xref: /spdk/lib/nvmf/ctrlr.c (revision 26ae3d66d89392c108a30e405ca8424617a03417)
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/trace.h"
43 #include "spdk/nvme_spec.h"
44 #include "spdk/nvmf_cmd.h"
45 #include "spdk/string.h"
46 #include "spdk/util.h"
47 #include "spdk/version.h"
48 
49 #include "spdk/log.h"
50 
51 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000
52 #define NVMF_DISC_KATO_IN_MS 120000
53 #define KAS_TIME_UNIT_IN_MS 100
54 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS)
55 
56 /*
57  * Report the SPDK version as the firmware revision.
58  * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts.
59  */
60 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING
61 
62 #define ANA_TRANSITION_TIME_IN_SEC 10
63 
64 /*
65  * Support for custom admin command handlers
66  */
67 struct spdk_nvmf_custom_admin_cmd {
68 	spdk_nvmf_custom_cmd_hdlr hdlr;
69 	uint32_t nsid; /* nsid to forward */
70 };
71 
72 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1];
73 
74 static void _nvmf_request_complete(void *ctx);
75 
76 static inline void
77 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp,
78 			      uint8_t iattr, uint16_t ipo)
79 {
80 	rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
81 	rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
82 	rsp->status_code_specific.invalid.iattr = iattr;
83 	rsp->status_code_specific.invalid.ipo = ipo;
84 }
85 
86 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field)	\
87 	nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field))
88 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field)	\
89 	nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field))
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 from subsystem %s due to keep alive timeout.\n",
186 			       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 
257 static void
258 _nvmf_ctrlr_add_admin_qpair(void *ctx)
259 {
260 	struct spdk_nvmf_request *req = ctx;
261 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
262 	struct spdk_nvmf_qpair *qpair = req->qpair;
263 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
264 
265 	ctrlr->admin_qpair = qpair;
266 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
267 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
268 	_nvmf_request_complete(req);
269 }
270 
271 static void
272 _nvmf_subsystem_add_ctrlr(void *ctx)
273 {
274 	struct spdk_nvmf_request *req = ctx;
275 	struct spdk_nvmf_qpair *qpair = req->qpair;
276 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
277 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
278 
279 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
280 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
281 		spdk_bit_array_free(&ctrlr->qpair_mask);
282 		free(ctrlr);
283 		qpair->ctrlr = NULL;
284 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
285 		spdk_nvmf_request_complete(req);
286 		return;
287 	}
288 
289 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
290 }
291 
292 static void
293 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
294 		      struct spdk_nvmf_ctrlr_data *cdata)
295 {
296 	cdata->kas = KAS_DEFAULT_VALUE;
297 	cdata->sgls.supported = 1;
298 	cdata->sgls.keyed_sgl = 1;
299 	cdata->sgls.sgl_offset = 1;
300 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
301 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
302 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
303 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
304 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
305 	cdata->nvmf_specific.msdbd = 1;
306 
307 	if (transport->ops->cdata_init) {
308 		transport->ops->cdata_init(transport, subsystem, cdata);
309 	}
310 }
311 
312 static struct spdk_nvmf_ctrlr *
313 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
314 		  struct spdk_nvmf_request *req,
315 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
316 		  struct spdk_nvmf_fabric_connect_data *connect_data)
317 {
318 	struct spdk_nvmf_ctrlr	*ctrlr;
319 	struct spdk_nvmf_transport *transport;
320 
321 	ctrlr = calloc(1, sizeof(*ctrlr));
322 	if (ctrlr == NULL) {
323 		SPDK_ERRLOG("Memory allocation failed\n");
324 		return NULL;
325 	}
326 
327 	TAILQ_INIT(&ctrlr->log_head);
328 	ctrlr->subsys = subsystem;
329 	ctrlr->thread = req->qpair->group->thread;
330 
331 	transport = req->qpair->transport;
332 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
333 	if (!ctrlr->qpair_mask) {
334 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
335 		free(ctrlr);
336 		return NULL;
337 	}
338 
339 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
340 
341 	/*
342 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
343 	 * If this field is cleared to 0h, then Keep Alive is not supported.
344 	 */
345 	if (ctrlr->cdata.kas) {
346 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
347 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
348 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
349 	}
350 
351 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
352 	if (ctrlr->subsys->flags.ana_reporting) {
353 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
354 	}
355 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
356 
357 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
358 		/*
359 		 * If keep-alive timeout is not set, discovery controllers use some
360 		 * arbitrary high value in order to cleanup stale discovery sessions
361 		 *
362 		 * From the 1.0a nvme-of spec:
363 		 * "The Keep Alive command is reserved for
364 		 * Discovery controllers. A transport may specify a
365 		 * fixed Discovery controller activity timeout value
366 		 * (e.g., 2 minutes). If no commands are received
367 		 * by a Discovery controller within that time
368 		 * period, the controller may perform the
369 		 * actions for Keep Alive Timer expiration".
370 		 *
371 		 * From the 1.1 nvme-of spec:
372 		 * "A host requests an explicit persistent connection
373 		 * to a Discovery controller and Asynchronous Event Notifications from
374 		 * the Discovery controller on that persistent connection by specifying
375 		 * a non-zero Keep Alive Timer value in the Connect command."
376 		 *
377 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
378 		 * otherwise we disable it and use default discovery controller KATO.
379 		 * KATO is in millisecond.
380 		 */
381 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
382 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
383 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
384 		} else {
385 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
386 		}
387 	}
388 
389 	/* Subtract 1 for admin queue, 1 for 0's based */
390 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
391 			1;
392 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
393 			1;
394 
395 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
396 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn));
397 
398 	ctrlr->vcprop.cap.raw = 0;
399 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
400 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
401 				      1; /* max queue depth */
402 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
403 	ctrlr->vcprop.cap.bits.to = 1; /* ready timeout - 500 msec units */
404 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
405 	ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
406 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
407 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
408 
409 	/* Version Supported: 1.3 */
410 	ctrlr->vcprop.vs.bits.mjr = 1;
411 	ctrlr->vcprop.vs.bits.mnr = 3;
412 	ctrlr->vcprop.vs.bits.ter = 0;
413 
414 	ctrlr->vcprop.cc.raw = 0;
415 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
416 
417 	ctrlr->vcprop.csts.raw = 0;
418 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
419 
420 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
421 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
422 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
423 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
424 
425 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
426 
427 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
428 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys,
429 				  req->qpair->trid);
430 		if (!ctrlr->listener) {
431 			SPDK_ERRLOG("Listener was not found\n");
432 			free(ctrlr);
433 			return NULL;
434 		}
435 	}
436 
437 	req->qpair->ctrlr = ctrlr;
438 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
439 
440 	return ctrlr;
441 }
442 
443 static void
444 _nvmf_ctrlr_destruct(void *ctx)
445 {
446 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
447 	struct spdk_nvmf_reservation_log *log, *log_tmp;
448 
449 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
450 	nvmf_ctrlr_stop_association_timer(ctrlr);
451 	spdk_bit_array_free(&ctrlr->qpair_mask);
452 
453 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
454 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
455 		free(log);
456 	}
457 	free(ctrlr);
458 }
459 
460 void
461 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
462 {
463 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
464 
465 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
466 }
467 
468 static void
469 nvmf_ctrlr_add_io_qpair(void *ctx)
470 {
471 	struct spdk_nvmf_request *req = ctx;
472 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
473 	struct spdk_nvmf_qpair *qpair = req->qpair;
474 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
475 
476 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
477 	  * For error case, the value should be NULL. So set it to NULL at first.
478 	  */
479 	qpair->ctrlr = NULL;
480 
481 	/* Make sure the controller is not being destroyed. */
482 	if (ctrlr->in_destruct) {
483 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
484 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
485 		goto end;
486 	}
487 
488 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
489 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
490 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
491 		goto end;
492 	}
493 
494 	if (!ctrlr->vcprop.cc.bits.en) {
495 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
496 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
497 		goto end;
498 	}
499 
500 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
501 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
502 			    ctrlr->vcprop.cc.bits.iosqes);
503 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
504 		goto end;
505 	}
506 
507 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
508 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
509 			    ctrlr->vcprop.cc.bits.iocqes);
510 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
511 		goto end;
512 	}
513 
514 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
515 end:
516 	spdk_nvmf_request_complete(req);
517 }
518 
519 static void
520 _nvmf_ctrlr_add_io_qpair(void *ctx)
521 {
522 	struct spdk_nvmf_request *req = ctx;
523 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
524 	struct spdk_nvmf_fabric_connect_data *data = req->data;
525 	struct spdk_nvmf_ctrlr *ctrlr;
526 	struct spdk_nvmf_qpair *qpair = req->qpair;
527 	struct spdk_nvmf_qpair *admin_qpair;
528 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
529 	struct spdk_nvmf_subsystem *subsystem;
530 	const struct spdk_nvmf_subsystem_listener *listener;
531 
532 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
533 
534 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
535 	/* We already checked this in spdk_nvmf_ctrlr_connect */
536 	assert(subsystem != NULL);
537 
538 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
539 	if (ctrlr == NULL) {
540 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
541 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
542 		spdk_nvmf_request_complete(req);
543 		return;
544 	}
545 
546 	/* fail before passing a message to the controller thread. */
547 	if (ctrlr->in_destruct) {
548 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
549 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
550 		spdk_nvmf_request_complete(req);
551 		return;
552 	}
553 
554 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
555 	if (subsystem->flags.ana_reporting) {
556 		listener = nvmf_subsystem_find_listener(subsystem, qpair->trid);
557 		if (listener != ctrlr->listener) {
558 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
559 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
560 			spdk_nvmf_request_complete(req);
561 			return;
562 		}
563 	}
564 
565 	admin_qpair = ctrlr->admin_qpair;
566 	qpair->ctrlr = ctrlr;
567 	spdk_thread_send_msg(admin_qpair->group->thread, nvmf_ctrlr_add_io_qpair, req);
568 }
569 
570 static bool
571 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
572 			  const char *hostnqn)
573 {
574 	struct spdk_nvme_transport_id listen_trid = {};
575 
576 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
577 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
578 		return false;
579 	}
580 
581 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
582 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
583 			    subsystem->subnqn);
584 		return false;
585 	}
586 
587 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
588 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
589 			    subsystem->subnqn, hostnqn);
590 		return false;
591 	}
592 
593 	return true;
594 }
595 
596 static int
597 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
598 {
599 	struct spdk_nvmf_fabric_connect_data *data = req->data;
600 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
601 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
602 	struct spdk_nvmf_qpair *qpair = req->qpair;
603 	struct spdk_nvmf_transport *transport = qpair->transport;
604 	struct spdk_nvmf_ctrlr *ctrlr;
605 	struct spdk_nvmf_subsystem *subsystem;
606 
607 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
608 		      cmd->recfmt, cmd->qid, cmd->sqsize);
609 
610 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
611 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
612 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
613 		      ntohl(*(uint32_t *)&data->hostid[0]),
614 		      ntohs(*(uint16_t *)&data->hostid[4]),
615 		      ntohs(*(uint16_t *)&data->hostid[6]),
616 		      data->hostid[8],
617 		      data->hostid[9],
618 		      ntohs(*(uint16_t *)&data->hostid[10]),
619 		      ntohl(*(uint32_t *)&data->hostid[12]));
620 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
621 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
622 
623 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
624 	if (!subsystem) {
625 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
626 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
627 	}
628 
629 	if (cmd->recfmt != 0) {
630 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
631 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
632 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
633 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
634 	}
635 
636 	/*
637 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
638 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
639 	 */
640 	if (cmd->sqsize == 0) {
641 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
642 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
643 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
644 	}
645 
646 	if (cmd->qid == 0) {
647 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
648 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
649 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
650 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
651 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
652 		}
653 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
654 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
655 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
656 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
657 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
658 	}
659 
660 	qpair->sq_head_max = cmd->sqsize;
661 	qpair->qid = cmd->qid;
662 
663 	if (0 == qpair->qid) {
664 		qpair->group->stat.admin_qpairs++;
665 	} else {
666 		qpair->group->stat.io_qpairs++;
667 	}
668 
669 	if (cmd->qid == 0) {
670 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
671 
672 		if (data->cntlid != 0xFFFF) {
673 			/* This NVMf target only supports dynamic mode. */
674 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
675 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
676 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
677 		}
678 
679 		/* Establish a new ctrlr */
680 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
681 		if (!ctrlr) {
682 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
683 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
684 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
685 		} else {
686 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
687 		}
688 	} else {
689 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
690 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
691 	}
692 }
693 
694 static inline bool
695 nvmf_request_is_fabric_connect(struct spdk_nvmf_request *req)
696 {
697 	return req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC &&
698 	       req->cmd->nvmf_cmd.fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT;
699 }
700 
701 static struct spdk_nvmf_subsystem_poll_group *
702 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
703 {
704 	struct spdk_nvmf_fabric_connect_data *data;
705 	struct spdk_nvmf_subsystem *subsystem;
706 	struct spdk_nvmf_tgt *tgt;
707 
708 	assert(nvmf_request_is_fabric_connect(req));
709 	assert(req->qpair->ctrlr == NULL);
710 
711 	data = req->data;
712 	tgt = req->qpair->transport->tgt;
713 
714 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
715 	if (subsystem == NULL) {
716 		return NULL;
717 	}
718 
719 	return &req->qpair->group->sgroups[subsystem->id];
720 }
721 
722 int
723 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
724 {
725 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
726 	struct spdk_nvmf_qpair *qpair = req->qpair;
727 	struct spdk_nvmf_subsystem_poll_group *sgroup;
728 	enum spdk_nvmf_request_exec_status status;
729 
730 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
731 	if (!sgroup) {
732 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
733 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
734 		goto out;
735 	}
736 
737 	sgroup->io_outstanding++;
738 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
739 
740 	status = _nvmf_ctrlr_connect(req);
741 
742 out:
743 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
744 		_nvmf_request_complete(req);
745 	}
746 
747 	return status;
748 }
749 
750 static int
751 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
752 {
753 	struct spdk_nvmf_fabric_connect_data *data = req->data;
754 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
755 	struct spdk_nvmf_transport *transport = req->qpair->transport;
756 	struct spdk_nvmf_subsystem *subsystem;
757 
758 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
759 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
760 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
761 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
762 	}
763 
764 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
765 	if (!subsystem) {
766 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
767 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
768 	}
769 
770 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
771 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
772 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
773 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
774 		SPDK_ERRLOG("Subsystem '%s' is not ready\n", subsystem->subnqn);
775 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
776 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_CONTROLLER_BUSY;
777 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
778 	}
779 
780 	/* Ensure that hostnqn is null terminated */
781 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
782 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
783 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
784 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
785 	}
786 
787 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
788 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
789 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
790 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
791 	}
792 
793 	return _nvmf_ctrlr_connect(req);
794 }
795 
796 static int
797 nvmf_ctrlr_association_remove(void *ctx)
798 {
799 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
800 	int rc;
801 
802 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
803 		      ctrlr->subsys->subnqn);
804 
805 	rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL);
806 	if (rc < 0) {
807 		SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
808 		assert(false);
809 	}
810 
811 	nvmf_ctrlr_stop_association_timer(ctrlr);
812 	return 1;
813 }
814 
815 static void
816 nvmf_ctrlr_cc_shn_done(struct spdk_io_channel_iter *i, int status)
817 {
818 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
819 
820 	if (status < 0) {
821 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
822 		assert(false);
823 	}
824 
825 	ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
826 
827 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
828 	 * between the host and controller shall be preserved for at least 2 minutes */
829 	if (ctrlr->association_timer) {
830 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
831 		nvmf_ctrlr_stop_association_timer(ctrlr);
832 	}
833 	ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
834 				   ctrlr->admin_qpair->transport->opts.association_timeout * 1000);
835 }
836 
837 static void
838 nvmf_ctrlr_cc_reset_done(struct spdk_io_channel_iter *i, int status)
839 {
840 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
841 
842 	if (status < 0) {
843 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
844 		assert(false);
845 	}
846 
847 	/* Only a subset of the registers are cleared out on a reset */
848 	ctrlr->vcprop.cc.raw = 0;
849 	ctrlr->vcprop.csts.raw = 0;
850 
851 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
852 	 * between the host and controller shall be preserved for at least 2 minutes */
853 	if (ctrlr->association_timer) {
854 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
855 		nvmf_ctrlr_stop_association_timer(ctrlr);
856 	}
857 	ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
858 				   ctrlr->admin_qpair->transport->opts.association_timeout * 1000);
859 }
860 
861 const struct spdk_nvmf_registers *
862 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
863 {
864 	return &ctrlr->vcprop;
865 }
866 
867 static uint64_t
868 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
869 {
870 	return ctrlr->vcprop.cap.raw;
871 }
872 
873 static uint64_t
874 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
875 {
876 	return ctrlr->vcprop.vs.raw;
877 }
878 
879 static uint64_t
880 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
881 {
882 	return ctrlr->vcprop.cc.raw;
883 }
884 
885 static bool
886 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
887 {
888 	union spdk_nvme_cc_register cc, diff;
889 
890 	cc.raw = value;
891 
892 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
893 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
894 
895 	/*
896 	 * Calculate which bits changed between the current and new CC.
897 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
898 	 */
899 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
900 
901 	if (diff.bits.en) {
902 		if (cc.bits.en) {
903 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
904 			nvmf_ctrlr_stop_association_timer(ctrlr);
905 
906 			ctrlr->vcprop.cc.bits.en = 1;
907 			ctrlr->vcprop.csts.bits.rdy = 1;
908 		} else {
909 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
910 			ctrlr->vcprop.cc.bits.en = 0;
911 			spdk_for_each_channel(ctrlr->subsys->tgt,
912 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
913 					      ctrlr,
914 					      nvmf_ctrlr_cc_reset_done);
915 		}
916 		diff.bits.en = 0;
917 	}
918 
919 	if (diff.bits.shn) {
920 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
921 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
922 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
923 				      cc.bits.shn >> 1, cc.bits.shn & 1);
924 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
925 			spdk_for_each_channel(ctrlr->subsys->tgt,
926 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
927 					      ctrlr,
928 					      nvmf_ctrlr_cc_shn_done);
929 
930 			/* From the time a shutdown is initiated the controller shall disable
931 			 * Keep Alive timer */
932 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
933 		} else if (cc.bits.shn == 0) {
934 			ctrlr->vcprop.cc.bits.shn = 0;
935 		} else {
936 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
937 				    cc.bits.shn >> 1, cc.bits.shn & 1);
938 			return false;
939 		}
940 		diff.bits.shn = 0;
941 	}
942 
943 	if (diff.bits.iosqes) {
944 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
945 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
946 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
947 		diff.bits.iosqes = 0;
948 	}
949 
950 	if (diff.bits.iocqes) {
951 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
952 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
953 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
954 		diff.bits.iocqes = 0;
955 	}
956 
957 	if (diff.bits.ams) {
958 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
959 		return false;
960 	}
961 
962 	if (diff.bits.mps) {
963 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
964 		return false;
965 	}
966 
967 	if (diff.bits.css) {
968 		SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
969 		return false;
970 	}
971 
972 	if (diff.raw != 0) {
973 		/* Print an error message, but don't fail the command in this case.
974 		 * If we did want to fail in this case, we'd need to ensure we acted
975 		 * on no other bits or the initiator gets confused. */
976 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
977 	}
978 
979 	return true;
980 }
981 
982 static uint64_t
983 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
984 {
985 	return ctrlr->vcprop.csts.raw;
986 }
987 
988 static uint64_t
989 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
990 {
991 	return ctrlr->vcprop.aqa.raw;
992 }
993 
994 static bool
995 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
996 {
997 	union spdk_nvme_aqa_register aqa;
998 
999 	aqa.raw = value;
1000 
1001 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1002 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1003 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1004 		return false;
1005 	}
1006 
1007 	ctrlr->vcprop.aqa.raw = value;
1008 
1009 	return true;
1010 }
1011 
1012 static uint64_t
1013 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1014 {
1015 	return ctrlr->vcprop.asq;
1016 }
1017 
1018 static bool
1019 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1020 {
1021 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1022 
1023 	return true;
1024 }
1025 
1026 static bool
1027 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1028 {
1029 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1030 
1031 	return true;
1032 }
1033 
1034 static uint64_t
1035 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1036 {
1037 	return ctrlr->vcprop.acq;
1038 }
1039 
1040 static bool
1041 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1042 {
1043 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1044 
1045 	return true;
1046 }
1047 
1048 static bool
1049 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1050 {
1051 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1052 
1053 	return true;
1054 }
1055 
1056 struct nvmf_prop {
1057 	uint32_t ofst;
1058 	uint8_t size;
1059 	char name[11];
1060 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1061 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1062 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1063 };
1064 
1065 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1066 	{ \
1067 		offsetof(struct spdk_nvme_registers, field), \
1068 		size, \
1069 		#field, \
1070 		get_cb, set_cb, set_upper_cb \
1071 	}
1072 
1073 static const struct nvmf_prop nvmf_props[] = {
1074 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1075 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1076 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1077 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1078 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1079 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1080 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1081 };
1082 
1083 static const struct nvmf_prop *
1084 find_prop(uint32_t ofst, uint8_t size)
1085 {
1086 	size_t i;
1087 
1088 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1089 		const struct nvmf_prop *prop = &nvmf_props[i];
1090 
1091 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1092 			return prop;
1093 		}
1094 	}
1095 
1096 	return NULL;
1097 }
1098 
1099 static int
1100 nvmf_property_get(struct spdk_nvmf_request *req)
1101 {
1102 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1103 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1104 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1105 	const struct nvmf_prop *prop;
1106 	uint8_t size;
1107 
1108 	response->status.sc = 0;
1109 	response->value.u64 = 0;
1110 
1111 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1112 		      cmd->attrib.size, cmd->ofst);
1113 
1114 	switch (cmd->attrib.size) {
1115 	case SPDK_NVMF_PROP_SIZE_4:
1116 		size = 4;
1117 		break;
1118 	case SPDK_NVMF_PROP_SIZE_8:
1119 		size = 8;
1120 		break;
1121 	default:
1122 		SPDK_ERRLOG("Invalid size value %d\n", cmd->attrib.size);
1123 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1124 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1125 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1126 	}
1127 
1128 	prop = find_prop(cmd->ofst, size);
1129 	if (prop == NULL || prop->get_cb == NULL) {
1130 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1131 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1132 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1133 	}
1134 
1135 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1136 
1137 	response->value.u64 = prop->get_cb(ctrlr);
1138 
1139 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1140 
1141 	if (size != prop->size) {
1142 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1143 		assert(size == 4);
1144 		assert(prop->size == 8);
1145 
1146 		if (cmd->ofst == prop->ofst) {
1147 			/* Keep bottom 4 bytes only */
1148 			response->value.u64 &= 0xFFFFFFFF;
1149 		} else {
1150 			/* Keep top 4 bytes only */
1151 			response->value.u64 >>= 32;
1152 		}
1153 	}
1154 
1155 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1156 }
1157 
1158 static int
1159 nvmf_property_set(struct spdk_nvmf_request *req)
1160 {
1161 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1162 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1163 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1164 	const struct nvmf_prop *prop;
1165 	uint64_t value;
1166 	uint8_t size;
1167 	bool ret;
1168 
1169 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1170 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1171 
1172 	switch (cmd->attrib.size) {
1173 	case SPDK_NVMF_PROP_SIZE_4:
1174 		size = 4;
1175 		break;
1176 	case SPDK_NVMF_PROP_SIZE_8:
1177 		size = 8;
1178 		break;
1179 	default:
1180 		SPDK_ERRLOG("Invalid size value %d\n", cmd->attrib.size);
1181 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1182 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1183 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1184 	}
1185 
1186 	prop = find_prop(cmd->ofst, size);
1187 	if (prop == NULL || prop->set_cb == NULL) {
1188 		SPDK_ERRLOG("Invalid offset 0x%x\n", cmd->ofst);
1189 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1190 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1191 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1192 	}
1193 
1194 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1195 
1196 	value = cmd->value.u64;
1197 
1198 	if (prop->size == 4) {
1199 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1200 	} else if (size != prop->size) {
1201 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1202 		assert(size == 4);
1203 		assert(prop->size == 8);
1204 
1205 		if (cmd->ofst == prop->ofst) {
1206 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1207 		} else {
1208 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1209 		}
1210 	} else {
1211 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1212 		if (ret) {
1213 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1214 		}
1215 	}
1216 
1217 	if (!ret) {
1218 		SPDK_ERRLOG("prop set_cb failed\n");
1219 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1220 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1221 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1222 	}
1223 
1224 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1225 }
1226 
1227 static int
1228 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1229 {
1230 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1231 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1232 
1233 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1234 
1235 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1236 	ctrlr->feat.arbitration.bits.reserved = 0;
1237 
1238 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1239 }
1240 
1241 static int
1242 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1243 {
1244 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1245 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1246 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1247 
1248 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1249 
1250 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1251 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1252 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1253 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1254 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1255 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1256 	}
1257 
1258 	ctrlr->feat.power_management.raw = cmd->cdw11;
1259 	ctrlr->feat.power_management.bits.reserved = 0;
1260 
1261 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1262 }
1263 
1264 static bool
1265 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1266 {
1267 	/*
1268 	 * Valid TMPSEL values:
1269 	 *  0000b - 1000b: temperature sensors
1270 	 *  1111b: set all implemented temperature sensors
1271 	 */
1272 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1273 		/* 1001b - 1110b: reserved */
1274 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1275 		return false;
1276 	}
1277 
1278 	/*
1279 	 * Valid THSEL values:
1280 	 *  00b: over temperature threshold
1281 	 *  01b: under temperature threshold
1282 	 */
1283 	if (opts->bits.thsel > 1) {
1284 		/* 10b - 11b: reserved */
1285 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1286 		return false;
1287 	}
1288 
1289 	return true;
1290 }
1291 
1292 static int
1293 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1294 {
1295 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1296 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1297 
1298 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1299 
1300 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1301 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1302 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1303 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1304 	}
1305 
1306 	/* TODO: no sensors implemented - ignore new values */
1307 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1308 }
1309 
1310 static int
1311 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1312 {
1313 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1314 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1315 
1316 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1317 
1318 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1319 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1320 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1321 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1322 	}
1323 
1324 	/* TODO: no sensors implemented - return 0 for all thresholds */
1325 	rsp->cdw0 = 0;
1326 
1327 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1328 }
1329 
1330 static int
1331 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1332 {
1333 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1334 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1335 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1336 
1337 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1338 
1339 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1340 		/*
1341 		 * Host is not allowed to set this bit, since we don't advertise it in
1342 		 * Identify Namespace.
1343 		 */
1344 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1345 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1346 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1347 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1348 	}
1349 
1350 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1351 	ctrlr->feat.error_recovery.bits.reserved = 0;
1352 
1353 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1354 }
1355 
1356 static int
1357 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1358 {
1359 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1360 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1361 
1362 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1363 
1364 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1365 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1366 
1367 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1368 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1369 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1370 }
1371 
1372 static int
1373 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1374 {
1375 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1376 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1377 
1378 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1379 
1380 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1381 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1382 
1383 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1384 }
1385 
1386 static int
1387 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1388 {
1389 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1390 
1391 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1392 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1393 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1394 }
1395 
1396 static int
1397 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1398 {
1399 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1400 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1401 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1402 
1403 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1404 
1405 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1406 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1407 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1408 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1409 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1410 	}
1411 
1412 	if (req->data == NULL || req->length < sizeof(ctrlr->hostid)) {
1413 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1414 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1415 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1416 	}
1417 
1418 	spdk_uuid_copy((struct spdk_uuid *)req->data, &ctrlr->hostid);
1419 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1420 }
1421 
1422 static int
1423 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1424 {
1425 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1426 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1427 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1428 	struct spdk_nvmf_ns *ns;
1429 
1430 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notificaton Mask\n");
1431 
1432 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1433 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1434 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1435 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1436 	}
1437 
1438 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1439 	if (ns == NULL) {
1440 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1441 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1442 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1443 	}
1444 	rsp->cdw0 = ns->mask;
1445 
1446 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1447 }
1448 
1449 static int
1450 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1451 {
1452 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1453 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1454 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1455 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1456 	struct spdk_nvmf_ns *ns;
1457 
1458 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notificaton Mask\n");
1459 
1460 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1461 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1462 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1463 			ns->mask = cmd->cdw11;
1464 		}
1465 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1466 	}
1467 
1468 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1469 	if (ns == NULL) {
1470 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1471 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1472 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1473 	}
1474 	ns->mask = cmd->cdw11;
1475 
1476 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1477 }
1478 
1479 static int
1480 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1481 {
1482 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1483 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1484 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1485 	struct spdk_nvmf_ns *ns;
1486 
1487 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1488 
1489 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1490 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1491 	if (ns == NULL) {
1492 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1493 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1494 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1495 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1496 	}
1497 
1498 	response->cdw0 = ns->ptpl_activated;
1499 
1500 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1501 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1502 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1503 }
1504 
1505 static int
1506 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req)
1507 {
1508 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1509 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1510 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1511 	struct spdk_nvmf_ns *ns;
1512 	bool ptpl;
1513 
1514 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1515 
1516 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1517 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1518 
1519 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && ns->ptpl_file) {
1520 		ns->ptpl_activated = ptpl;
1521 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1522 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns && ns->ptpl_file;
1523 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1524 			ns->ptpl_activated = ptpl;
1525 		}
1526 	} else {
1527 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1528 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1529 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1530 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1531 	}
1532 
1533 	/* TODO: Feature not changeable for now */
1534 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1535 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1536 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1537 }
1538 
1539 static int
1540 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
1541 {
1542 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1543 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1544 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1545 
1546 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
1547 
1548 	/*
1549 	 * if attempts to disable keep alive by setting kato to 0h
1550 	 * a status value of keep alive invalid shall be returned
1551 	 */
1552 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
1553 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
1554 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
1555 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
1556 	} else {
1557 		/* round up to milliseconds */
1558 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
1559 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
1560 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
1561 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
1562 	}
1563 
1564 	/*
1565 	 * if change the keep alive timeout value successfully
1566 	 * update the keep alive poller.
1567 	 */
1568 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
1569 		if (ctrlr->keep_alive_poller != NULL) {
1570 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
1571 		}
1572 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
1573 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
1574 	}
1575 
1576 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
1577 		      ctrlr->feat.keep_alive_timer.bits.kato);
1578 
1579 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1580 }
1581 
1582 static int
1583 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
1584 {
1585 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1586 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1587 	uint32_t count;
1588 
1589 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
1590 		      req->cmd->nvme_cmd.cdw11);
1591 
1592 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1593 	/* verify that the controller is ready to process commands */
1594 	if (count > 1) {
1595 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
1596 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1597 	} else {
1598 		/*
1599 		 * Ignore the value requested by the host -
1600 		 * always return the pre-configured value based on max_qpairs_allowed.
1601 		 */
1602 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
1603 	}
1604 
1605 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1606 }
1607 
1608 static int
1609 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
1610 {
1611 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1612 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1613 
1614 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
1615 		      cmd->cdw11);
1616 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
1617 	ctrlr->feat.async_event_configuration.bits.reserved = 0;
1618 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1619 }
1620 
1621 static int
1622 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
1623 {
1624 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1625 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1626 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1627 
1628 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
1629 
1630 	/* AER cmd is an exception */
1631 	sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
1632 	assert(sgroup != NULL);
1633 	sgroup->io_outstanding--;
1634 
1635 	/* Four asynchronous events are supported for now */
1636 	if (ctrlr->nr_aer_reqs >= NVMF_MAX_ASYNC_EVENTS) {
1637 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
1638 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1639 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
1640 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1641 	}
1642 
1643 	if (ctrlr->notice_event.bits.async_event_type ==
1644 	    SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) {
1645 		rsp->cdw0 = ctrlr->notice_event.raw;
1646 		ctrlr->notice_event.raw = 0;
1647 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1648 	}
1649 
1650 	if (ctrlr->reservation_event.bits.async_event_type ==
1651 	    SPDK_NVME_ASYNC_EVENT_TYPE_IO) {
1652 		rsp->cdw0 = ctrlr->reservation_event.raw;
1653 		ctrlr->reservation_event.raw = 0;
1654 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1655 	}
1656 
1657 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
1658 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
1659 }
1660 
1661 static void
1662 nvmf_get_firmware_slot_log_page(void *buffer, uint64_t offset, uint32_t length)
1663 {
1664 	struct spdk_nvme_firmware_page fw_page;
1665 	size_t copy_len;
1666 
1667 	memset(&fw_page, 0, sizeof(fw_page));
1668 	fw_page.afi.active_slot = 1;
1669 	fw_page.afi.next_reset_slot = 0;
1670 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
1671 
1672 	if (offset < sizeof(fw_page)) {
1673 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
1674 		if (copy_len > 0) {
1675 			memcpy(buffer, (const char *)&fw_page + offset, copy_len);
1676 		}
1677 	}
1678 }
1679 
1680 #define SPDK_NVMF_ANA_DESC_SIZE	(sizeof(struct spdk_nvme_ana_group_descriptor) +	\
1681 				 sizeof(uint32_t))
1682 static void
1683 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, void *data,
1684 		      uint64_t offset, uint32_t length)
1685 {
1686 	char *buf = data;
1687 	struct spdk_nvme_ana_page ana_hdr;
1688 	char _ana_desc[SPDK_NVMF_ANA_DESC_SIZE];
1689 	struct spdk_nvme_ana_group_descriptor *ana_desc;
1690 	size_t copy_len;
1691 	uint32_t num_ns = 0;
1692 	struct spdk_nvmf_ns *ns;
1693 
1694 	if (length == 0) {
1695 		return;
1696 	}
1697 
1698 	if (offset >= sizeof(ana_hdr)) {
1699 		offset -= sizeof(ana_hdr);
1700 	} else {
1701 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1702 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1703 			num_ns++;
1704 		}
1705 
1706 		memset(&ana_hdr, 0, sizeof(ana_hdr));
1707 
1708 		ana_hdr.num_ana_group_desc = num_ns;
1709 		/* TODO: Support Change Count. */
1710 		ana_hdr.change_count = 0;
1711 
1712 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
1713 		memcpy(buf, (const char *)&ana_hdr + offset, copy_len);
1714 		length -= copy_len;
1715 		buf += copy_len;
1716 		offset = 0;
1717 	}
1718 
1719 	if (length == 0) {
1720 		return;
1721 	}
1722 
1723 	ana_desc = (void *)_ana_desc;
1724 
1725 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1726 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1727 		if (offset >= SPDK_NVMF_ANA_DESC_SIZE) {
1728 			offset -= SPDK_NVMF_ANA_DESC_SIZE;
1729 			continue;
1730 		}
1731 
1732 		memset(ana_desc, 0, SPDK_NVMF_ANA_DESC_SIZE);
1733 
1734 		ana_desc->ana_group_id = ns->nsid;
1735 		ana_desc->num_of_nsid = 1;
1736 		ana_desc->ana_state = ctrlr->listener->ana_state;
1737 		ana_desc->nsid[0] = ns->nsid;
1738 		/* TODO: Support Change Count. */
1739 		ana_desc->change_count = 0;
1740 
1741 		copy_len = spdk_min(SPDK_NVMF_ANA_DESC_SIZE - offset, length);
1742 		memcpy(buf, (const char *)ana_desc + offset, copy_len);
1743 		length -= copy_len;
1744 		buf += copy_len;
1745 		offset = 0;
1746 
1747 		if (length == 0) {
1748 			return;
1749 		}
1750 	}
1751 }
1752 
1753 void
1754 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
1755 {
1756 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
1757 	uint16_t i;
1758 	bool found = false;
1759 
1760 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
1761 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
1762 			/* nsid is already in the list */
1763 			found = true;
1764 			break;
1765 		}
1766 	}
1767 
1768 	if (!found) {
1769 		if (ctrlr->changed_ns_list_count == max_changes) {
1770 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
1771 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
1772 			for (i = 1; i < max_changes; i++) {
1773 				ctrlr->changed_ns_list.ns_list[i] = 0;
1774 			}
1775 		} else {
1776 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
1777 		}
1778 	}
1779 }
1780 
1781 static void
1782 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
1783 				  void *buffer, uint64_t offset, uint32_t length)
1784 {
1785 	size_t copy_length;
1786 
1787 	if (offset < sizeof(ctrlr->changed_ns_list)) {
1788 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
1789 		if (copy_length) {
1790 			memcpy(buffer, (char *)&ctrlr->changed_ns_list + offset, copy_length);
1791 		}
1792 	}
1793 
1794 	/* Clear log page each time it is read */
1795 	ctrlr->changed_ns_list_count = 0;
1796 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
1797 }
1798 
1799 /* The structure can be modified if we provide support for other commands in future */
1800 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
1801 	.admin_cmds_supported = {
1802 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
1803 		/* Get Log Page */
1804 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
1805 		/* Identify */
1806 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
1807 		/* Abort */
1808 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
1809 		/* Set Features */
1810 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
1811 		/* Get Features */
1812 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
1813 		/* Async Event Request */
1814 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
1815 		/* Keep Alive */
1816 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
1817 	},
1818 	.io_cmds_supported = {
1819 		/* FLUSH */
1820 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
1821 		/* WRITE */
1822 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
1823 		/* READ */
1824 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
1825 		/* WRITE ZEROES */
1826 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
1827 		/* DATASET MANAGEMENT */
1828 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
1829 		/* COMPARE */
1830 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
1831 	},
1832 };
1833 
1834 static void
1835 nvmf_get_cmds_and_effects_log_page(void *buffer,
1836 				   uint64_t offset, uint32_t length)
1837 {
1838 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
1839 	size_t copy_len = 0;
1840 	size_t zero_len = length;
1841 
1842 	if (offset < page_size) {
1843 		copy_len = spdk_min(page_size - offset, length);
1844 		zero_len -= copy_len;
1845 		memcpy(buffer, (char *)(&g_cmds_and_effect_log_page) + offset, copy_len);
1846 	}
1847 
1848 	if (zero_len) {
1849 		memset((char *)buffer + copy_len, 0, zero_len);
1850 	}
1851 }
1852 
1853 static void
1854 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
1855 		void *data, uint64_t offset, uint32_t length)
1856 {
1857 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
1858 	struct spdk_nvmf_reservation_log *log, *log_tmp;
1859 	uint8_t *buf = data;
1860 
1861 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
1862 	/* No available log, return 1 zeroed log page */
1863 	if (!ctrlr->num_avail_log_pages) {
1864 		memset(buf, 0, spdk_min(length, unit_log_len));
1865 		return;
1866 	}
1867 
1868 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
1869 	if (offset >= avail_log_len) {
1870 		return;
1871 	}
1872 
1873 	next_pos = 0;
1874 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
1875 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
1876 		ctrlr->num_avail_log_pages--;
1877 
1878 		next_pos += unit_log_len;
1879 		if (next_pos > offset) {
1880 			copy_len = spdk_min(next_pos - offset, length);
1881 			memcpy(buf, &log->log, copy_len);
1882 			length -= copy_len;
1883 			offset += copy_len;
1884 			buf += copy_len;
1885 		}
1886 		free(log);
1887 
1888 		if (length == 0) {
1889 			break;
1890 		}
1891 	}
1892 	return;
1893 }
1894 
1895 static int
1896 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
1897 {
1898 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1899 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1900 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1901 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1902 	uint64_t offset, len;
1903 	uint32_t numdl, numdu;
1904 	uint8_t lid;
1905 
1906 	if (req->data == NULL) {
1907 		SPDK_ERRLOG("get log command with no buffer\n");
1908 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1909 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1910 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1911 	}
1912 
1913 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
1914 	if (offset & 3) {
1915 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
1916 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1917 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1918 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1919 	}
1920 
1921 	numdl = cmd->cdw10_bits.get_log_page.numdl;
1922 	numdu = cmd->cdw11_bits.get_log_page.numdu;
1923 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
1924 	if (len > req->length) {
1925 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
1926 			    len, req->length);
1927 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1928 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1929 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1930 	}
1931 
1932 	lid = cmd->cdw10_bits.get_log_page.lid;
1933 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 "\n",
1934 		      lid, offset, len);
1935 
1936 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
1937 		switch (lid) {
1938 		case SPDK_NVME_LOG_DISCOVERY:
1939 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt, offset,
1940 						    len);
1941 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1942 		default:
1943 			goto invalid_log_page;
1944 		}
1945 	} else {
1946 		switch (lid) {
1947 		case SPDK_NVME_LOG_ERROR:
1948 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
1949 			/* TODO: actually fill out log page data */
1950 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1951 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
1952 			nvmf_get_firmware_slot_log_page(req->data, offset, len);
1953 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1954 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
1955 			if (subsystem->flags.ana_reporting) {
1956 				nvmf_get_ana_log_page(ctrlr, req->data, offset, len);
1957 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1958 			} else {
1959 				goto invalid_log_page;
1960 			}
1961 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
1962 			nvmf_get_cmds_and_effects_log_page(req->data, offset, len);
1963 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1964 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
1965 			nvmf_get_changed_ns_list_log_page(ctrlr, req->data, offset, len);
1966 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1967 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
1968 			nvmf_get_reservation_notification_log_page(ctrlr, req->data, offset, len);
1969 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1970 		default:
1971 			goto invalid_log_page;
1972 		}
1973 	}
1974 
1975 invalid_log_page:
1976 	SPDK_ERRLOG("Unsupported Get Log Page 0x%02X\n", lid);
1977 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1978 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1979 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1980 }
1981 
1982 int
1983 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
1984 			    struct spdk_nvme_cmd *cmd,
1985 			    struct spdk_nvme_cpl *rsp,
1986 			    struct spdk_nvme_ns_data *nsdata)
1987 {
1988 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1989 	struct spdk_nvmf_ns *ns;
1990 	uint32_t max_num_blocks;
1991 
1992 	if (cmd->nsid == 0 || cmd->nsid > subsystem->max_nsid) {
1993 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", cmd->nsid);
1994 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1995 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
1996 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1997 	}
1998 
1999 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2000 	if (ns == NULL || ns->bdev == NULL) {
2001 		/*
2002 		 * Inactive namespaces should return a zero filled data structure.
2003 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2004 		 * so we can just return early here.
2005 		 */
2006 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", cmd->nsid);
2007 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2008 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2009 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2010 	}
2011 
2012 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2013 
2014 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2015 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2016 			 (1U << nsdata->lbaf[nsdata->flbas.format].lbads);
2017 	if (nsdata->noiob > max_num_blocks) {
2018 		nsdata->noiob = max_num_blocks;
2019 	}
2020 
2021 	if (subsystem->flags.ana_reporting) {
2022 		/* ANA group ID matches NSID. */
2023 		nsdata->anagrpid = ns->nsid;
2024 
2025 		if (ctrlr->listener->ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2026 		    ctrlr->listener->ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2027 			nsdata->nuse = 0;
2028 		}
2029 	}
2030 
2031 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2032 }
2033 
2034 static void
2035 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2036 			 struct spdk_nvme_ctrlr_data *cdata)
2037 {
2038 	cdata->oacs.virtualization_management =
2039 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2040 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2041 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2042 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2043 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2044 	cdata->oacs.device_self_test =
2045 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2046 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2047 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2048 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2049 			       NULL
2050 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2051 	cdata->oacs.format =
2052 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2053 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2054 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2055 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2056 				     NULL;
2057 }
2058 
2059 int
2060 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2061 {
2062 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2063 	struct spdk_nvmf_transport *transport = ctrlr->admin_qpair->transport;
2064 
2065 	/*
2066 	 * Common fields for discovery and NVM subsystems
2067 	 */
2068 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2069 	assert((transport->opts.max_io_size % 4096) == 0);
2070 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2071 	cdata->cntlid = ctrlr->cntlid;
2072 	cdata->ver = ctrlr->vcprop.vs;
2073 	cdata->aerl = NVMF_MAX_ASYNC_EVENTS - 1;
2074 	cdata->lpa.edlp = 1;
2075 	cdata->elpe = 127;
2076 	cdata->maxcmd = transport->opts.max_queue_depth;
2077 	cdata->sgls = ctrlr->cdata.sgls;
2078 	cdata->fuses.compare_and_write = 1;
2079 	cdata->acwu = 1;
2080 	if (subsystem->flags.ana_reporting) {
2081 		cdata->mnan = subsystem->max_nsid;
2082 	}
2083 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2084 
2085 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2086 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2087 
2088 
2089 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2090 		/*
2091 		 * NVM Discovery subsystem fields
2092 		 */
2093 		cdata->oaes.discovery_log_change_notices = 1;
2094 	} else {
2095 		/*
2096 		 * NVM subsystem fields (reserved for discovery subsystems)
2097 		 */
2098 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2099 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2100 		cdata->kas = ctrlr->cdata.kas;
2101 
2102 		cdata->rab = 6;
2103 		cdata->cmic.multi_port = 1;
2104 		cdata->cmic.multi_host = 1;
2105 		if (subsystem->flags.ana_reporting) {
2106 			/* Asymmetric Namespace Access Reporting is supported. */
2107 			cdata->cmic.ana_reporting = 1;
2108 		}
2109 		cdata->oaes.ns_attribute_notices = 1;
2110 		if (subsystem->flags.ana_reporting) {
2111 			cdata->oaes.ana_change_notices = 1;
2112 		}
2113 		cdata->ctratt.host_id_exhid_supported = 1;
2114 		/* TODO: Concurrent execution of multiple abort commands. */
2115 		cdata->acl = 0;
2116 		cdata->aerl = 0;
2117 		cdata->frmw.slot1_ro = 1;
2118 		cdata->frmw.num_slots = 1;
2119 
2120 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2121 
2122 		cdata->sqes.min = 6;
2123 		cdata->sqes.max = 6;
2124 		cdata->cqes.min = 4;
2125 		cdata->cqes.max = 4;
2126 		cdata->nn = subsystem->max_nsid;
2127 		cdata->vwc.present = 1;
2128 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
2129 
2130 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
2131 
2132 		cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr);
2133 		cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr);
2134 		cdata->oncs.reservations = 1;
2135 		if (subsystem->flags.ana_reporting) {
2136 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
2137 			/* ANA Change state is not used, and ANA Persistent Loss state
2138 			 * is not supported for now.
2139 			 */
2140 			cdata->anacap.ana_optimized_state = 1;
2141 			cdata->anacap.ana_non_optimized_state = 1;
2142 			cdata->anacap.ana_inaccessible_state = 1;
2143 			/* ANAGRPID does not change while namespace is attached to controller */
2144 			cdata->anacap.no_change_anagrpid = 1;
2145 			cdata->anagrpmax = subsystem->max_nsid;
2146 			cdata->nanagrpid = subsystem->max_nsid;
2147 		}
2148 
2149 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
2150 
2151 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
2152 			      cdata->nvmf_specific.ioccsz);
2153 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
2154 			      cdata->nvmf_specific.iorcsz);
2155 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
2156 			      cdata->nvmf_specific.icdoff);
2157 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
2158 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
2159 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
2160 			      cdata->nvmf_specific.msdbd);
2161 	}
2162 
2163 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2164 }
2165 
2166 static int
2167 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_subsystem *subsystem,
2168 				   struct spdk_nvme_cmd *cmd,
2169 				   struct spdk_nvme_cpl *rsp,
2170 				   struct spdk_nvme_ns_list *ns_list)
2171 {
2172 	struct spdk_nvmf_ns *ns;
2173 	uint32_t count = 0;
2174 
2175 	if (cmd->nsid >= 0xfffffffeUL) {
2176 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
2177 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2178 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2179 	}
2180 
2181 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
2182 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2183 		if (ns->opts.nsid <= cmd->nsid) {
2184 			continue;
2185 		}
2186 
2187 		ns_list->ns_list[count++] = ns->opts.nsid;
2188 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
2189 			break;
2190 		}
2191 	}
2192 
2193 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2194 }
2195 
2196 static void
2197 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
2198 		enum spdk_nvme_nidt type,
2199 		const void *data, size_t data_size)
2200 {
2201 	struct spdk_nvme_ns_id_desc *desc;
2202 	size_t desc_size = sizeof(*desc) + data_size;
2203 
2204 	/*
2205 	 * These should never fail in practice, since all valid NS ID descriptors
2206 	 * should be defined so that they fit in the available 4096-byte buffer.
2207 	 */
2208 	assert(data_size > 0);
2209 	assert(data_size <= UINT8_MAX);
2210 	assert(desc_size < *buf_remain);
2211 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
2212 		return;
2213 	}
2214 
2215 	desc = *buf_ptr;
2216 	desc->nidt = type;
2217 	desc->nidl = data_size;
2218 	memcpy(desc->nid, data, data_size);
2219 
2220 	*buf_ptr += desc_size;
2221 	*buf_remain -= desc_size;
2222 }
2223 
2224 static int
2225 nvmf_ctrlr_identify_ns_id_descriptor_list(
2226 	struct spdk_nvmf_subsystem *subsystem,
2227 	struct spdk_nvme_cmd *cmd,
2228 	struct spdk_nvme_cpl *rsp,
2229 	void *id_desc_list, size_t id_desc_list_size)
2230 {
2231 	struct spdk_nvmf_ns *ns;
2232 	size_t buf_remain = id_desc_list_size;
2233 	void *buf_ptr = id_desc_list;
2234 
2235 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2236 	if (ns == NULL || ns->bdev == NULL) {
2237 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2238 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2239 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2240 	}
2241 
2242 #define ADD_ID_DESC(type, data, size) \
2243 	do { \
2244 		if (!spdk_mem_all_zero(data, size)) { \
2245 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
2246 		} \
2247 	} while (0)
2248 
2249 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
2250 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
2251 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
2252 
2253 	/*
2254 	 * The list is automatically 0-terminated because controller to host buffers in
2255 	 * admin commands always get zeroed in nvmf_ctrlr_process_admin_cmd().
2256 	 */
2257 
2258 #undef ADD_ID_DESC
2259 
2260 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2261 }
2262 
2263 static int
2264 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
2265 {
2266 	uint8_t cns;
2267 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2268 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2269 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2270 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2271 
2272 	if (req->data == NULL || req->length < 4096) {
2273 		SPDK_ERRLOG("identify command with invalid buffer\n");
2274 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2275 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2276 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2277 	}
2278 
2279 	cns = cmd->cdw10_bits.identify.cns;
2280 
2281 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY &&
2282 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
2283 		/* Discovery controllers only support Identify Controller */
2284 		goto invalid_cns;
2285 	}
2286 
2287 	switch (cns) {
2288 	case SPDK_NVME_IDENTIFY_NS:
2289 		return spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, req->data);
2290 	case SPDK_NVME_IDENTIFY_CTRLR:
2291 		return spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, req->data);
2292 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
2293 		return nvmf_ctrlr_identify_active_ns_list(subsystem, cmd, rsp, req->data);
2294 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
2295 		return nvmf_ctrlr_identify_ns_id_descriptor_list(subsystem, cmd, rsp, req->data, req->length);
2296 	default:
2297 		goto invalid_cns;
2298 	}
2299 
2300 invalid_cns:
2301 	SPDK_ERRLOG("Identify command with unsupported CNS 0x%02x\n", cns);
2302 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2303 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2304 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2305 }
2306 
2307 static bool
2308 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
2309 {
2310 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
2311 	struct spdk_nvmf_request *req;
2312 	int i;
2313 
2314 	if (!nvmf_qpair_is_admin_queue(qpair)) {
2315 		return false;
2316 	}
2317 
2318 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2319 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
2320 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
2321 			req = ctrlr->aer_req[i];
2322 			ctrlr->aer_req[i] = NULL;
2323 			ctrlr->nr_aer_reqs--;
2324 
2325 			/* Move the last req to the aborting position for making aer_reqs
2326 			 * in continuous
2327 			 */
2328 			if (i < ctrlr->nr_aer_reqs) {
2329 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
2330 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
2331 			}
2332 
2333 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2334 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2335 			_nvmf_request_complete(req);
2336 			return true;
2337 		}
2338 	}
2339 
2340 	return false;
2341 }
2342 
2343 static void
2344 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
2345 {
2346 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
2347 
2348 	if (nvmf_qpair_abort_aer(qpair, cid)) {
2349 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
2350 			      qpair->ctrlr, qpair->qid, cid);
2351 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
2352 
2353 		spdk_nvmf_request_complete(req);
2354 		return;
2355 	}
2356 
2357 	nvmf_transport_qpair_abort_request(qpair, req);
2358 }
2359 
2360 static void
2361 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
2362 {
2363 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2364 
2365 	if (status == 0) {
2366 		/* There was no qpair whose ID matches SQID of the abort command.
2367 		 * Hence call _nvmf_request_complete() here.
2368 		 */
2369 		_nvmf_request_complete(req);
2370 	}
2371 }
2372 
2373 static void
2374 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
2375 {
2376 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2377 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
2378 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
2379 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
2380 	struct spdk_nvmf_qpair *qpair;
2381 
2382 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
2383 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
2384 			/* Found the qpair */
2385 
2386 			nvmf_qpair_abort_request(qpair, req);
2387 
2388 			/* Return -1 for the status so the iteration across threads stops. */
2389 			spdk_for_each_channel_continue(i, -1);
2390 			return;
2391 		}
2392 	}
2393 
2394 	spdk_for_each_channel_continue(i, 0);
2395 }
2396 
2397 static int
2398 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
2399 {
2400 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2401 
2402 	rsp->cdw0 = 1U; /* Command not aborted */
2403 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2404 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2405 
2406 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
2407 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
2408 			      nvmf_ctrlr_abort_on_pg,
2409 			      req,
2410 			      nvmf_ctrlr_abort_done
2411 			     );
2412 
2413 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2414 }
2415 
2416 int
2417 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
2418 {
2419 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
2420 	struct spdk_bdev *bdev;
2421 	struct spdk_bdev_desc *desc;
2422 	struct spdk_io_channel *ch;
2423 	int rc;
2424 
2425 	assert(req_to_abort != NULL);
2426 
2427 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
2428 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
2429 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
2430 	}
2431 
2432 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
2433 					&bdev, &desc, &ch);
2434 	if (rc != 0) {
2435 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2436 	}
2437 
2438 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
2439 }
2440 
2441 static int
2442 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
2443 {
2444 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2445 
2446 	rsp->cdw0 = cdw0;
2447 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2448 }
2449 
2450 /* we have to use the typedef in the function declaration to appease astyle. */
2451 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
2452 
2453 static spdk_nvme_path_status_code_t
2454 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
2455 {
2456 	switch (ana_state) {
2457 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2458 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
2459 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2460 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
2461 	case SPDK_NVME_ANA_CHANGE_STATE:
2462 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
2463 	default:
2464 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
2465 	}
2466 }
2467 
2468 static int
2469 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
2470 {
2471 	uint8_t feature;
2472 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2473 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2474 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2475 	enum spdk_nvme_ana_state ana_state;
2476 
2477 	feature = cmd->cdw10_bits.get_features.fid;
2478 
2479 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2480 		/*
2481 		 * Features supported by Discovery controller
2482 		 */
2483 		switch (feature) {
2484 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2485 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
2486 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2487 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
2488 		default:
2489 			SPDK_ERRLOG("Get Features command with unsupported feature ID 0x%02x\n", feature);
2490 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2491 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2492 		}
2493 	}
2494 	/*
2495 	 * Process Get Features command for non-discovery controller
2496 	 */
2497 	ana_state = ctrlr->listener->ana_state;
2498 	switch (ana_state) {
2499 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2500 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2501 	case SPDK_NVME_ANA_CHANGE_STATE:
2502 		switch (feature) {
2503 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
2504 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2505 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2506 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2507 			response->status.sct = SPDK_NVME_SCT_PATH;
2508 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2509 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2510 		default:
2511 			break;
2512 		}
2513 		break;
2514 	default:
2515 		break;
2516 	}
2517 
2518 	switch (feature) {
2519 	case SPDK_NVME_FEAT_ARBITRATION:
2520 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
2521 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
2522 		return get_features_generic(req, ctrlr->feat.power_management.raw);
2523 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
2524 		return nvmf_ctrlr_get_features_temperature_threshold(req);
2525 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
2526 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
2527 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
2528 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
2529 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
2530 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
2531 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2532 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
2533 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2534 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
2535 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2536 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
2537 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2538 		return nvmf_ctrlr_get_features_host_identifier(req);
2539 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2540 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
2541 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2542 		return nvmf_ctrlr_get_features_reservation_persistence(req);
2543 	default:
2544 		SPDK_ERRLOG("Get Features command with unsupported feature ID 0x%02x\n", feature);
2545 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2546 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2547 	}
2548 }
2549 
2550 static int
2551 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
2552 {
2553 	uint8_t feature, save;
2554 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2555 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2556 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2557 	enum spdk_nvme_ana_state ana_state;
2558 	/*
2559 	 * Features are not saveable by the controller as indicated by
2560 	 * ONCS field of the Identify Controller data.
2561 	 * */
2562 	save = cmd->cdw10_bits.set_features.sv;
2563 	if (save) {
2564 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
2565 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
2566 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2567 	}
2568 
2569 	feature = cmd->cdw10_bits.set_features.fid;
2570 
2571 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2572 		/*
2573 		 * Features supported by Discovery controller
2574 		 */
2575 		switch (feature) {
2576 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2577 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
2578 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2579 			return nvmf_ctrlr_set_features_async_event_configuration(req);
2580 		default:
2581 			SPDK_ERRLOG("Set Features command with unsupported feature ID 0x%02x\n", feature);
2582 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2583 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2584 		}
2585 	}
2586 	/*
2587 	 * Process Set Features command for non-discovery controller
2588 	 */
2589 	ana_state = ctrlr->listener->ana_state;
2590 	switch (ana_state) {
2591 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
2592 	case SPDK_NVME_ANA_CHANGE_STATE:
2593 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
2594 			response->status.sct = SPDK_NVME_SCT_PATH;
2595 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2596 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2597 		} else {
2598 			switch (feature) {
2599 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
2600 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2601 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2602 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2603 				response->status.sct = SPDK_NVME_SCT_PATH;
2604 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
2605 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2606 			default:
2607 				break;
2608 			}
2609 		}
2610 		break;
2611 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
2612 		response->status.sct = SPDK_NVME_SCT_PATH;
2613 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
2614 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2615 	default:
2616 		break;
2617 	}
2618 
2619 	switch (feature) {
2620 	case SPDK_NVME_FEAT_ARBITRATION:
2621 		return nvmf_ctrlr_set_features_arbitration(req);
2622 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
2623 		return nvmf_ctrlr_set_features_power_management(req);
2624 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
2625 		return nvmf_ctrlr_set_features_temperature_threshold(req);
2626 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
2627 		return nvmf_ctrlr_set_features_error_recovery(req);
2628 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
2629 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
2630 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
2631 		return nvmf_ctrlr_set_features_number_of_queues(req);
2632 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
2633 		return nvmf_ctrlr_set_features_write_atomicity(req);
2634 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
2635 		return nvmf_ctrlr_set_features_async_event_configuration(req);
2636 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
2637 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
2638 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
2639 		return nvmf_ctrlr_set_features_host_identifier(req);
2640 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
2641 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
2642 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
2643 		return nvmf_ctrlr_set_features_reservation_persistence(req);
2644 	default:
2645 		SPDK_ERRLOG("Set Features command with unsupported feature ID 0x%02x\n", feature);
2646 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2647 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2648 	}
2649 }
2650 
2651 static int
2652 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
2653 {
2654 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2655 
2656 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
2657 	/*
2658 	 * To handle keep alive just clear or reset the
2659 	 * ctrlr based keep alive duration counter.
2660 	 * When added, a separate timer based process
2661 	 * will monitor if the time since last recorded
2662 	 * keep alive has exceeded the max duration and
2663 	 * take appropriate action.
2664 	 */
2665 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
2666 
2667 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2668 }
2669 
2670 int
2671 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
2672 {
2673 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2674 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2675 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2676 	int rc;
2677 
2678 	if (ctrlr == NULL) {
2679 		SPDK_ERRLOG("Admin command sent before CONNECT\n");
2680 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2681 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2682 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2683 	}
2684 
2685 	if (ctrlr->vcprop.cc.bits.en != 1) {
2686 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
2687 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2688 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2689 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2690 	}
2691 
2692 	if (req->data && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
2693 		memset(req->data, 0, req->length);
2694 	}
2695 
2696 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2697 		/* Discovery controllers only support these admin OPS. */
2698 		switch (cmd->opc) {
2699 		case SPDK_NVME_OPC_IDENTIFY:
2700 		case SPDK_NVME_OPC_GET_LOG_PAGE:
2701 		case SPDK_NVME_OPC_KEEP_ALIVE:
2702 		case SPDK_NVME_OPC_SET_FEATURES:
2703 		case SPDK_NVME_OPC_GET_FEATURES:
2704 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
2705 			break;
2706 		default:
2707 			goto invalid_opcode;
2708 		}
2709 	}
2710 
2711 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
2712 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
2713 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
2714 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
2715 			/* The handler took care of this commmand */
2716 			return rc;
2717 		}
2718 	}
2719 
2720 	switch (cmd->opc) {
2721 	case SPDK_NVME_OPC_GET_LOG_PAGE:
2722 		return nvmf_ctrlr_get_log_page(req);
2723 	case SPDK_NVME_OPC_IDENTIFY:
2724 		return nvmf_ctrlr_identify(req);
2725 	case SPDK_NVME_OPC_ABORT:
2726 		return nvmf_ctrlr_abort(req);
2727 	case SPDK_NVME_OPC_GET_FEATURES:
2728 		return nvmf_ctrlr_get_features(req);
2729 	case SPDK_NVME_OPC_SET_FEATURES:
2730 		return nvmf_ctrlr_set_features(req);
2731 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
2732 		return nvmf_ctrlr_async_event_request(req);
2733 	case SPDK_NVME_OPC_KEEP_ALIVE:
2734 		return nvmf_ctrlr_keep_alive(req);
2735 
2736 	case SPDK_NVME_OPC_CREATE_IO_SQ:
2737 	case SPDK_NVME_OPC_CREATE_IO_CQ:
2738 	case SPDK_NVME_OPC_DELETE_IO_SQ:
2739 	case SPDK_NVME_OPC_DELETE_IO_CQ:
2740 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
2741 		goto invalid_opcode;
2742 
2743 	default:
2744 		goto invalid_opcode;
2745 	}
2746 
2747 invalid_opcode:
2748 	SPDK_ERRLOG("Unsupported admin opcode 0x%x\n", cmd->opc);
2749 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2750 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
2751 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2752 }
2753 
2754 static int
2755 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
2756 {
2757 	struct spdk_nvmf_qpair *qpair = req->qpair;
2758 	struct spdk_nvmf_capsule_cmd *cap_hdr;
2759 
2760 	cap_hdr = &req->cmd->nvmf_cmd;
2761 
2762 	if (qpair->ctrlr == NULL) {
2763 		/* No ctrlr established yet; the only valid command is Connect */
2764 		if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
2765 			return nvmf_ctrlr_cmd_connect(req);
2766 		} else {
2767 			SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n",
2768 				      cap_hdr->fctype);
2769 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2770 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
2771 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2772 		}
2773 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
2774 		/*
2775 		 * Controller session is established, and this is an admin queue.
2776 		 * Disallow Connect and allow other fabrics commands.
2777 		 */
2778 		switch (cap_hdr->fctype) {
2779 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
2780 			return nvmf_property_set(req);
2781 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
2782 			return nvmf_property_get(req);
2783 		default:
2784 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
2785 				      cap_hdr->fctype);
2786 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2787 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
2788 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2789 		}
2790 	} else {
2791 		/* Controller session is established, and this is an I/O queue */
2792 		/* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */
2793 		SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
2794 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2795 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
2796 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2797 	}
2798 }
2799 
2800 static inline int
2801 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
2802 				    union spdk_nvme_async_event_completion *event)
2803 {
2804 	struct spdk_nvmf_request *req;
2805 	struct spdk_nvme_cpl *rsp;
2806 
2807 	assert(ctrlr->nr_aer_reqs > 0);
2808 
2809 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
2810 	rsp = &req->rsp->nvme_cpl;
2811 
2812 	rsp->cdw0 = event->raw;
2813 
2814 	_nvmf_request_complete(req);
2815 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
2816 
2817 	return 0;
2818 }
2819 
2820 int
2821 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
2822 {
2823 	union spdk_nvme_async_event_completion event = {0};
2824 
2825 	/* Users may disable the event notification */
2826 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
2827 		return 0;
2828 	}
2829 
2830 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
2831 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
2832 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
2833 
2834 	/* If there is no outstanding AER request, queue the event.  Then
2835 	 * if an AER is later submitted, this event can be sent as a
2836 	 * response.
2837 	 */
2838 	if (ctrlr->nr_aer_reqs == 0) {
2839 		if (ctrlr->notice_event.bits.async_event_type ==
2840 		    SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) {
2841 			return 0;
2842 		}
2843 
2844 		ctrlr->notice_event.raw = event.raw;
2845 		return 0;
2846 	}
2847 
2848 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
2849 }
2850 
2851 int
2852 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
2853 {
2854 	union spdk_nvme_async_event_completion event = {0};
2855 
2856 	/* Users may disable the event notification */
2857 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
2858 		return 0;
2859 	}
2860 
2861 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
2862 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
2863 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
2864 
2865 	/* If there is no outstanding AER request, queue the event.  Then
2866 	 * if an AER is later submitted, this event can be sent as a
2867 	 * response.
2868 	 */
2869 	if (ctrlr->nr_aer_reqs == 0) {
2870 		if (ctrlr->notice_event.bits.async_event_type ==
2871 		    SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) {
2872 			return 0;
2873 		}
2874 
2875 		ctrlr->notice_event.raw = event.raw;
2876 		return 0;
2877 	}
2878 
2879 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
2880 }
2881 
2882 void
2883 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
2884 {
2885 	union spdk_nvme_async_event_completion event = {0};
2886 
2887 	if (!ctrlr->num_avail_log_pages) {
2888 		return;
2889 	}
2890 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
2891 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
2892 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
2893 
2894 	/* If there is no outstanding AER request, queue the event.  Then
2895 	 * if an AER is later submitted, this event can be sent as a
2896 	 * response.
2897 	 */
2898 	if (ctrlr->nr_aer_reqs == 0) {
2899 		if (ctrlr->reservation_event.bits.async_event_type ==
2900 		    SPDK_NVME_ASYNC_EVENT_TYPE_IO) {
2901 			return;
2902 		}
2903 
2904 		ctrlr->reservation_event.raw = event.raw;
2905 		return;
2906 	}
2907 
2908 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
2909 }
2910 
2911 int
2912 nvmf_ctrlr_async_event_discovery_log_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
2913 {
2914 	union spdk_nvme_async_event_completion event = {0};
2915 
2916 	/* Users may disable the event notification manually or
2917 	 * it may not be enabled due to keep alive timeout
2918 	 * not being set in connect command to discovery controller.
2919 	 */
2920 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
2921 		return 0;
2922 	}
2923 
2924 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
2925 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
2926 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
2927 
2928 	/* If there is no outstanding AER request, queue the event.  Then
2929 	 * if an AER is later submitted, this event can be sent as a
2930 	 * response.
2931 	 */
2932 	if (ctrlr->nr_aer_reqs == 0) {
2933 		if (ctrlr->notice_event.bits.async_event_type ==
2934 		    SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) {
2935 			return 0;
2936 		}
2937 
2938 		ctrlr->notice_event.raw = event.raw;
2939 		return 0;
2940 	}
2941 
2942 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
2943 }
2944 
2945 void
2946 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
2947 {
2948 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
2949 	int i;
2950 
2951 	if (!nvmf_qpair_is_admin_queue(qpair)) {
2952 		return;
2953 	}
2954 
2955 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2956 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
2957 		ctrlr->aer_req[i] = NULL;
2958 	}
2959 
2960 	ctrlr->nr_aer_reqs = 0;
2961 }
2962 
2963 void
2964 nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
2965 {
2966 	struct spdk_nvmf_request *req;
2967 	int i;
2968 
2969 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2970 		req = ctrlr->aer_req[i];
2971 
2972 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2973 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2974 		_nvmf_request_complete(req);
2975 
2976 		ctrlr->aer_req[i] = NULL;
2977 	}
2978 
2979 	ctrlr->nr_aer_reqs = 0;
2980 }
2981 
2982 static void
2983 _nvmf_ctrlr_add_reservation_log(void *ctx)
2984 {
2985 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
2986 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
2987 
2988 	ctrlr->log_page_count++;
2989 
2990 	/* Maximum number of queued log pages is 255 */
2991 	if (ctrlr->num_avail_log_pages == 0xff) {
2992 		struct spdk_nvmf_reservation_log *entry;
2993 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
2994 		entry->log.log_page_count = ctrlr->log_page_count;
2995 		free(log);
2996 		return;
2997 	}
2998 
2999 	log->log.log_page_count = ctrlr->log_page_count;
3000 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
3001 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
3002 
3003 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
3004 }
3005 
3006 void
3007 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
3008 				  struct spdk_nvmf_ns *ns,
3009 				  enum spdk_nvme_reservation_notification_log_page_type type)
3010 {
3011 	struct spdk_nvmf_reservation_log *log;
3012 
3013 	switch (type) {
3014 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
3015 		return;
3016 	case SPDK_NVME_REGISTRATION_PREEMPTED:
3017 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
3018 			return;
3019 		}
3020 		break;
3021 	case SPDK_NVME_RESERVATION_RELEASED:
3022 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
3023 			return;
3024 		}
3025 		break;
3026 	case SPDK_NVME_RESERVATION_PREEMPTED:
3027 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
3028 			return;
3029 		}
3030 		break;
3031 	default:
3032 		return;
3033 	}
3034 
3035 	log = calloc(1, sizeof(*log));
3036 	if (!log) {
3037 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
3038 		return;
3039 	}
3040 	log->ctrlr = ctrlr;
3041 	log->log.type = type;
3042 	log->log.nsid = ns->nsid;
3043 
3044 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
3045 }
3046 
3047 /* Check from subsystem poll group's namespace information data structure */
3048 static bool
3049 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3050 				 struct spdk_nvmf_ctrlr *ctrlr)
3051 {
3052 	uint32_t i;
3053 
3054 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
3055 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
3056 			return true;
3057 		}
3058 	}
3059 
3060 	return false;
3061 }
3062 
3063 /*
3064  * Check the NVMe command is permitted or not for current controller(Host).
3065  */
3066 static int
3067 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3068 				  struct spdk_nvmf_ctrlr *ctrlr,
3069 				  struct spdk_nvmf_request *req)
3070 {
3071 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3072 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
3073 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3074 	uint8_t racqa;
3075 	bool is_registrant;
3076 
3077 	/* No valid reservation */
3078 	if (!rtype) {
3079 		return 0;
3080 	}
3081 
3082 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
3083 	/* All registrants type and current ctrlr is a valid registrant */
3084 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3085 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
3086 		return 0;
3087 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
3088 		return 0;
3089 	}
3090 
3091 	/* Non-holder for current controller */
3092 	switch (cmd->opc) {
3093 	case SPDK_NVME_OPC_READ:
3094 	case SPDK_NVME_OPC_COMPARE:
3095 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3096 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3097 			goto exit;
3098 		}
3099 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
3100 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
3101 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3102 		}
3103 		break;
3104 	case SPDK_NVME_OPC_FLUSH:
3105 	case SPDK_NVME_OPC_WRITE:
3106 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
3107 	case SPDK_NVME_OPC_WRITE_ZEROES:
3108 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3109 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
3110 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3111 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3112 			goto exit;
3113 		}
3114 		if (!is_registrant) {
3115 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3116 		}
3117 		break;
3118 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3119 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
3120 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
3121 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3122 			goto exit;
3123 		}
3124 		if (!is_registrant) {
3125 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3126 		}
3127 		break;
3128 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3129 		if (!is_registrant) {
3130 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3131 		}
3132 		break;
3133 	default:
3134 		break;
3135 	}
3136 
3137 exit:
3138 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3139 	req->rsp->nvme_cpl.status.sc = status;
3140 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
3141 		return -EPERM;
3142 	}
3143 
3144 	return 0;
3145 }
3146 
3147 static int
3148 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
3149 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
3150 {
3151 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3152 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3153 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
3154 	int rc;
3155 
3156 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
3157 		/* first fused operation (should be compare) */
3158 		if (first_fused_req != NULL) {
3159 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3160 
3161 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3162 
3163 			/* abort req->qpair->first_fused_request and continue with new fused command */
3164 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3165 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3166 			_nvmf_request_complete(first_fused_req);
3167 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
3168 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3169 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3170 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3171 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3172 		}
3173 
3174 		req->qpair->first_fused_req = req;
3175 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3176 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
3177 		/* second fused operation (should be write) */
3178 		if (first_fused_req == NULL) {
3179 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3180 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3181 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3182 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3183 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
3184 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3185 
3186 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3187 
3188 			/* abort req->qpair->first_fused_request and fail current command */
3189 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3190 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3191 			_nvmf_request_complete(first_fused_req);
3192 
3193 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3194 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3195 			req->qpair->first_fused_req = NULL;
3196 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3197 		}
3198 
3199 		/* save request of first command to generate response later */
3200 		req->first_fused_req = first_fused_req;
3201 		req->qpair->first_fused_req = NULL;
3202 	} else {
3203 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
3204 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3205 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3206 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3207 	}
3208 
3209 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
3210 
3211 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3212 		if (spdk_nvme_cpl_is_error(rsp)) {
3213 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3214 
3215 			fused_response->status = rsp->status;
3216 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3217 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
3218 			/* Complete first of fused commands. Second will be completed by upper layer */
3219 			_nvmf_request_complete(first_fused_req);
3220 			req->first_fused_req = NULL;
3221 		}
3222 	}
3223 
3224 	return rc;
3225 }
3226 
3227 int
3228 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
3229 {
3230 	uint32_t nsid;
3231 	struct spdk_nvmf_ns *ns;
3232 	struct spdk_bdev *bdev;
3233 	struct spdk_bdev_desc *desc;
3234 	struct spdk_io_channel *ch;
3235 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3236 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3237 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3238 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3239 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3240 	enum spdk_nvme_ana_state ana_state;
3241 
3242 	/* pre-set response details for this command */
3243 	response->status.sc = SPDK_NVME_SC_SUCCESS;
3244 	nsid = cmd->nsid;
3245 
3246 	if (spdk_unlikely(ctrlr == NULL)) {
3247 		SPDK_ERRLOG("I/O command sent before CONNECT\n");
3248 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3249 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3250 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3251 	}
3252 
3253 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
3254 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
3255 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3256 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3257 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3258 	}
3259 
3260 	/* It will be lower overhead to check if ANA state is optimized or
3261 	 * non-optimized.
3262 	 */
3263 	ana_state = ctrlr->listener->ana_state;
3264 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
3265 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
3266 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
3267 			      ana_state);
3268 		response->status.sct = SPDK_NVME_SCT_PATH;
3269 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3270 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3271 	}
3272 
3273 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3274 	if (ns == NULL || ns->bdev == NULL) {
3275 		SPDK_ERRLOG("Unsuccessful query for nsid %u\n", cmd->nsid);
3276 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3277 		response->status.dnr = 1;
3278 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3279 	}
3280 
3281 	/* scan-build falsely reporting dereference of null pointer */
3282 	assert(group != NULL && group->sgroups != NULL);
3283 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3284 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
3285 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
3286 			      cmd->nsid, cmd->opc);
3287 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3288 	}
3289 
3290 	bdev = ns->bdev;
3291 	desc = ns->desc;
3292 	ch = ns_info->channel;
3293 
3294 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
3295 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
3296 	} else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) {
3297 		struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl;
3298 
3299 		SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n");
3300 
3301 		/* abort req->qpair->first_fused_request and continue with new command */
3302 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3303 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3304 		_nvmf_request_complete(req->qpair->first_fused_req);
3305 		req->qpair->first_fused_req = NULL;
3306 	}
3307 
3308 	switch (cmd->opc) {
3309 	case SPDK_NVME_OPC_READ:
3310 		return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
3311 	case SPDK_NVME_OPC_WRITE:
3312 		return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
3313 	case SPDK_NVME_OPC_COMPARE:
3314 		return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
3315 	case SPDK_NVME_OPC_WRITE_ZEROES:
3316 		return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
3317 	case SPDK_NVME_OPC_FLUSH:
3318 		return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
3319 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3320 		return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
3321 	case SPDK_NVME_OPC_RESERVATION_REGISTER:
3322 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3323 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3324 	case SPDK_NVME_OPC_RESERVATION_REPORT:
3325 		spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
3326 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3327 	default:
3328 		return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
3329 	}
3330 }
3331 
3332 static void
3333 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
3334 {
3335 	if (qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING) {
3336 		assert(qpair->state_cb != NULL);
3337 
3338 		if (TAILQ_EMPTY(&qpair->outstanding)) {
3339 			qpair->state_cb(qpair->state_cb_arg, 0);
3340 		}
3341 	}
3342 }
3343 
3344 int
3345 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
3346 {
3347 	struct spdk_nvmf_qpair *qpair = req->qpair;
3348 
3349 	TAILQ_REMOVE(&qpair->outstanding, req, link);
3350 	if (nvmf_transport_req_free(req)) {
3351 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
3352 	}
3353 
3354 	nvmf_qpair_request_cleanup(qpair);
3355 
3356 	return 0;
3357 }
3358 
3359 static void
3360 _nvmf_request_complete(void *ctx)
3361 {
3362 	struct spdk_nvmf_request *req = ctx;
3363 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3364 	struct spdk_nvmf_qpair *qpair;
3365 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3366 	bool is_aer = false;
3367 
3368 	rsp->sqid = 0;
3369 	rsp->status.p = 0;
3370 	rsp->cid = req->cmd->nvme_cmd.cid;
3371 
3372 	qpair = req->qpair;
3373 	if (qpair->ctrlr) {
3374 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3375 		assert(sgroup != NULL);
3376 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
3377 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3378 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3379 	}
3380 
3381 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
3382 		spdk_nvme_print_completion(qpair->qid, rsp);
3383 	}
3384 
3385 	TAILQ_REMOVE(&qpair->outstanding, req, link);
3386 	if (nvmf_transport_req_complete(req)) {
3387 		SPDK_ERRLOG("Transport request completion error!\n");
3388 	}
3389 
3390 	/* AER cmd is an exception */
3391 	if (sgroup && !is_aer) {
3392 		assert(sgroup->io_outstanding > 0);
3393 		sgroup->io_outstanding--;
3394 		if (sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
3395 		    sgroup->io_outstanding == 0) {
3396 			sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
3397 			sgroup->cb_fn(sgroup->cb_arg, 0);
3398 		}
3399 	}
3400 
3401 	nvmf_qpair_request_cleanup(qpair);
3402 }
3403 
3404 int
3405 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
3406 {
3407 	struct spdk_nvmf_qpair *qpair = req->qpair;
3408 
3409 	if (spdk_likely(qpair->group->thread == spdk_get_thread())) {
3410 		_nvmf_request_complete(req);
3411 	} else {
3412 		spdk_thread_send_msg(qpair->group->thread,
3413 				     _nvmf_request_complete, req);
3414 	}
3415 
3416 	return 0;
3417 }
3418 
3419 static void
3420 _nvmf_request_exec(struct spdk_nvmf_request *req,
3421 		   struct spdk_nvmf_subsystem_poll_group *sgroup)
3422 {
3423 	struct spdk_nvmf_qpair *qpair = req->qpair;
3424 	enum spdk_nvmf_request_exec_status status;
3425 
3426 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
3427 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
3428 	}
3429 
3430 	if (sgroup) {
3431 		sgroup->io_outstanding++;
3432 	}
3433 
3434 	/* Place the request on the outstanding list so we can keep track of it */
3435 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3436 
3437 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
3438 		status = nvmf_ctrlr_process_fabrics_cmd(req);
3439 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
3440 		status = nvmf_ctrlr_process_admin_cmd(req);
3441 	} else {
3442 		status = nvmf_ctrlr_process_io_cmd(req);
3443 	}
3444 
3445 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3446 		_nvmf_request_complete(req);
3447 	}
3448 }
3449 
3450 void
3451 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
3452 {
3453 	struct spdk_nvmf_qpair *qpair = req->qpair;
3454 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
3455 
3456 	if (qpair->ctrlr) {
3457 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
3458 		assert(sgroup != NULL);
3459 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
3460 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
3461 	}
3462 
3463 	if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
3464 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3465 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3466 		/* Place the request on the outstanding list so we can keep track of it */
3467 		TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
3468 		/* Still increment io_outstanding because request_complete decrements it */
3469 		if (sgroup != NULL) {
3470 			sgroup->io_outstanding++;
3471 		}
3472 		_nvmf_request_complete(req);
3473 		return;
3474 	}
3475 
3476 	/* Check if the subsystem is paused (if there is a subsystem) */
3477 	if (sgroup != NULL) {
3478 		if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
3479 			/* The subsystem is not currently active. Queue this request. */
3480 			TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
3481 			return;
3482 		}
3483 	}
3484 
3485 	_nvmf_request_exec(req, sgroup);
3486 }
3487 
3488 static bool
3489 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
3490 		       struct spdk_dif_ctx *dif_ctx)
3491 {
3492 	struct spdk_nvmf_ns *ns;
3493 	struct spdk_bdev *bdev;
3494 
3495 	if (ctrlr == NULL || cmd == NULL) {
3496 		return false;
3497 	}
3498 
3499 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
3500 	if (ns == NULL || ns->bdev == NULL) {
3501 		return false;
3502 	}
3503 
3504 	bdev = ns->bdev;
3505 
3506 	switch (cmd->opc) {
3507 	case SPDK_NVME_OPC_READ:
3508 	case SPDK_NVME_OPC_WRITE:
3509 	case SPDK_NVME_OPC_COMPARE:
3510 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
3511 	default:
3512 		break;
3513 	}
3514 
3515 	return false;
3516 }
3517 
3518 bool
3519 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
3520 {
3521 	struct spdk_nvmf_qpair *qpair = req->qpair;
3522 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3523 
3524 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
3525 		return false;
3526 	}
3527 
3528 	if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) {
3529 		return false;
3530 	}
3531 
3532 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
3533 		return false;
3534 	}
3535 
3536 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
3537 		return false;
3538 	}
3539 
3540 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
3541 }
3542 
3543 void
3544 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
3545 {
3546 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
3547 }
3548 
3549 static int
3550 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
3551 {
3552 	struct spdk_bdev *bdev;
3553 	struct spdk_bdev_desc *desc;
3554 	struct spdk_io_channel *ch;
3555 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
3556 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
3557 	uint32_t bdev_nsid;
3558 	int rc;
3559 
3560 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid == 0) {
3561 		bdev_nsid = cmd->nsid;
3562 	} else {
3563 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
3564 	}
3565 
3566 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
3567 	if (rc) {
3568 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3569 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3570 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3571 	}
3572 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
3573 }
3574 
3575 void
3576 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
3577 {
3578 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
3579 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
3580 }
3581 
3582 int
3583 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
3584 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
3585 {
3586 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3587 	struct spdk_nvmf_ns *ns;
3588 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3589 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3590 
3591 	*bdev = NULL;
3592 	*desc = NULL;
3593 	*ch = NULL;
3594 
3595 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3596 	if (ns == NULL || ns->bdev == NULL) {
3597 		return -EINVAL;
3598 	}
3599 
3600 	assert(group != NULL && group->sgroups != NULL);
3601 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3602 	*bdev = ns->bdev;
3603 	*desc = ns->desc;
3604 	*ch = ns_info->channel;
3605 
3606 	return 0;
3607 }
3608 
3609 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
3610 {
3611 	return req->qpair->ctrlr;
3612 }
3613 
3614 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
3615 {
3616 	return &req->cmd->nvme_cmd;
3617 }
3618 
3619 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
3620 {
3621 	return &req->rsp->nvme_cpl;
3622 }
3623 
3624 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
3625 {
3626 	return req->qpair->ctrlr->subsys;
3627 }
3628 
3629 void spdk_nvmf_request_get_data(struct spdk_nvmf_request *req, void **data, uint32_t *length)
3630 {
3631 	*data = req->data;
3632 	*length = req->length;
3633 }
3634 
3635 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
3636 {
3637 	return ctrlr->subsys;
3638 }
3639 
3640 uint16_t spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
3641 {
3642 	return ctrlr->cntlid;
3643 }
3644 
3645 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
3646 {
3647 	return req->req_to_abort;
3648 }
3649