xref: /spdk/lib/nvmf/ctrlr.c (revision cc0d05b427038cc877539c12fb0f8bc9053300c0)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "nvmf_internal.h"
37 #include "transport.h"
38 
39 #include "spdk/bit_array.h"
40 #include "spdk/endian.h"
41 #include "spdk/thread.h"
42 #include "spdk/nvme_spec.h"
43 #include "spdk/nvmf_cmd.h"
44 #include "spdk/string.h"
45 #include "spdk/util.h"
46 #include "spdk/version.h"
47 #include "spdk/log.h"
48 #include "spdk_internal/usdt.h"
49 
50 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000
51 #define NVMF_DISC_KATO_IN_MS 120000
52 #define KAS_TIME_UNIT_IN_MS 100
53 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS)
54 
55 /*
56  * Report the SPDK version as the firmware revision.
57  * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts.
58  */
59 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING
60 
61 #define ANA_TRANSITION_TIME_IN_SEC 10
62 
63 /*
64  * Support for custom admin command handlers
65  */
66 struct spdk_nvmf_custom_admin_cmd {
67 	spdk_nvmf_custom_cmd_hdlr hdlr;
68 	uint32_t nsid; /* nsid to forward */
69 };
70 
71 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1];
72 
73 static void _nvmf_request_complete(void *ctx);
74 
75 static inline void
76 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp,
77 			      uint8_t iattr, uint16_t ipo)
78 {
79 	rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
80 	rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
81 	rsp->status_code_specific.invalid.iattr = iattr;
82 	rsp->status_code_specific.invalid.ipo = ipo;
83 }
84 
85 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field)	\
86 	nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field))
87 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field)	\
88 	nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field))
89 
90 
91 static void
92 nvmf_ctrlr_stop_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
93 {
94 	if (!ctrlr) {
95 		SPDK_ERRLOG("Controller is NULL\n");
96 		return;
97 	}
98 
99 	if (ctrlr->keep_alive_poller == NULL) {
100 		return;
101 	}
102 
103 	SPDK_DEBUGLOG(nvmf, "Stop keep alive poller\n");
104 	spdk_poller_unregister(&ctrlr->keep_alive_poller);
105 }
106 
107 static void
108 nvmf_ctrlr_stop_association_timer(struct spdk_nvmf_ctrlr *ctrlr)
109 {
110 	if (!ctrlr) {
111 		SPDK_ERRLOG("Controller is NULL\n");
112 		assert(false);
113 		return;
114 	}
115 
116 	if (ctrlr->association_timer == NULL) {
117 		return;
118 	}
119 
120 	SPDK_DEBUGLOG(nvmf, "Stop association timer\n");
121 	spdk_poller_unregister(&ctrlr->association_timer);
122 }
123 
124 static void
125 nvmf_ctrlr_disconnect_qpairs_done(struct spdk_io_channel_iter *i, int status)
126 {
127 	if (status == 0) {
128 		SPDK_DEBUGLOG(nvmf, "ctrlr disconnect qpairs complete successfully\n");
129 	} else {
130 		SPDK_ERRLOG("Fail to disconnect ctrlr qpairs\n");
131 	}
132 }
133 
134 static int
135 _nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i, bool include_admin)
136 {
137 	int rc = 0;
138 	struct spdk_nvmf_ctrlr *ctrlr;
139 	struct spdk_nvmf_qpair *qpair, *temp_qpair;
140 	struct spdk_io_channel *ch;
141 	struct spdk_nvmf_poll_group *group;
142 
143 	ctrlr = spdk_io_channel_iter_get_ctx(i);
144 	ch = spdk_io_channel_iter_get_channel(i);
145 	group = spdk_io_channel_get_ctx(ch);
146 
147 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, temp_qpair) {
148 		if (qpair->ctrlr == ctrlr && (include_admin || !nvmf_qpair_is_admin_queue(qpair))) {
149 			rc = spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
150 			if (rc) {
151 				SPDK_ERRLOG("Qpair disconnect failed\n");
152 				return rc;
153 			}
154 		}
155 	}
156 
157 	return rc;
158 }
159 
160 static void
161 nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i)
162 {
163 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, true));
164 }
165 
166 static void
167 nvmf_ctrlr_disconnect_io_qpairs_on_pg(struct spdk_io_channel_iter *i)
168 {
169 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, false));
170 }
171 
172 static int
173 nvmf_ctrlr_keep_alive_poll(void *ctx)
174 {
175 	uint64_t keep_alive_timeout_tick;
176 	uint64_t now = spdk_get_ticks();
177 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
178 
179 	SPDK_DEBUGLOG(nvmf, "Polling ctrlr keep alive timeout\n");
180 
181 	/* If the Keep alive feature is in use and the timer expires */
182 	keep_alive_timeout_tick = ctrlr->last_keep_alive_tick +
183 				  ctrlr->feat.keep_alive_timer.bits.kato * spdk_get_ticks_hz() / UINT64_C(1000);
184 	if (now > keep_alive_timeout_tick) {
185 		SPDK_NOTICELOG("Disconnecting host %s from subsystem %s due to keep alive timeout.\n",
186 			       ctrlr->hostnqn, ctrlr->subsys->subnqn);
187 		/* set the Controller Fatal Status bit to '1' */
188 		if (ctrlr->vcprop.csts.bits.cfs == 0) {
189 			ctrlr->vcprop.csts.bits.cfs = 1;
190 
191 			/*
192 			 * disconnect qpairs, terminate Transport connection
193 			 * destroy ctrlr, break the host to controller association
194 			 * disconnect qpairs with qpair->ctrlr == ctrlr
195 			 */
196 			spdk_for_each_channel(ctrlr->subsys->tgt,
197 					      nvmf_ctrlr_disconnect_qpairs_on_pg,
198 					      ctrlr,
199 					      nvmf_ctrlr_disconnect_qpairs_done);
200 		}
201 	}
202 
203 	return SPDK_POLLER_BUSY;
204 }
205 
206 static void
207 nvmf_ctrlr_start_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
208 {
209 	if (!ctrlr) {
210 		SPDK_ERRLOG("Controller is NULL\n");
211 		return;
212 	}
213 
214 	/* if cleared to 0 then the Keep Alive Timer is disabled */
215 	if (ctrlr->feat.keep_alive_timer.bits.kato != 0) {
216 
217 		ctrlr->last_keep_alive_tick = spdk_get_ticks();
218 
219 		SPDK_DEBUGLOG(nvmf, "Ctrlr add keep alive poller\n");
220 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
221 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
222 	}
223 }
224 
225 static void
226 ctrlr_add_qpair_and_update_rsp(struct spdk_nvmf_qpair *qpair,
227 			       struct spdk_nvmf_ctrlr *ctrlr,
228 			       struct spdk_nvmf_fabric_connect_rsp *rsp)
229 {
230 	assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
231 
232 	/* check if we would exceed ctrlr connection limit */
233 	if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) {
234 		SPDK_ERRLOG("Requested QID %u but Max QID is %u\n",
235 			    qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1);
236 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
237 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
238 		return;
239 	}
240 
241 	if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) {
242 		SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid);
243 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
244 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
245 		return;
246 	}
247 
248 	qpair->ctrlr = ctrlr;
249 	spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid);
250 
251 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
252 	rsp->status_code_specific.success.cntlid = ctrlr->cntlid;
253 	SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n",
254 		      rsp->status_code_specific.success.cntlid);
255 
256 	SPDK_DTRACE_PROBE4(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn,
257 			   ctrlr->hostnqn);
258 }
259 
260 static void
261 _nvmf_ctrlr_add_admin_qpair(void *ctx)
262 {
263 	struct spdk_nvmf_request *req = ctx;
264 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
265 	struct spdk_nvmf_qpair *qpair = req->qpair;
266 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
267 
268 	ctrlr->admin_qpair = qpair;
269 	ctrlr->association_timeout = qpair->transport->opts.association_timeout;
270 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
271 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
272 	_nvmf_request_complete(req);
273 }
274 
275 static void
276 _nvmf_subsystem_add_ctrlr(void *ctx)
277 {
278 	struct spdk_nvmf_request *req = ctx;
279 	struct spdk_nvmf_qpair *qpair = req->qpair;
280 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
281 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
282 
283 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
284 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
285 		spdk_bit_array_free(&ctrlr->qpair_mask);
286 		free(ctrlr);
287 		qpair->ctrlr = NULL;
288 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
289 		spdk_nvmf_request_complete(req);
290 		return;
291 	}
292 
293 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
294 }
295 
296 static void
297 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
298 		      struct spdk_nvmf_ctrlr_data *cdata)
299 {
300 	cdata->kas = KAS_DEFAULT_VALUE;
301 	cdata->sgls.supported = 1;
302 	cdata->sgls.keyed_sgl = 1;
303 	cdata->sgls.sgl_offset = 1;
304 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
305 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
306 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
307 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
308 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
309 	cdata->nvmf_specific.msdbd = 1;
310 
311 	if (transport->ops->cdata_init) {
312 		transport->ops->cdata_init(transport, subsystem, cdata);
313 	}
314 }
315 
316 static struct spdk_nvmf_ctrlr *
317 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
318 		  struct spdk_nvmf_request *req,
319 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
320 		  struct spdk_nvmf_fabric_connect_data *connect_data)
321 {
322 	struct spdk_nvmf_ctrlr	*ctrlr;
323 	struct spdk_nvmf_transport *transport;
324 	struct spdk_nvme_transport_id listen_trid = {};
325 
326 	ctrlr = calloc(1, sizeof(*ctrlr));
327 	if (ctrlr == NULL) {
328 		SPDK_ERRLOG("Memory allocation failed\n");
329 		return NULL;
330 	}
331 
332 	STAILQ_INIT(&ctrlr->async_events);
333 	TAILQ_INIT(&ctrlr->log_head);
334 	ctrlr->subsys = subsystem;
335 	ctrlr->thread = req->qpair->group->thread;
336 	ctrlr->disconnect_in_progress = false;
337 
338 	transport = req->qpair->transport;
339 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
340 	if (!ctrlr->qpair_mask) {
341 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
342 		goto err_qpair_mask;
343 	}
344 
345 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
346 
347 	/*
348 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
349 	 * If this field is cleared to 0h, then Keep Alive is not supported.
350 	 */
351 	if (ctrlr->cdata.kas) {
352 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
353 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
354 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
355 	}
356 
357 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
358 	if (ctrlr->subsys->flags.ana_reporting) {
359 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
360 	}
361 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
362 	/* Coalescing Disable */
363 	ctrlr->feat.interrupt_vector_configuration.bits.cd = 1;
364 
365 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
366 		/*
367 		 * If keep-alive timeout is not set, discovery controllers use some
368 		 * arbitrary high value in order to cleanup stale discovery sessions
369 		 *
370 		 * From the 1.0a nvme-of spec:
371 		 * "The Keep Alive command is reserved for
372 		 * Discovery controllers. A transport may specify a
373 		 * fixed Discovery controller activity timeout value
374 		 * (e.g., 2 minutes). If no commands are received
375 		 * by a Discovery controller within that time
376 		 * period, the controller may perform the
377 		 * actions for Keep Alive Timer expiration".
378 		 *
379 		 * From the 1.1 nvme-of spec:
380 		 * "A host requests an explicit persistent connection
381 		 * to a Discovery controller and Asynchronous Event Notifications from
382 		 * the Discovery controller on that persistent connection by specifying
383 		 * a non-zero Keep Alive Timer value in the Connect command."
384 		 *
385 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
386 		 * otherwise we disable it and use default discovery controller KATO.
387 		 * KATO is in millisecond.
388 		 */
389 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
390 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
391 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
392 		} else {
393 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
394 		}
395 	}
396 
397 	/* Subtract 1 for admin queue, 1 for 0's based */
398 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
399 			1;
400 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
401 			1;
402 
403 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
404 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn));
405 
406 	ctrlr->vcprop.cap.raw = 0;
407 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
408 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
409 				      1; /* max queue depth */
410 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
411 	ctrlr->vcprop.cap.bits.to = 1; /* ready timeout - 500 msec units */
412 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
413 	ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
414 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
415 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
416 
417 	/* Version Supported: 1.3 */
418 	ctrlr->vcprop.vs.bits.mjr = 1;
419 	ctrlr->vcprop.vs.bits.mnr = 3;
420 	ctrlr->vcprop.vs.bits.ter = 0;
421 
422 	ctrlr->vcprop.cc.raw = 0;
423 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
424 
425 	ctrlr->vcprop.csts.raw = 0;
426 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
427 
428 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
429 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
430 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
431 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
432 
433 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
434 
435 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
436 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
437 			SPDK_ERRLOG("Could not get listener transport ID\n");
438 			goto err_listener;
439 		}
440 
441 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid);
442 		if (!ctrlr->listener) {
443 			SPDK_ERRLOG("Listener was not found\n");
444 			goto err_listener;
445 		}
446 	}
447 
448 	req->qpair->ctrlr = ctrlr;
449 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
450 
451 	return ctrlr;
452 err_listener:
453 	spdk_bit_array_free(&ctrlr->qpair_mask);
454 err_qpair_mask:
455 	free(ctrlr);
456 	return NULL;
457 }
458 
459 static void
460 _nvmf_ctrlr_destruct(void *ctx)
461 {
462 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
463 	struct spdk_nvmf_reservation_log *log, *log_tmp;
464 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
465 
466 	if (ctrlr->disconnect_in_progress) {
467 		SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n");
468 		spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
469 		return;
470 	}
471 
472 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
473 	nvmf_ctrlr_stop_association_timer(ctrlr);
474 	spdk_bit_array_free(&ctrlr->qpair_mask);
475 
476 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
477 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
478 		free(log);
479 	}
480 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
481 		STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link);
482 		free(event);
483 	}
484 	free(ctrlr);
485 }
486 
487 void
488 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
489 {
490 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
491 
492 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
493 }
494 
495 static void
496 nvmf_ctrlr_add_io_qpair(void *ctx)
497 {
498 	struct spdk_nvmf_request *req = ctx;
499 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
500 	struct spdk_nvmf_qpair *qpair = req->qpair;
501 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
502 	struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair;
503 
504 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
505 	  * For error case, the value should be NULL. So set it to NULL at first.
506 	  */
507 	qpair->ctrlr = NULL;
508 
509 	/* Make sure the controller is not being destroyed. */
510 	if (ctrlr->in_destruct) {
511 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
512 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
513 		goto end;
514 	}
515 
516 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
517 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
518 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
519 		goto end;
520 	}
521 
522 	if (!ctrlr->vcprop.cc.bits.en) {
523 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
524 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
525 		goto end;
526 	}
527 
528 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
529 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
530 			    ctrlr->vcprop.cc.bits.iosqes);
531 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
532 		goto end;
533 	}
534 
535 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
536 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
537 			    ctrlr->vcprop.cc.bits.iocqes);
538 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
539 		goto end;
540 	}
541 
542 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
543 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
544 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
545 		 * state to DEACTIVATING and removing it from poll group */
546 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
547 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
548 		goto end;
549 	}
550 
551 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
552 end:
553 	spdk_nvmf_request_complete(req);
554 }
555 
556 static void
557 _nvmf_ctrlr_add_io_qpair(void *ctx)
558 {
559 	struct spdk_nvmf_request *req = ctx;
560 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
561 	struct spdk_nvmf_fabric_connect_data *data = req->data;
562 	struct spdk_nvmf_ctrlr *ctrlr;
563 	struct spdk_nvmf_qpair *qpair = req->qpair;
564 	struct spdk_nvmf_qpair *admin_qpair;
565 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
566 	struct spdk_nvmf_subsystem *subsystem;
567 	struct spdk_nvme_transport_id listen_trid = {};
568 	const struct spdk_nvmf_subsystem_listener *listener;
569 
570 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
571 
572 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
573 	/* We already checked this in spdk_nvmf_ctrlr_connect */
574 	assert(subsystem != NULL);
575 
576 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
577 	if (ctrlr == NULL) {
578 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
579 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
580 		spdk_nvmf_request_complete(req);
581 		return;
582 	}
583 
584 	/* fail before passing a message to the controller thread. */
585 	if (ctrlr->in_destruct) {
586 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
587 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
588 		spdk_nvmf_request_complete(req);
589 		return;
590 	}
591 
592 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
593 	if (subsystem->flags.ana_reporting) {
594 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
595 			SPDK_ERRLOG("Could not get listener transport ID\n");
596 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
597 			spdk_nvmf_request_complete(req);
598 			return;
599 		}
600 
601 		listener = nvmf_subsystem_find_listener(subsystem, &listen_trid);
602 		if (listener != ctrlr->listener) {
603 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
604 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
605 			spdk_nvmf_request_complete(req);
606 			return;
607 		}
608 	}
609 
610 	admin_qpair = ctrlr->admin_qpair;
611 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
612 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
613 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
614 		 * state to DEACTIVATING and removing it from poll group */
615 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
616 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
617 		spdk_nvmf_request_complete(req);
618 		return;
619 	}
620 	qpair->ctrlr = ctrlr;
621 	spdk_thread_send_msg(admin_qpair->group->thread, nvmf_ctrlr_add_io_qpair, req);
622 }
623 
624 static bool
625 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
626 			  const char *hostnqn)
627 {
628 	struct spdk_nvme_transport_id listen_trid = {};
629 
630 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
631 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
632 		return false;
633 	}
634 
635 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
636 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
637 			    subsystem->subnqn);
638 		return false;
639 	}
640 
641 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
642 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
643 			    subsystem->subnqn, hostnqn);
644 		return false;
645 	}
646 
647 	return true;
648 }
649 
650 static int
651 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
652 {
653 	struct spdk_nvmf_fabric_connect_data *data = req->data;
654 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
655 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
656 	struct spdk_nvmf_qpair *qpair = req->qpair;
657 	struct spdk_nvmf_transport *transport = qpair->transport;
658 	struct spdk_nvmf_ctrlr *ctrlr;
659 	struct spdk_nvmf_subsystem *subsystem;
660 
661 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
662 		      cmd->recfmt, cmd->qid, cmd->sqsize);
663 
664 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
665 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
666 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
667 		      ntohl(*(uint32_t *)&data->hostid[0]),
668 		      ntohs(*(uint16_t *)&data->hostid[4]),
669 		      ntohs(*(uint16_t *)&data->hostid[6]),
670 		      data->hostid[8],
671 		      data->hostid[9],
672 		      ntohs(*(uint16_t *)&data->hostid[10]),
673 		      ntohl(*(uint32_t *)&data->hostid[12]));
674 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
675 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
676 
677 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
678 	if (!subsystem) {
679 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
680 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
681 	}
682 
683 	if (cmd->recfmt != 0) {
684 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
685 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
686 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
687 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
688 	}
689 
690 	/*
691 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
692 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
693 	 */
694 	if (cmd->sqsize == 0) {
695 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
696 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
697 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
698 	}
699 
700 	if (cmd->qid == 0) {
701 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
702 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
703 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
704 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
705 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
706 		}
707 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
708 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
709 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
710 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
711 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
712 	}
713 
714 	qpair->sq_head_max = cmd->sqsize;
715 	qpair->qid = cmd->qid;
716 
717 	if (0 == qpair->qid) {
718 		qpair->group->stat.admin_qpairs++;
719 		qpair->group->stat.current_admin_qpairs++;
720 	} else {
721 		qpair->group->stat.io_qpairs++;
722 		qpair->group->stat.current_io_qpairs++;
723 	}
724 
725 	if (cmd->qid == 0) {
726 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
727 
728 		if (data->cntlid != 0xFFFF) {
729 			/* This NVMf target only supports dynamic mode. */
730 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
731 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
732 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
733 		}
734 
735 		/* Establish a new ctrlr */
736 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
737 		if (!ctrlr) {
738 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
739 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
740 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
741 		} else {
742 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
743 		}
744 	} else {
745 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
746 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
747 	}
748 }
749 
750 static inline bool
751 nvmf_request_is_fabric_connect(struct spdk_nvmf_request *req)
752 {
753 	return req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC &&
754 	       req->cmd->nvmf_cmd.fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT;
755 }
756 
757 static struct spdk_nvmf_subsystem_poll_group *
758 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
759 {
760 	struct spdk_nvmf_fabric_connect_data *data;
761 	struct spdk_nvmf_subsystem *subsystem;
762 	struct spdk_nvmf_tgt *tgt;
763 
764 	assert(nvmf_request_is_fabric_connect(req));
765 	assert(req->qpair->ctrlr == NULL);
766 
767 	data = req->data;
768 	tgt = req->qpair->transport->tgt;
769 
770 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
771 	if (subsystem == NULL) {
772 		return NULL;
773 	}
774 
775 	return &req->qpair->group->sgroups[subsystem->id];
776 }
777 
778 static void
779 nvmf_add_to_outstanding_queue(struct spdk_nvmf_request *req)
780 {
781 	if (!spdk_nvmf_using_zcopy(req->zcopy_phase)) {
782 		/* if using zcopy then request has been added when the start zcopy was actioned */
783 		struct spdk_nvmf_qpair *qpair = req->qpair;
784 		TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
785 	}
786 }
787 
788 int
789 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
790 {
791 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
792 	struct spdk_nvmf_subsystem_poll_group *sgroup;
793 	enum spdk_nvmf_request_exec_status status;
794 
795 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
796 	if (!sgroup) {
797 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
798 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
799 		goto out;
800 	}
801 
802 	sgroup->mgmt_io_outstanding++;
803 	nvmf_add_to_outstanding_queue(req);
804 
805 	status = _nvmf_ctrlr_connect(req);
806 
807 out:
808 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
809 		_nvmf_request_complete(req);
810 	}
811 
812 	return status;
813 }
814 
815 static int nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req);
816 
817 static int
818 retry_connect(void *arg)
819 {
820 	struct spdk_nvmf_request *req = arg;
821 	struct spdk_nvmf_subsystem_poll_group *sgroup;
822 	int rc;
823 
824 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
825 	assert(sgroup != NULL);
826 	sgroup->mgmt_io_outstanding++;
827 	spdk_poller_unregister(&req->poller);
828 	rc = nvmf_ctrlr_cmd_connect(req);
829 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
830 		_nvmf_request_complete(req);
831 	}
832 	return SPDK_POLLER_BUSY;
833 }
834 
835 static int
836 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
837 {
838 	struct spdk_nvmf_fabric_connect_data *data = req->data;
839 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
840 	struct spdk_nvmf_transport *transport = req->qpair->transport;
841 	struct spdk_nvmf_subsystem *subsystem;
842 
843 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
844 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
845 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
846 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
847 	}
848 
849 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
850 	if (!subsystem) {
851 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
852 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
853 	}
854 
855 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
856 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
857 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
858 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
859 		struct spdk_nvmf_subsystem_poll_group *sgroup;
860 
861 		if (req->timeout_tsc == 0) {
862 			/* We will only retry the request up to 1 second. */
863 			req->timeout_tsc = spdk_get_ticks() + spdk_get_ticks_hz();
864 		} else if (spdk_get_ticks() > req->timeout_tsc) {
865 			SPDK_ERRLOG("Subsystem '%s' was not ready for 1 second\n", subsystem->subnqn);
866 			rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
867 			rsp->status.sc = SPDK_NVMF_FABRIC_SC_CONTROLLER_BUSY;
868 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
869 		}
870 
871 		/* Subsystem is not ready to handle a connect. Use a poller to retry it
872 		 * again later. Decrement the mgmt_io_outstanding to avoid the
873 		 * subsystem waiting for this command to complete before unpausing.
874 		 */
875 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
876 		assert(sgroup != NULL);
877 		sgroup->mgmt_io_outstanding--;
878 		SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn);
879 		req->poller = SPDK_POLLER_REGISTER(retry_connect, req, 100);
880 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
881 	}
882 
883 	/* Ensure that hostnqn is null terminated */
884 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
885 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
886 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
887 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
888 	}
889 
890 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
891 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
892 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
893 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
894 	}
895 
896 	return _nvmf_ctrlr_connect(req);
897 }
898 
899 static int
900 nvmf_ctrlr_association_remove(void *ctx)
901 {
902 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
903 	int rc;
904 
905 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
906 		      ctrlr->subsys->subnqn);
907 
908 	rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL);
909 	if (rc < 0) {
910 		SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
911 		assert(false);
912 	}
913 
914 	nvmf_ctrlr_stop_association_timer(ctrlr);
915 	return 1;
916 }
917 
918 static void
919 nvmf_ctrlr_cc_shn_done(struct spdk_io_channel_iter *i, int status)
920 {
921 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
922 
923 	if (status < 0) {
924 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
925 		assert(false);
926 	}
927 
928 	ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
929 
930 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
931 	 * between the host and controller shall be preserved for at least 2 minutes */
932 	if (ctrlr->association_timer) {
933 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
934 		nvmf_ctrlr_stop_association_timer(ctrlr);
935 	}
936 	if (ctrlr->association_timeout) {
937 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
938 					   ctrlr->association_timeout * 1000);
939 	}
940 	ctrlr->disconnect_in_progress = false;
941 }
942 
943 static void
944 nvmf_ctrlr_cc_reset_done(struct spdk_io_channel_iter *i, int status)
945 {
946 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
947 
948 	if (status < 0) {
949 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
950 		assert(false);
951 	}
952 
953 	/* Only a subset of the registers are cleared out on a reset */
954 	ctrlr->vcprop.cc.raw = 0;
955 	ctrlr->vcprop.csts.raw = 0;
956 
957 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
958 	 * between the host and controller shall be preserved for at least 2 minutes */
959 	if (ctrlr->association_timer) {
960 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
961 		nvmf_ctrlr_stop_association_timer(ctrlr);
962 	}
963 	if (ctrlr->association_timeout) {
964 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
965 					   ctrlr->association_timeout * 1000);
966 	}
967 	ctrlr->disconnect_in_progress = false;
968 }
969 
970 const struct spdk_nvmf_registers *
971 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
972 {
973 	return &ctrlr->vcprop;
974 }
975 
976 static uint64_t
977 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
978 {
979 	return ctrlr->vcprop.cap.raw;
980 }
981 
982 static uint64_t
983 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
984 {
985 	return ctrlr->vcprop.vs.raw;
986 }
987 
988 static uint64_t
989 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
990 {
991 	return ctrlr->vcprop.cc.raw;
992 }
993 
994 static bool
995 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
996 {
997 	union spdk_nvme_cc_register cc, diff;
998 
999 	cc.raw = value;
1000 
1001 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
1002 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
1003 
1004 	/*
1005 	 * Calculate which bits changed between the current and new CC.
1006 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
1007 	 */
1008 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
1009 
1010 	if (diff.bits.en) {
1011 		if (cc.bits.en) {
1012 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
1013 			nvmf_ctrlr_stop_association_timer(ctrlr);
1014 
1015 			ctrlr->vcprop.cc.bits.en = 1;
1016 			ctrlr->vcprop.csts.bits.rdy = 1;
1017 		} else {
1018 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
1019 			ctrlr->vcprop.cc.bits.en = 0;
1020 			ctrlr->disconnect_in_progress = true;
1021 			spdk_for_each_channel(ctrlr->subsys->tgt,
1022 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1023 					      ctrlr,
1024 					      nvmf_ctrlr_cc_reset_done);
1025 		}
1026 		diff.bits.en = 0;
1027 	}
1028 
1029 	if (diff.bits.shn) {
1030 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
1031 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
1032 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
1033 				      cc.bits.shn >> 1, cc.bits.shn & 1);
1034 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
1035 			ctrlr->disconnect_in_progress = true;
1036 			spdk_for_each_channel(ctrlr->subsys->tgt,
1037 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1038 					      ctrlr,
1039 					      nvmf_ctrlr_cc_shn_done);
1040 
1041 			/* From the time a shutdown is initiated the controller shall disable
1042 			 * Keep Alive timer */
1043 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
1044 		} else if (cc.bits.shn == 0) {
1045 			ctrlr->vcprop.cc.bits.shn = 0;
1046 		} else {
1047 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
1048 				    cc.bits.shn >> 1, cc.bits.shn & 1);
1049 			return false;
1050 		}
1051 		diff.bits.shn = 0;
1052 	}
1053 
1054 	if (diff.bits.iosqes) {
1055 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
1056 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
1057 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
1058 		diff.bits.iosqes = 0;
1059 	}
1060 
1061 	if (diff.bits.iocqes) {
1062 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
1063 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
1064 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
1065 		diff.bits.iocqes = 0;
1066 	}
1067 
1068 	if (diff.bits.ams) {
1069 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
1070 		return false;
1071 	}
1072 
1073 	if (diff.bits.mps) {
1074 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
1075 		return false;
1076 	}
1077 
1078 	if (diff.bits.css) {
1079 		SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
1080 		return false;
1081 	}
1082 
1083 	if (diff.raw != 0) {
1084 		/* Print an error message, but don't fail the command in this case.
1085 		 * If we did want to fail in this case, we'd need to ensure we acted
1086 		 * on no other bits or the initiator gets confused. */
1087 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
1088 	}
1089 
1090 	return true;
1091 }
1092 
1093 static uint64_t
1094 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
1095 {
1096 	return ctrlr->vcprop.csts.raw;
1097 }
1098 
1099 static uint64_t
1100 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
1101 {
1102 	return ctrlr->vcprop.aqa.raw;
1103 }
1104 
1105 static bool
1106 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1107 {
1108 	union spdk_nvme_aqa_register aqa;
1109 
1110 	aqa.raw = value;
1111 
1112 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1113 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1114 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1115 		return false;
1116 	}
1117 
1118 	ctrlr->vcprop.aqa.raw = value;
1119 
1120 	return true;
1121 }
1122 
1123 static uint64_t
1124 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1125 {
1126 	return ctrlr->vcprop.asq;
1127 }
1128 
1129 static bool
1130 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1131 {
1132 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1133 
1134 	return true;
1135 }
1136 
1137 static bool
1138 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1139 {
1140 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1141 
1142 	return true;
1143 }
1144 
1145 static uint64_t
1146 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1147 {
1148 	return ctrlr->vcprop.acq;
1149 }
1150 
1151 static bool
1152 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1153 {
1154 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1155 
1156 	return true;
1157 }
1158 
1159 static bool
1160 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1161 {
1162 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1163 
1164 	return true;
1165 }
1166 
1167 struct nvmf_prop {
1168 	uint32_t ofst;
1169 	uint8_t size;
1170 	char name[11];
1171 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1172 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1173 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1174 };
1175 
1176 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1177 	{ \
1178 		offsetof(struct spdk_nvme_registers, field), \
1179 		size, \
1180 		#field, \
1181 		get_cb, set_cb, set_upper_cb \
1182 	}
1183 
1184 static const struct nvmf_prop nvmf_props[] = {
1185 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1186 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1187 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1188 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1189 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1190 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1191 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1192 };
1193 
1194 static const struct nvmf_prop *
1195 find_prop(uint32_t ofst, uint8_t size)
1196 {
1197 	size_t i;
1198 
1199 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1200 		const struct nvmf_prop *prop = &nvmf_props[i];
1201 
1202 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1203 			return prop;
1204 		}
1205 	}
1206 
1207 	return NULL;
1208 }
1209 
1210 static int
1211 nvmf_property_get(struct spdk_nvmf_request *req)
1212 {
1213 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1214 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1215 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1216 	const struct nvmf_prop *prop;
1217 	uint8_t size;
1218 
1219 	response->status.sc = 0;
1220 	response->value.u64 = 0;
1221 
1222 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1223 		      cmd->attrib.size, cmd->ofst);
1224 
1225 	switch (cmd->attrib.size) {
1226 	case SPDK_NVMF_PROP_SIZE_4:
1227 		size = 4;
1228 		break;
1229 	case SPDK_NVMF_PROP_SIZE_8:
1230 		size = 8;
1231 		break;
1232 	default:
1233 		SPDK_ERRLOG("Invalid size value %d\n", cmd->attrib.size);
1234 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1235 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1236 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1237 	}
1238 
1239 	prop = find_prop(cmd->ofst, size);
1240 	if (prop == NULL || prop->get_cb == NULL) {
1241 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1242 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1243 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1244 	}
1245 
1246 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1247 
1248 	response->value.u64 = prop->get_cb(ctrlr);
1249 
1250 	if (size != prop->size) {
1251 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1252 		assert(size == 4);
1253 		assert(prop->size == 8);
1254 
1255 		if (cmd->ofst == prop->ofst) {
1256 			/* Keep bottom 4 bytes only */
1257 			response->value.u64 &= 0xFFFFFFFF;
1258 		} else {
1259 			/* Keep top 4 bytes only */
1260 			response->value.u64 >>= 32;
1261 		}
1262 	}
1263 
1264 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1265 
1266 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1267 }
1268 
1269 static int
1270 nvmf_property_set(struct spdk_nvmf_request *req)
1271 {
1272 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1273 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1274 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1275 	const struct nvmf_prop *prop;
1276 	uint64_t value;
1277 	uint8_t size;
1278 	bool ret;
1279 
1280 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1281 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1282 
1283 	switch (cmd->attrib.size) {
1284 	case SPDK_NVMF_PROP_SIZE_4:
1285 		size = 4;
1286 		break;
1287 	case SPDK_NVMF_PROP_SIZE_8:
1288 		size = 8;
1289 		break;
1290 	default:
1291 		SPDK_ERRLOG("Invalid size value %d\n", cmd->attrib.size);
1292 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1293 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1294 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1295 	}
1296 
1297 	prop = find_prop(cmd->ofst, size);
1298 	if (prop == NULL || prop->set_cb == NULL) {
1299 		SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst);
1300 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1301 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1302 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1303 	}
1304 
1305 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1306 
1307 	value = cmd->value.u64;
1308 
1309 	if (prop->size == 4) {
1310 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1311 	} else if (size != prop->size) {
1312 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1313 		assert(size == 4);
1314 		assert(prop->size == 8);
1315 
1316 		if (cmd->ofst == prop->ofst) {
1317 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1318 		} else {
1319 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1320 		}
1321 	} else {
1322 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1323 		if (ret) {
1324 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1325 		}
1326 	}
1327 
1328 	if (!ret) {
1329 		SPDK_ERRLOG("prop set_cb failed\n");
1330 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1331 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1332 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1333 	}
1334 
1335 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1336 }
1337 
1338 static int
1339 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1340 {
1341 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1342 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1343 
1344 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1345 
1346 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1347 	ctrlr->feat.arbitration.bits.reserved = 0;
1348 
1349 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1350 }
1351 
1352 static int
1353 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1354 {
1355 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1356 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1357 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1358 
1359 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1360 
1361 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1362 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1363 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1364 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1365 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1366 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1367 	}
1368 
1369 	ctrlr->feat.power_management.raw = cmd->cdw11;
1370 	ctrlr->feat.power_management.bits.reserved = 0;
1371 
1372 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1373 }
1374 
1375 static bool
1376 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1377 {
1378 	/*
1379 	 * Valid TMPSEL values:
1380 	 *  0000b - 1000b: temperature sensors
1381 	 *  1111b: set all implemented temperature sensors
1382 	 */
1383 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1384 		/* 1001b - 1110b: reserved */
1385 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1386 		return false;
1387 	}
1388 
1389 	/*
1390 	 * Valid THSEL values:
1391 	 *  00b: over temperature threshold
1392 	 *  01b: under temperature threshold
1393 	 */
1394 	if (opts->bits.thsel > 1) {
1395 		/* 10b - 11b: reserved */
1396 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1397 		return false;
1398 	}
1399 
1400 	return true;
1401 }
1402 
1403 static int
1404 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1405 {
1406 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1407 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1408 
1409 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1410 
1411 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1412 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1413 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1414 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1415 	}
1416 
1417 	/* TODO: no sensors implemented - ignore new values */
1418 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1419 }
1420 
1421 static int
1422 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1423 {
1424 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1425 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1426 
1427 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1428 
1429 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1430 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1431 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1432 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1433 	}
1434 
1435 	/* TODO: no sensors implemented - return 0 for all thresholds */
1436 	rsp->cdw0 = 0;
1437 
1438 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1439 }
1440 
1441 static int
1442 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req)
1443 {
1444 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1445 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1446 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1447 	union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {};
1448 
1449 	SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11);
1450 
1451 	iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv;
1452 	iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd;
1453 	rsp->cdw0 = iv_conf.raw;
1454 
1455 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1456 }
1457 
1458 static int
1459 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1460 {
1461 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1462 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1463 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1464 
1465 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1466 
1467 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1468 		/*
1469 		 * Host is not allowed to set this bit, since we don't advertise it in
1470 		 * Identify Namespace.
1471 		 */
1472 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1473 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1474 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1475 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1476 	}
1477 
1478 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1479 	ctrlr->feat.error_recovery.bits.reserved = 0;
1480 
1481 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1482 }
1483 
1484 static int
1485 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1486 {
1487 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1488 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1489 
1490 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1491 
1492 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1493 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1494 
1495 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1496 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1497 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1498 }
1499 
1500 static int
1501 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1502 {
1503 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1504 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1505 
1506 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1507 
1508 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1509 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1510 
1511 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1512 }
1513 
1514 static int
1515 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1516 {
1517 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1518 
1519 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1520 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1521 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1522 }
1523 
1524 static int
1525 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1526 {
1527 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1528 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1529 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1530 
1531 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1532 
1533 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1534 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1535 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1536 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1537 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1538 	}
1539 
1540 	if (req->data == NULL || req->length < sizeof(ctrlr->hostid)) {
1541 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1542 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1543 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1544 	}
1545 
1546 	spdk_uuid_copy((struct spdk_uuid *)req->data, &ctrlr->hostid);
1547 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1548 }
1549 
1550 static int
1551 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1552 {
1553 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1554 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1555 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1556 	struct spdk_nvmf_ns *ns;
1557 
1558 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notificaton Mask\n");
1559 
1560 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1561 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1562 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1563 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1564 	}
1565 
1566 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1567 	if (ns == NULL) {
1568 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1569 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1570 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1571 	}
1572 	rsp->cdw0 = ns->mask;
1573 
1574 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1575 }
1576 
1577 static int
1578 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1579 {
1580 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1581 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1582 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1583 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1584 	struct spdk_nvmf_ns *ns;
1585 
1586 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notificaton Mask\n");
1587 
1588 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1589 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1590 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1591 			ns->mask = cmd->cdw11;
1592 		}
1593 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1594 	}
1595 
1596 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1597 	if (ns == NULL) {
1598 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1599 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1600 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1601 	}
1602 	ns->mask = cmd->cdw11;
1603 
1604 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1605 }
1606 
1607 static int
1608 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1609 {
1610 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1611 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1612 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1613 	struct spdk_nvmf_ns *ns;
1614 
1615 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1616 
1617 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1618 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1619 	if (ns == NULL) {
1620 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1621 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1622 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1623 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1624 	}
1625 
1626 	response->cdw0 = ns->ptpl_activated;
1627 
1628 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1629 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1630 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1631 }
1632 
1633 static int
1634 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req)
1635 {
1636 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1637 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1638 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1639 	struct spdk_nvmf_ns *ns;
1640 	bool ptpl;
1641 
1642 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1643 
1644 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1645 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1646 
1647 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && ns->ptpl_file) {
1648 		ns->ptpl_activated = ptpl;
1649 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1650 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns && ns->ptpl_file;
1651 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1652 			ns->ptpl_activated = ptpl;
1653 		}
1654 	} else {
1655 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1656 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1657 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1658 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1659 	}
1660 
1661 	/* TODO: Feature not changeable for now */
1662 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1663 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1664 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1665 }
1666 
1667 static int
1668 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req)
1669 {
1670 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1671 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1672 	struct spdk_nvme_host_behavior *host_behavior;
1673 
1674 	SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n");
1675 	if (req->iovcnt != 1) {
1676 		SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt);
1677 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1678 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1679 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1680 	}
1681 	if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) {
1682 		SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len);
1683 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1684 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1685 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1686 	}
1687 
1688 	host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base;
1689 	if (host_behavior->acre == 0) {
1690 		ctrlr->acre_enabled = false;
1691 	} else if (host_behavior->acre == 1) {
1692 		ctrlr->acre_enabled = true;
1693 	} else {
1694 		SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre);
1695 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1696 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1697 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1698 	}
1699 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1700 }
1701 
1702 static int
1703 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
1704 {
1705 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1706 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1707 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1708 
1709 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
1710 
1711 	/*
1712 	 * if attempts to disable keep alive by setting kato to 0h
1713 	 * a status value of keep alive invalid shall be returned
1714 	 */
1715 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
1716 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
1717 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
1718 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
1719 	} else {
1720 		/* round up to milliseconds */
1721 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
1722 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
1723 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
1724 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
1725 	}
1726 
1727 	/*
1728 	 * if change the keep alive timeout value successfully
1729 	 * update the keep alive poller.
1730 	 */
1731 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
1732 		if (ctrlr->keep_alive_poller != NULL) {
1733 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
1734 		}
1735 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
1736 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
1737 	}
1738 
1739 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
1740 		      ctrlr->feat.keep_alive_timer.bits.kato);
1741 
1742 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1743 }
1744 
1745 static int
1746 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
1747 {
1748 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1749 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1750 	uint32_t count;
1751 
1752 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
1753 		      req->cmd->nvme_cmd.cdw11);
1754 
1755 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1756 	/* verify that the controller is ready to process commands */
1757 	if (count > 1) {
1758 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
1759 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1760 	} else {
1761 		/*
1762 		 * Ignore the value requested by the host -
1763 		 * always return the pre-configured value based on max_qpairs_allowed.
1764 		 */
1765 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
1766 	}
1767 
1768 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1769 }
1770 
1771 static int
1772 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
1773 {
1774 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1775 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1776 
1777 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
1778 		      cmd->cdw11);
1779 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
1780 	ctrlr->feat.async_event_configuration.bits.reserved1 = 0;
1781 	ctrlr->feat.async_event_configuration.bits.reserved2 = 0;
1782 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1783 }
1784 
1785 static int
1786 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
1787 {
1788 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1789 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1790 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1791 	struct spdk_nvmf_async_event_completion *pending_event;
1792 
1793 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
1794 
1795 	/* AER cmd is an exception */
1796 	sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
1797 	assert(sgroup != NULL);
1798 	sgroup->mgmt_io_outstanding--;
1799 
1800 	/* Four asynchronous events are supported for now */
1801 	if (ctrlr->nr_aer_reqs >= NVMF_MAX_ASYNC_EVENTS) {
1802 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
1803 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1804 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
1805 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1806 	}
1807 
1808 	if (!STAILQ_EMPTY(&ctrlr->async_events)) {
1809 		pending_event = STAILQ_FIRST(&ctrlr->async_events);
1810 		rsp->cdw0 = pending_event->event.raw;
1811 		STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link);
1812 		free(pending_event);
1813 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1814 	}
1815 
1816 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
1817 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
1818 }
1819 
1820 struct copy_iovs_ctx {
1821 	struct iovec *iovs;
1822 	int iovcnt;
1823 	int cur_iov_idx;
1824 	size_t cur_iov_offset;
1825 };
1826 
1827 static void
1828 _init_copy_iovs_ctx(struct copy_iovs_ctx *copy_ctx, struct iovec *iovs, int iovcnt)
1829 {
1830 	int iov_idx = 0;
1831 	struct iovec *iov;
1832 
1833 	copy_ctx->iovs = iovs;
1834 	copy_ctx->iovcnt = iovcnt;
1835 	copy_ctx->cur_iov_idx = 0;
1836 	copy_ctx->cur_iov_offset = 0;
1837 
1838 	while (iov_idx < copy_ctx->iovcnt) {
1839 		iov = &copy_ctx->iovs[iov_idx];
1840 		memset(iov->iov_base, 0, iov->iov_len);
1841 		iov_idx++;
1842 	}
1843 }
1844 
1845 static size_t
1846 _copy_buf_to_iovs(struct copy_iovs_ctx *copy_ctx, const void *buf, size_t buf_len)
1847 {
1848 	size_t len, iov_remain_len, copied_len = 0;
1849 	struct iovec *iov;
1850 
1851 	if (buf_len == 0) {
1852 		return 0;
1853 	}
1854 
1855 	while (copy_ctx->cur_iov_idx < copy_ctx->iovcnt) {
1856 		iov = &copy_ctx->iovs[copy_ctx->cur_iov_idx];
1857 		iov_remain_len = iov->iov_len - copy_ctx->cur_iov_offset;
1858 		if (iov_remain_len == 0) {
1859 			copy_ctx->cur_iov_idx++;
1860 			copy_ctx->cur_iov_offset = 0;
1861 			continue;
1862 		}
1863 
1864 		len = spdk_min(iov_remain_len, buf_len - copied_len);
1865 		memcpy((char *)iov->iov_base + copy_ctx->cur_iov_offset,
1866 		       (const char *)buf + copied_len,
1867 		       len);
1868 		copied_len += len;
1869 		copy_ctx->cur_iov_offset += len;
1870 
1871 		if (buf_len == copied_len) {
1872 			return copied_len;
1873 		}
1874 	}
1875 
1876 	return copied_len;
1877 }
1878 
1879 static void
1880 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length)
1881 {
1882 	struct spdk_nvme_firmware_page fw_page;
1883 	size_t copy_len;
1884 	struct copy_iovs_ctx copy_ctx;
1885 
1886 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
1887 
1888 	memset(&fw_page, 0, sizeof(fw_page));
1889 	fw_page.afi.active_slot = 1;
1890 	fw_page.afi.next_reset_slot = 0;
1891 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
1892 
1893 	if (offset < sizeof(fw_page)) {
1894 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
1895 		if (copy_len > 0) {
1896 			_copy_buf_to_iovs(&copy_ctx, (const char *)&fw_page + offset, copy_len);
1897 		}
1898 	}
1899 }
1900 
1901 /*
1902  * Asynchronous Event Mask Bit
1903  */
1904 enum spdk_nvme_async_event_mask_bit {
1905 	/* Mask Namespace Change Notificaton */
1906 	SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT		= 0,
1907 	/* Mask Asymmetric Namespace Access Change Notification */
1908 	SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT		= 1,
1909 	/* Mask Discovery Log Change Notification */
1910 	SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT	= 2,
1911 	/* Mask Reservation Log Page Available Notification */
1912 	SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT	= 3,
1913 
1914 	/* 4 - 63 Reserved */
1915 };
1916 
1917 static inline void
1918 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr,
1919 		      enum spdk_nvme_async_event_mask_bit mask)
1920 {
1921 	ctrlr->notice_aen_mask &= ~(1 << mask);
1922 }
1923 
1924 static inline bool
1925 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr,
1926 		    enum spdk_nvme_async_event_mask_bit mask)
1927 {
1928 	if (ctrlr->notice_aen_mask & (1 << mask)) {
1929 		return false;
1930 	} else {
1931 		ctrlr->notice_aen_mask |= (1 << mask);
1932 		return true;
1933 	}
1934 }
1935 
1936 #define SPDK_NVMF_ANA_DESC_SIZE	(sizeof(struct spdk_nvme_ana_group_descriptor) +	\
1937 				 sizeof(uint32_t))
1938 static void
1939 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
1940 		      uint64_t offset, uint32_t length, uint32_t rae)
1941 {
1942 	struct spdk_nvme_ana_page ana_hdr;
1943 	char _ana_desc[SPDK_NVMF_ANA_DESC_SIZE];
1944 	struct spdk_nvme_ana_group_descriptor *ana_desc;
1945 	size_t copy_len, copied_len;
1946 	uint32_t num_ns = 0;
1947 	struct spdk_nvmf_ns *ns;
1948 	struct copy_iovs_ctx copy_ctx;
1949 
1950 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
1951 
1952 	if (length == 0) {
1953 		return;
1954 	}
1955 
1956 	if (offset >= sizeof(ana_hdr)) {
1957 		offset -= sizeof(ana_hdr);
1958 	} else {
1959 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1960 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1961 			num_ns++;
1962 		}
1963 
1964 		memset(&ana_hdr, 0, sizeof(ana_hdr));
1965 
1966 		ana_hdr.num_ana_group_desc = num_ns;
1967 		/* TODO: Support Change Count. */
1968 		ana_hdr.change_count = 0;
1969 
1970 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
1971 		copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ana_hdr + offset, copy_len);
1972 		assert(copied_len == copy_len);
1973 		length -= copied_len;
1974 		offset = 0;
1975 	}
1976 
1977 	if (length == 0) {
1978 		return;
1979 	}
1980 
1981 	ana_desc = (void *)_ana_desc;
1982 
1983 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1984 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1985 		if (offset >= SPDK_NVMF_ANA_DESC_SIZE) {
1986 			offset -= SPDK_NVMF_ANA_DESC_SIZE;
1987 			continue;
1988 		}
1989 
1990 		memset(ana_desc, 0, SPDK_NVMF_ANA_DESC_SIZE);
1991 
1992 		ana_desc->ana_group_id = ns->nsid;
1993 		ana_desc->num_of_nsid = 1;
1994 		ana_desc->ana_state = ctrlr->listener->ana_state;
1995 		ana_desc->nsid[0] = ns->nsid;
1996 		/* TODO: Support Change Count. */
1997 		ana_desc->change_count = 0;
1998 
1999 		copy_len = spdk_min(SPDK_NVMF_ANA_DESC_SIZE - offset, length);
2000 		copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)ana_desc + offset, copy_len);
2001 		assert(copied_len == copy_len);
2002 		length -= copied_len;
2003 		offset = 0;
2004 
2005 		if (length == 0) {
2006 			goto done;
2007 		}
2008 	}
2009 
2010 done:
2011 	if (!rae) {
2012 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT);
2013 	}
2014 }
2015 
2016 void
2017 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2018 {
2019 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
2020 	uint16_t i;
2021 	bool found = false;
2022 
2023 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
2024 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
2025 			/* nsid is already in the list */
2026 			found = true;
2027 			break;
2028 		}
2029 	}
2030 
2031 	if (!found) {
2032 		if (ctrlr->changed_ns_list_count == max_changes) {
2033 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
2034 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
2035 			for (i = 1; i < max_changes; i++) {
2036 				ctrlr->changed_ns_list.ns_list[i] = 0;
2037 			}
2038 		} else {
2039 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
2040 		}
2041 	}
2042 }
2043 
2044 static void
2045 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2046 				  struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2047 {
2048 	size_t copy_length;
2049 	struct copy_iovs_ctx copy_ctx;
2050 
2051 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2052 
2053 	if (offset < sizeof(ctrlr->changed_ns_list)) {
2054 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
2055 		if (copy_length) {
2056 			_copy_buf_to_iovs(&copy_ctx, (char *)&ctrlr->changed_ns_list + offset, copy_length);
2057 		}
2058 	}
2059 
2060 	/* Clear log page each time it is read */
2061 	ctrlr->changed_ns_list_count = 0;
2062 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
2063 
2064 	if (!rae) {
2065 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT);
2066 	}
2067 }
2068 
2069 /* The structure can be modified if we provide support for other commands in future */
2070 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
2071 	.admin_cmds_supported = {
2072 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
2073 		/* Get Log Page */
2074 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2075 		/* Identify */
2076 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
2077 		/* Abort */
2078 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
2079 		/* Set Features */
2080 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2081 		/* Get Features */
2082 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2083 		/* Async Event Request */
2084 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
2085 		/* Keep Alive */
2086 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2087 	},
2088 	.io_cmds_supported = {
2089 		/* FLUSH */
2090 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
2091 		/* WRITE */
2092 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
2093 		/* READ */
2094 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
2095 		/* WRITE ZEROES */
2096 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
2097 		/* DATASET MANAGEMENT */
2098 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
2099 		/* COMPARE */
2100 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
2101 	},
2102 };
2103 
2104 static void
2105 nvmf_get_cmds_and_effects_log_page(struct iovec *iovs, int iovcnt,
2106 				   uint64_t offset, uint32_t length)
2107 {
2108 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
2109 	size_t copy_len = 0;
2110 	struct copy_iovs_ctx copy_ctx;
2111 
2112 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2113 
2114 	if (offset < page_size) {
2115 		copy_len = spdk_min(page_size - offset, length);
2116 		_copy_buf_to_iovs(&copy_ctx, (char *)(&g_cmds_and_effect_log_page) + offset, copy_len);
2117 	}
2118 }
2119 
2120 static void
2121 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2122 		struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2123 {
2124 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
2125 	struct spdk_nvmf_reservation_log *log, *log_tmp;
2126 	struct copy_iovs_ctx copy_ctx;
2127 
2128 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2129 
2130 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
2131 	/* No available log, return zeroed log pages */
2132 	if (!ctrlr->num_avail_log_pages) {
2133 		return;
2134 	}
2135 
2136 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
2137 	if (offset >= avail_log_len) {
2138 		return;
2139 	}
2140 
2141 	next_pos = 0;
2142 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
2143 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
2144 		ctrlr->num_avail_log_pages--;
2145 
2146 		next_pos += unit_log_len;
2147 		if (next_pos > offset) {
2148 			copy_len = spdk_min(next_pos - offset, length);
2149 			_copy_buf_to_iovs(&copy_ctx, &log->log, copy_len);
2150 			length -= copy_len;
2151 			offset += copy_len;
2152 		}
2153 		free(log);
2154 
2155 		if (length == 0) {
2156 			break;
2157 		}
2158 	}
2159 
2160 	if (!rae) {
2161 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT);
2162 	}
2163 	return;
2164 }
2165 
2166 static int
2167 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
2168 {
2169 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2170 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2171 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2172 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2173 	uint64_t offset, len;
2174 	uint32_t rae, numdl, numdu;
2175 	uint8_t lid;
2176 
2177 	if (req->data == NULL) {
2178 		SPDK_ERRLOG("get log command with no buffer\n");
2179 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2180 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2181 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2182 	}
2183 
2184 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
2185 	if (offset & 3) {
2186 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
2187 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2188 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2189 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2190 	}
2191 
2192 	rae = cmd->cdw10_bits.get_log_page.rae;
2193 	numdl = cmd->cdw10_bits.get_log_page.numdl;
2194 	numdu = cmd->cdw11_bits.get_log_page.numdu;
2195 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
2196 	if (len > req->length) {
2197 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
2198 			    len, req->length);
2199 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2200 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2201 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2202 	}
2203 
2204 	lid = cmd->cdw10_bits.get_log_page.lid;
2205 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n",
2206 		      lid, offset, len, rae);
2207 
2208 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2209 		switch (lid) {
2210 		case SPDK_NVME_LOG_DISCOVERY:
2211 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt, offset,
2212 						    len);
2213 			if (!rae) {
2214 				nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT);
2215 			}
2216 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2217 		default:
2218 			goto invalid_log_page;
2219 		}
2220 	} else {
2221 		switch (lid) {
2222 		case SPDK_NVME_LOG_ERROR:
2223 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
2224 			/* TODO: actually fill out log page data */
2225 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2226 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
2227 			nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len);
2228 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2229 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
2230 			if (subsystem->flags.ana_reporting) {
2231 				nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2232 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2233 			} else {
2234 				goto invalid_log_page;
2235 			}
2236 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
2237 			nvmf_get_cmds_and_effects_log_page(req->iov, req->iovcnt, offset, len);
2238 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2239 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
2240 			nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2241 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2242 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
2243 			nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2244 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2245 		default:
2246 			goto invalid_log_page;
2247 		}
2248 	}
2249 
2250 invalid_log_page:
2251 	SPDK_DEBUGLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid);
2252 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2253 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2254 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2255 }
2256 
2257 int
2258 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2259 			    struct spdk_nvme_cmd *cmd,
2260 			    struct spdk_nvme_cpl *rsp,
2261 			    struct spdk_nvme_ns_data *nsdata)
2262 {
2263 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2264 	struct spdk_nvmf_ns *ns;
2265 	uint32_t max_num_blocks;
2266 
2267 	if (cmd->nsid == 0 || cmd->nsid > subsystem->max_nsid) {
2268 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", cmd->nsid);
2269 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2270 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2271 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2272 	}
2273 
2274 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2275 	if (ns == NULL || ns->bdev == NULL) {
2276 		/*
2277 		 * Inactive namespaces should return a zero filled data structure.
2278 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2279 		 * so we can just return early here.
2280 		 */
2281 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", cmd->nsid);
2282 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2283 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2284 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2285 	}
2286 
2287 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2288 
2289 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2290 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2291 			 (1U << nsdata->lbaf[nsdata->flbas.format].lbads);
2292 	if (nsdata->noiob > max_num_blocks) {
2293 		nsdata->noiob = max_num_blocks;
2294 	}
2295 
2296 	if (subsystem->flags.ana_reporting) {
2297 		/* ANA group ID matches NSID. */
2298 		nsdata->anagrpid = ns->nsid;
2299 
2300 		if (ctrlr->listener->ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2301 		    ctrlr->listener->ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2302 			nsdata->nuse = 0;
2303 		}
2304 	}
2305 
2306 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2307 }
2308 
2309 static void
2310 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2311 			 struct spdk_nvme_ctrlr_data *cdata)
2312 {
2313 	cdata->oacs.virtualization_management =
2314 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2315 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2316 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2317 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2318 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2319 	cdata->oacs.device_self_test =
2320 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2321 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2322 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2323 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2324 			       NULL
2325 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2326 	cdata->oacs.format =
2327 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2328 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2329 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2330 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2331 				     NULL;
2332 }
2333 
2334 int
2335 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2336 {
2337 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2338 	struct spdk_nvmf_transport *transport = ctrlr->admin_qpair->transport;
2339 
2340 	/*
2341 	 * Common fields for discovery and NVM subsystems
2342 	 */
2343 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2344 	assert((transport->opts.max_io_size % 4096) == 0);
2345 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2346 	cdata->cntlid = ctrlr->cntlid;
2347 	cdata->ver = ctrlr->vcprop.vs;
2348 	cdata->aerl = NVMF_MAX_ASYNC_EVENTS - 1;
2349 	cdata->lpa.edlp = 1;
2350 	cdata->elpe = 127;
2351 	cdata->maxcmd = transport->opts.max_queue_depth;
2352 	cdata->sgls = ctrlr->cdata.sgls;
2353 	cdata->fuses.compare_and_write = 1;
2354 	cdata->acwu = 1;
2355 	if (subsystem->flags.ana_reporting) {
2356 		cdata->mnan = subsystem->max_nsid;
2357 	}
2358 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2359 
2360 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2361 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2362 
2363 
2364 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2365 		/*
2366 		 * NVM Discovery subsystem fields
2367 		 */
2368 		cdata->oaes.discovery_log_change_notices = 1;
2369 	} else {
2370 		/*
2371 		 * NVM subsystem fields (reserved for discovery subsystems)
2372 		 */
2373 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2374 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2375 		cdata->kas = ctrlr->cdata.kas;
2376 
2377 		cdata->rab = 6;
2378 		cdata->cmic.multi_port = 1;
2379 		cdata->cmic.multi_ctrlr = 1;
2380 		if (subsystem->flags.ana_reporting) {
2381 			/* Asymmetric Namespace Access Reporting is supported. */
2382 			cdata->cmic.ana_reporting = 1;
2383 		}
2384 		cdata->oaes.ns_attribute_notices = 1;
2385 		if (subsystem->flags.ana_reporting) {
2386 			cdata->oaes.ana_change_notices = 1;
2387 		}
2388 		cdata->ctratt.host_id_exhid_supported = 1;
2389 		/* TODO: Concurrent execution of multiple abort commands. */
2390 		cdata->acl = 0;
2391 		cdata->frmw.slot1_ro = 1;
2392 		cdata->frmw.num_slots = 1;
2393 
2394 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2395 
2396 		cdata->sqes.min = 6;
2397 		cdata->sqes.max = 6;
2398 		cdata->cqes.min = 4;
2399 		cdata->cqes.max = 4;
2400 		cdata->nn = subsystem->max_nsid;
2401 		cdata->vwc.present = 1;
2402 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
2403 
2404 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
2405 
2406 		cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr);
2407 		cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr);
2408 		cdata->oncs.reservations = 1;
2409 		if (subsystem->flags.ana_reporting) {
2410 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
2411 			/* ANA Change state is not used, and ANA Persistent Loss state
2412 			 * is not supported for now.
2413 			 */
2414 			cdata->anacap.ana_optimized_state = 1;
2415 			cdata->anacap.ana_non_optimized_state = 1;
2416 			cdata->anacap.ana_inaccessible_state = 1;
2417 			/* ANAGRPID does not change while namespace is attached to controller */
2418 			cdata->anacap.no_change_anagrpid = 1;
2419 			cdata->anagrpmax = subsystem->max_nsid;
2420 			cdata->nanagrpid = subsystem->max_nsid;
2421 		}
2422 
2423 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
2424 
2425 		assert(subsystem->tgt != NULL);
2426 		cdata->crdt[0] = subsystem->tgt->crdt[0];
2427 		cdata->crdt[1] = subsystem->tgt->crdt[1];
2428 		cdata->crdt[2] = subsystem->tgt->crdt[2];
2429 
2430 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
2431 			      cdata->nvmf_specific.ioccsz);
2432 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
2433 			      cdata->nvmf_specific.iorcsz);
2434 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
2435 			      cdata->nvmf_specific.icdoff);
2436 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
2437 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
2438 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
2439 			      cdata->nvmf_specific.msdbd);
2440 	}
2441 
2442 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2443 }
2444 
2445 static int
2446 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_subsystem *subsystem,
2447 				   struct spdk_nvme_cmd *cmd,
2448 				   struct spdk_nvme_cpl *rsp,
2449 				   struct spdk_nvme_ns_list *ns_list)
2450 {
2451 	struct spdk_nvmf_ns *ns;
2452 	uint32_t count = 0;
2453 
2454 	if (cmd->nsid >= 0xfffffffeUL) {
2455 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
2456 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2457 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2458 	}
2459 
2460 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
2461 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2462 		if (ns->opts.nsid <= cmd->nsid) {
2463 			continue;
2464 		}
2465 
2466 		ns_list->ns_list[count++] = ns->opts.nsid;
2467 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
2468 			break;
2469 		}
2470 	}
2471 
2472 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2473 }
2474 
2475 static void
2476 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
2477 		enum spdk_nvme_nidt type,
2478 		const void *data, size_t data_size)
2479 {
2480 	struct spdk_nvme_ns_id_desc *desc;
2481 	size_t desc_size = sizeof(*desc) + data_size;
2482 
2483 	/*
2484 	 * These should never fail in practice, since all valid NS ID descriptors
2485 	 * should be defined so that they fit in the available 4096-byte buffer.
2486 	 */
2487 	assert(data_size > 0);
2488 	assert(data_size <= UINT8_MAX);
2489 	assert(desc_size < *buf_remain);
2490 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
2491 		return;
2492 	}
2493 
2494 	desc = *buf_ptr;
2495 	desc->nidt = type;
2496 	desc->nidl = data_size;
2497 	memcpy(desc->nid, data, data_size);
2498 
2499 	*buf_ptr += desc_size;
2500 	*buf_remain -= desc_size;
2501 }
2502 
2503 static int
2504 nvmf_ctrlr_identify_ns_id_descriptor_list(
2505 	struct spdk_nvmf_subsystem *subsystem,
2506 	struct spdk_nvme_cmd *cmd,
2507 	struct spdk_nvme_cpl *rsp,
2508 	void *id_desc_list, size_t id_desc_list_size)
2509 {
2510 	struct spdk_nvmf_ns *ns;
2511 	size_t buf_remain = id_desc_list_size;
2512 	void *buf_ptr = id_desc_list;
2513 
2514 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2515 	if (ns == NULL || ns->bdev == NULL) {
2516 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2517 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2518 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2519 	}
2520 
2521 #define ADD_ID_DESC(type, data, size) \
2522 	do { \
2523 		if (!spdk_mem_all_zero(data, size)) { \
2524 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
2525 		} \
2526 	} while (0)
2527 
2528 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
2529 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
2530 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
2531 
2532 	/*
2533 	 * The list is automatically 0-terminated because controller to host buffers in
2534 	 * admin commands always get zeroed in nvmf_ctrlr_process_admin_cmd().
2535 	 */
2536 
2537 #undef ADD_ID_DESC
2538 
2539 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2540 }
2541 
2542 static int
2543 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
2544 {
2545 	uint8_t cns;
2546 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2547 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2548 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2549 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2550 
2551 	if (req->data == NULL || req->length < 4096) {
2552 		SPDK_ERRLOG("identify command with invalid buffer\n");
2553 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2554 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2555 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2556 	}
2557 
2558 	cns = cmd->cdw10_bits.identify.cns;
2559 
2560 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY &&
2561 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
2562 		/* Discovery controllers only support Identify Controller */
2563 		goto invalid_cns;
2564 	}
2565 
2566 	switch (cns) {
2567 	case SPDK_NVME_IDENTIFY_NS:
2568 		return spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, req->data);
2569 	case SPDK_NVME_IDENTIFY_CTRLR:
2570 		return spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, req->data);
2571 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
2572 		return nvmf_ctrlr_identify_active_ns_list(subsystem, cmd, rsp, req->data);
2573 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
2574 		return nvmf_ctrlr_identify_ns_id_descriptor_list(subsystem, cmd, rsp, req->data, req->length);
2575 	default:
2576 		goto invalid_cns;
2577 	}
2578 
2579 invalid_cns:
2580 	SPDK_ERRLOG("Identify command with unsupported CNS 0x%02x\n", cns);
2581 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2582 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2583 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2584 }
2585 
2586 static bool
2587 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
2588 {
2589 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
2590 	struct spdk_nvmf_request *req;
2591 	int i;
2592 
2593 	if (!nvmf_qpair_is_admin_queue(qpair)) {
2594 		return false;
2595 	}
2596 
2597 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2598 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
2599 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
2600 			req = ctrlr->aer_req[i];
2601 			ctrlr->aer_req[i] = NULL;
2602 			ctrlr->nr_aer_reqs--;
2603 
2604 			/* Move the last req to the aborting position for making aer_reqs
2605 			 * in continuous
2606 			 */
2607 			if (i < ctrlr->nr_aer_reqs) {
2608 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
2609 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
2610 			}
2611 
2612 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2613 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2614 			_nvmf_request_complete(req);
2615 			return true;
2616 		}
2617 	}
2618 
2619 	return false;
2620 }
2621 
2622 static void
2623 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
2624 {
2625 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
2626 
2627 	if (nvmf_qpair_abort_aer(qpair, cid)) {
2628 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
2629 			      qpair->ctrlr, qpair->qid, cid);
2630 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
2631 
2632 		spdk_nvmf_request_complete(req);
2633 		return;
2634 	}
2635 
2636 	nvmf_transport_qpair_abort_request(qpair, req);
2637 }
2638 
2639 static void
2640 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
2641 {
2642 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2643 
2644 	if (status == 0) {
2645 		/* There was no qpair whose ID matches SQID of the abort command.
2646 		 * Hence call _nvmf_request_complete() here.
2647 		 */
2648 		_nvmf_request_complete(req);
2649 	}
2650 }
2651 
2652 static void
2653 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
2654 {
2655 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2656 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
2657 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
2658 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
2659 	struct spdk_nvmf_qpair *qpair;
2660 
2661 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
2662 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
2663 			/* Found the qpair */
2664 
2665 			nvmf_qpair_abort_request(qpair, req);
2666 
2667 			/* Return -1 for the status so the iteration across threads stops. */
2668 			spdk_for_each_channel_continue(i, -1);
2669 			return;
2670 		}
2671 	}
2672 
2673 	spdk_for_each_channel_continue(i, 0);
2674 }
2675 
2676 static int
2677 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
2678 {
2679 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2680 
2681 	rsp->cdw0 = 1U; /* Command not aborted */
2682 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2683 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2684 
2685 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
2686 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
2687 			      nvmf_ctrlr_abort_on_pg,
2688 			      req,
2689 			      nvmf_ctrlr_abort_done
2690 			     );
2691 
2692 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2693 }
2694 
2695 int
2696 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
2697 {
2698 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
2699 	struct spdk_bdev *bdev;
2700 	struct spdk_bdev_desc *desc;
2701 	struct spdk_io_channel *ch;
2702 	int rc;
2703 
2704 	assert(req_to_abort != NULL);
2705 
2706 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
2707 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
2708 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
2709 	}
2710 
2711 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
2712 					&bdev, &desc, &ch);
2713 	if (rc != 0) {
2714 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2715 	}
2716 
2717 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
2718 }
2719 
2720 static int
2721 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
2722 {
2723 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2724 
2725 	rsp->cdw0 = cdw0;
2726 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2727 }
2728 
2729 /* we have to use the typedef in the function declaration to appease astyle. */
2730 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
2731 
2732 static spdk_nvme_path_status_code_t
2733 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
2734 {
2735 	switch (ana_state) {
2736 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2737 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
2738 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2739 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
2740 	case SPDK_NVME_ANA_CHANGE_STATE:
2741 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
2742 	default:
2743 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
2744 	}
2745 }
2746 
2747 static int
2748 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
2749 {
2750 	uint8_t feature;
2751 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2752 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2753 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2754 	enum spdk_nvme_ana_state ana_state;
2755 
2756 	feature = cmd->cdw10_bits.get_features.fid;
2757 
2758 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2759 		/*
2760 		 * Features supported by Discovery controller
2761 		 */
2762 		switch (feature) {
2763 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2764 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
2765 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2766 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
2767 		default:
2768 			SPDK_DEBUGLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
2769 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2770 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2771 		}
2772 	}
2773 	/*
2774 	 * Process Get Features command for non-discovery controller
2775 	 */
2776 	ana_state = ctrlr->listener->ana_state;
2777 	switch (ana_state) {
2778 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2779 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2780 	case SPDK_NVME_ANA_CHANGE_STATE:
2781 		switch (feature) {
2782 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
2783 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2784 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2785 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2786 			response->status.sct = SPDK_NVME_SCT_PATH;
2787 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2788 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2789 		default:
2790 			break;
2791 		}
2792 		break;
2793 	default:
2794 		break;
2795 	}
2796 
2797 	switch (feature) {
2798 	case SPDK_NVME_FEAT_ARBITRATION:
2799 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
2800 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
2801 		return get_features_generic(req, ctrlr->feat.power_management.raw);
2802 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
2803 		return nvmf_ctrlr_get_features_temperature_threshold(req);
2804 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
2805 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
2806 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
2807 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
2808 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
2809 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
2810 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
2811 		return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw);
2812 	case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION:
2813 		return nvmf_ctrlr_get_features_interrupt_vector_configuration(req);
2814 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2815 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
2816 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2817 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
2818 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2819 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
2820 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2821 		return nvmf_ctrlr_get_features_host_identifier(req);
2822 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2823 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
2824 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2825 		return nvmf_ctrlr_get_features_reservation_persistence(req);
2826 	default:
2827 		SPDK_ERRLOG("Get Features command with unsupported feature ID 0x%02x\n", feature);
2828 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2829 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2830 	}
2831 }
2832 
2833 static int
2834 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
2835 {
2836 	uint8_t feature, save;
2837 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2838 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2839 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2840 	enum spdk_nvme_ana_state ana_state;
2841 	/*
2842 	 * Features are not saveable by the controller as indicated by
2843 	 * ONCS field of the Identify Controller data.
2844 	 * */
2845 	save = cmd->cdw10_bits.set_features.sv;
2846 	if (save) {
2847 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
2848 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
2849 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2850 	}
2851 
2852 	feature = cmd->cdw10_bits.set_features.fid;
2853 
2854 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2855 		/*
2856 		 * Features supported by Discovery controller
2857 		 */
2858 		switch (feature) {
2859 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2860 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
2861 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2862 			return nvmf_ctrlr_set_features_async_event_configuration(req);
2863 		default:
2864 			SPDK_ERRLOG("Set Features command with unsupported feature ID 0x%02x\n", feature);
2865 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2866 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2867 		}
2868 	}
2869 	/*
2870 	 * Process Set Features command for non-discovery controller
2871 	 */
2872 	ana_state = ctrlr->listener->ana_state;
2873 	switch (ana_state) {
2874 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2875 	case SPDK_NVME_ANA_CHANGE_STATE:
2876 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
2877 			response->status.sct = SPDK_NVME_SCT_PATH;
2878 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2879 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2880 		} else {
2881 			switch (feature) {
2882 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
2883 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2884 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2885 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2886 				response->status.sct = SPDK_NVME_SCT_PATH;
2887 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2888 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2889 			default:
2890 				break;
2891 			}
2892 		}
2893 		break;
2894 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2895 		response->status.sct = SPDK_NVME_SCT_PATH;
2896 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
2897 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2898 	default:
2899 		break;
2900 	}
2901 
2902 	switch (feature) {
2903 	case SPDK_NVME_FEAT_ARBITRATION:
2904 		return nvmf_ctrlr_set_features_arbitration(req);
2905 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
2906 		return nvmf_ctrlr_set_features_power_management(req);
2907 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
2908 		return nvmf_ctrlr_set_features_temperature_threshold(req);
2909 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
2910 		return nvmf_ctrlr_set_features_error_recovery(req);
2911 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
2912 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
2913 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
2914 		return nvmf_ctrlr_set_features_number_of_queues(req);
2915 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2916 		return nvmf_ctrlr_set_features_write_atomicity(req);
2917 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2918 		return nvmf_ctrlr_set_features_async_event_configuration(req);
2919 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2920 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
2921 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2922 		return nvmf_ctrlr_set_features_host_identifier(req);
2923 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2924 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
2925 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2926 		return nvmf_ctrlr_set_features_reservation_persistence(req);
2927 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
2928 		return nvmf_ctrlr_set_features_host_behavior_support(req);
2929 	default:
2930 		SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
2931 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2932 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2933 	}
2934 }
2935 
2936 static int
2937 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
2938 {
2939 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2940 
2941 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
2942 	/*
2943 	 * To handle keep alive just clear or reset the
2944 	 * ctrlr based keep alive duration counter.
2945 	 * When added, a separate timer based process
2946 	 * will monitor if the time since last recorded
2947 	 * keep alive has exceeded the max duration and
2948 	 * take appropriate action.
2949 	 */
2950 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
2951 
2952 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2953 }
2954 
2955 int
2956 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
2957 {
2958 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2959 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2960 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2961 	int rc;
2962 
2963 	if (ctrlr == NULL) {
2964 		SPDK_ERRLOG("Admin command sent before CONNECT\n");
2965 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2966 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2967 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2968 	}
2969 
2970 	if (ctrlr->vcprop.cc.bits.en != 1) {
2971 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
2972 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2973 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2974 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2975 	}
2976 
2977 	if (req->data && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2978 		memset(req->data, 0, req->length);
2979 	}
2980 
2981 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2982 		/* Discovery controllers only support these admin OPS. */
2983 		switch (cmd->opc) {
2984 		case SPDK_NVME_OPC_IDENTIFY:
2985 		case SPDK_NVME_OPC_GET_LOG_PAGE:
2986 		case SPDK_NVME_OPC_KEEP_ALIVE:
2987 		case SPDK_NVME_OPC_SET_FEATURES:
2988 		case SPDK_NVME_OPC_GET_FEATURES:
2989 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
2990 			break;
2991 		default:
2992 			goto invalid_opcode;
2993 		}
2994 	}
2995 
2996 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
2997 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
2998 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
2999 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3000 			/* The handler took care of this commmand */
3001 			return rc;
3002 		}
3003 	}
3004 
3005 	switch (cmd->opc) {
3006 	case SPDK_NVME_OPC_GET_LOG_PAGE:
3007 		return nvmf_ctrlr_get_log_page(req);
3008 	case SPDK_NVME_OPC_IDENTIFY:
3009 		return nvmf_ctrlr_identify(req);
3010 	case SPDK_NVME_OPC_ABORT:
3011 		return nvmf_ctrlr_abort(req);
3012 	case SPDK_NVME_OPC_GET_FEATURES:
3013 		return nvmf_ctrlr_get_features(req);
3014 	case SPDK_NVME_OPC_SET_FEATURES:
3015 		return nvmf_ctrlr_set_features(req);
3016 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3017 		return nvmf_ctrlr_async_event_request(req);
3018 	case SPDK_NVME_OPC_KEEP_ALIVE:
3019 		return nvmf_ctrlr_keep_alive(req);
3020 
3021 	case SPDK_NVME_OPC_CREATE_IO_SQ:
3022 	case SPDK_NVME_OPC_CREATE_IO_CQ:
3023 	case SPDK_NVME_OPC_DELETE_IO_SQ:
3024 	case SPDK_NVME_OPC_DELETE_IO_CQ:
3025 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
3026 		goto invalid_opcode;
3027 
3028 	default:
3029 		goto invalid_opcode;
3030 	}
3031 
3032 invalid_opcode:
3033 	SPDK_ERRLOG("Unsupported admin opcode 0x%x\n", cmd->opc);
3034 	response->status.sct = SPDK_NVME_SCT_GENERIC;
3035 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3036 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3037 }
3038 
3039 static int
3040 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
3041 {
3042 	struct spdk_nvmf_qpair *qpair = req->qpair;
3043 	struct spdk_nvmf_capsule_cmd *cap_hdr;
3044 
3045 	cap_hdr = &req->cmd->nvmf_cmd;
3046 
3047 	if (qpair->ctrlr == NULL) {
3048 		/* No ctrlr established yet; the only valid command is Connect */
3049 		if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
3050 			return nvmf_ctrlr_cmd_connect(req);
3051 		} else {
3052 			SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n",
3053 				      cap_hdr->fctype);
3054 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3055 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3056 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3057 		}
3058 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
3059 		/*
3060 		 * Controller session is established, and this is an admin queue.
3061 		 * Disallow Connect and allow other fabrics commands.
3062 		 */
3063 		switch (cap_hdr->fctype) {
3064 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
3065 			return nvmf_property_set(req);
3066 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
3067 			return nvmf_property_get(req);
3068 		default:
3069 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
3070 				      cap_hdr->fctype);
3071 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3072 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3073 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3074 		}
3075 	} else {
3076 		/* Controller session is established, and this is an I/O queue */
3077 		/* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */
3078 		SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
3079 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3080 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3081 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3082 	}
3083 }
3084 
3085 static inline int
3086 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
3087 				    union spdk_nvme_async_event_completion *event)
3088 {
3089 	struct spdk_nvmf_request *req;
3090 	struct spdk_nvme_cpl *rsp;
3091 
3092 	assert(ctrlr->nr_aer_reqs > 0);
3093 
3094 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
3095 	rsp = &req->rsp->nvme_cpl;
3096 
3097 	rsp->cdw0 = event->raw;
3098 
3099 	_nvmf_request_complete(req);
3100 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3101 
3102 	return 0;
3103 }
3104 
3105 static inline void
3106 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr,
3107 				     union spdk_nvme_async_event_completion *event)
3108 {
3109 	struct spdk_nvmf_async_event_completion *nvmf_event;
3110 
3111 	nvmf_event = calloc(1, sizeof(*nvmf_event));
3112 	if (!nvmf_event) {
3113 		SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n");
3114 		return;
3115 	}
3116 	nvmf_event->event.raw = event->raw;
3117 	STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link);
3118 }
3119 
3120 int
3121 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
3122 {
3123 	union spdk_nvme_async_event_completion event = {0};
3124 
3125 	/* Users may disable the event notification */
3126 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
3127 		return 0;
3128 	}
3129 
3130 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) {
3131 		return 0;
3132 	}
3133 
3134 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3135 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
3136 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
3137 
3138 	/* If there is no outstanding AER request, queue the event.  Then
3139 	 * if an AER is later submitted, this event can be sent as a
3140 	 * response.
3141 	 */
3142 	if (ctrlr->nr_aer_reqs == 0) {
3143 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3144 		return 0;
3145 	}
3146 
3147 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3148 }
3149 
3150 int
3151 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3152 {
3153 	union spdk_nvme_async_event_completion event = {0};
3154 
3155 	/* Users may disable the event notification */
3156 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
3157 		return 0;
3158 	}
3159 
3160 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) {
3161 		return 0;
3162 	}
3163 
3164 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3165 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
3166 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
3167 
3168 	/* If there is no outstanding AER request, queue the event.  Then
3169 	 * if an AER is later submitted, this event can be sent as a
3170 	 * response.
3171 	 */
3172 	if (ctrlr->nr_aer_reqs == 0) {
3173 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3174 		return 0;
3175 	}
3176 
3177 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3178 }
3179 
3180 void
3181 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
3182 {
3183 	union spdk_nvme_async_event_completion event = {0};
3184 
3185 	if (!ctrlr->num_avail_log_pages) {
3186 		return;
3187 	}
3188 
3189 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) {
3190 		return;
3191 	}
3192 
3193 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
3194 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
3195 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
3196 
3197 	/* If there is no outstanding AER request, queue the event.  Then
3198 	 * if an AER is later submitted, this event can be sent as a
3199 	 * response.
3200 	 */
3201 	if (ctrlr->nr_aer_reqs == 0) {
3202 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3203 		return;
3204 	}
3205 
3206 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3207 }
3208 
3209 int
3210 nvmf_ctrlr_async_event_discovery_log_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3211 {
3212 	union spdk_nvme_async_event_completion event = {0};
3213 
3214 	/* Users may disable the event notification manually or
3215 	 * it may not be enabled due to keep alive timeout
3216 	 * not being set in connect command to discovery controller.
3217 	 */
3218 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
3219 		return 0;
3220 	}
3221 
3222 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) {
3223 		return 0;
3224 	}
3225 
3226 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3227 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
3228 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
3229 
3230 	/* If there is no outstanding AER request, queue the event.  Then
3231 	 * if an AER is later submitted, this event can be sent as a
3232 	 * response.
3233 	 */
3234 	if (ctrlr->nr_aer_reqs == 0) {
3235 		nvmf_ctrlr_queue_pending_async_event(ctrlr, &event);
3236 		return 0;
3237 	}
3238 
3239 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3240 }
3241 
3242 void
3243 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
3244 {
3245 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3246 	int i;
3247 
3248 	if (!nvmf_qpair_is_admin_queue(qpair)) {
3249 		return;
3250 	}
3251 
3252 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3253 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
3254 		ctrlr->aer_req[i] = NULL;
3255 	}
3256 
3257 	ctrlr->nr_aer_reqs = 0;
3258 }
3259 
3260 void
3261 nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
3262 {
3263 	struct spdk_nvmf_request *req;
3264 	int i;
3265 
3266 	if (!ctrlr->nr_aer_reqs) {
3267 		return;
3268 	}
3269 
3270 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3271 		req = ctrlr->aer_req[i];
3272 
3273 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3274 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3275 		_nvmf_request_complete(req);
3276 
3277 		ctrlr->aer_req[i] = NULL;
3278 	}
3279 
3280 	ctrlr->nr_aer_reqs = 0;
3281 }
3282 
3283 static void
3284 _nvmf_ctrlr_add_reservation_log(void *ctx)
3285 {
3286 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
3287 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
3288 
3289 	ctrlr->log_page_count++;
3290 
3291 	/* Maximum number of queued log pages is 255 */
3292 	if (ctrlr->num_avail_log_pages == 0xff) {
3293 		struct spdk_nvmf_reservation_log *entry;
3294 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
3295 		entry->log.log_page_count = ctrlr->log_page_count;
3296 		free(log);
3297 		return;
3298 	}
3299 
3300 	log->log.log_page_count = ctrlr->log_page_count;
3301 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
3302 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
3303 
3304 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
3305 }
3306 
3307 void
3308 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
3309 				  struct spdk_nvmf_ns *ns,
3310 				  enum spdk_nvme_reservation_notification_log_page_type type)
3311 {
3312 	struct spdk_nvmf_reservation_log *log;
3313 
3314 	switch (type) {
3315 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
3316 		return;
3317 	case SPDK_NVME_REGISTRATION_PREEMPTED:
3318 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
3319 			return;
3320 		}
3321 		break;
3322 	case SPDK_NVME_RESERVATION_RELEASED:
3323 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
3324 			return;
3325 		}
3326 		break;
3327 	case SPDK_NVME_RESERVATION_PREEMPTED:
3328 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
3329 			return;
3330 		}
3331 		break;
3332 	default:
3333 		return;
3334 	}
3335 
3336 	log = calloc(1, sizeof(*log));
3337 	if (!log) {
3338 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
3339 		return;
3340 	}
3341 	log->ctrlr = ctrlr;
3342 	log->log.type = type;
3343 	log->log.nsid = ns->nsid;
3344 
3345 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
3346 }
3347 
3348 /* Check from subsystem poll group's namespace information data structure */
3349 static bool
3350 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3351 				 struct spdk_nvmf_ctrlr *ctrlr)
3352 {
3353 	uint32_t i;
3354 
3355 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
3356 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
3357 			return true;
3358 		}
3359 	}
3360 
3361 	return false;
3362 }
3363 
3364 /*
3365  * Check the NVMe command is permitted or not for current controller(Host).
3366  */
3367 static int
3368 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3369 				  struct spdk_nvmf_ctrlr *ctrlr,
3370 				  struct spdk_nvmf_request *req)
3371 {
3372 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3373 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
3374 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3375 	uint8_t racqa;
3376 	bool is_registrant;
3377 
3378 	/* No valid reservation */
3379 	if (!rtype) {
3380 		return 0;
3381 	}
3382 
3383 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
3384 	/* All registrants type and current ctrlr is a valid registrant */
3385 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3386 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
3387 		return 0;
3388 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
3389 		return 0;
3390 	}
3391 
3392 	/* Non-holder for current controller */
3393 	switch (cmd->opc) {
3394 	case SPDK_NVME_OPC_READ:
3395 	case SPDK_NVME_OPC_COMPARE:
3396 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3397 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3398 			goto exit;
3399 		}
3400 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
3401 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
3402 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3403 		}
3404 		break;
3405 	case SPDK_NVME_OPC_FLUSH:
3406 	case SPDK_NVME_OPC_WRITE:
3407 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
3408 	case SPDK_NVME_OPC_WRITE_ZEROES:
3409 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3410 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
3411 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3412 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3413 			goto exit;
3414 		}
3415 		if (!is_registrant) {
3416 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3417 		}
3418 		break;
3419 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3420 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
3421 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
3422 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3423 			goto exit;
3424 		}
3425 		if (!is_registrant) {
3426 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3427 		}
3428 		break;
3429 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3430 		if (!is_registrant) {
3431 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3432 		}
3433 		break;
3434 	default:
3435 		break;
3436 	}
3437 
3438 exit:
3439 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3440 	req->rsp->nvme_cpl.status.sc = status;
3441 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
3442 		return -EPERM;
3443 	}
3444 
3445 	return 0;
3446 }
3447 
3448 static int
3449 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
3450 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
3451 {
3452 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3453 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3454 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
3455 	int rc;
3456 
3457 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
3458 		/* first fused operation (should be compare) */
3459 		if (first_fused_req != NULL) {
3460 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3461 
3462 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3463 
3464 			/* abort req->qpair->first_fused_request and continue with new fused command */
3465 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3466 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3467 			_nvmf_request_complete(first_fused_req);
3468 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
3469 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3470 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3471 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3472 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3473 		}
3474 
3475 		req->qpair->first_fused_req = req;
3476 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3477 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
3478 		/* second fused operation (should be write) */
3479 		if (first_fused_req == NULL) {
3480 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3481 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3482 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3483 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3484 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
3485 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3486 
3487 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3488 
3489 			/* abort req->qpair->first_fused_request and fail current command */
3490 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3491 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3492 			_nvmf_request_complete(first_fused_req);
3493 
3494 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3495 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3496 			req->qpair->first_fused_req = NULL;
3497 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3498 		}
3499 
3500 		/* save request of first command to generate response later */
3501 		req->first_fused_req = first_fused_req;
3502 		req->qpair->first_fused_req = NULL;
3503 	} else {
3504 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
3505 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3506 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3507 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3508 	}
3509 
3510 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
3511 
3512 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3513 		if (spdk_nvme_cpl_is_error(rsp)) {
3514 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3515 
3516 			fused_response->status = rsp->status;
3517 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3518 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
3519 			/* Complete first of fused commands. Second will be completed by upper layer */
3520 			_nvmf_request_complete(first_fused_req);
3521 			req->first_fused_req = NULL;
3522 		}
3523 	}
3524 
3525 	return rc;
3526 }
3527 
3528 bool
3529 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req)
3530 {
3531 	struct spdk_nvmf_ns *ns;
3532 
3533 	req->zcopy_phase = NVMF_ZCOPY_PHASE_NONE;
3534 
3535 	if (nvmf_qpair_is_admin_queue(req->qpair)) {
3536 		/* Admin queue */
3537 		return false;
3538 	}
3539 
3540 	if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) &&
3541 	    (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) {
3542 		/* Not a READ or WRITE command */
3543 		return false;
3544 	}
3545 
3546 	if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) {
3547 		/* Fused commands dont use zcopy buffers */
3548 		return false;
3549 	}
3550 
3551 	ns = _nvmf_subsystem_get_ns(req->qpair->ctrlr->subsys, req->cmd->nvme_cmd.nsid);
3552 	if (ns == NULL || ns->bdev == NULL || !ns->zcopy) {
3553 		return false;
3554 	}
3555 
3556 	req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT;
3557 	return true;
3558 }
3559 
3560 /* If this function returns a non-zero value the request
3561  * reverts to using SPDK buffers
3562  */
3563 int
3564 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req)
3565 {
3566 	struct spdk_nvmf_qpair *qpair = req->qpair;
3567 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3568 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3569 	uint32_t nsid;
3570 	struct spdk_bdev *bdev;
3571 	struct spdk_bdev_desc *desc;
3572 	struct spdk_io_channel *ch;
3573 	int rc;
3574 
3575 	if (!qpair->ctrlr) {
3576 		goto end;
3577 	}
3578 
3579 	if (qpair->group->sgroups == NULL) {
3580 		goto end;
3581 	}
3582 
3583 	rc = spdk_nvmf_request_get_bdev(req->cmd->nvme_cmd.nsid, req,
3584 					&bdev, &desc, &ch);
3585 	if (rc != 0) {
3586 		goto end;
3587 	}
3588 
3589 	if (ch == NULL) {
3590 		goto end;
3591 	}
3592 
3593 	nsid = req->cmd->nvme_cmd.nsid;
3594 	sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3595 	ns_info = &sgroup->ns_info[nsid - 1];
3596 	if (ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
3597 		goto end;
3598 	}
3599 
3600 	if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
3601 		goto end;
3602 	}
3603 
3604 	/* backward compatible */
3605 	req->data = req->iov[0].iov_base;
3606 
3607 	/* Set iovcnt to be the maximum number of
3608 	 * iovs that the ZCOPY can use
3609 	 */
3610 	req->iovcnt = NVMF_REQ_MAX_BUFFERS;
3611 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3612 	rc = nvmf_bdev_ctrlr_start_zcopy(bdev, desc, ch, req);
3613 	if (rc == 0) {
3614 		ns_info->io_outstanding++;
3615 		return 0;
3616 	}
3617 	TAILQ_REMOVE(&qpair->outstanding, req, link);
3618 
3619 end:
3620 	/* An error occurred, the subsystem is paused, or the qpair is not active.
3621 	 * Revert to using SPDK buffers
3622 	 */
3623 	req->zcopy_phase = NVMF_ZCOPY_PHASE_NONE;
3624 	return -1;
3625 }
3626 
3627 int
3628 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req)
3629 {
3630 	req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING;
3631 	return nvmf_bdev_ctrlr_end_zcopy(req);
3632 }
3633 
3634 int
3635 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
3636 {
3637 	uint32_t nsid;
3638 	struct spdk_nvmf_ns *ns;
3639 	struct spdk_bdev *bdev;
3640 	struct spdk_bdev_desc *desc;
3641 	struct spdk_io_channel *ch;
3642 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3643 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3644 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3645 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3646 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3647 	enum spdk_nvme_ana_state ana_state;
3648 
3649 	/* pre-set response details for this command */
3650 	response->status.sc = SPDK_NVME_SC_SUCCESS;
3651 	nsid = cmd->nsid;
3652 
3653 	if (spdk_unlikely(ctrlr == NULL)) {
3654 		SPDK_ERRLOG("I/O command sent before CONNECT\n");
3655 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3656 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3657 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3658 	}
3659 
3660 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
3661 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
3662 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3663 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3664 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3665 	}
3666 
3667 	/* It will be lower overhead to check if ANA state is optimized or
3668 	 * non-optimized.
3669 	 */
3670 	ana_state = ctrlr->listener->ana_state;
3671 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
3672 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
3673 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
3674 			      ana_state);
3675 		response->status.sct = SPDK_NVME_SCT_PATH;
3676 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3677 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3678 	}
3679 
3680 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3681 	if (ns == NULL || ns->bdev == NULL) {
3682 		SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid);
3683 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3684 		response->status.dnr = 1;
3685 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3686 	}
3687 
3688 	/* scan-build falsely reporting dereference of null pointer */
3689 	assert(group != NULL && group->sgroups != NULL);
3690 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3691 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
3692 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
3693 			      cmd->nsid, cmd->opc);
3694 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3695 	}
3696 
3697 	bdev = ns->bdev;
3698 	desc = ns->desc;
3699 	ch = ns_info->channel;
3700 
3701 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
3702 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
3703 	} else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) {
3704 		struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl;
3705 
3706 		SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n");
3707 
3708 		/* abort req->qpair->first_fused_request and continue with new command */
3709 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3710 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3711 		_nvmf_request_complete(req->qpair->first_fused_req);
3712 		req->qpair->first_fused_req = NULL;
3713 	}
3714 
3715 	switch (cmd->opc) {
3716 	case SPDK_NVME_OPC_READ:
3717 		return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
3718 	case SPDK_NVME_OPC_WRITE:
3719 		return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
3720 	case SPDK_NVME_OPC_COMPARE:
3721 		return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
3722 	case SPDK_NVME_OPC_WRITE_ZEROES:
3723 		return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
3724 	case SPDK_NVME_OPC_FLUSH:
3725 		return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
3726 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3727 		return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
3728 	case SPDK_NVME_OPC_RESERVATION_REGISTER:
3729 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3730 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3731 	case SPDK_NVME_OPC_RESERVATION_REPORT:
3732 		spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
3733 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3734 	default:
3735 		return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
3736 	}
3737 }
3738 
3739 static void
3740 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
3741 {
3742 	if (qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING) {
3743 		assert(qpair->state_cb != NULL);
3744 
3745 		if (TAILQ_EMPTY(&qpair->outstanding)) {
3746 			qpair->state_cb(qpair->state_cb_arg, 0);
3747 		}
3748 	}
3749 }
3750 
3751 int
3752 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
3753 {
3754 	struct spdk_nvmf_qpair *qpair = req->qpair;
3755 
3756 	TAILQ_REMOVE(&qpair->outstanding, req, link);
3757 	if (nvmf_transport_req_free(req)) {
3758 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
3759 	}
3760 
3761 	nvmf_qpair_request_cleanup(qpair);
3762 
3763 	return 0;
3764 }
3765 
3766 static void
3767 _nvmf_request_complete(void *ctx)
3768 {
3769 	struct spdk_nvmf_request *req = ctx;
3770 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3771 	struct spdk_nvmf_qpair *qpair;
3772 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3773 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3774 	bool is_aer = false;
3775 	uint32_t nsid;
3776 	bool paused;
3777 	uint8_t opcode;
3778 
3779 	rsp->sqid = 0;
3780 	rsp->status.p = 0;
3781 	rsp->cid = req->cmd->nvme_cmd.cid;
3782 	nsid = req->cmd->nvme_cmd.nsid;
3783 	opcode = req->cmd->nvmf_cmd.opcode;
3784 
3785 	qpair = req->qpair;
3786 	if (qpair->ctrlr) {
3787 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3788 		assert(sgroup != NULL);
3789 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
3790 
3791 		/*
3792 		 * Set the crd value.
3793 		 * If the the IO has any error, and dnr (DoNotRetry) is not 1,
3794 		 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT.
3795 		 */
3796 		if (spdk_nvme_cpl_is_error(rsp) &&
3797 		    rsp->status.dnr == 0 &&
3798 		    qpair->ctrlr->acre_enabled) {
3799 			rsp->status.crd = 1;
3800 		}
3801 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3802 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3803 	}
3804 
3805 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
3806 		spdk_nvme_print_completion(qpair->qid, rsp);
3807 	}
3808 
3809 	switch (req->zcopy_phase) {
3810 	case NVMF_ZCOPY_PHASE_NONE:
3811 		TAILQ_REMOVE(&qpair->outstanding, req, link);
3812 		break;
3813 	case NVMF_ZCOPY_PHASE_INIT:
3814 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) {
3815 			/* The START failed or was aborted so revert to a normal IO */
3816 			req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED;
3817 			TAILQ_REMOVE(&qpair->outstanding, req, link);
3818 		} else {
3819 			req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE;
3820 		}
3821 		break;
3822 	case NVMF_ZCOPY_PHASE_EXECUTE:
3823 		break;
3824 	case NVMF_ZCOPY_PHASE_END_PENDING:
3825 		TAILQ_REMOVE(&qpair->outstanding, req, link);
3826 		req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE;
3827 		break;
3828 	default:
3829 		SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase);
3830 		break;
3831 	}
3832 
3833 	if (nvmf_transport_req_complete(req)) {
3834 		SPDK_ERRLOG("Transport request completion error!\n");
3835 	}
3836 
3837 	/* AER cmd is an exception */
3838 	if (sgroup && !is_aer) {
3839 		if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC ||
3840 				  nvmf_qpair_is_admin_queue(qpair))) {
3841 			assert(sgroup->mgmt_io_outstanding > 0);
3842 			sgroup->mgmt_io_outstanding--;
3843 		} else {
3844 			if ((req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE) ||
3845 			    (req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE)) {
3846 				/* End of request */
3847 
3848 				/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
3849 				if (spdk_likely(nsid - 1 < sgroup->num_ns)) {
3850 					sgroup->ns_info[nsid - 1].io_outstanding--;
3851 				}
3852 			}
3853 		}
3854 
3855 		if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
3856 				  sgroup->mgmt_io_outstanding == 0)) {
3857 			paused = true;
3858 			for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
3859 				ns_info = &sgroup->ns_info[nsid];
3860 
3861 				if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
3862 				    ns_info->io_outstanding > 0) {
3863 					paused = false;
3864 					break;
3865 				}
3866 			}
3867 
3868 			if (paused) {
3869 				sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
3870 				sgroup->cb_fn(sgroup->cb_arg, 0);
3871 				sgroup->cb_fn = NULL;
3872 				sgroup->cb_arg = NULL;
3873 			}
3874 		}
3875 
3876 	}
3877 
3878 	nvmf_qpair_request_cleanup(qpair);
3879 }
3880 
3881 int
3882 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
3883 {
3884 	struct spdk_nvmf_qpair *qpair = req->qpair;
3885 
3886 	if (spdk_likely(qpair->group->thread == spdk_get_thread())) {
3887 		_nvmf_request_complete(req);
3888 	} else {
3889 		spdk_thread_send_msg(qpair->group->thread,
3890 				     _nvmf_request_complete, req);
3891 	}
3892 
3893 	return 0;
3894 }
3895 
3896 void
3897 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req)
3898 {
3899 	struct spdk_nvmf_qpair *qpair = req->qpair;
3900 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3901 	enum spdk_nvmf_request_exec_status status;
3902 
3903 	if (qpair->ctrlr) {
3904 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3905 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3906 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3907 	}
3908 
3909 	assert(sgroup != NULL);
3910 	sgroup->mgmt_io_outstanding++;
3911 
3912 	/* Place the request on the outstanding list so we can keep track of it */
3913 	nvmf_add_to_outstanding_queue(req);
3914 
3915 	assert(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC);
3916 	status = nvmf_ctrlr_process_fabrics_cmd(req);
3917 
3918 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3919 		_nvmf_request_complete(req);
3920 	}
3921 }
3922 
3923 static bool nvmf_check_subsystem_active(struct spdk_nvmf_request *req)
3924 {
3925 	struct spdk_nvmf_qpair *qpair = req->qpair;
3926 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3927 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3928 	uint32_t nsid;
3929 
3930 	if (qpair->ctrlr) {
3931 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3932 		assert(sgroup != NULL);
3933 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3934 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3935 	}
3936 
3937 	/* Check if the subsystem is paused (if there is a subsystem) */
3938 	if (sgroup != NULL) {
3939 		if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC ||
3940 				  nvmf_qpair_is_admin_queue(qpair))) {
3941 			if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
3942 				/* The subsystem is not currently active. Queue this request. */
3943 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
3944 				return false;
3945 			}
3946 			sgroup->mgmt_io_outstanding++;
3947 		} else {
3948 			nsid = req->cmd->nvme_cmd.nsid;
3949 
3950 			/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
3951 			if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) {
3952 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3953 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3954 				req->rsp->nvme_cpl.status.dnr = 1;
3955 				nvmf_add_to_outstanding_queue(req);
3956 				_nvmf_request_complete(req);
3957 				return false;
3958 			}
3959 
3960 			ns_info = &sgroup->ns_info[nsid - 1];
3961 			if (ns_info->channel == NULL) {
3962 				/* This can can happen if host sends I/O to a namespace that is
3963 				 * in the process of being added, but before the full addition
3964 				 * process is complete.  Report invalid namespace in that case.
3965 				 */
3966 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3967 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3968 				req->rsp->nvme_cpl.status.dnr = 1;
3969 				nvmf_add_to_outstanding_queue(req);
3970 				ns_info->io_outstanding++;
3971 				_nvmf_request_complete(req);
3972 				return false;
3973 			}
3974 
3975 			if (ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
3976 				/* The namespace is not currently active. Queue this request. */
3977 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
3978 				return false;
3979 			}
3980 
3981 			ns_info->io_outstanding++;
3982 		}
3983 
3984 		if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
3985 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3986 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3987 			nvmf_add_to_outstanding_queue(req);
3988 			_nvmf_request_complete(req);
3989 			return false;
3990 		}
3991 	}
3992 
3993 	return true;
3994 }
3995 
3996 void
3997 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
3998 {
3999 	struct spdk_nvmf_qpair *qpair = req->qpair;
4000 	enum spdk_nvmf_request_exec_status status;
4001 
4002 	if (!spdk_nvmf_using_zcopy(req->zcopy_phase)) {
4003 		if (!nvmf_check_subsystem_active(req)) {
4004 			return;
4005 		}
4006 	}
4007 
4008 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4009 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
4010 	}
4011 
4012 	/* Place the request on the outstanding list so we can keep track of it */
4013 	nvmf_add_to_outstanding_queue(req);
4014 
4015 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4016 		status = nvmf_ctrlr_process_fabrics_cmd(req);
4017 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4018 		status = nvmf_ctrlr_process_admin_cmd(req);
4019 	} else {
4020 		status = nvmf_ctrlr_process_io_cmd(req);
4021 	}
4022 
4023 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4024 		_nvmf_request_complete(req);
4025 	}
4026 }
4027 
4028 static bool
4029 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
4030 		       struct spdk_dif_ctx *dif_ctx)
4031 {
4032 	struct spdk_nvmf_ns *ns;
4033 	struct spdk_bdev *bdev;
4034 
4035 	if (ctrlr == NULL || cmd == NULL) {
4036 		return false;
4037 	}
4038 
4039 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
4040 	if (ns == NULL || ns->bdev == NULL) {
4041 		return false;
4042 	}
4043 
4044 	bdev = ns->bdev;
4045 
4046 	switch (cmd->opc) {
4047 	case SPDK_NVME_OPC_READ:
4048 	case SPDK_NVME_OPC_WRITE:
4049 	case SPDK_NVME_OPC_COMPARE:
4050 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
4051 	default:
4052 		break;
4053 	}
4054 
4055 	return false;
4056 }
4057 
4058 bool
4059 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
4060 {
4061 	struct spdk_nvmf_qpair *qpair = req->qpair;
4062 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4063 
4064 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
4065 		return false;
4066 	}
4067 
4068 	if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) {
4069 		return false;
4070 	}
4071 
4072 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4073 		return false;
4074 	}
4075 
4076 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4077 		return false;
4078 	}
4079 
4080 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
4081 }
4082 
4083 void
4084 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
4085 {
4086 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
4087 }
4088 
4089 static int
4090 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
4091 {
4092 	struct spdk_bdev *bdev;
4093 	struct spdk_bdev_desc *desc;
4094 	struct spdk_io_channel *ch;
4095 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
4096 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
4097 	uint32_t bdev_nsid;
4098 	int rc;
4099 
4100 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid == 0) {
4101 		bdev_nsid = cmd->nsid;
4102 	} else {
4103 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
4104 	}
4105 
4106 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
4107 	if (rc) {
4108 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4109 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4110 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4111 	}
4112 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
4113 }
4114 
4115 void
4116 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
4117 {
4118 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
4119 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
4120 }
4121 
4122 int
4123 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
4124 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
4125 {
4126 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
4127 	struct spdk_nvmf_ns *ns;
4128 	struct spdk_nvmf_poll_group *group = req->qpair->group;
4129 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4130 
4131 	*bdev = NULL;
4132 	*desc = NULL;
4133 	*ch = NULL;
4134 
4135 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
4136 	if (ns == NULL || ns->bdev == NULL) {
4137 		return -EINVAL;
4138 	}
4139 
4140 	assert(group != NULL && group->sgroups != NULL);
4141 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
4142 	*bdev = ns->bdev;
4143 	*desc = ns->desc;
4144 	*ch = ns_info->channel;
4145 
4146 	return 0;
4147 }
4148 
4149 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
4150 {
4151 	return req->qpair->ctrlr;
4152 }
4153 
4154 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
4155 {
4156 	return &req->cmd->nvme_cmd;
4157 }
4158 
4159 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
4160 {
4161 	return &req->rsp->nvme_cpl;
4162 }
4163 
4164 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
4165 {
4166 	return req->qpair->ctrlr->subsys;
4167 }
4168 
4169 void spdk_nvmf_request_get_data(struct spdk_nvmf_request *req, void **data, uint32_t *length)
4170 {
4171 	*data = req->data;
4172 	*length = req->length;
4173 }
4174 
4175 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
4176 {
4177 	return ctrlr->subsys;
4178 }
4179 
4180 uint16_t spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
4181 {
4182 	return ctrlr->cntlid;
4183 }
4184 
4185 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
4186 {
4187 	return req->req_to_abort;
4188 }
4189