2721 			cdata->anacap.ana_inaccessible_state = 1;
2722 			/* ANAGRPID does not change while namespace is attached to controller */
2723 			cdata->anacap.no_change_anagrpid = 1;
2724 			cdata->anagrpmax = subsystem->max_nsid;
2725 			cdata->nanagrpid = subsystem->max_nsid;
2726 		}
2727 
2728 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
2729 
2730 		assert(subsystem->tgt != NULL);
2731 		cdata->crdt[0] = subsystem->tgt->crdt[0];
2732 		cdata->crdt[1] = subsystem->tgt->crdt[1];
2733 		cdata->crdt[2] = subsystem->tgt->crdt[2];
2734 
2735 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
2736 			      cdata->nvmf_specific.ioccsz);
2737 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
2738 			      cdata->nvmf_specific.iorcsz);
2739 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
2740 			      cdata->nvmf_specific.icdoff);
2741 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
2742 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
2743 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
2744 			      cdata->nvmf_specific.msdbd);
2745 	}
2746 
2747 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2748 }
2749 
2750 static int
2751 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_subsystem *subsystem,
2752 				   struct spdk_nvme_cmd *cmd,
2753 				   struct spdk_nvme_cpl *rsp,
2754 				   struct spdk_nvme_ns_list *ns_list)
2755 {
2756 	struct spdk_nvmf_ns *ns;
2757 	uint32_t count = 0;
2758 
2759 	if (cmd->nsid >= 0xfffffffeUL) {
2760 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
2761 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2762 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2763 	}
2764 
2765 	memset(ns_list, 0, sizeof(*ns_list));
2766 
2767 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
2768 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2769 		if (ns->opts.nsid <= cmd->nsid) {
2770 			continue;
2771 		}
2772 
2773 		ns_list->ns_list[count++] = ns->opts.nsid;
2774 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
2775 			break;
2776 		}
2777 	}
2778 
2779 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2780 }
2781 
2782 static void
2783 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
2784 		enum spdk_nvme_nidt type,
2785 		const void *data, size_t data_size)
2786 {
2787 	struct spdk_nvme_ns_id_desc *desc;
2788 	size_t desc_size = sizeof(*desc) + data_size;
2789 
2790 	/*
2791 	 * These should never fail in practice, since all valid NS ID descriptors
2792 	 * should be defined so that they fit in the available 4096-byte buffer.
2793 	 */
2794 	assert(data_size > 0);
2795 	assert(data_size <= UINT8_MAX);
2796 	assert(desc_size < *buf_remain);
2797 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
2798 		return;
2799 	}
2800 
2801 	desc = *buf_ptr;
2802 	desc->nidt = type;
2803 	desc->nidl = data_size;
2804 	memcpy(desc->nid, data, data_size);
2805 
2806 	*buf_ptr += desc_size;
2807 	*buf_remain -= desc_size;
2808 }
2809 
2810 static int
2811 nvmf_ctrlr_identify_ns_id_descriptor_list(
2812 	struct spdk_nvmf_subsystem *subsystem,
2813 	struct spdk_nvme_cmd *cmd,
2814 	struct spdk_nvme_cpl *rsp,
2815 	void *id_desc_list, size_t id_desc_list_size)
2816 {
2817 	struct spdk_nvmf_ns *ns;
2818 	size_t buf_remain = id_desc_list_size;
2819 	void *buf_ptr = id_desc_list;
2820 
2821 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2822 	if (ns == NULL || ns->bdev == NULL) {
2823 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2824 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2825 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2826 	}
2827 
2828 #define ADD_ID_DESC(type, data, size) \
2829 	do { \
2830 		if (!spdk_mem_all_zero(data, size)) { \
2831 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
2832 		} \
2833 	} while (0)
2834 
2835 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
2836 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
2837 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
2838 
2839 	/*
2840 	 * The list is automatically 0-terminated because controller to host buffers in
2841 	 * admin commands always get zeroed in nvmf_ctrlr_process_admin_cmd().
2842 	 */
2843 
2844 #undef ADD_ID_DESC
2845 
2846 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2847 }
2848 
2849 static int
2850 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
2851 {
2852 	uint8_t cns;
2853 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2854 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2855 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2856 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2857 
2858 	if (req->data == NULL || req->length < 4096) {
2859 		SPDK_DEBUGLOG(nvmf, "identify command with invalid buffer\n");
2860 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2861 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2862 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2863 	}
2864 
2865 	cns = cmd->cdw10_bits.identify.cns;
2866 
2867 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY &&
2868 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
2869 		/* Discovery controllers only support Identify Controller */
2870 		goto invalid_cns;
2871 	}
2872 
2873 	switch (cns) {
2874 	case SPDK_NVME_IDENTIFY_NS:
2875 		return spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, req->data);
2876 	case SPDK_NVME_IDENTIFY_CTRLR:
2877 		return spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, req->data);
2878 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
2879 		return nvmf_ctrlr_identify_active_ns_list(subsystem, cmd, rsp, req->data);
2880 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
2881 		return nvmf_ctrlr_identify_ns_id_descriptor_list(subsystem, cmd, rsp, req->data, req->length);
2882 	default:
2883 		goto invalid_cns;
2884 	}
2885 
2886 invalid_cns:
2887 	SPDK_INFOLOG(nvmf, "Identify command with unsupported CNS 0x%02x\n", cns);
2888 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2889 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2890 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2891 }
2892 
2893 static bool
2894 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
2895 {
2896 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
2897 	struct spdk_nvmf_request *req;
2898 	int i;
2899 
2900 	if (!nvmf_qpair_is_admin_queue(qpair)) {
2901 		return false;
2902 	}
2903 
2904 	assert(spdk_get_thread() == ctrlr->thread);
2905 
2906 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2907 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
2908 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
2909 			req = ctrlr->aer_req[i];
2910 			ctrlr->aer_req[i] = NULL;
2911 			ctrlr->nr_aer_reqs--;
2912 
2913 			/* Move the last req to the aborting position for making aer_reqs
2914 			 * in continuous
2915 			 */
2916 			if (i < ctrlr->nr_aer_reqs) {
2917 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
2918 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
2919 			}
2920 
2921 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2922 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2923 			_nvmf_request_complete(req);
2924 			return true;
2925 		}
2926 	}
2927 
2928 	return false;
2929 }
2930 
2931 void
2932 nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair)
2933 {
2934 	struct spdk_nvmf_request *req, *tmp;
2935 
2936 	TAILQ_FOREACH_SAFE(req, &qpair->outstanding, link, tmp) {
2937 		if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) {
2938 			/* Zero-copy requests are kept on the outstanding queue from the moment
2939 			 * zcopy_start is sent until a zcopy_end callback is received.  Therefore,
2940 			 * we can't remove them from the outstanding queue here, but need to rely on
2941 			 * the transport to do a zcopy_end to release their buffers and, in turn,
2942 			 * remove them from the queue.
2943 			 */
2944 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2945 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2946 			nvmf_transport_req_free(req);
2947 		}
2948 	}
2949 }
2950 
2951 static void
2952 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
2953 {
2954 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
2955 
2956 	if (nvmf_qpair_abort_aer(qpair, cid)) {
2957 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
2958 			      qpair->ctrlr, qpair->qid, cid);
2959 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
2960 
2961 		spdk_nvmf_request_complete(req);
2962 		return;
2963 	}
2964 
2965 	nvmf_transport_qpair_abort_request(qpair, req);
2966 }
2967 
2968 static void
2969 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
2970 {
2971 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2972 
2973 	if (status == 0) {
2974 		/* There was no qpair whose ID matches SQID of the abort command.
2975 		 * Hence call _nvmf_request_complete() here.
2976 		 */
2977 		_nvmf_request_complete(req);
2978 	}
2979 }
2980 
2981 static void
2982 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
2983 {
2984 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2985 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
2986 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
2987 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
2988 	struct spdk_nvmf_qpair *qpair;
2989 
2990 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
2991 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
2992 			/* Found the qpair */
2993 
2994 			nvmf_qpair_abort_request(qpair, req);
2995 
2996 			/* Return -1 for the status so the iteration across threads stops. */
2997 			spdk_for_each_channel_continue(i, -1);
2998 			return;
2999 		}
3000 	}
3001 
3002 	spdk_for_each_channel_continue(i, 0);
3003 }
3004 
3005 static int
3006 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
3007 {
3008 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3009 
3010 	rsp->cdw0 = 1U; /* Command not aborted */
3011 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3012 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
3013 
3014 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
3015 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
3016 			      nvmf_ctrlr_abort_on_pg,
3017 			      req,
3018 			      nvmf_ctrlr_abort_done
3019 			     );
3020 
3021 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3022 }
3023 
3024 int
3025 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
3026 {
3027 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
3028 	struct spdk_bdev *bdev;
3029 	struct spdk_bdev_desc *desc;
3030 	struct spdk_io_channel *ch;
3031 	int rc;
3032 
3033 	assert(req_to_abort != NULL);
3034 
3035 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
3036 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
3037 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
3038 	}
3039 
3040 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
3041 					&bdev, &desc, &ch);
3042 	if (rc != 0) {
3043 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3044 	}
3045 
3046 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
3047 }
3048 
3049 static int
3050 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
3051 {
3052 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3053 
3054 	rsp->cdw0 = cdw0;
3055 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3056 }
3057 
3058 /* we have to use the typedef in the function declaration to appease astyle. */
3059 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
3060 
3061 static spdk_nvme_path_status_code_t
3062 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
3063 {
3064 	switch (ana_state) {
3065 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3066 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
3067 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3068 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3069 	case SPDK_NVME_ANA_CHANGE_STATE:
3070 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
3071 	default:
3072 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
3073 	}
3074 }
3075 
3076 static int
3077 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
3078 {
3079 	uint8_t feature;
3080 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3081 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3082 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3083 	enum spdk_nvme_ana_state ana_state;
3084 
3085 	feature = cmd->cdw10_bits.get_features.fid;
3086 
3087 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
3088 		/*
3089 		 * Features supported by Discovery controller
3090 		 */
3091 		switch (feature) {
3092 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3093 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3094 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3095 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3096 		default:
3097 			SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3098 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3099 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3100 		}
3101 	}
3102 	/*
3103 	 * Process Get Features command for non-discovery controller
3104 	 */
3105 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3106 	switch (ana_state) {
3107 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3108 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3109 	case SPDK_NVME_ANA_CHANGE_STATE:
3110 		switch (feature) {
3111 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
3112 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3113 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3114 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3115 			response->status.sct = SPDK_NVME_SCT_PATH;
3116 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3117 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3118 		default:
3119 			break;
3120 		}
3121 		break;
3122 	default:
3123 		break;
3124 	}
3125 
3126 	switch (feature) {
3127 	case SPDK_NVME_FEAT_ARBITRATION:
3128 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
3129 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3130 		return get_features_generic(req, ctrlr->feat.power_management.raw);
3131 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3132 		return nvmf_ctrlr_get_features_temperature_threshold(req);
3133 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3134 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
3135 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3136 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
3137 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3138 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
3139 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3140 		return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw);
3141 	case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION:
3142 		return nvmf_ctrlr_get_features_interrupt_vector_configuration(req);
3143 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3144 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
3145 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3146 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3147 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3148 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3149 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3150 		return nvmf_ctrlr_get_features_host_identifier(req);
3151 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3152 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
3153 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3154 		return nvmf_ctrlr_get_features_reservation_persistence(req);
3155 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3156 		return nvmf_ctrlr_get_features_host_behavior_support(req);
3157 	default:
3158 		SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3159 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3160 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3161 	}
3162 }
3163 
3164 static int
3165 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
3166 {
3167 	uint8_t feature, save;
3168 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3169 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3170 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3171 	enum spdk_nvme_ana_state ana_state;
3172 	/*
3173 	 * Features are not saveable by the controller as indicated by
3174 	 * ONCS field of the Identify Controller data.
3175 	 * */
3176 	save = cmd->cdw10_bits.set_features.sv;
3177 	if (save) {
3178 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
3179 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3180 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3181 	}
3182 
3183 	feature = cmd->cdw10_bits.set_features.fid;
3184 
3185 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
3186 		/*
3187 		 * Features supported by Discovery controller
3188 		 */
3189 		switch (feature) {
3190 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3191 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
3192 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3193 			return nvmf_ctrlr_set_features_async_event_configuration(req);
3194 		default:
3195 			SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3196 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3197 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3198 		}
3199 	}
3200 	/*
3201 	 * Process Set Features command for non-discovery controller
3202 	 */
3203 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3204 	switch (ana_state) {
3205 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3206 	case SPDK_NVME_ANA_CHANGE_STATE:
3207 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
3208 			response->status.sct = SPDK_NVME_SCT_PATH;
3209 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3210 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3211 		} else {
3212 			switch (feature) {
3213 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
3214 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3215 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3216 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3217 				response->status.sct = SPDK_NVME_SCT_PATH;
3218 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3219 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3220 			default:
3221 				break;
3222 			}
3223 		}
3224 		break;
3225 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3226 		response->status.sct = SPDK_NVME_SCT_PATH;
3227 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3228 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3229 	default:
3230 		break;
3231 	}
3232 
3233 	switch (feature) {
3234 	case SPDK_NVME_FEAT_ARBITRATION:
3235 		return nvmf_ctrlr_set_features_arbitration(req);
3236 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3237 		return nvmf_ctrlr_set_features_power_management(req);
3238 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3239 		return nvmf_ctrlr_set_features_temperature_threshold(req);
3240 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3241 		return nvmf_ctrlr_set_features_error_recovery(req);
3242 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3243 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
3244 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3245 		return nvmf_ctrlr_set_features_number_of_queues(req);
3246 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3247 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3248 		response->status.sc = SPDK_NVME_SC_FEATURE_NOT_CHANGEABLE;
3249 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3250 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3251 		return nvmf_ctrlr_set_features_write_atomicity(req);
3252 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3253 		return nvmf_ctrlr_set_features_async_event_configuration(req);
3254 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3255 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
3256 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3257 		return nvmf_ctrlr_set_features_host_identifier(req);
3258 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3259 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
3260 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3261 		return nvmf_ctrlr_set_features_reservation_persistence(req);
3262 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3263 		return nvmf_ctrlr_set_features_host_behavior_support(req);
3264 	default:
3265 		SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3266 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3267 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3268 	}
3269 }
3270 
3271 static int
3272 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
3273 {
3274 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3275 
3276 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
3277 	/*
3278 	 * To handle keep alive just clear or reset the
3279 	 * ctrlr based keep alive duration counter.
3280 	 * When added, a separate timer based process
3281 	 * will monitor if the time since last recorded
3282 	 * keep alive has exceeded the max duration and
3283 	 * take appropriate action.
3284 	 */
3285 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
3286 
3287 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3288 }
3289 
3290 int
3291 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
3292 {
3293 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3294 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3295 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3296 	int rc;
3297 
3298 	if (ctrlr == NULL) {
3299 		SPDK_ERRLOG("Admin command sent before CONNECT\n");
3300 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3301 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3302 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3303 	}
3304 
3305 	assert(spdk_get_thread() == ctrlr->thread);
3306 
3307 	if (cmd->fuse != 0) {
3308 		/* Fused admin commands are not supported. */
3309 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3310 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3311 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3312 	}
3313 
3314 	if (ctrlr->vcprop.cc.bits.en != 1) {
3315 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
3316 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3317 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3318 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3319 	}
3320 
3321 	if (req->data && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
3322 		memset(req->data, 0, req->length);
3323 	}
3324 
3325 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
3326 		/* Discovery controllers only support these admin OPS. */
3327 		switch (cmd->opc) {
3328 		case SPDK_NVME_OPC_IDENTIFY:
3329 		case SPDK_NVME_OPC_GET_LOG_PAGE:
3330 		case SPDK_NVME_OPC_KEEP_ALIVE:
3331 		case SPDK_NVME_OPC_SET_FEATURES:
3332 		case SPDK_NVME_OPC_GET_FEATURES:
3333 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3334 			break;
3335 		default:
3336 			goto invalid_opcode;
3337 		}
3338 	}
3339 
3340 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
3341 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
3342 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
3343 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3344 			/* The handler took care of this command */
3345 			return rc;
3346 		}
3347 	}
3348 
3349 	switch (cmd->opc) {
3350 	case SPDK_NVME_OPC_GET_LOG_PAGE:
3351 		return nvmf_ctrlr_get_log_page(req);
3352 	case SPDK_NVME_OPC_IDENTIFY:
3353 		return nvmf_ctrlr_identify(req);
3354 	case SPDK_NVME_OPC_ABORT:
3355 		return nvmf_ctrlr_abort(req);
3356 	case SPDK_NVME_OPC_GET_FEATURES:
3357 		return nvmf_ctrlr_get_features(req);
3358 	case SPDK_NVME_OPC_SET_FEATURES:
3359 		return nvmf_ctrlr_set_features(req);
3360 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3361 		return nvmf_ctrlr_async_event_request(req);
3362 	case SPDK_NVME_OPC_KEEP_ALIVE:
3363 		return nvmf_ctrlr_keep_alive(req);
3364 
3365 	case SPDK_NVME_OPC_CREATE_IO_SQ:
3366 	case SPDK_NVME_OPC_CREATE_IO_CQ:
3367 	case SPDK_NVME_OPC_DELETE_IO_SQ:
3368 	case SPDK_NVME_OPC_DELETE_IO_CQ:
3369 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
3370 		goto invalid_opcode;
3371 
3372 	default:
3373 		goto invalid_opcode;
3374 	}
3375 
3376 invalid_opcode:
3377 	SPDK_INFOLOG(nvmf, "Unsupported admin opcode 0x%x\n", cmd->opc);
3378 	response->status.sct = SPDK_NVME_SCT_GENERIC;
3379 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3380 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3381 }
3382 
3383 static int
3384 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
3385 {
3386 	struct spdk_nvmf_qpair *qpair = req->qpair;
3387 	struct spdk_nvmf_capsule_cmd *cap_hdr;
3388 
3389 	cap_hdr = &req->cmd->nvmf_cmd;
3390 
3391 	if (qpair->ctrlr == NULL) {
3392 		/* No ctrlr established yet; the only valid command is Connect */
3393 		if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
3394 			return nvmf_ctrlr_cmd_connect(req);
3395 		} else {
3396 			SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n",
3397 				      cap_hdr->fctype);
3398 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3399 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3400 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3401 		}
3402 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
3403 		/*
3404 		 * Controller session is established, and this is an admin queue.
3405 		 * Disallow Connect and allow other fabrics commands.
3406 		 */
3407 		switch (cap_hdr->fctype) {
3408 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
3409 			return nvmf_property_set(req);
3410 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
3411 			return nvmf_property_get(req);
3412 		default:
3413 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
3414 				      cap_hdr->fctype);
3415 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3416 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3417 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3418 		}
3419 	} else {
3420 		/* Controller session is established, and this is an I/O queue */
3421 		/* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */
3422 		SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
3423 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3424 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3425 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3426 	}
3427 }
3428 
3429 static inline void
3430 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr,
3431 				     union spdk_nvme_async_event_completion *event)
3432 {
3433 	struct spdk_nvmf_async_event_completion *nvmf_event;
3434 
3435 	nvmf_event = calloc(1, sizeof(*nvmf_event));
3436 	if (!nvmf_event) {
3437 		SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n");
3438 		return;
3439 	}
3440 	nvmf_event->event.raw = event->raw;
3441 	STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link);
3442 }
3443 
3444 static inline int
3445 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
3446 				    union spdk_nvme_async_event_completion *event)
3447 {
3448 	struct spdk_nvmf_request *req;
3449 	struct spdk_nvme_cpl *rsp;
3450 
3451 	assert(spdk_get_thread() == ctrlr->thread);
3452 
3453 	/* If there is no outstanding AER request, queue the event.  Then
3454 	 * if an AER is later submitted, this event can be sent as a
3455 	 * response.
3456 	 */
3457 	if (ctrlr->nr_aer_reqs == 0) {
3458 		nvmf_ctrlr_queue_pending_async_event(ctrlr, event);
3459 		return 0;
3460 	}
3461 
3462 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
3463 	rsp = &req->rsp->nvme_cpl;
3464 
3465 	rsp->cdw0 = event->raw;
3466 
3467 	_nvmf_request_complete(req);
3468 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3469 
3470 	return 0;
3471 }
3472 
3473 int
3474 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
3475 {
3476 	union spdk_nvme_async_event_completion event = {0};
3477 
3478 	/* Users may disable the event notification */
3479 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
3480 		return 0;
3481 	}
3482 
3483 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) {
3484 		return 0;
3485 	}
3486 
3487 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3488 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
3489 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
3490 
3491 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3492 }
3493 
3494 int
3495 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3496 {
3497 	union spdk_nvme_async_event_completion event = {0};
3498 
3499 	/* Users may disable the event notification */
3500 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
3501 		return 0;
3502 	}
3503 
3504 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) {
3505 		return 0;
3506 	}
3507 
3508 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3509 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
3510 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
3511 
3512 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3513 }
3514 
3515 void
3516 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
3517 {
3518 	union spdk_nvme_async_event_completion event = {0};
3519 
3520 	if (!ctrlr->num_avail_log_pages) {
3521 		return;
3522 	}
3523 
3524 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) {
3525 		return;
3526 	}
3527 
3528 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
3529 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
3530 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
3531 
3532 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3533 }
3534 
3535 void
3536 nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx)
3537 {
3538 	union spdk_nvme_async_event_completion event = {0};
3539 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
3540 
3541 	/* Users may disable the event notification manually or
3542 	 * it may not be enabled due to keep alive timeout
3543 	 * not being set in connect command to discovery controller.
3544 	 */
3545 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
3546 		return;
3547 	}
3548 
3549 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) {
3550 		return;
3551 	}
3552 
3553 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3554 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
3555 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
3556 
3557 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3558 }
3559 
3560 int
3561 nvmf_ctrlr_async_event_error_event(struct spdk_nvmf_ctrlr *ctrlr,
3562 				   union spdk_nvme_async_event_completion event)
3563 {
3564 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT)) {
3565 		return 0;
3566 	}
3567 
3568 	if (event.bits.async_event_type != SPDK_NVME_ASYNC_EVENT_TYPE_ERROR ||
3569 	    event.bits.async_event_info > SPDK_NVME_ASYNC_EVENT_FW_IMAGE_LOAD) {
3570 		return 0;
3571 	}
3572 
3573 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3574 }
3575 
3576 void
3577 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
3578 {
3579 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3580 	int i;
3581 
3582 	if (!nvmf_qpair_is_admin_queue(qpair)) {
3583 		return;
3584 	}
3585 
3586 	assert(spdk_get_thread() == ctrlr->thread);
3587 
3588 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3589 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
3590 		ctrlr->aer_req[i] = NULL;
3591 	}
3592 
3593 	ctrlr->nr_aer_reqs = 0;
3594 }
3595 
3596 void
3597 nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
3598 {
3599 	struct spdk_nvmf_request *req;
3600 	int i;
3601 
3602 	assert(spdk_get_thread() == ctrlr->thread);
3603 
3604 	if (!ctrlr->nr_aer_reqs) {
3605 		return;
3606 	}
3607 
3608 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3609 		req = ctrlr->aer_req[i];
3610 
3611 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3612 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3613 		_nvmf_request_complete(req);
3614 
3615 		ctrlr->aer_req[i] = NULL;
3616 	}
3617 
3618 	ctrlr->nr_aer_reqs = 0;
3619 }
3620 
3621 static void
3622 _nvmf_ctrlr_add_reservation_log(void *ctx)
3623 {
3624 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
3625 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
3626 
3627 	ctrlr->log_page_count++;
3628 
3629 	/* Maximum number of queued log pages is 255 */
3630 	if (ctrlr->num_avail_log_pages == 0xff) {
3631 		struct spdk_nvmf_reservation_log *entry;
3632 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
3633 		entry->log.log_page_count = ctrlr->log_page_count;
3634 		free(log);
3635 		return;
3636 	}
3637 
3638 	log->log.log_page_count = ctrlr->log_page_count;
3639 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
3640 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
3641 
3642 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
3643 }
3644 
3645 void
3646 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
3647 				  struct spdk_nvmf_ns *ns,
3648 				  enum spdk_nvme_reservation_notification_log_page_type type)
3649 {
3650 	struct spdk_nvmf_reservation_log *log;
3651 
3652 	switch (type) {
3653 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
3654 		return;
3655 	case SPDK_NVME_REGISTRATION_PREEMPTED:
3656 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
3657 			return;
3658 		}
3659 		break;
3660 	case SPDK_NVME_RESERVATION_RELEASED:
3661 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
3662 			return;
3663 		}
3664 		break;
3665 	case SPDK_NVME_RESERVATION_PREEMPTED:
3666 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
3667 			return;
3668 		}
3669 		break;
3670 	default:
3671 		return;
3672 	}
3673 
3674 	log = calloc(1, sizeof(*log));
3675 	if (!log) {
3676 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
3677 		return;
3678 	}
3679 	log->ctrlr = ctrlr;
3680 	log->log.type = type;
3681 	log->log.nsid = ns->nsid;
3682 
3683 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
3684 }
3685 
3686 /* Check from subsystem poll group's namespace information data structure */
3687 static bool
3688 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3689 				 struct spdk_nvmf_ctrlr *ctrlr)
3690 {
3691 	uint32_t i;
3692 
3693 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
3694 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
3695 			return true;
3696 		}
3697 	}
3698 
3699 	return false;
3700 }
3701 
3702 /*
3703  * Check the NVMe command is permitted or not for current controller(Host).
3704  */
3705 static int
3706 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3707 				  struct spdk_nvmf_ctrlr *ctrlr,
3708 				  struct spdk_nvmf_request *req)
3709 {
3710 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3711 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
3712 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3713 	uint8_t racqa;
3714 	bool is_registrant;
3715 
3716 	/* No valid reservation */
3717 	if (!rtype) {
3718 		return 0;
3719 	}
3720 
3721 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
3722 	/* All registrants type and current ctrlr is a valid registrant */
3723 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3724 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
3725 		return 0;
3726 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
3727 		return 0;
3728 	}
3729 
3730 	/* Non-holder for current controller */
3731 	switch (cmd->opc) {
3732 	case SPDK_NVME_OPC_READ:
3733 	case SPDK_NVME_OPC_COMPARE:
3734 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3735 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3736 			goto exit;
3737 		}
3738 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
3739 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
3740 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3741 		}
3742 		break;
3743 	case SPDK_NVME_OPC_FLUSH:
3744 	case SPDK_NVME_OPC_WRITE:
3745 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
3746 	case SPDK_NVME_OPC_WRITE_ZEROES:
3747 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3748 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
3749 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3750 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3751 			goto exit;
3752 		}
3753 		if (!is_registrant) {
3754 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3755 		}
3756 		break;
3757 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3758 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
3759 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
3760 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3761 			goto exit;
3762 		}
3763 		if (!is_registrant) {
3764 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3765 		}
3766 		break;
3767 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3768 		if (!is_registrant) {
3769 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3770 		}
3771 		break;
3772 	default:
3773 		break;
3774 	}
3775 
3776 exit:
3777 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3778 	req->rsp->nvme_cpl.status.sc = status;
3779 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
3780 		return -EPERM;
3781 	}
3782 
3783 	return 0;
3784 }
3785 
3786 static int
3787 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
3788 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
3789 {
3790 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3791 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3792 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
3793 	int rc;
3794 
3795 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
3796 		/* first fused operation (should be compare) */
3797 		if (first_fused_req != NULL) {
3798 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3799 
3800 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3801 
3802 			/* abort req->qpair->first_fused_request and continue with new fused command */
3803 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3804 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3805 			_nvmf_request_complete(first_fused_req);
3806 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
3807 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3808 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3809 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3810 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3811 		}
3812 
3813 		req->qpair->first_fused_req = req;
3814 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3815 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
3816 		/* second fused operation (should be write) */
3817 		if (first_fused_req == NULL) {
3818 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3819 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3820 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3821 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3822 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
3823 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3824 
3825 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3826 
3827 			/* abort req->qpair->first_fused_request and fail current command */
3828 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3829 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3830 			_nvmf_request_complete(first_fused_req);
3831 
3832 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3833 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3834 			req->qpair->first_fused_req = NULL;
3835 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3836 		}
3837 
3838 		/* save request of first command to generate response later */
3839 		req->first_fused_req = first_fused_req;
3840 		req->qpair->first_fused_req = NULL;
3841 	} else {
3842 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
3843 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3844 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3845 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3846 	}
3847 
3848 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
3849 
3850 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3851 		if (spdk_nvme_cpl_is_error(rsp)) {
3852 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3853 
3854 			fused_response->status = rsp->status;
3855 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3856 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
3857 			/* Complete first of fused commands. Second will be completed by upper layer */
3858 			_nvmf_request_complete(first_fused_req);
3859 			req->first_fused_req = NULL;
3860 		}
3861 	}
3862 
3863 	return rc;
3864 }
3865 
3866 bool
3867 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req)
3868 {
3869 	struct spdk_nvmf_transport *transport = req->qpair->transport;
3870 	struct spdk_nvmf_ns *ns;
3871 
3872 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE);
3873 
3874 	if (!transport->opts.zcopy) {
3875 		return false;
3876 	}
3877 
3878 	if (nvmf_qpair_is_admin_queue(req->qpair)) {
3879 		/* Admin queue */
3880 		return false;
3881 	}
3882 
3883 	if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) &&
3884 	    (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) {
3885 		/* Not a READ or WRITE command */
3886 		return false;
3887 	}
3888 
3889 	if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) {
3890 		/* Fused commands dont use zcopy buffers */
3891 		return false;
3892 	}
3893 
3894 	ns = _nvmf_subsystem_get_ns(req->qpair->ctrlr->subsys, req->cmd->nvme_cmd.nsid);
3895 	if (ns == NULL || ns->bdev == NULL || !ns->zcopy) {
3896 		return false;
3897 	}
3898 
3899 	req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT;
3900 	return true;
3901 }
3902 
3903 void
3904 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req)
3905 {
3906 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
3907 
3908 	/* Set iovcnt to be the maximum number of iovs that the ZCOPY can use */
3909 	req->iovcnt = NVMF_REQ_MAX_BUFFERS;
3910 
3911 	spdk_nvmf_request_exec(req);
3912 }
3913 
3914 void
3915 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req, bool commit)
3916 {
3917 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE);
3918 	req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING;
3919 
3920 	nvmf_bdev_ctrlr_zcopy_end(req, commit);
3921 }
3922 
3923 int
3924 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
3925 {
3926 	uint32_t nsid;
3927 	struct spdk_nvmf_ns *ns;
3928 	struct spdk_bdev *bdev;
3929 	struct spdk_bdev_desc *desc;
3930 	struct spdk_io_channel *ch;
3931 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3932 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3933 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3934 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3935 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3936 	enum spdk_nvme_ana_state ana_state;
3937 
3938 	/* pre-set response details for this command */
3939 	response->status.sc = SPDK_NVME_SC_SUCCESS;
3940 	nsid = cmd->nsid;
3941 
3942 	if (spdk_unlikely(ctrlr == NULL)) {
3943 		SPDK_ERRLOG("I/O command sent before CONNECT\n");
3944 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3945 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3946 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3947 	}
3948 
3949 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
3950 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
3951 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3952 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3953 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3954 	}
3955 
3956 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3957 	if (ns == NULL || ns->bdev == NULL) {
3958 		SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid);
3959 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3960 		response->status.dnr = 1;
3961 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3962 	}
3963 
3964 	ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
3965 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
3966 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
3967 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
3968 			      ana_state);
3969 		response->status.sct = SPDK_NVME_SCT_PATH;
3970 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3971 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3972 	}
3973 
3974 	if (spdk_likely(ctrlr->listener != NULL)) {
3975 		SPDK_DTRACE_PROBE3(nvmf_request_io_exec_path, req,
3976 				   ctrlr->listener->trid->traddr,
3977 				   ctrlr->listener->trid->trsvcid);
3978 	}
3979 
3980 	/* scan-build falsely reporting dereference of null pointer */
3981 	assert(group != NULL && group->sgroups != NULL);
3982 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3983 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
3984 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
3985 			      cmd->nsid, cmd->opc);
3986 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3987 	}
3988 
3989 	bdev = ns->bdev;
3990 	desc = ns->desc;
3991 	ch = ns_info->channel;
3992 
3993 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
3994 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
3995 	} else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) {
3996 		struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl;
3997 
3998 		SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n");
3999 
4000 		/* abort req->qpair->first_fused_request and continue with new command */
4001 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
4002 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
4003 		_nvmf_request_complete(req->qpair->first_fused_req);
4004 		req->qpair->first_fused_req = NULL;
4005 	}
4006 
4007 	if (spdk_nvmf_request_using_zcopy(req)) {
4008 		assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
4009 		return nvmf_bdev_ctrlr_zcopy_start(bdev, desc, ch, req);
4010 	} else {
4011 		switch (cmd->opc) {
4012 		case SPDK_NVME_OPC_READ:
4013 			return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
4014 		case SPDK_NVME_OPC_WRITE:
4015 			return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
4016 		case SPDK_NVME_OPC_COMPARE:
4017 			return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
4018 		case SPDK_NVME_OPC_WRITE_ZEROES:
4019 			return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
4020 		case SPDK_NVME_OPC_FLUSH:
4021 			return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
4022 		case SPDK_NVME_OPC_DATASET_MANAGEMENT:
4023 			return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
4024 		case SPDK_NVME_OPC_RESERVATION_REGISTER:
4025 		case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
4026 		case SPDK_NVME_OPC_RESERVATION_RELEASE:
4027 		case SPDK_NVME_OPC_RESERVATION_REPORT:
4028 			spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
4029 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
4030 		default:
4031 			return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
4032 		}
4033 	}
4034 }
4035 
4036 static void
4037 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
4038 {
4039 	if (qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING) {
4040 		assert(qpair->state_cb != NULL);
4041 
4042 		if (TAILQ_EMPTY(&qpair->outstanding)) {
4043 			qpair->state_cb(qpair->state_cb_arg, 0);
4044 		}
4045 	}
4046 }
4047 
4048 int
4049 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
4050 {
4051 	struct spdk_nvmf_qpair *qpair = req->qpair;
4052 
4053 	TAILQ_REMOVE(&qpair->outstanding, req, link);
4054 	if (nvmf_transport_req_free(req)) {
4055 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
4056 	}
4057 
4058 	nvmf_qpair_request_cleanup(qpair);
4059 
4060 	return 0;
4061 }
4062 
4063 static void
4064 _nvmf_request_complete(void *ctx)
4065 {
4066 	struct spdk_nvmf_request *req = ctx;
4067 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
4068 	struct spdk_nvmf_qpair *qpair;
4069 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4070 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4071 	bool is_aer = false;
4072 	uint32_t nsid;
4073 	bool paused;
4074 	uint8_t opcode;
4075 
4076 	rsp->sqid = 0;
4077 	rsp->status.p = 0;
4078 	rsp->cid = req->cmd->nvme_cmd.cid;
4079 	nsid = req->cmd->nvme_cmd.nsid;
4080 	opcode = req->cmd->nvmf_cmd.opcode;
4081 
4082 	qpair = req->qpair;
4083 	if (qpair->ctrlr) {
4084 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4085 		assert(sgroup != NULL);
4086 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
4087 
4088 		/*
4089 		 * Set the crd value.
4090 		 * If the the IO has any error, and dnr (DoNotRetry) is not 1,
4091 		 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT.
4092 		 */
4093 		if (spdk_nvme_cpl_is_error(rsp) &&
4094 		    rsp->status.dnr == 0 &&
4095 		    qpair->ctrlr->acre_enabled) {
4096 			rsp->status.crd = 1;
4097 		}
4098 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4099 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4100 	}
4101 
4102 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4103 		spdk_nvme_print_completion(qpair->qid, rsp);
4104 	}
4105 
4106 	switch (req->zcopy_phase) {
4107 	case NVMF_ZCOPY_PHASE_NONE:
4108 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4109 		break;
4110 	case NVMF_ZCOPY_PHASE_INIT:
4111 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) {
4112 			req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED;
4113 			TAILQ_REMOVE(&qpair->outstanding, req, link);
4114 		} else {
4115 			req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE;
4116 		}
4117 		break;
4118 	case NVMF_ZCOPY_PHASE_EXECUTE:
4119 		break;
4120 	case NVMF_ZCOPY_PHASE_END_PENDING:
4121 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4122 		req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE;
4123 		break;
4124 	default:
4125 		SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase);
4126 		break;
4127 	}
4128 
4129 	if (nvmf_transport_req_complete(req)) {
4130 		SPDK_ERRLOG("Transport request completion error!\n");
4131 	}
4132 
4133 	/* AER cmd is an exception */
4134 	if (sgroup && !is_aer) {
4135 		if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC ||
4136 				  nvmf_qpair_is_admin_queue(qpair))) {
4137 			assert(sgroup->mgmt_io_outstanding > 0);
4138 			sgroup->mgmt_io_outstanding--;
4139 		} else {
4140 			if (req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE ||
4141 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE ||
4142 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT_FAILED) {
4143 				/* End of request */
4144 
4145 				/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4146 				if (spdk_likely(nsid - 1 < sgroup->num_ns)) {
4147 					sgroup->ns_info[nsid - 1].io_outstanding--;
4148 				}
4149 			}
4150 		}
4151 
4152 		if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4153 				  sgroup->mgmt_io_outstanding == 0)) {
4154 			paused = true;
4155 			for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
4156 				ns_info = &sgroup->ns_info[nsid];
4157 
4158 				if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4159 				    ns_info->io_outstanding > 0) {
4160 					paused = false;
4161 					break;
4162 				}
4163 			}
4164 
4165 			if (paused) {
4166 				sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
4167 				sgroup->cb_fn(sgroup->cb_arg, 0);
4168 				sgroup->cb_fn = NULL;
4169 				sgroup->cb_arg = NULL;
4170 			}
4171 		}
4172 
4173 	}
4174 
4175 	nvmf_qpair_request_cleanup(qpair);
4176 }
4177 
4178 int
4179 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
4180 {
4181 	struct spdk_nvmf_qpair *qpair = req->qpair;
4182 
4183 	spdk_thread_exec_msg(qpair->group->thread, _nvmf_request_complete, req);
4184 
4185 	return 0;
4186 }
4187 
4188 void
4189 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req)
4190 {
4191 	struct spdk_nvmf_qpair *qpair = req->qpair;
4192 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4193 	enum spdk_nvmf_request_exec_status status;
4194 
4195 	if (qpair->ctrlr) {
4196 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4197 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4198 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4199 	}
4200 
4201 	assert(sgroup != NULL);
4202 	sgroup->mgmt_io_outstanding++;
4203 
4204 	/* Place the request on the outstanding list so we can keep track of it */
4205 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4206 
4207 	assert(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC);
4208 	status = nvmf_ctrlr_process_fabrics_cmd(req);
4209 
4210 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4211 		_nvmf_request_complete(req);
4212 	}
4213 }
4214 
4215 static bool nvmf_check_subsystem_active(struct spdk_nvmf_request *req)
4216 {
4217 	struct spdk_nvmf_qpair *qpair = req->qpair;
4218 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4219 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4220 	uint32_t nsid;
4221 
4222 	if (qpair->ctrlr) {
4223 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4224 		assert(sgroup != NULL);
4225 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4226 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4227 	}
4228 
4229 	/* Check if the subsystem is paused (if there is a subsystem) */
4230 	if (sgroup != NULL) {
4231 		if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC ||
4232 				  nvmf_qpair_is_admin_queue(qpair))) {
4233 			if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
4234 				/* The subsystem is not currently active. Queue this request. */
4235 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4236 				return false;
4237 			}
4238 			sgroup->mgmt_io_outstanding++;
4239 		} else {
4240 			nsid = req->cmd->nvme_cmd.nsid;
4241 
4242 			/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4243 			if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) {
4244 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4245 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4246 				req->rsp->nvme_cpl.status.dnr = 1;
4247 				TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4248 				_nvmf_request_complete(req);
4249 				return false;
4250 			}
4251 
4252 			ns_info = &sgroup->ns_info[nsid - 1];
4253 			if (ns_info->channel == NULL) {
4254 				/* This can can happen if host sends I/O to a namespace that is
4255 				 * in the process of being added, but before the full addition
4256 				 * process is complete.  Report invalid namespace in that case.
4257 				 */
4258 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4259 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4260 				req->rsp->nvme_cpl.status.dnr = 1;
4261 				TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4262 				ns_info->io_outstanding++;
4263 				_nvmf_request_complete(req);
4264 				return false;
4265 			}
4266 
4267 			if (ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
4268 				/* The namespace is not currently active. Queue this request. */
4269 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4270 				return false;
4271 			}
4272 
4273 			ns_info->io_outstanding++;
4274 		}
4275 
4276 		if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
4277 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4278 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
4279 			TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4280 			_nvmf_request_complete(req);
4281 			return false;
4282 		}
4283 	}
4284 
4285 	return true;
4286 }
4287 
4288 void
4289 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
4290 {
4291 	struct spdk_nvmf_qpair *qpair = req->qpair;
4292 	struct spdk_nvmf_transport *transport = qpair->transport;
4293 	enum spdk_nvmf_request_exec_status status;
4294 
4295 	if (!nvmf_check_subsystem_active(req)) {
4296 		return;
4297 	}
4298 
4299 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4300 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
4301 	}
4302 
4303 	/* Place the request on the outstanding list so we can keep track of it */
4304 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4305 
4306 	if (spdk_unlikely((req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC) &&
4307 			  spdk_nvme_trtype_is_fabrics(transport->ops->type))) {
4308 		status = nvmf_ctrlr_process_fabrics_cmd(req);
4309 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4310 		status = nvmf_ctrlr_process_admin_cmd(req);
4311 	} else {
4312 		status = nvmf_ctrlr_process_io_cmd(req);
4313 	}
4314 
4315 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4316 		_nvmf_request_complete(req);
4317 	}
4318 }
4319 
4320 static bool
4321 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
4322 		       struct spdk_dif_ctx *dif_ctx)
4323 {
4324 	struct spdk_nvmf_ns *ns;
4325 	struct spdk_bdev *bdev;
4326 
4327 	if (ctrlr == NULL || cmd == NULL) {
4328 		return false;
4329 	}
4330 
4331 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
4332 	if (ns == NULL || ns->bdev == NULL) {
4333 		return false;
4334 	}
4335 
4336 	bdev = ns->bdev;
4337 
4338 	switch (cmd->opc) {
4339 	case SPDK_NVME_OPC_READ:
4340 	case SPDK_NVME_OPC_WRITE:
4341 	case SPDK_NVME_OPC_COMPARE:
4342 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
4343 	default:
4344 		break;
4345 	}
4346 
4347 	return false;
4348 }
4349 
4350 bool
4351 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
4352 {
4353 	struct spdk_nvmf_qpair *qpair = req->qpair;
4354 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4355 
4356 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
4357 		return false;
4358 	}
4359 
4360 	if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) {
4361 		return false;
4362 	}
4363 
4364 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4365 		return false;
4366 	}
4367 
4368 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4369 		return false;
4370 	}
4371 
4372 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
4373 }
4374 
4375 void
4376 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
4377 {
4378 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
4379 }
4380 
4381 static int
4382 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
4383 {
4384 	struct spdk_bdev *bdev;
4385 	struct spdk_bdev_desc *desc;
4386 	struct spdk_io_channel *ch;
4387 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
4388 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
4389 	uint32_t bdev_nsid;
4390 	int rc;
4391 
4392 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid == 0) {
4393 		bdev_nsid = cmd->nsid;
4394 	} else {
4395 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
4396 	}
4397 
4398 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
4399 	if (rc) {
4400 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4401 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4402 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4403 	}
4404 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
4405 }
4406 
4407 void
4408 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
4409 {
4410 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
4411 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
4412 }
4413 
4414 int
4415 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
4416 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
4417 {
4418 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
4419 	struct spdk_nvmf_ns *ns;
4420 	struct spdk_nvmf_poll_group *group = req->qpair->group;
4421 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4422 
4423 	*bdev = NULL;
4424 	*desc = NULL;
4425 	*ch = NULL;
4426 
4427 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
4428 	if (ns == NULL || ns->bdev == NULL) {
4429 		return -EINVAL;
4430 	}
4431 
4432 	assert(group != NULL && group->sgroups != NULL);
4433 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
4434 	*bdev = ns->bdev;
4435 	*desc = ns->desc;
4436 	*ch = ns_info->channel;
4437 
4438 	return 0;
4439 }
4440 
4441 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
4442 {
4443 	return req->qpair->ctrlr;
4444 }
4445 
4446 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
4447 {
4448 	return &req->cmd->nvme_cmd;
4449 }
4450 
4451 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
4452 {
4453 	return &req->rsp->nvme_cpl;
4454 }
4455 
4456 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
4457 {
4458 	return req->qpair->ctrlr->subsys;
4459 }
4460 
4461 void spdk_nvmf_request_get_data(struct spdk_nvmf_request *req, void **data, uint32_t *length)
4462 {
4463 	*data = req->data;
4464 	*length = req->length;
4465 }
4466 
4467 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
4468 {
4469 	return ctrlr->subsys;
4470 }
4471 
4472 uint16_t spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
4473 {
4474 	return ctrlr->cntlid;
4475 }
4476 
4477 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
4478 {
4479 	return req->req_to_abort;
4480 }
4481