xref: /spdk/lib/nvmf/ctrlr.c (revision 488570ebd418ba07c9e69e65106dcc964f3bb41b)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (c) Intel Corporation. All rights reserved.
3  *   Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved.
4  *   Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5  */
6 
7 #include "spdk/stdinc.h"
8 
9 #include "nvmf_internal.h"
10 #include "transport.h"
11 
12 #include "spdk/bit_array.h"
13 #include "spdk/endian.h"
14 #include "spdk/thread.h"
15 #include "spdk/nvme_spec.h"
16 #include "spdk/nvmf_cmd.h"
17 #include "spdk/string.h"
18 #include "spdk/util.h"
19 #include "spdk/version.h"
20 #include "spdk/log.h"
21 #include "spdk_internal/usdt.h"
22 
23 #define MIN_KEEP_ALIVE_TIMEOUT_IN_MS 10000
24 #define NVMF_DISC_KATO_IN_MS 120000
25 #define KAS_TIME_UNIT_IN_MS 100
26 #define KAS_DEFAULT_VALUE (MIN_KEEP_ALIVE_TIMEOUT_IN_MS / KAS_TIME_UNIT_IN_MS)
27 
28 #define NVMF_CC_RESET_SHN_TIMEOUT_IN_MS	10000
29 
30 #define NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS	(NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + 5000)
31 
32 /*
33  * Report the SPDK version as the firmware revision.
34  * SPDK_VERSION_STRING won't fit into FR (only 8 bytes), so try to fit the most important parts.
35  */
36 #define FW_VERSION SPDK_VERSION_MAJOR_STRING SPDK_VERSION_MINOR_STRING SPDK_VERSION_PATCH_STRING
37 
38 #define ANA_TRANSITION_TIME_IN_SEC 10
39 
40 #define NVMF_ABORT_COMMAND_LIMIT 3
41 
42 /*
43  * Support for custom admin command handlers
44  */
45 struct spdk_nvmf_custom_admin_cmd {
46 	spdk_nvmf_custom_cmd_hdlr hdlr;
47 	uint32_t nsid; /* nsid to forward */
48 };
49 
50 static struct spdk_nvmf_custom_admin_cmd g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_MAX_OPC + 1];
51 
52 static void _nvmf_request_complete(void *ctx);
53 
54 static inline void
55 nvmf_invalid_connect_response(struct spdk_nvmf_fabric_connect_rsp *rsp,
56 			      uint8_t iattr, uint16_t ipo)
57 {
58 	rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
59 	rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
60 	rsp->status_code_specific.invalid.iattr = iattr;
61 	rsp->status_code_specific.invalid.ipo = ipo;
62 }
63 
64 #define SPDK_NVMF_INVALID_CONNECT_CMD(rsp, field)	\
65 	nvmf_invalid_connect_response(rsp, 0, offsetof(struct spdk_nvmf_fabric_connect_cmd, field))
66 #define SPDK_NVMF_INVALID_CONNECT_DATA(rsp, field)	\
67 	nvmf_invalid_connect_response(rsp, 1, offsetof(struct spdk_nvmf_fabric_connect_data, field))
68 
69 
70 static void
71 nvmf_ctrlr_stop_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
72 {
73 	if (!ctrlr) {
74 		SPDK_ERRLOG("Controller is NULL\n");
75 		return;
76 	}
77 
78 	if (ctrlr->keep_alive_poller == NULL) {
79 		return;
80 	}
81 
82 	SPDK_DEBUGLOG(nvmf, "Stop keep alive poller\n");
83 	spdk_poller_unregister(&ctrlr->keep_alive_poller);
84 }
85 
86 static void
87 nvmf_ctrlr_stop_association_timer(struct spdk_nvmf_ctrlr *ctrlr)
88 {
89 	if (!ctrlr) {
90 		SPDK_ERRLOG("Controller is NULL\n");
91 		assert(false);
92 		return;
93 	}
94 
95 	if (ctrlr->association_timer == NULL) {
96 		return;
97 	}
98 
99 	SPDK_DEBUGLOG(nvmf, "Stop association timer\n");
100 	spdk_poller_unregister(&ctrlr->association_timer);
101 }
102 
103 static void
104 nvmf_ctrlr_disconnect_qpairs_done(struct spdk_io_channel_iter *i, int status)
105 {
106 	if (status == 0) {
107 		SPDK_DEBUGLOG(nvmf, "ctrlr disconnect qpairs complete successfully\n");
108 	} else {
109 		SPDK_ERRLOG("Fail to disconnect ctrlr qpairs\n");
110 	}
111 }
112 
113 static int
114 _nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i, bool include_admin)
115 {
116 	int rc = 0;
117 	struct spdk_nvmf_ctrlr *ctrlr;
118 	struct spdk_nvmf_qpair *qpair, *temp_qpair;
119 	struct spdk_io_channel *ch;
120 	struct spdk_nvmf_poll_group *group;
121 
122 	ctrlr = spdk_io_channel_iter_get_ctx(i);
123 	ch = spdk_io_channel_iter_get_channel(i);
124 	group = spdk_io_channel_get_ctx(ch);
125 
126 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, temp_qpair) {
127 		if (qpair->ctrlr == ctrlr && (include_admin || !nvmf_qpair_is_admin_queue(qpair))) {
128 			rc = spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
129 			if (rc) {
130 				SPDK_ERRLOG("Qpair disconnect failed\n");
131 				return rc;
132 			}
133 		}
134 	}
135 
136 	return rc;
137 }
138 
139 static void
140 nvmf_ctrlr_disconnect_qpairs_on_pg(struct spdk_io_channel_iter *i)
141 {
142 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, true));
143 }
144 
145 static void
146 nvmf_ctrlr_disconnect_io_qpairs_on_pg(struct spdk_io_channel_iter *i)
147 {
148 	spdk_for_each_channel_continue(i, _nvmf_ctrlr_disconnect_qpairs_on_pg(i, false));
149 }
150 
151 static int
152 nvmf_ctrlr_keep_alive_poll(void *ctx)
153 {
154 	uint64_t keep_alive_timeout_tick;
155 	uint64_t now = spdk_get_ticks();
156 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
157 
158 	if (ctrlr->in_destruct) {
159 		nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
160 		return SPDK_POLLER_IDLE;
161 	}
162 
163 	SPDK_DEBUGLOG(nvmf, "Polling ctrlr keep alive timeout\n");
164 
165 	/* If the Keep alive feature is in use and the timer expires */
166 	keep_alive_timeout_tick = ctrlr->last_keep_alive_tick +
167 				  ctrlr->feat.keep_alive_timer.bits.kato * spdk_get_ticks_hz() / UINT64_C(1000);
168 	if (now > keep_alive_timeout_tick) {
169 		SPDK_NOTICELOG("Disconnecting host %s from subsystem %s due to keep alive timeout.\n",
170 			       ctrlr->hostnqn, ctrlr->subsys->subnqn);
171 		/* set the Controller Fatal Status bit to '1' */
172 		if (ctrlr->vcprop.csts.bits.cfs == 0) {
173 			nvmf_ctrlr_set_fatal_status(ctrlr);
174 
175 			/*
176 			 * disconnect qpairs, terminate Transport connection
177 			 * destroy ctrlr, break the host to controller association
178 			 * disconnect qpairs with qpair->ctrlr == ctrlr
179 			 */
180 			spdk_for_each_channel(ctrlr->subsys->tgt,
181 					      nvmf_ctrlr_disconnect_qpairs_on_pg,
182 					      ctrlr,
183 					      nvmf_ctrlr_disconnect_qpairs_done);
184 			return SPDK_POLLER_BUSY;
185 		}
186 	}
187 
188 	return SPDK_POLLER_IDLE;
189 }
190 
191 static void
192 nvmf_ctrlr_start_keep_alive_timer(struct spdk_nvmf_ctrlr *ctrlr)
193 {
194 	if (!ctrlr) {
195 		SPDK_ERRLOG("Controller is NULL\n");
196 		return;
197 	}
198 
199 	/* if cleared to 0 then the Keep Alive Timer is disabled */
200 	if (ctrlr->feat.keep_alive_timer.bits.kato != 0) {
201 
202 		ctrlr->last_keep_alive_tick = spdk_get_ticks();
203 
204 		SPDK_DEBUGLOG(nvmf, "Ctrlr add keep alive poller\n");
205 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
206 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
207 	}
208 }
209 
210 static void
211 ctrlr_add_qpair_and_update_rsp(struct spdk_nvmf_qpair *qpair,
212 			       struct spdk_nvmf_ctrlr *ctrlr,
213 			       struct spdk_nvmf_fabric_connect_rsp *rsp)
214 {
215 	assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
216 
217 	/* check if we would exceed ctrlr connection limit */
218 	if (qpair->qid >= spdk_bit_array_capacity(ctrlr->qpair_mask)) {
219 		SPDK_ERRLOG("Requested QID %u but Max QID is %u\n",
220 			    qpair->qid, spdk_bit_array_capacity(ctrlr->qpair_mask) - 1);
221 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
222 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
223 		return;
224 	}
225 
226 	if (spdk_bit_array_get(ctrlr->qpair_mask, qpair->qid)) {
227 		SPDK_ERRLOG("Got I/O connect with duplicate QID %u\n", qpair->qid);
228 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
229 		rsp->status.sc = SPDK_NVME_SC_INVALID_QUEUE_IDENTIFIER;
230 		return;
231 	}
232 
233 	qpair->ctrlr = ctrlr;
234 	spdk_bit_array_set(ctrlr->qpair_mask, qpair->qid);
235 
236 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
237 	rsp->status_code_specific.success.cntlid = ctrlr->cntlid;
238 	SPDK_DEBUGLOG(nvmf, "connect capsule response: cntlid = 0x%04x\n",
239 		      rsp->status_code_specific.success.cntlid);
240 
241 	SPDK_DTRACE_PROBE4(nvmf_ctrlr_add_qpair, qpair, qpair->qid, ctrlr->subsys->subnqn,
242 			   ctrlr->hostnqn);
243 }
244 
245 static void
246 _nvmf_ctrlr_add_admin_qpair(void *ctx)
247 {
248 	struct spdk_nvmf_request *req = ctx;
249 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
250 	struct spdk_nvmf_qpair *qpair = req->qpair;
251 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
252 
253 	ctrlr->admin_qpair = qpair;
254 	ctrlr->association_timeout = qpair->transport->opts.association_timeout;
255 	nvmf_ctrlr_start_keep_alive_timer(ctrlr);
256 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
257 	_nvmf_request_complete(req);
258 }
259 
260 static void
261 _nvmf_subsystem_add_ctrlr(void *ctx)
262 {
263 	struct spdk_nvmf_request *req = ctx;
264 	struct spdk_nvmf_qpair *qpair = req->qpair;
265 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
266 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
267 
268 	if (nvmf_subsystem_add_ctrlr(ctrlr->subsys, ctrlr)) {
269 		SPDK_ERRLOG("Unable to add controller to subsystem\n");
270 		spdk_bit_array_free(&ctrlr->qpair_mask);
271 		free(ctrlr);
272 		qpair->ctrlr = NULL;
273 		rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
274 		spdk_nvmf_request_complete(req);
275 		return;
276 	}
277 
278 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_admin_qpair, req);
279 }
280 
281 static void
282 nvmf_ctrlr_cdata_init(struct spdk_nvmf_transport *transport, struct spdk_nvmf_subsystem *subsystem,
283 		      struct spdk_nvmf_ctrlr_data *cdata)
284 {
285 	cdata->aerl = NVMF_MAX_ASYNC_EVENTS - 1;
286 	cdata->kas = KAS_DEFAULT_VALUE;
287 	cdata->vid = SPDK_PCI_VID_INTEL;
288 	cdata->ssvid = SPDK_PCI_VID_INTEL;
289 	/* INTEL OUI */
290 	cdata->ieee[0] = 0xe4;
291 	cdata->ieee[1] = 0xd2;
292 	cdata->ieee[2] = 0x5c;
293 	cdata->oncs.reservations = 1;
294 	cdata->sgls.supported = 1;
295 	cdata->sgls.keyed_sgl = 1;
296 	cdata->sgls.sgl_offset = 1;
297 	cdata->nvmf_specific.ioccsz = sizeof(struct spdk_nvme_cmd) / 16;
298 	cdata->nvmf_specific.ioccsz += transport->opts.in_capsule_data_size / 16;
299 	cdata->nvmf_specific.iorcsz = sizeof(struct spdk_nvme_cpl) / 16;
300 	cdata->nvmf_specific.icdoff = 0; /* offset starts directly after SQE */
301 	cdata->nvmf_specific.ctrattr.ctrlr_model = SPDK_NVMF_CTRLR_MODEL_DYNAMIC;
302 	cdata->nvmf_specific.msdbd = 1;
303 
304 	if (transport->ops->cdata_init) {
305 		transport->ops->cdata_init(transport, subsystem, cdata);
306 	}
307 }
308 
309 static struct spdk_nvmf_ctrlr *
310 nvmf_ctrlr_create(struct spdk_nvmf_subsystem *subsystem,
311 		  struct spdk_nvmf_request *req,
312 		  struct spdk_nvmf_fabric_connect_cmd *connect_cmd,
313 		  struct spdk_nvmf_fabric_connect_data *connect_data)
314 {
315 	struct spdk_nvmf_ctrlr *ctrlr;
316 	struct spdk_nvmf_transport *transport = req->qpair->transport;
317 	struct spdk_nvme_transport_id listen_trid = {};
318 
319 	ctrlr = calloc(1, sizeof(*ctrlr));
320 	if (ctrlr == NULL) {
321 		SPDK_ERRLOG("Memory allocation failed\n");
322 		return NULL;
323 	}
324 
325 	if (spdk_nvme_trtype_is_fabrics(transport->ops->type)) {
326 		ctrlr->dynamic_ctrlr = true;
327 	} else {
328 		ctrlr->cntlid = connect_data->cntlid;
329 	}
330 
331 	SPDK_DTRACE_PROBE3(nvmf_ctrlr_create, ctrlr, subsystem->subnqn,
332 			   spdk_thread_get_id(req->qpair->group->thread));
333 
334 	STAILQ_INIT(&ctrlr->async_events);
335 	TAILQ_INIT(&ctrlr->log_head);
336 	ctrlr->subsys = subsystem;
337 	ctrlr->thread = req->qpair->group->thread;
338 	ctrlr->disconnect_in_progress = false;
339 
340 	ctrlr->qpair_mask = spdk_bit_array_create(transport->opts.max_qpairs_per_ctrlr);
341 	if (!ctrlr->qpair_mask) {
342 		SPDK_ERRLOG("Failed to allocate controller qpair mask\n");
343 		goto err_qpair_mask;
344 	}
345 
346 	nvmf_ctrlr_cdata_init(transport, subsystem, &ctrlr->cdata);
347 
348 	/*
349 	 * KAS: This field indicates the granularity of the Keep Alive Timer in 100ms units.
350 	 * If this field is cleared to 0h, then Keep Alive is not supported.
351 	 */
352 	if (ctrlr->cdata.kas) {
353 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(connect_cmd->kato,
354 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
355 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
356 	}
357 
358 	ctrlr->feat.async_event_configuration.bits.ns_attr_notice = 1;
359 	if (ctrlr->subsys->flags.ana_reporting) {
360 		ctrlr->feat.async_event_configuration.bits.ana_change_notice = 1;
361 	}
362 	ctrlr->feat.volatile_write_cache.bits.wce = 1;
363 	/* Coalescing Disable */
364 	ctrlr->feat.interrupt_vector_configuration.bits.cd = 1;
365 
366 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
367 		/*
368 		 * If keep-alive timeout is not set, discovery controllers use some
369 		 * arbitrary high value in order to cleanup stale discovery sessions
370 		 *
371 		 * From the 1.0a nvme-of spec:
372 		 * "The Keep Alive command is reserved for
373 		 * Discovery controllers. A transport may specify a
374 		 * fixed Discovery controller activity timeout value
375 		 * (e.g., 2 minutes). If no commands are received
376 		 * by a Discovery controller within that time
377 		 * period, the controller may perform the
378 		 * actions for Keep Alive Timer expiration".
379 		 *
380 		 * From the 1.1 nvme-of spec:
381 		 * "A host requests an explicit persistent connection
382 		 * to a Discovery controller and Asynchronous Event Notifications from
383 		 * the Discovery controller on that persistent connection by specifying
384 		 * a non-zero Keep Alive Timer value in the Connect command."
385 		 *
386 		 * In case non-zero KATO is used, we enable discovery_log_change_notice
387 		 * otherwise we disable it and use default discovery controller KATO.
388 		 * KATO is in millisecond.
389 		 */
390 		if (ctrlr->feat.keep_alive_timer.bits.kato == 0) {
391 			ctrlr->feat.keep_alive_timer.bits.kato = NVMF_DISC_KATO_IN_MS;
392 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 0;
393 		} else {
394 			ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice = 1;
395 		}
396 	}
397 
398 	/* Subtract 1 for admin queue, 1 for 0's based */
399 	ctrlr->feat.number_of_queues.bits.ncqr = transport->opts.max_qpairs_per_ctrlr - 1 -
400 			1;
401 	ctrlr->feat.number_of_queues.bits.nsqr = transport->opts.max_qpairs_per_ctrlr - 1 -
402 			1;
403 
404 	spdk_uuid_copy(&ctrlr->hostid, (struct spdk_uuid *)connect_data->hostid);
405 	memcpy(ctrlr->hostnqn, connect_data->hostnqn, sizeof(ctrlr->hostnqn));
406 
407 	ctrlr->vcprop.cap.raw = 0;
408 	ctrlr->vcprop.cap.bits.cqr = 1; /* NVMe-oF specification required */
409 	ctrlr->vcprop.cap.bits.mqes = transport->opts.max_queue_depth -
410 				      1; /* max queue depth */
411 	ctrlr->vcprop.cap.bits.ams = 0; /* optional arb mechanisms */
412 	/* ready timeout - 500 msec units */
413 	ctrlr->vcprop.cap.bits.to = NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS / 500;
414 	ctrlr->vcprop.cap.bits.dstrd = 0; /* fixed to 0 for NVMe-oF */
415 	ctrlr->vcprop.cap.bits.css = SPDK_NVME_CAP_CSS_NVM; /* NVM command set */
416 	ctrlr->vcprop.cap.bits.mpsmin = 0; /* 2 ^ (12 + mpsmin) == 4k */
417 	ctrlr->vcprop.cap.bits.mpsmax = 0; /* 2 ^ (12 + mpsmax) == 4k */
418 
419 	/* Version Supported: 1.3 */
420 	ctrlr->vcprop.vs.bits.mjr = 1;
421 	ctrlr->vcprop.vs.bits.mnr = 3;
422 	ctrlr->vcprop.vs.bits.ter = 0;
423 
424 	ctrlr->vcprop.cc.raw = 0;
425 	ctrlr->vcprop.cc.bits.en = 0; /* Init controller disabled */
426 
427 	ctrlr->vcprop.csts.raw = 0;
428 	ctrlr->vcprop.csts.bits.rdy = 0; /* Init controller as not ready */
429 
430 	SPDK_DEBUGLOG(nvmf, "cap 0x%" PRIx64 "\n", ctrlr->vcprop.cap.raw);
431 	SPDK_DEBUGLOG(nvmf, "vs 0x%x\n", ctrlr->vcprop.vs.raw);
432 	SPDK_DEBUGLOG(nvmf, "cc 0x%x\n", ctrlr->vcprop.cc.raw);
433 	SPDK_DEBUGLOG(nvmf, "csts 0x%x\n", ctrlr->vcprop.csts.raw);
434 
435 	ctrlr->dif_insert_or_strip = transport->opts.dif_insert_or_strip;
436 
437 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_NVME) {
438 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
439 			SPDK_ERRLOG("Could not get listener transport ID\n");
440 			goto err_listener;
441 		}
442 
443 		ctrlr->listener = nvmf_subsystem_find_listener(ctrlr->subsys, &listen_trid);
444 		if (!ctrlr->listener) {
445 			SPDK_ERRLOG("Listener was not found\n");
446 			goto err_listener;
447 		}
448 	}
449 
450 	req->qpair->ctrlr = ctrlr;
451 	spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_add_ctrlr, req);
452 
453 	return ctrlr;
454 err_listener:
455 	spdk_bit_array_free(&ctrlr->qpair_mask);
456 err_qpair_mask:
457 	free(ctrlr);
458 	return NULL;
459 }
460 
461 static void
462 _nvmf_ctrlr_destruct(void *ctx)
463 {
464 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
465 	struct spdk_nvmf_reservation_log *log, *log_tmp;
466 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
467 
468 	SPDK_DTRACE_PROBE3(nvmf_ctrlr_destruct, ctrlr, ctrlr->subsys->subnqn,
469 			   spdk_thread_get_id(ctrlr->thread));
470 
471 	assert(spdk_get_thread() == ctrlr->thread);
472 	assert(ctrlr->in_destruct);
473 
474 	SPDK_DEBUGLOG(nvmf, "Destroy ctrlr 0x%hx\n", ctrlr->cntlid);
475 	if (ctrlr->disconnect_in_progress) {
476 		SPDK_ERRLOG("freeing ctrlr with disconnect in progress\n");
477 		spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
478 		return;
479 	}
480 
481 	nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
482 	nvmf_ctrlr_stop_association_timer(ctrlr);
483 	spdk_bit_array_free(&ctrlr->qpair_mask);
484 
485 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
486 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
487 		free(log);
488 	}
489 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
490 		STAILQ_REMOVE(&ctrlr->async_events, event, spdk_nvmf_async_event_completion, link);
491 		free(event);
492 	}
493 	free(ctrlr);
494 }
495 
496 void
497 nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr)
498 {
499 	nvmf_subsystem_remove_ctrlr(ctrlr->subsys, ctrlr);
500 
501 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_destruct, ctrlr);
502 }
503 
504 static void
505 nvmf_ctrlr_add_io_qpair(void *ctx)
506 {
507 	struct spdk_nvmf_request *req = ctx;
508 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
509 	struct spdk_nvmf_qpair *qpair = req->qpair;
510 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
511 	struct spdk_nvmf_qpair *admin_qpair = ctrlr->admin_qpair;
512 
513 	SPDK_DTRACE_PROBE4(nvmf_ctrlr_add_io_qpair, ctrlr, req->qpair, req->qpair->qid,
514 			   spdk_thread_get_id(ctrlr->thread));
515 
516 	/* Unit test will check qpair->ctrlr after calling spdk_nvmf_ctrlr_connect.
517 	  * For error case, the value should be NULL. So set it to NULL at first.
518 	  */
519 	qpair->ctrlr = NULL;
520 
521 	/* Make sure the controller is not being destroyed. */
522 	if (ctrlr->in_destruct) {
523 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
524 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
525 		goto end;
526 	}
527 
528 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
529 		SPDK_ERRLOG("I/O connect not allowed on discovery controller\n");
530 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
531 		goto end;
532 	}
533 
534 	if (!ctrlr->vcprop.cc.bits.en) {
535 		SPDK_ERRLOG("Got I/O connect before ctrlr was enabled\n");
536 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
537 		goto end;
538 	}
539 
540 	if (1u << ctrlr->vcprop.cc.bits.iosqes != sizeof(struct spdk_nvme_cmd)) {
541 		SPDK_ERRLOG("Got I/O connect with invalid IOSQES %u\n",
542 			    ctrlr->vcprop.cc.bits.iosqes);
543 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
544 		goto end;
545 	}
546 
547 	if (1u << ctrlr->vcprop.cc.bits.iocqes != sizeof(struct spdk_nvme_cpl)) {
548 		SPDK_ERRLOG("Got I/O connect with invalid IOCQES %u\n",
549 			    ctrlr->vcprop.cc.bits.iocqes);
550 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
551 		goto end;
552 	}
553 
554 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
555 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
556 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
557 		 * state to DEACTIVATING and removing it from poll group */
558 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
559 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
560 		goto end;
561 	}
562 
563 	ctrlr_add_qpair_and_update_rsp(qpair, ctrlr, rsp);
564 end:
565 	spdk_nvmf_request_complete(req);
566 }
567 
568 static void
569 _nvmf_ctrlr_add_io_qpair(void *ctx)
570 {
571 	struct spdk_nvmf_request *req = ctx;
572 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
573 	struct spdk_nvmf_fabric_connect_data *data = req->data;
574 	struct spdk_nvmf_ctrlr *ctrlr;
575 	struct spdk_nvmf_qpair *qpair = req->qpair;
576 	struct spdk_nvmf_qpair *admin_qpair;
577 	struct spdk_nvmf_tgt *tgt = qpair->transport->tgt;
578 	struct spdk_nvmf_subsystem *subsystem;
579 	struct spdk_nvme_transport_id listen_trid = {};
580 	const struct spdk_nvmf_subsystem_listener *listener;
581 
582 	SPDK_DEBUGLOG(nvmf, "Connect I/O Queue for controller id 0x%x\n", data->cntlid);
583 
584 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
585 	/* We already checked this in spdk_nvmf_ctrlr_connect */
586 	assert(subsystem != NULL);
587 
588 	ctrlr = nvmf_subsystem_get_ctrlr(subsystem, data->cntlid);
589 	if (ctrlr == NULL) {
590 		SPDK_ERRLOG("Unknown controller ID 0x%x\n", data->cntlid);
591 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
592 		spdk_nvmf_request_complete(req);
593 		return;
594 	}
595 
596 	/* fail before passing a message to the controller thread. */
597 	if (ctrlr->in_destruct) {
598 		SPDK_ERRLOG("Got I/O connect while ctrlr was being destroyed.\n");
599 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
600 		spdk_nvmf_request_complete(req);
601 		return;
602 	}
603 
604 	/* If ANA reporting is enabled, check if I/O connect is on the same listener. */
605 	if (subsystem->flags.ana_reporting) {
606 		if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &listen_trid) != 0) {
607 			SPDK_ERRLOG("Could not get listener transport ID\n");
608 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
609 			spdk_nvmf_request_complete(req);
610 			return;
611 		}
612 
613 		listener = nvmf_subsystem_find_listener(subsystem, &listen_trid);
614 		if (listener != ctrlr->listener) {
615 			SPDK_ERRLOG("I/O connect is on a listener different from admin connect\n");
616 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
617 			spdk_nvmf_request_complete(req);
618 			return;
619 		}
620 	}
621 
622 	admin_qpair = ctrlr->admin_qpair;
623 	if (admin_qpair->state != SPDK_NVMF_QPAIR_ACTIVE || admin_qpair->group == NULL) {
624 		/* There is a chance that admin qpair is being destroyed at this moment due to e.g.
625 		 * expired keep alive timer. Part of the qpair destruction process is change of qpair's
626 		 * state to DEACTIVATING and removing it from poll group */
627 		SPDK_ERRLOG("Inactive admin qpair (state %d, group %p)\n", admin_qpair->state, admin_qpair->group);
628 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, qid);
629 		spdk_nvmf_request_complete(req);
630 		return;
631 	}
632 	qpair->ctrlr = ctrlr;
633 	spdk_thread_send_msg(admin_qpair->group->thread, nvmf_ctrlr_add_io_qpair, req);
634 }
635 
636 static bool
637 nvmf_qpair_access_allowed(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_subsystem *subsystem,
638 			  const char *hostnqn)
639 {
640 	struct spdk_nvme_transport_id listen_trid = {};
641 
642 	if (!spdk_nvmf_subsystem_host_allowed(subsystem, hostnqn)) {
643 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s'\n", subsystem->subnqn, hostnqn);
644 		return false;
645 	}
646 
647 	if (spdk_nvmf_qpair_get_listen_trid(qpair, &listen_trid)) {
648 		SPDK_ERRLOG("Subsystem '%s' is unable to enforce access control due to an internal error.\n",
649 			    subsystem->subnqn);
650 		return false;
651 	}
652 
653 	if (!spdk_nvmf_subsystem_listener_allowed(subsystem, &listen_trid)) {
654 		SPDK_ERRLOG("Subsystem '%s' does not allow host '%s' to connect at this address.\n",
655 			    subsystem->subnqn, hostnqn);
656 		return false;
657 	}
658 
659 	return true;
660 }
661 
662 static int
663 _nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
664 {
665 	struct spdk_nvmf_fabric_connect_data *data = req->data;
666 	struct spdk_nvmf_fabric_connect_cmd *cmd = &req->cmd->connect_cmd;
667 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
668 	struct spdk_nvmf_qpair *qpair = req->qpair;
669 	struct spdk_nvmf_transport *transport = qpair->transport;
670 	struct spdk_nvmf_ctrlr *ctrlr;
671 	struct spdk_nvmf_subsystem *subsystem;
672 
673 	SPDK_DEBUGLOG(nvmf, "recfmt 0x%x qid %u sqsize %u\n",
674 		      cmd->recfmt, cmd->qid, cmd->sqsize);
675 
676 	SPDK_DEBUGLOG(nvmf, "Connect data:\n");
677 	SPDK_DEBUGLOG(nvmf, "  cntlid:  0x%04x\n", data->cntlid);
678 	SPDK_DEBUGLOG(nvmf, "  hostid: %08x-%04x-%04x-%02x%02x-%04x%08x ***\n",
679 		      ntohl(*(uint32_t *)&data->hostid[0]),
680 		      ntohs(*(uint16_t *)&data->hostid[4]),
681 		      ntohs(*(uint16_t *)&data->hostid[6]),
682 		      data->hostid[8],
683 		      data->hostid[9],
684 		      ntohs(*(uint16_t *)&data->hostid[10]),
685 		      ntohl(*(uint32_t *)&data->hostid[12]));
686 	SPDK_DEBUGLOG(nvmf, "  subnqn: \"%s\"\n", data->subnqn);
687 	SPDK_DEBUGLOG(nvmf, "  hostnqn: \"%s\"\n", data->hostnqn);
688 
689 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
690 	if (!subsystem) {
691 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
692 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
693 	}
694 
695 	if (cmd->recfmt != 0) {
696 		SPDK_ERRLOG("Connect command unsupported RECFMT %u\n", cmd->recfmt);
697 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
698 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INCOMPATIBLE_FORMAT;
699 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
700 	}
701 
702 	/*
703 	 * SQSIZE is a 0-based value, so it must be at least 1 (minimum queue depth is 2) and
704 	 * strictly less than max_aq_depth (admin queues) or max_queue_depth (io queues).
705 	 */
706 	if (cmd->sqsize == 0) {
707 		SPDK_ERRLOG("Invalid SQSIZE = 0\n");
708 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
709 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
710 	}
711 
712 	if (cmd->qid == 0) {
713 		if (cmd->sqsize >= transport->opts.max_aq_depth) {
714 			SPDK_ERRLOG("Invalid SQSIZE for admin queue %u (min 1, max %u)\n",
715 				    cmd->sqsize, transport->opts.max_aq_depth - 1);
716 			SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
717 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
718 		}
719 	} else if (cmd->sqsize >= transport->opts.max_queue_depth) {
720 		SPDK_ERRLOG("Invalid SQSIZE %u (min 1, max %u)\n",
721 			    cmd->sqsize, transport->opts.max_queue_depth - 1);
722 		SPDK_NVMF_INVALID_CONNECT_CMD(rsp, sqsize);
723 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
724 	}
725 
726 	qpair->sq_head_max = cmd->sqsize;
727 	qpair->qid = cmd->qid;
728 
729 	if (0 == qpair->qid) {
730 		qpair->group->stat.admin_qpairs++;
731 		qpair->group->stat.current_admin_qpairs++;
732 	} else {
733 		qpair->group->stat.io_qpairs++;
734 		qpair->group->stat.current_io_qpairs++;
735 	}
736 
737 	if (cmd->qid == 0) {
738 		SPDK_DEBUGLOG(nvmf, "Connect Admin Queue for controller ID 0x%x\n", data->cntlid);
739 
740 		if (spdk_nvme_trtype_is_fabrics(transport->ops->type) && data->cntlid != 0xFFFF) {
741 			/* This NVMf target only supports dynamic mode. */
742 			SPDK_ERRLOG("The NVMf target only supports dynamic mode (CNTLID = 0x%x).\n", data->cntlid);
743 			SPDK_NVMF_INVALID_CONNECT_DATA(rsp, cntlid);
744 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
745 		}
746 
747 		/* Establish a new ctrlr */
748 		ctrlr = nvmf_ctrlr_create(subsystem, req, cmd, data);
749 		if (!ctrlr) {
750 			SPDK_ERRLOG("nvmf_ctrlr_create() failed\n");
751 			rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
752 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
753 		} else {
754 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
755 		}
756 	} else {
757 		spdk_thread_send_msg(subsystem->thread, _nvmf_ctrlr_add_io_qpair, req);
758 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
759 	}
760 }
761 
762 static inline bool
763 nvmf_request_is_fabric_connect(struct spdk_nvmf_request *req)
764 {
765 	return req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC &&
766 	       req->cmd->nvmf_cmd.fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT;
767 }
768 
769 static struct spdk_nvmf_subsystem_poll_group *
770 nvmf_subsystem_pg_from_connect_cmd(struct spdk_nvmf_request *req)
771 {
772 	struct spdk_nvmf_fabric_connect_data *data;
773 	struct spdk_nvmf_subsystem *subsystem;
774 	struct spdk_nvmf_tgt *tgt;
775 
776 	assert(nvmf_request_is_fabric_connect(req));
777 	assert(req->qpair->ctrlr == NULL);
778 
779 	data = req->data;
780 	tgt = req->qpair->transport->tgt;
781 
782 	subsystem = spdk_nvmf_tgt_find_subsystem(tgt, data->subnqn);
783 	if (subsystem == NULL) {
784 		return NULL;
785 	}
786 
787 	return &req->qpair->group->sgroups[subsystem->id];
788 }
789 
790 int
791 spdk_nvmf_ctrlr_connect(struct spdk_nvmf_request *req)
792 {
793 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
794 	struct spdk_nvmf_subsystem_poll_group *sgroup;
795 	struct spdk_nvmf_qpair *qpair = req->qpair;
796 	enum spdk_nvmf_request_exec_status status;
797 
798 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
799 	if (!sgroup) {
800 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
801 		status = SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
802 		goto out;
803 	}
804 
805 	sgroup->mgmt_io_outstanding++;
806 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
807 
808 	status = _nvmf_ctrlr_connect(req);
809 
810 out:
811 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
812 		_nvmf_request_complete(req);
813 	}
814 
815 	return status;
816 }
817 
818 static int nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req);
819 
820 static int
821 retry_connect(void *arg)
822 {
823 	struct spdk_nvmf_request *req = arg;
824 	struct spdk_nvmf_subsystem_poll_group *sgroup;
825 	int rc;
826 
827 	sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
828 	assert(sgroup != NULL);
829 	sgroup->mgmt_io_outstanding++;
830 	spdk_poller_unregister(&req->poller);
831 	rc = nvmf_ctrlr_cmd_connect(req);
832 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
833 		_nvmf_request_complete(req);
834 	}
835 	return SPDK_POLLER_BUSY;
836 }
837 
838 static int
839 nvmf_ctrlr_cmd_connect(struct spdk_nvmf_request *req)
840 {
841 	struct spdk_nvmf_fabric_connect_data *data = req->data;
842 	struct spdk_nvmf_fabric_connect_rsp *rsp = &req->rsp->connect_rsp;
843 	struct spdk_nvmf_transport *transport = req->qpair->transport;
844 	struct spdk_nvmf_subsystem *subsystem;
845 
846 	if (req->length < sizeof(struct spdk_nvmf_fabric_connect_data)) {
847 		SPDK_ERRLOG("Connect command data length 0x%x too small\n", req->length);
848 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
849 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
850 	}
851 
852 	subsystem = spdk_nvmf_tgt_find_subsystem(transport->tgt, data->subnqn);
853 	if (!subsystem) {
854 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, subnqn);
855 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
856 	}
857 
858 	if ((subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE) ||
859 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSING) ||
860 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED) ||
861 	    (subsystem->state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING)) {
862 		struct spdk_nvmf_subsystem_poll_group *sgroup;
863 
864 		if (req->timeout_tsc == 0) {
865 			/* We will only retry the request up to 1 second. */
866 			req->timeout_tsc = spdk_get_ticks() + spdk_get_ticks_hz();
867 		} else if (spdk_get_ticks() > req->timeout_tsc) {
868 			SPDK_ERRLOG("Subsystem '%s' was not ready for 1 second\n", subsystem->subnqn);
869 			rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
870 			rsp->status.sc = SPDK_NVMF_FABRIC_SC_CONTROLLER_BUSY;
871 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
872 		}
873 
874 		/* Subsystem is not ready to handle a connect. Use a poller to retry it
875 		 * again later. Decrement the mgmt_io_outstanding to avoid the
876 		 * subsystem waiting for this command to complete before unpausing.
877 		 */
878 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
879 		assert(sgroup != NULL);
880 		sgroup->mgmt_io_outstanding--;
881 		SPDK_DEBUGLOG(nvmf, "Subsystem '%s' is not ready for connect, retrying...\n", subsystem->subnqn);
882 		req->poller = SPDK_POLLER_REGISTER(retry_connect, req, 100);
883 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
884 	}
885 
886 	/* Ensure that hostnqn is null terminated */
887 	if (!memchr(data->hostnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
888 		SPDK_ERRLOG("Connect HOSTNQN is not null terminated\n");
889 		SPDK_NVMF_INVALID_CONNECT_DATA(rsp, hostnqn);
890 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
891 	}
892 
893 	if (!nvmf_qpair_access_allowed(req->qpair, subsystem, data->hostnqn)) {
894 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
895 		rsp->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_HOST;
896 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
897 	}
898 
899 	return _nvmf_ctrlr_connect(req);
900 }
901 
902 static int
903 nvmf_ctrlr_association_remove(void *ctx)
904 {
905 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
906 	int rc;
907 
908 	nvmf_ctrlr_stop_association_timer(ctrlr);
909 
910 	if (ctrlr->in_destruct) {
911 		return SPDK_POLLER_IDLE;
912 	}
913 	SPDK_DEBUGLOG(nvmf, "Disconnecting host from subsystem %s due to association timeout.\n",
914 		      ctrlr->subsys->subnqn);
915 
916 	if (ctrlr->admin_qpair) {
917 		rc = spdk_nvmf_qpair_disconnect(ctrlr->admin_qpair, NULL, NULL);
918 		if (rc < 0) {
919 			SPDK_ERRLOG("Fail to disconnect admin ctrlr qpair\n");
920 			assert(false);
921 		}
922 	}
923 
924 	return SPDK_POLLER_BUSY;
925 }
926 
927 static int
928 _nvmf_ctrlr_cc_reset_shn_done(void *ctx)
929 {
930 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
931 	uint64_t now = spdk_get_ticks();
932 	uint32_t count;
933 
934 	if (ctrlr->cc_timer) {
935 		spdk_poller_unregister(&ctrlr->cc_timer);
936 	}
937 
938 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
939 	SPDK_DEBUGLOG(nvmf, "ctrlr %p active queue count %u\n", ctrlr, count);
940 
941 	if (count > 1) {
942 		if (now < ctrlr->cc_timeout_tsc) {
943 			/* restart cc timer */
944 			ctrlr->cc_timer = SPDK_POLLER_REGISTER(_nvmf_ctrlr_cc_reset_shn_done, ctrlr, 100 * 1000);
945 			return SPDK_POLLER_IDLE;
946 		} else {
947 			/* controller fatal status */
948 			SPDK_WARNLOG("IO timeout, ctrlr %p is in fatal status\n", ctrlr);
949 			nvmf_ctrlr_set_fatal_status(ctrlr);
950 		}
951 	}
952 
953 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
954 
955 	if (ctrlr->disconnect_is_shn) {
956 		ctrlr->vcprop.csts.bits.shst = SPDK_NVME_SHST_COMPLETE;
957 		ctrlr->disconnect_is_shn = false;
958 	} else {
959 		/* Only a subset of the registers are cleared out on a reset */
960 		ctrlr->vcprop.cc.raw = 0;
961 		ctrlr->vcprop.csts.raw = 0;
962 	}
963 
964 	/* After CC.EN transitions to 0 (due to shutdown or reset), the association
965 	 * between the host and controller shall be preserved for at least 2 minutes */
966 	if (ctrlr->association_timer) {
967 		SPDK_DEBUGLOG(nvmf, "Association timer already set\n");
968 		nvmf_ctrlr_stop_association_timer(ctrlr);
969 	}
970 	if (ctrlr->association_timeout) {
971 		ctrlr->association_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_association_remove, ctrlr,
972 					   ctrlr->association_timeout * 1000);
973 	}
974 	ctrlr->disconnect_in_progress = false;
975 	return SPDK_POLLER_BUSY;
976 }
977 
978 static void
979 nvmf_ctrlr_cc_reset_shn_done(struct spdk_io_channel_iter *i, int status)
980 {
981 	struct spdk_nvmf_ctrlr *ctrlr = spdk_io_channel_iter_get_ctx(i);
982 
983 	if (status < 0) {
984 		SPDK_ERRLOG("Fail to disconnect io ctrlr qpairs\n");
985 		assert(false);
986 	}
987 
988 	_nvmf_ctrlr_cc_reset_shn_done((void *)ctrlr);
989 }
990 
991 static void
992 nvmf_bdev_complete_reset(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
993 {
994 	SPDK_NOTICELOG("Resetting bdev done with %s\n", success ? "success" : "failure");
995 
996 	spdk_bdev_free_io(bdev_io);
997 }
998 
999 
1000 static int
1001 nvmf_ctrlr_cc_timeout(void *ctx)
1002 {
1003 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
1004 	struct spdk_nvmf_poll_group *group = ctrlr->admin_qpair->group;
1005 	struct spdk_nvmf_ns *ns;
1006 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1007 
1008 	assert(group != NULL && group->sgroups != NULL);
1009 	spdk_poller_unregister(&ctrlr->cc_timeout_timer);
1010 	SPDK_DEBUGLOG(nvmf, "Ctrlr %p reset or shutdown timeout\n", ctrlr);
1011 
1012 	for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
1013 	     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1014 		if (ns->bdev == NULL) {
1015 			continue;
1016 		}
1017 		ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[ns->opts.nsid - 1];
1018 		SPDK_NOTICELOG("Ctrlr %p resetting NSID %u\n", ctrlr, ns->opts.nsid);
1019 		spdk_bdev_reset(ns->desc, ns_info->channel, nvmf_bdev_complete_reset, NULL);
1020 	}
1021 
1022 	return SPDK_POLLER_BUSY;
1023 }
1024 
1025 const struct spdk_nvmf_registers *
1026 spdk_nvmf_ctrlr_get_regs(struct spdk_nvmf_ctrlr *ctrlr)
1027 {
1028 	return &ctrlr->vcprop;
1029 }
1030 
1031 void
1032 nvmf_ctrlr_set_fatal_status(struct spdk_nvmf_ctrlr *ctrlr)
1033 {
1034 	ctrlr->vcprop.csts.bits.cfs = 1;
1035 }
1036 
1037 static uint64_t
1038 nvmf_prop_get_cap(struct spdk_nvmf_ctrlr *ctrlr)
1039 {
1040 	return ctrlr->vcprop.cap.raw;
1041 }
1042 
1043 static uint64_t
1044 nvmf_prop_get_vs(struct spdk_nvmf_ctrlr *ctrlr)
1045 {
1046 	return ctrlr->vcprop.vs.raw;
1047 }
1048 
1049 static uint64_t
1050 nvmf_prop_get_cc(struct spdk_nvmf_ctrlr *ctrlr)
1051 {
1052 	return ctrlr->vcprop.cc.raw;
1053 }
1054 
1055 static bool
1056 nvmf_prop_set_cc(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1057 {
1058 	union spdk_nvme_cc_register cc, diff;
1059 	uint32_t cc_timeout_ms;
1060 
1061 	cc.raw = value;
1062 
1063 	SPDK_DEBUGLOG(nvmf, "cur CC: 0x%08x\n", ctrlr->vcprop.cc.raw);
1064 	SPDK_DEBUGLOG(nvmf, "new CC: 0x%08x\n", cc.raw);
1065 
1066 	/*
1067 	 * Calculate which bits changed between the current and new CC.
1068 	 * Mark each bit as 0 once it is handled to determine if any unhandled bits were changed.
1069 	 */
1070 	diff.raw = cc.raw ^ ctrlr->vcprop.cc.raw;
1071 
1072 	if (diff.bits.en) {
1073 		if (cc.bits.en) {
1074 			SPDK_DEBUGLOG(nvmf, "Property Set CC Enable!\n");
1075 			nvmf_ctrlr_stop_association_timer(ctrlr);
1076 
1077 			ctrlr->vcprop.cc.bits.en = 1;
1078 			ctrlr->vcprop.csts.bits.rdy = 1;
1079 		} else {
1080 			SPDK_DEBUGLOG(nvmf, "Property Set CC Disable!\n");
1081 			if (ctrlr->disconnect_in_progress) {
1082 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1083 				return true;
1084 			}
1085 
1086 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1087 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1088 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1089 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1090 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1091 
1092 			ctrlr->vcprop.cc.bits.en = 0;
1093 			ctrlr->disconnect_in_progress = true;
1094 			ctrlr->disconnect_is_shn = false;
1095 			spdk_for_each_channel(ctrlr->subsys->tgt,
1096 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1097 					      ctrlr,
1098 					      nvmf_ctrlr_cc_reset_shn_done);
1099 		}
1100 		diff.bits.en = 0;
1101 	}
1102 
1103 	if (diff.bits.shn) {
1104 		if (cc.bits.shn == SPDK_NVME_SHN_NORMAL ||
1105 		    cc.bits.shn == SPDK_NVME_SHN_ABRUPT) {
1106 			SPDK_DEBUGLOG(nvmf, "Property Set CC Shutdown %u%ub!\n",
1107 				      cc.bits.shn >> 1, cc.bits.shn & 1);
1108 			if (ctrlr->disconnect_in_progress) {
1109 				SPDK_DEBUGLOG(nvmf, "Disconnect in progress\n");
1110 				return true;
1111 			}
1112 
1113 			ctrlr->cc_timeout_timer = SPDK_POLLER_REGISTER(nvmf_ctrlr_cc_timeout, ctrlr,
1114 						  NVMF_CC_RESET_SHN_TIMEOUT_IN_MS * 1000);
1115 			/* Make sure cc_timeout_ms is between cc_timeout_timer and Host reset/shutdown timeout */
1116 			cc_timeout_ms = (NVMF_CC_RESET_SHN_TIMEOUT_IN_MS + NVMF_CTRLR_RESET_SHN_TIMEOUT_IN_MS) / 2;
1117 			ctrlr->cc_timeout_tsc = spdk_get_ticks() + cc_timeout_ms * spdk_get_ticks_hz() / (uint64_t)1000;
1118 
1119 			ctrlr->vcprop.cc.bits.shn = cc.bits.shn;
1120 			ctrlr->disconnect_in_progress = true;
1121 			ctrlr->disconnect_is_shn = true;
1122 			spdk_for_each_channel(ctrlr->subsys->tgt,
1123 					      nvmf_ctrlr_disconnect_io_qpairs_on_pg,
1124 					      ctrlr,
1125 					      nvmf_ctrlr_cc_reset_shn_done);
1126 
1127 			/* From the time a shutdown is initiated the controller shall disable
1128 			 * Keep Alive timer */
1129 			nvmf_ctrlr_stop_keep_alive_timer(ctrlr);
1130 		} else if (cc.bits.shn == 0) {
1131 			ctrlr->vcprop.cc.bits.shn = 0;
1132 		} else {
1133 			SPDK_ERRLOG("Prop Set CC: Invalid SHN value %u%ub\n",
1134 				    cc.bits.shn >> 1, cc.bits.shn & 1);
1135 			return false;
1136 		}
1137 		diff.bits.shn = 0;
1138 	}
1139 
1140 	if (diff.bits.iosqes) {
1141 		SPDK_DEBUGLOG(nvmf, "Prop Set IOSQES = %u (%u bytes)\n",
1142 			      cc.bits.iosqes, 1u << cc.bits.iosqes);
1143 		ctrlr->vcprop.cc.bits.iosqes = cc.bits.iosqes;
1144 		diff.bits.iosqes = 0;
1145 	}
1146 
1147 	if (diff.bits.iocqes) {
1148 		SPDK_DEBUGLOG(nvmf, "Prop Set IOCQES = %u (%u bytes)\n",
1149 			      cc.bits.iocqes, 1u << cc.bits.iocqes);
1150 		ctrlr->vcprop.cc.bits.iocqes = cc.bits.iocqes;
1151 		diff.bits.iocqes = 0;
1152 	}
1153 
1154 	if (diff.bits.ams) {
1155 		SPDK_ERRLOG("Arbitration Mechanism Selected (AMS) 0x%x not supported!\n", cc.bits.ams);
1156 		return false;
1157 	}
1158 
1159 	if (diff.bits.mps) {
1160 		SPDK_ERRLOG("Memory Page Size (MPS) %u KiB not supported!\n", (1 << (2 + cc.bits.mps)));
1161 		return false;
1162 	}
1163 
1164 	if (diff.bits.css) {
1165 		SPDK_ERRLOG("I/O Command Set Selected (CSS) 0x%x not supported!\n", cc.bits.css);
1166 		return false;
1167 	}
1168 
1169 	if (diff.raw != 0) {
1170 		/* Print an error message, but don't fail the command in this case.
1171 		 * If we did want to fail in this case, we'd need to ensure we acted
1172 		 * on no other bits or the initiator gets confused. */
1173 		SPDK_ERRLOG("Prop Set CC toggled reserved bits 0x%x!\n", diff.raw);
1174 	}
1175 
1176 	return true;
1177 }
1178 
1179 static uint64_t
1180 nvmf_prop_get_csts(struct spdk_nvmf_ctrlr *ctrlr)
1181 {
1182 	return ctrlr->vcprop.csts.raw;
1183 }
1184 
1185 static uint64_t
1186 nvmf_prop_get_aqa(struct spdk_nvmf_ctrlr *ctrlr)
1187 {
1188 	return ctrlr->vcprop.aqa.raw;
1189 }
1190 
1191 static bool
1192 nvmf_prop_set_aqa(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1193 {
1194 	union spdk_nvme_aqa_register aqa;
1195 
1196 	aqa.raw = value;
1197 
1198 	/*
1199 	 * We don't need to explicitly check for maximum size, as the fields are
1200 	 * limited to 12 bits (4096).
1201 	 */
1202 	if (aqa.bits.asqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1203 	    aqa.bits.acqs < SPDK_NVME_ADMIN_QUEUE_MIN_ENTRIES - 1 ||
1204 	    aqa.bits.reserved1 != 0 || aqa.bits.reserved2 != 0) {
1205 		return false;
1206 	}
1207 
1208 	ctrlr->vcprop.aqa.raw = value;
1209 
1210 	return true;
1211 }
1212 
1213 static uint64_t
1214 nvmf_prop_get_asq(struct spdk_nvmf_ctrlr *ctrlr)
1215 {
1216 	return ctrlr->vcprop.asq;
1217 }
1218 
1219 static bool
1220 nvmf_prop_set_asq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1221 {
1222 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & (0xFFFFFFFFULL << 32ULL)) | value;
1223 
1224 	return true;
1225 }
1226 
1227 static bool
1228 nvmf_prop_set_asq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1229 {
1230 	ctrlr->vcprop.asq = (ctrlr->vcprop.asq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1231 
1232 	return true;
1233 }
1234 
1235 static uint64_t
1236 nvmf_prop_get_acq(struct spdk_nvmf_ctrlr *ctrlr)
1237 {
1238 	return ctrlr->vcprop.acq;
1239 }
1240 
1241 static bool
1242 nvmf_prop_set_acq_lower(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1243 {
1244 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & (0xFFFFFFFFULL << 32ULL)) | value;
1245 
1246 	return true;
1247 }
1248 
1249 static bool
1250 nvmf_prop_set_acq_upper(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value)
1251 {
1252 	ctrlr->vcprop.acq = (ctrlr->vcprop.acq & 0xFFFFFFFFULL) | ((uint64_t)value << 32ULL);
1253 
1254 	return true;
1255 }
1256 
1257 struct nvmf_prop {
1258 	uint32_t ofst;
1259 	uint8_t size;
1260 	char name[11];
1261 	uint64_t (*get_cb)(struct spdk_nvmf_ctrlr *ctrlr);
1262 	bool (*set_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1263 	bool (*set_upper_cb)(struct spdk_nvmf_ctrlr *ctrlr, uint32_t value);
1264 };
1265 
1266 #define PROP(field, size, get_cb, set_cb, set_upper_cb) \
1267 	{ \
1268 		offsetof(struct spdk_nvme_registers, field), \
1269 		size, \
1270 		#field, \
1271 		get_cb, set_cb, set_upper_cb \
1272 	}
1273 
1274 static const struct nvmf_prop nvmf_props[] = {
1275 	PROP(cap,  8, nvmf_prop_get_cap,  NULL,                    NULL),
1276 	PROP(vs,   4, nvmf_prop_get_vs,   NULL,                    NULL),
1277 	PROP(cc,   4, nvmf_prop_get_cc,   nvmf_prop_set_cc,        NULL),
1278 	PROP(csts, 4, nvmf_prop_get_csts, NULL,                    NULL),
1279 	PROP(aqa,  4, nvmf_prop_get_aqa,  nvmf_prop_set_aqa,       NULL),
1280 	PROP(asq,  8, nvmf_prop_get_asq,  nvmf_prop_set_asq_lower, nvmf_prop_set_asq_upper),
1281 	PROP(acq,  8, nvmf_prop_get_acq,  nvmf_prop_set_acq_lower, nvmf_prop_set_acq_upper),
1282 };
1283 
1284 static const struct nvmf_prop *
1285 find_prop(uint32_t ofst, uint8_t size)
1286 {
1287 	size_t i;
1288 
1289 	for (i = 0; i < SPDK_COUNTOF(nvmf_props); i++) {
1290 		const struct nvmf_prop *prop = &nvmf_props[i];
1291 
1292 		if ((ofst >= prop->ofst) && (ofst + size <= prop->ofst + prop->size)) {
1293 			return prop;
1294 		}
1295 	}
1296 
1297 	return NULL;
1298 }
1299 
1300 static int
1301 nvmf_property_get(struct spdk_nvmf_request *req)
1302 {
1303 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1304 	struct spdk_nvmf_fabric_prop_get_cmd *cmd = &req->cmd->prop_get_cmd;
1305 	struct spdk_nvmf_fabric_prop_get_rsp *response = &req->rsp->prop_get_rsp;
1306 	const struct nvmf_prop *prop;
1307 	uint8_t size;
1308 
1309 	response->status.sc = 0;
1310 	response->value.u64 = 0;
1311 
1312 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x\n",
1313 		      cmd->attrib.size, cmd->ofst);
1314 
1315 	switch (cmd->attrib.size) {
1316 	case SPDK_NVMF_PROP_SIZE_4:
1317 		size = 4;
1318 		break;
1319 	case SPDK_NVMF_PROP_SIZE_8:
1320 		size = 8;
1321 		break;
1322 	default:
1323 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1324 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1325 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1326 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1327 	}
1328 
1329 	prop = find_prop(cmd->ofst, size);
1330 	if (prop == NULL || prop->get_cb == NULL) {
1331 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1332 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1333 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1334 	}
1335 
1336 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1337 
1338 	response->value.u64 = prop->get_cb(ctrlr);
1339 
1340 	if (size != prop->size) {
1341 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to read. */
1342 		assert(size == 4);
1343 		assert(prop->size == 8);
1344 
1345 		if (cmd->ofst == prop->ofst) {
1346 			/* Keep bottom 4 bytes only */
1347 			response->value.u64 &= 0xFFFFFFFF;
1348 		} else {
1349 			/* Keep top 4 bytes only */
1350 			response->value.u64 >>= 32;
1351 		}
1352 	}
1353 
1354 	SPDK_DEBUGLOG(nvmf, "response value: 0x%" PRIx64 "\n", response->value.u64);
1355 
1356 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1357 }
1358 
1359 static int
1360 nvmf_property_set(struct spdk_nvmf_request *req)
1361 {
1362 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1363 	struct spdk_nvmf_fabric_prop_set_cmd *cmd = &req->cmd->prop_set_cmd;
1364 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1365 	const struct nvmf_prop *prop;
1366 	uint64_t value;
1367 	uint8_t size;
1368 	bool ret;
1369 
1370 	SPDK_DEBUGLOG(nvmf, "size %d, offset 0x%x, value 0x%" PRIx64 "\n",
1371 		      cmd->attrib.size, cmd->ofst, cmd->value.u64);
1372 
1373 	switch (cmd->attrib.size) {
1374 	case SPDK_NVMF_PROP_SIZE_4:
1375 		size = 4;
1376 		break;
1377 	case SPDK_NVMF_PROP_SIZE_8:
1378 		size = 8;
1379 		break;
1380 	default:
1381 		SPDK_DEBUGLOG(nvmf, "Invalid size value %d\n", cmd->attrib.size);
1382 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1383 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1384 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1385 	}
1386 
1387 	prop = find_prop(cmd->ofst, size);
1388 	if (prop == NULL || prop->set_cb == NULL) {
1389 		SPDK_INFOLOG(nvmf, "Invalid offset 0x%x\n", cmd->ofst);
1390 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1391 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1392 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1393 	}
1394 
1395 	SPDK_DEBUGLOG(nvmf, "name: %s\n", prop->name);
1396 
1397 	value = cmd->value.u64;
1398 
1399 	if (prop->size == 4) {
1400 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1401 	} else if (size != prop->size) {
1402 		/* The size must be 4 and the prop->size is 8. Figure out which part of the property to write. */
1403 		assert(size == 4);
1404 		assert(prop->size == 8);
1405 
1406 		if (cmd->ofst == prop->ofst) {
1407 			ret = prop->set_cb(ctrlr, (uint32_t)value);
1408 		} else {
1409 			ret = prop->set_upper_cb(ctrlr, (uint32_t)value);
1410 		}
1411 	} else {
1412 		ret = prop->set_cb(ctrlr, (uint32_t)value);
1413 		if (ret) {
1414 			ret = prop->set_upper_cb(ctrlr, (uint32_t)(value >> 32));
1415 		}
1416 	}
1417 
1418 	if (!ret) {
1419 		SPDK_ERRLOG("prop set_cb failed\n");
1420 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1421 		response->status.sc = SPDK_NVMF_FABRIC_SC_INVALID_PARAM;
1422 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1423 	}
1424 
1425 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1426 }
1427 
1428 static int
1429 nvmf_ctrlr_set_features_arbitration(struct spdk_nvmf_request *req)
1430 {
1431 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1432 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1433 
1434 	SPDK_DEBUGLOG(nvmf, "Set Features - Arbitration (cdw11 = 0x%0x)\n", cmd->cdw11);
1435 
1436 	ctrlr->feat.arbitration.raw = cmd->cdw11;
1437 	ctrlr->feat.arbitration.bits.reserved = 0;
1438 
1439 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1440 }
1441 
1442 static int
1443 nvmf_ctrlr_set_features_power_management(struct spdk_nvmf_request *req)
1444 {
1445 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1446 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1447 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1448 
1449 	SPDK_DEBUGLOG(nvmf, "Set Features - Power Management (cdw11 = 0x%0x)\n", cmd->cdw11);
1450 
1451 	/* Only PS = 0 is allowed, since we report NPSS = 0 */
1452 	if (cmd->cdw11_bits.feat_power_management.bits.ps != 0) {
1453 		SPDK_ERRLOG("Invalid power state %u\n", cmd->cdw11_bits.feat_power_management.bits.ps);
1454 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1455 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1456 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1457 	}
1458 
1459 	ctrlr->feat.power_management.raw = cmd->cdw11;
1460 	ctrlr->feat.power_management.bits.reserved = 0;
1461 
1462 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1463 }
1464 
1465 static bool
1466 temp_threshold_opts_valid(const union spdk_nvme_feat_temperature_threshold *opts)
1467 {
1468 	/*
1469 	 * Valid TMPSEL values:
1470 	 *  0000b - 1000b: temperature sensors
1471 	 *  1111b: set all implemented temperature sensors
1472 	 */
1473 	if (opts->bits.tmpsel >= 9 && opts->bits.tmpsel != 15) {
1474 		/* 1001b - 1110b: reserved */
1475 		SPDK_ERRLOG("Invalid TMPSEL %u\n", opts->bits.tmpsel);
1476 		return false;
1477 	}
1478 
1479 	/*
1480 	 * Valid THSEL values:
1481 	 *  00b: over temperature threshold
1482 	 *  01b: under temperature threshold
1483 	 */
1484 	if (opts->bits.thsel > 1) {
1485 		/* 10b - 11b: reserved */
1486 		SPDK_ERRLOG("Invalid THSEL %u\n", opts->bits.thsel);
1487 		return false;
1488 	}
1489 
1490 	return true;
1491 }
1492 
1493 static int
1494 nvmf_ctrlr_set_features_temperature_threshold(struct spdk_nvmf_request *req)
1495 {
1496 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1497 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1498 
1499 	SPDK_DEBUGLOG(nvmf, "Set Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1500 
1501 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1502 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1503 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1504 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1505 	}
1506 
1507 	/* TODO: no sensors implemented - ignore new values */
1508 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1509 }
1510 
1511 static int
1512 nvmf_ctrlr_get_features_temperature_threshold(struct spdk_nvmf_request *req)
1513 {
1514 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1515 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1516 
1517 	SPDK_DEBUGLOG(nvmf, "Get Features - Temperature Threshold (cdw11 = 0x%0x)\n", cmd->cdw11);
1518 
1519 	if (!temp_threshold_opts_valid(&cmd->cdw11_bits.feat_temp_threshold)) {
1520 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1521 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1522 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1523 	}
1524 
1525 	/* TODO: no sensors implemented - return 0 for all thresholds */
1526 	rsp->cdw0 = 0;
1527 
1528 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1529 }
1530 
1531 static int
1532 nvmf_ctrlr_get_features_interrupt_vector_configuration(struct spdk_nvmf_request *req)
1533 {
1534 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1535 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1536 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1537 	union spdk_nvme_feat_interrupt_vector_configuration iv_conf = {};
1538 
1539 	SPDK_DEBUGLOG(nvmf, "Get Features - Interrupt Vector Configuration (cdw11 = 0x%0x)\n", cmd->cdw11);
1540 
1541 	iv_conf.bits.iv = cmd->cdw11_bits.feat_interrupt_vector_configuration.bits.iv;
1542 	iv_conf.bits.cd = ctrlr->feat.interrupt_vector_configuration.bits.cd;
1543 	rsp->cdw0 = iv_conf.raw;
1544 
1545 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1546 }
1547 
1548 static int
1549 nvmf_ctrlr_set_features_error_recovery(struct spdk_nvmf_request *req)
1550 {
1551 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1552 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1553 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1554 
1555 	SPDK_DEBUGLOG(nvmf, "Set Features - Error Recovery (cdw11 = 0x%0x)\n", cmd->cdw11);
1556 
1557 	if (cmd->cdw11_bits.feat_error_recovery.bits.dulbe) {
1558 		/*
1559 		 * Host is not allowed to set this bit, since we don't advertise it in
1560 		 * Identify Namespace.
1561 		 */
1562 		SPDK_ERRLOG("Host set unsupported DULBE bit\n");
1563 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
1564 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1565 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1566 	}
1567 
1568 	ctrlr->feat.error_recovery.raw = cmd->cdw11;
1569 	ctrlr->feat.error_recovery.bits.reserved = 0;
1570 
1571 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1572 }
1573 
1574 static int
1575 nvmf_ctrlr_set_features_volatile_write_cache(struct spdk_nvmf_request *req)
1576 {
1577 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1578 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1579 
1580 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache (cdw11 = 0x%0x)\n", cmd->cdw11);
1581 
1582 	ctrlr->feat.volatile_write_cache.raw = cmd->cdw11;
1583 	ctrlr->feat.volatile_write_cache.bits.reserved = 0;
1584 
1585 	SPDK_DEBUGLOG(nvmf, "Set Features - Volatile Write Cache %s\n",
1586 		      ctrlr->feat.volatile_write_cache.bits.wce ? "Enabled" : "Disabled");
1587 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1588 }
1589 
1590 static int
1591 nvmf_ctrlr_set_features_write_atomicity(struct spdk_nvmf_request *req)
1592 {
1593 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1594 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1595 
1596 	SPDK_DEBUGLOG(nvmf, "Set Features - Write Atomicity (cdw11 = 0x%0x)\n", cmd->cdw11);
1597 
1598 	ctrlr->feat.write_atomicity.raw = cmd->cdw11;
1599 	ctrlr->feat.write_atomicity.bits.reserved = 0;
1600 
1601 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1602 }
1603 
1604 static int
1605 nvmf_ctrlr_set_features_host_identifier(struct spdk_nvmf_request *req)
1606 {
1607 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1608 
1609 	SPDK_ERRLOG("Set Features - Host Identifier not allowed\n");
1610 	response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1611 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1612 }
1613 
1614 static int
1615 nvmf_ctrlr_get_features_host_identifier(struct spdk_nvmf_request *req)
1616 {
1617 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1618 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1619 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1620 
1621 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Identifier\n");
1622 
1623 	if (!cmd->cdw11_bits.feat_host_identifier.bits.exhid) {
1624 		/* NVMe over Fabrics requires EXHID=1 (128-bit/16-byte host ID) */
1625 		SPDK_ERRLOG("Get Features - Host Identifier with EXHID=0 not allowed\n");
1626 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1627 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1628 	}
1629 
1630 	if (req->data == NULL || req->length < sizeof(ctrlr->hostid)) {
1631 		SPDK_ERRLOG("Invalid data buffer for Get Features - Host Identifier\n");
1632 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1633 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1634 	}
1635 
1636 	spdk_uuid_copy((struct spdk_uuid *)req->data, &ctrlr->hostid);
1637 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1638 }
1639 
1640 static int
1641 nvmf_ctrlr_get_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1642 {
1643 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1644 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1645 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1646 	struct spdk_nvmf_ns *ns;
1647 
1648 	SPDK_DEBUGLOG(nvmf, "get Features - Reservation Notification Mask\n");
1649 
1650 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1651 		SPDK_ERRLOG("get Features - Invalid Namespace ID\n");
1652 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1653 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1654 	}
1655 
1656 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1657 	if (ns == NULL) {
1658 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1659 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1660 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1661 	}
1662 	rsp->cdw0 = ns->mask;
1663 
1664 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1665 }
1666 
1667 static int
1668 nvmf_ctrlr_set_features_reservation_notification_mask(struct spdk_nvmf_request *req)
1669 {
1670 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1671 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
1672 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1673 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1674 	struct spdk_nvmf_ns *ns;
1675 
1676 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Notification Mask\n");
1677 
1678 	if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1679 		for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
1680 		     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
1681 			ns->mask = cmd->cdw11;
1682 		}
1683 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1684 	}
1685 
1686 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1687 	if (ns == NULL) {
1688 		SPDK_ERRLOG("Set Features - Invalid Namespace ID\n");
1689 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1690 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1691 	}
1692 	ns->mask = cmd->cdw11;
1693 
1694 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1695 }
1696 
1697 static int
1698 nvmf_ctrlr_get_features_reservation_persistence(struct spdk_nvmf_request *req)
1699 {
1700 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1701 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1702 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1703 	struct spdk_nvmf_ns *ns;
1704 
1705 	SPDK_DEBUGLOG(nvmf, "Get Features - Reservation Persistence\n");
1706 
1707 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1708 	/* NSID with SPDK_NVME_GLOBAL_NS_TAG (=0xffffffff) also included */
1709 	if (ns == NULL) {
1710 		SPDK_ERRLOG("Get Features - Invalid Namespace ID\n");
1711 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1712 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1713 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1714 	}
1715 
1716 	response->cdw0 = ns->ptpl_activated;
1717 
1718 	response->status.sct = SPDK_NVME_SCT_GENERIC;
1719 	response->status.sc = SPDK_NVME_SC_SUCCESS;
1720 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1721 }
1722 
1723 static int
1724 nvmf_ctrlr_set_features_reservation_persistence(struct spdk_nvmf_request *req)
1725 {
1726 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1727 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1728 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1729 	struct spdk_nvmf_ns *ns;
1730 	bool ptpl;
1731 
1732 	SPDK_DEBUGLOG(nvmf, "Set Features - Reservation Persistence\n");
1733 
1734 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
1735 	ptpl = cmd->cdw11_bits.feat_rsv_persistence.bits.ptpl;
1736 
1737 	if (cmd->nsid != SPDK_NVME_GLOBAL_NS_TAG && ns && ns->ptpl_file) {
1738 		ns->ptpl_activated = ptpl;
1739 	} else if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1740 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns && ns->ptpl_file;
1741 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
1742 			ns->ptpl_activated = ptpl;
1743 		}
1744 	} else {
1745 		SPDK_ERRLOG("Set Features - Invalid Namespace ID or Reservation Configuration\n");
1746 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1747 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1748 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1749 	}
1750 
1751 	/* TODO: Feature not changeable for now */
1752 	response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1753 	response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
1754 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1755 }
1756 
1757 static int
1758 nvmf_ctrlr_get_features_host_behavior_support(struct spdk_nvmf_request *req)
1759 {
1760 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1761 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1762 	struct spdk_nvme_host_behavior host_behavior = {};
1763 
1764 	SPDK_DEBUGLOG(nvmf, "Get Features - Host Behavior Support\n");
1765 
1766 	if (req->data == NULL || req->length < sizeof(struct spdk_nvme_host_behavior)) {
1767 		SPDK_ERRLOG("invalid data buffer for Host Behavior Support\n");
1768 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1769 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1770 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1771 	}
1772 
1773 	host_behavior.acre = ctrlr->acre_enabled;
1774 	memcpy(req->data, &host_behavior, sizeof(host_behavior));
1775 
1776 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1777 }
1778 
1779 static int
1780 nvmf_ctrlr_set_features_host_behavior_support(struct spdk_nvmf_request *req)
1781 {
1782 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1783 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
1784 	struct spdk_nvme_host_behavior *host_behavior;
1785 
1786 	SPDK_DEBUGLOG(nvmf, "Set Features - Host Behavior Support\n");
1787 	if (req->iovcnt != 1) {
1788 		SPDK_ERRLOG("Host Behavior Support invalid iovcnt: %d\n", req->iovcnt);
1789 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1790 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1791 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1792 	}
1793 	if (req->iov[0].iov_len != sizeof(struct spdk_nvme_host_behavior)) {
1794 		SPDK_ERRLOG("Host Behavior Support invalid iov_len: %zd\n", req->iov[0].iov_len);
1795 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1796 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1797 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1798 	}
1799 
1800 	host_behavior = (struct spdk_nvme_host_behavior *)req->iov[0].iov_base;
1801 	if (host_behavior->acre == 0) {
1802 		ctrlr->acre_enabled = false;
1803 	} else if (host_behavior->acre == 1) {
1804 		ctrlr->acre_enabled = true;
1805 	} else {
1806 		SPDK_ERRLOG("Host Behavior Support invalid acre: 0x%02x\n", host_behavior->acre);
1807 		response->status.sct = SPDK_NVME_SCT_GENERIC;
1808 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1809 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1810 	}
1811 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1812 }
1813 
1814 static int
1815 nvmf_ctrlr_set_features_keep_alive_timer(struct spdk_nvmf_request *req)
1816 {
1817 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1818 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1819 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1820 
1821 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer (%u ms)\n", cmd->cdw11);
1822 
1823 	/*
1824 	 * if attempts to disable keep alive by setting kato to 0h
1825 	 * a status value of keep alive invalid shall be returned
1826 	 */
1827 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato == 0) {
1828 		rsp->status.sc = SPDK_NVME_SC_KEEP_ALIVE_INVALID;
1829 	} else if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato < MIN_KEEP_ALIVE_TIMEOUT_IN_MS) {
1830 		ctrlr->feat.keep_alive_timer.bits.kato = MIN_KEEP_ALIVE_TIMEOUT_IN_MS;
1831 	} else {
1832 		/* round up to milliseconds */
1833 		ctrlr->feat.keep_alive_timer.bits.kato = spdk_divide_round_up(
1834 					cmd->cdw11_bits.feat_keep_alive_timer.bits.kato,
1835 					KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS) *
1836 				KAS_DEFAULT_VALUE * KAS_TIME_UNIT_IN_MS;
1837 	}
1838 
1839 	/*
1840 	 * if change the keep alive timeout value successfully
1841 	 * update the keep alive poller.
1842 	 */
1843 	if (cmd->cdw11_bits.feat_keep_alive_timer.bits.kato != 0) {
1844 		if (ctrlr->keep_alive_poller != NULL) {
1845 			spdk_poller_unregister(&ctrlr->keep_alive_poller);
1846 		}
1847 		ctrlr->keep_alive_poller = SPDK_POLLER_REGISTER(nvmf_ctrlr_keep_alive_poll, ctrlr,
1848 					   ctrlr->feat.keep_alive_timer.bits.kato * 1000);
1849 	}
1850 
1851 	SPDK_DEBUGLOG(nvmf, "Set Features - Keep Alive Timer set to %u ms\n",
1852 		      ctrlr->feat.keep_alive_timer.bits.kato);
1853 
1854 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1855 }
1856 
1857 static int
1858 nvmf_ctrlr_set_features_number_of_queues(struct spdk_nvmf_request *req)
1859 {
1860 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1861 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1862 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1863 	uint32_t count;
1864 
1865 	SPDK_DEBUGLOG(nvmf, "Set Features - Number of Queues, cdw11 0x%x\n",
1866 		      req->cmd->nvme_cmd.cdw11);
1867 
1868 	if (cmd->cdw11_bits.feat_num_of_queues.bits.ncqr == UINT16_MAX ||
1869 	    cmd->cdw11_bits.feat_num_of_queues.bits.nsqr == UINT16_MAX) {
1870 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
1871 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1872 	}
1873 
1874 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
1875 	/* verify that the controller is ready to process commands */
1876 	if (count > 1) {
1877 		SPDK_DEBUGLOG(nvmf, "Queue pairs already active!\n");
1878 		rsp->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
1879 	} else {
1880 		/*
1881 		 * Ignore the value requested by the host -
1882 		 * always return the pre-configured value based on max_qpairs_allowed.
1883 		 */
1884 		rsp->cdw0 = ctrlr->feat.number_of_queues.raw;
1885 	}
1886 
1887 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1888 }
1889 
1890 int
1891 nvmf_ctrlr_save_aers(struct spdk_nvmf_ctrlr *ctrlr, uint16_t *aer_cids,
1892 		     uint16_t max_aers)
1893 {
1894 	struct spdk_nvmf_request *req;
1895 	uint16_t i;
1896 
1897 	if (!aer_cids || max_aers < ctrlr->nr_aer_reqs) {
1898 		return -EINVAL;
1899 	}
1900 
1901 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
1902 		req = ctrlr->aer_req[i];
1903 		aer_cids[i] = req->cmd->nvme_cmd.cid;
1904 	}
1905 
1906 	return ctrlr->nr_aer_reqs;
1907 }
1908 
1909 int
1910 nvmf_ctrlr_save_migr_data(struct spdk_nvmf_ctrlr *ctrlr, struct nvmf_ctrlr_migr_data *data)
1911 {
1912 	uint32_t num_async_events = 0;
1913 	struct spdk_nvmf_async_event_completion *event, *event_tmp;
1914 
1915 	memcpy(&data->feat, &ctrlr->feat, sizeof(struct spdk_nvmf_ctrlr_feat));
1916 	data->cntlid = ctrlr->cntlid;
1917 	data->acre_enabled = ctrlr->acre_enabled;
1918 	data->notice_aen_mask = ctrlr->notice_aen_mask;
1919 
1920 	STAILQ_FOREACH_SAFE(event, &ctrlr->async_events, link, event_tmp) {
1921 		data->async_events[num_async_events++].raw = event->event.raw;
1922 		if (num_async_events == NVMF_MIGR_MAX_PENDING_AERS) {
1923 			SPDK_ERRLOG("%p has too many pending AERs\n", ctrlr);
1924 			break;
1925 		}
1926 	}
1927 	data->num_async_events = num_async_events;
1928 
1929 	return 0;
1930 }
1931 
1932 int
1933 nvmf_ctrlr_restore_migr_data(struct spdk_nvmf_ctrlr *ctrlr, struct nvmf_ctrlr_migr_data *data)
1934 {
1935 	struct spdk_nvmf_async_event_completion *event;
1936 	uint32_t i;
1937 
1938 	memcpy(&ctrlr->feat, &data->feat, sizeof(struct spdk_nvmf_ctrlr_feat));
1939 	ctrlr->acre_enabled = data->acre_enabled;
1940 	ctrlr->notice_aen_mask = data->notice_aen_mask;
1941 
1942 	for (i = 0; i < data->num_async_events; i++) {
1943 		event = calloc(1, sizeof(struct spdk_nvmf_async_event_completion));
1944 		if (!event) {
1945 			return -ENOMEM;
1946 		}
1947 		event->event.raw = data->async_events[i].raw;
1948 		STAILQ_INSERT_TAIL(&ctrlr->async_events, event, link);
1949 	}
1950 
1951 	return 0;
1952 }
1953 
1954 static int
1955 nvmf_ctrlr_set_features_async_event_configuration(struct spdk_nvmf_request *req)
1956 {
1957 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1958 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
1959 
1960 	SPDK_DEBUGLOG(nvmf, "Set Features - Async Event Configuration, cdw11 0x%08x\n",
1961 		      cmd->cdw11);
1962 	ctrlr->feat.async_event_configuration.raw = cmd->cdw11;
1963 	ctrlr->feat.async_event_configuration.bits.reserved1 = 0;
1964 	ctrlr->feat.async_event_configuration.bits.reserved2 = 0;
1965 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1966 }
1967 
1968 static int
1969 nvmf_ctrlr_async_event_request(struct spdk_nvmf_request *req)
1970 {
1971 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
1972 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
1973 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1974 	struct spdk_nvmf_async_event_completion *pending_event;
1975 
1976 	SPDK_DEBUGLOG(nvmf, "Async Event Request\n");
1977 
1978 	/* AER cmd is an exception */
1979 	sgroup = &req->qpair->group->sgroups[ctrlr->subsys->id];
1980 	assert(sgroup != NULL);
1981 	sgroup->mgmt_io_outstanding--;
1982 
1983 	/* Four asynchronous events are supported for now */
1984 	if (ctrlr->nr_aer_reqs >= NVMF_MAX_ASYNC_EVENTS) {
1985 		SPDK_DEBUGLOG(nvmf, "AERL exceeded\n");
1986 		rsp->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
1987 		rsp->status.sc = SPDK_NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED;
1988 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1989 	}
1990 
1991 	if (!STAILQ_EMPTY(&ctrlr->async_events)) {
1992 		pending_event = STAILQ_FIRST(&ctrlr->async_events);
1993 		rsp->cdw0 = pending_event->event.raw;
1994 		STAILQ_REMOVE(&ctrlr->async_events, pending_event, spdk_nvmf_async_event_completion, link);
1995 		free(pending_event);
1996 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
1997 	}
1998 
1999 	ctrlr->aer_req[ctrlr->nr_aer_reqs++] = req;
2000 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
2001 }
2002 
2003 struct copy_iovs_ctx {
2004 	struct iovec *iovs;
2005 	int iovcnt;
2006 	int cur_iov_idx;
2007 	size_t cur_iov_offset;
2008 };
2009 
2010 static void
2011 _clear_iovs(struct iovec *iovs, int iovcnt)
2012 {
2013 	int iov_idx = 0;
2014 	struct iovec *iov;
2015 
2016 	while (iov_idx < iovcnt) {
2017 		iov = &iovs[iov_idx];
2018 		memset(iov->iov_base, 0, iov->iov_len);
2019 		iov_idx++;
2020 	}
2021 }
2022 
2023 static void
2024 _init_copy_iovs_ctx(struct copy_iovs_ctx *copy_ctx, struct iovec *iovs, int iovcnt)
2025 {
2026 	copy_ctx->iovs = iovs;
2027 	copy_ctx->iovcnt = iovcnt;
2028 	copy_ctx->cur_iov_idx = 0;
2029 	copy_ctx->cur_iov_offset = 0;
2030 }
2031 
2032 static size_t
2033 _copy_buf_to_iovs(struct copy_iovs_ctx *copy_ctx, const void *buf, size_t buf_len)
2034 {
2035 	size_t len, iov_remain_len, copied_len = 0;
2036 	struct iovec *iov;
2037 
2038 	if (buf_len == 0) {
2039 		return 0;
2040 	}
2041 
2042 	while (copy_ctx->cur_iov_idx < copy_ctx->iovcnt) {
2043 		iov = &copy_ctx->iovs[copy_ctx->cur_iov_idx];
2044 		iov_remain_len = iov->iov_len - copy_ctx->cur_iov_offset;
2045 		if (iov_remain_len == 0) {
2046 			copy_ctx->cur_iov_idx++;
2047 			copy_ctx->cur_iov_offset = 0;
2048 			continue;
2049 		}
2050 
2051 		len = spdk_min(iov_remain_len, buf_len - copied_len);
2052 		memcpy((char *)iov->iov_base + copy_ctx->cur_iov_offset,
2053 		       (const char *)buf + copied_len,
2054 		       len);
2055 		copied_len += len;
2056 		copy_ctx->cur_iov_offset += len;
2057 
2058 		if (buf_len == copied_len) {
2059 			return copied_len;
2060 		}
2061 	}
2062 
2063 	return copied_len;
2064 }
2065 
2066 static void
2067 nvmf_get_firmware_slot_log_page(struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length)
2068 {
2069 	struct spdk_nvme_firmware_page fw_page;
2070 	size_t copy_len;
2071 	struct copy_iovs_ctx copy_ctx;
2072 
2073 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2074 
2075 	memset(&fw_page, 0, sizeof(fw_page));
2076 	fw_page.afi.active_slot = 1;
2077 	fw_page.afi.next_reset_slot = 0;
2078 	spdk_strcpy_pad(fw_page.revision[0], FW_VERSION, sizeof(fw_page.revision[0]), ' ');
2079 
2080 	if (offset < sizeof(fw_page)) {
2081 		copy_len = spdk_min(sizeof(fw_page) - offset, length);
2082 		if (copy_len > 0) {
2083 			_copy_buf_to_iovs(&copy_ctx, (const char *)&fw_page + offset, copy_len);
2084 		}
2085 	}
2086 }
2087 
2088 /*
2089  * Asynchronous Event Mask Bit
2090  */
2091 enum spdk_nvme_async_event_mask_bit {
2092 	/* Mask Namespace Change Notification */
2093 	SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT		= 0,
2094 	/* Mask Asymmetric Namespace Access Change Notification */
2095 	SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT		= 1,
2096 	/* Mask Discovery Log Change Notification */
2097 	SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT	= 2,
2098 	/* Mask Reservation Log Page Available Notification */
2099 	SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT	= 3,
2100 	/* Mask Error Event */
2101 	SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT			= 4,
2102 	/* 4 - 63 Reserved */
2103 };
2104 
2105 static inline void
2106 nvmf_ctrlr_unmask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2107 		      enum spdk_nvme_async_event_mask_bit mask)
2108 {
2109 	ctrlr->notice_aen_mask &= ~(1 << mask);
2110 }
2111 
2112 static inline bool
2113 nvmf_ctrlr_mask_aen(struct spdk_nvmf_ctrlr *ctrlr,
2114 		    enum spdk_nvme_async_event_mask_bit mask)
2115 {
2116 	if (ctrlr->notice_aen_mask & (1 << mask)) {
2117 		return false;
2118 	} else {
2119 		ctrlr->notice_aen_mask |= (1 << mask);
2120 		return true;
2121 	}
2122 }
2123 
2124 /* we have to use the typedef in the function declaration to appease astyle. */
2125 typedef enum spdk_nvme_ana_state spdk_nvme_ana_state_t;
2126 
2127 static inline spdk_nvme_ana_state_t
2128 nvmf_ctrlr_get_ana_state(struct spdk_nvmf_ctrlr *ctrlr, uint32_t anagrpid)
2129 {
2130 	if (!ctrlr->subsys->flags.ana_reporting) {
2131 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2132 	}
2133 
2134 	if (spdk_unlikely(ctrlr->listener == NULL)) {
2135 		return SPDK_NVME_ANA_INACCESSIBLE_STATE;
2136 	}
2137 
2138 	assert(anagrpid - 1 < ctrlr->subsys->max_nsid);
2139 	return ctrlr->listener->ana_state[anagrpid - 1];
2140 }
2141 
2142 static spdk_nvme_ana_state_t
2143 nvmf_ctrlr_get_ana_state_from_nsid(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2144 {
2145 	struct spdk_nvmf_ns *ns;
2146 
2147 	/* We do not have NVM subsystem specific ANA state. Hence if NSID is either
2148 	 * SPDK_NVMF_GLOBAL_NS_TAG, invalid, or for inactive namespace, return
2149 	 * the optimized state.
2150 	 */
2151 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
2152 	if (ns == NULL) {
2153 		return SPDK_NVME_ANA_OPTIMIZED_STATE;
2154 	}
2155 
2156 	return nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2157 }
2158 
2159 static void
2160 nvmf_get_error_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2161 			uint64_t offset, uint32_t length, uint32_t rae)
2162 {
2163 	if (!rae) {
2164 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT);
2165 	}
2166 
2167 	/* TODO: actually fill out log page data */
2168 }
2169 
2170 static void
2171 nvmf_get_ana_log_page(struct spdk_nvmf_ctrlr *ctrlr, struct iovec *iovs, int iovcnt,
2172 		      uint64_t offset, uint32_t length, uint32_t rae)
2173 {
2174 	struct spdk_nvme_ana_page ana_hdr;
2175 	struct spdk_nvme_ana_group_descriptor ana_desc;
2176 	size_t copy_len, copied_len;
2177 	uint32_t num_anagrp = 0, anagrpid;
2178 	struct spdk_nvmf_ns *ns;
2179 	struct copy_iovs_ctx copy_ctx;
2180 
2181 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2182 
2183 	if (length == 0) {
2184 		goto done;
2185 	}
2186 
2187 	if (offset >= sizeof(ana_hdr)) {
2188 		offset -= sizeof(ana_hdr);
2189 	} else {
2190 		for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2191 			if (ctrlr->subsys->ana_group[anagrpid - 1] > 0) {
2192 				num_anagrp++;
2193 			}
2194 		}
2195 
2196 		memset(&ana_hdr, 0, sizeof(ana_hdr));
2197 
2198 		ana_hdr.num_ana_group_desc = num_anagrp;
2199 		/* TODO: Support Change Count. */
2200 		ana_hdr.change_count = 0;
2201 
2202 		copy_len = spdk_min(sizeof(ana_hdr) - offset, length);
2203 		copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ana_hdr + offset, copy_len);
2204 		assert(copied_len == copy_len);
2205 		length -= copied_len;
2206 		offset = 0;
2207 	}
2208 
2209 	if (length == 0) {
2210 		goto done;
2211 	}
2212 
2213 	for (anagrpid = 1; anagrpid <= ctrlr->subsys->max_nsid; anagrpid++) {
2214 		if (ctrlr->subsys->ana_group[anagrpid - 1] == 0) {
2215 			continue;
2216 		}
2217 
2218 		if (offset >= sizeof(ana_desc)) {
2219 			offset -= sizeof(ana_desc);
2220 		} else {
2221 			memset(&ana_desc, 0, sizeof(ana_desc));
2222 
2223 			ana_desc.ana_group_id = anagrpid;
2224 			ana_desc.num_of_nsid = ctrlr->subsys->ana_group[anagrpid - 1];
2225 			ana_desc.ana_state = nvmf_ctrlr_get_ana_state(ctrlr, anagrpid);
2226 
2227 			copy_len = spdk_min(sizeof(ana_desc) - offset, length);
2228 			copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ana_desc + offset,
2229 						       copy_len);
2230 			assert(copied_len == copy_len);
2231 			length -= copied_len;
2232 			offset = 0;
2233 
2234 			if (length == 0) {
2235 				goto done;
2236 			}
2237 		}
2238 
2239 		/* TODO: Revisit here about O(n^2) cost if we have subsystem with
2240 		 * many namespaces in the future.
2241 		 */
2242 		for (ns = spdk_nvmf_subsystem_get_first_ns(ctrlr->subsys); ns != NULL;
2243 		     ns = spdk_nvmf_subsystem_get_next_ns(ctrlr->subsys, ns)) {
2244 			if (ns->anagrpid != anagrpid) {
2245 				continue;
2246 			}
2247 
2248 			if (offset >= sizeof(uint32_t)) {
2249 				offset -= sizeof(uint32_t);
2250 				continue;
2251 			}
2252 
2253 			copy_len = spdk_min(sizeof(uint32_t) - offset, length);
2254 			copied_len = _copy_buf_to_iovs(&copy_ctx, (const char *)&ns->nsid + offset,
2255 						       copy_len);
2256 			assert(copied_len == copy_len);
2257 			length -= copied_len;
2258 			offset = 0;
2259 
2260 			if (length == 0) {
2261 				goto done;
2262 			}
2263 		}
2264 	}
2265 
2266 done:
2267 	if (!rae) {
2268 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT);
2269 	}
2270 }
2271 
2272 void
2273 nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid)
2274 {
2275 	uint16_t max_changes = SPDK_COUNTOF(ctrlr->changed_ns_list.ns_list);
2276 	uint16_t i;
2277 	bool found = false;
2278 
2279 	for (i = 0; i < ctrlr->changed_ns_list_count; i++) {
2280 		if (ctrlr->changed_ns_list.ns_list[i] == nsid) {
2281 			/* nsid is already in the list */
2282 			found = true;
2283 			break;
2284 		}
2285 	}
2286 
2287 	if (!found) {
2288 		if (ctrlr->changed_ns_list_count == max_changes) {
2289 			/* Out of space - set first entry to FFFFFFFFh and zero-fill the rest. */
2290 			ctrlr->changed_ns_list.ns_list[0] = 0xFFFFFFFFu;
2291 			for (i = 1; i < max_changes; i++) {
2292 				ctrlr->changed_ns_list.ns_list[i] = 0;
2293 			}
2294 		} else {
2295 			ctrlr->changed_ns_list.ns_list[ctrlr->changed_ns_list_count++] = nsid;
2296 		}
2297 	}
2298 }
2299 
2300 static void
2301 nvmf_get_changed_ns_list_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2302 				  struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2303 {
2304 	size_t copy_length;
2305 	struct copy_iovs_ctx copy_ctx;
2306 
2307 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2308 
2309 	if (offset < sizeof(ctrlr->changed_ns_list)) {
2310 		copy_length = spdk_min(length, sizeof(ctrlr->changed_ns_list) - offset);
2311 		if (copy_length) {
2312 			_copy_buf_to_iovs(&copy_ctx, (char *)&ctrlr->changed_ns_list + offset, copy_length);
2313 		}
2314 	}
2315 
2316 	/* Clear log page each time it is read */
2317 	ctrlr->changed_ns_list_count = 0;
2318 	memset(&ctrlr->changed_ns_list, 0, sizeof(ctrlr->changed_ns_list));
2319 
2320 	if (!rae) {
2321 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT);
2322 	}
2323 }
2324 
2325 /* The structure can be modified if we provide support for other commands in future */
2326 static const struct spdk_nvme_cmds_and_effect_log_page g_cmds_and_effect_log_page = {
2327 	.admin_cmds_supported = {
2328 		/* CSUPP, LBCC, NCC, NIC, CCC, CSE */
2329 		/* Get Log Page */
2330 		[SPDK_NVME_OPC_GET_LOG_PAGE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2331 		/* Identify */
2332 		[SPDK_NVME_OPC_IDENTIFY]		= {1, 0, 0, 0, 0, 0, 0, 0},
2333 		/* Abort */
2334 		[SPDK_NVME_OPC_ABORT]			= {1, 0, 0, 0, 0, 0, 0, 0},
2335 		/* Set Features */
2336 		[SPDK_NVME_OPC_SET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2337 		/* Get Features */
2338 		[SPDK_NVME_OPC_GET_FEATURES]		= {1, 0, 0, 0, 0, 0, 0, 0},
2339 		/* Async Event Request */
2340 		[SPDK_NVME_OPC_ASYNC_EVENT_REQUEST]	= {1, 0, 0, 0, 0, 0, 0, 0},
2341 		/* Keep Alive */
2342 		[SPDK_NVME_OPC_KEEP_ALIVE]		= {1, 0, 0, 0, 0, 0, 0, 0},
2343 	},
2344 	.io_cmds_supported = {
2345 		/* FLUSH */
2346 		[SPDK_NVME_OPC_FLUSH]			= {1, 1, 0, 0, 0, 0, 0, 0},
2347 		/* WRITE */
2348 		[SPDK_NVME_OPC_WRITE]			= {1, 1, 0, 0, 0, 0, 0, 0},
2349 		/* READ */
2350 		[SPDK_NVME_OPC_READ]			= {1, 0, 0, 0, 0, 0, 0, 0},
2351 		/* WRITE ZEROES */
2352 		[SPDK_NVME_OPC_WRITE_ZEROES]		= {1, 1, 0, 0, 0, 0, 0, 0},
2353 		/* DATASET MANAGEMENT */
2354 		[SPDK_NVME_OPC_DATASET_MANAGEMENT]	= {1, 1, 0, 0, 0, 0, 0, 0},
2355 		/* COMPARE */
2356 		[SPDK_NVME_OPC_COMPARE]			= {1, 0, 0, 0, 0, 0, 0, 0},
2357 	},
2358 };
2359 
2360 static void
2361 nvmf_get_cmds_and_effects_log_page(struct iovec *iovs, int iovcnt,
2362 				   uint64_t offset, uint32_t length)
2363 {
2364 	uint32_t page_size = sizeof(struct spdk_nvme_cmds_and_effect_log_page);
2365 	size_t copy_len = 0;
2366 	struct copy_iovs_ctx copy_ctx;
2367 
2368 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2369 
2370 	if (offset < page_size) {
2371 		copy_len = spdk_min(page_size - offset, length);
2372 		_copy_buf_to_iovs(&copy_ctx, (char *)(&g_cmds_and_effect_log_page) + offset, copy_len);
2373 	}
2374 }
2375 
2376 static void
2377 nvmf_get_reservation_notification_log_page(struct spdk_nvmf_ctrlr *ctrlr,
2378 		struct iovec *iovs, int iovcnt, uint64_t offset, uint32_t length, uint32_t rae)
2379 {
2380 	uint32_t unit_log_len, avail_log_len, next_pos, copy_len;
2381 	struct spdk_nvmf_reservation_log *log, *log_tmp;
2382 	struct copy_iovs_ctx copy_ctx;
2383 
2384 	_init_copy_iovs_ctx(&copy_ctx, iovs, iovcnt);
2385 
2386 	unit_log_len = sizeof(struct spdk_nvme_reservation_notification_log);
2387 	/* No available log, return zeroed log pages */
2388 	if (!ctrlr->num_avail_log_pages) {
2389 		return;
2390 	}
2391 
2392 	avail_log_len = ctrlr->num_avail_log_pages * unit_log_len;
2393 	if (offset >= avail_log_len) {
2394 		return;
2395 	}
2396 
2397 	next_pos = 0;
2398 	TAILQ_FOREACH_SAFE(log, &ctrlr->log_head, link, log_tmp) {
2399 		TAILQ_REMOVE(&ctrlr->log_head, log, link);
2400 		ctrlr->num_avail_log_pages--;
2401 
2402 		next_pos += unit_log_len;
2403 		if (next_pos > offset) {
2404 			copy_len = spdk_min(next_pos - offset, length);
2405 			_copy_buf_to_iovs(&copy_ctx, &log->log, copy_len);
2406 			length -= copy_len;
2407 			offset += copy_len;
2408 		}
2409 		free(log);
2410 
2411 		if (length == 0) {
2412 			break;
2413 		}
2414 	}
2415 
2416 	if (!rae) {
2417 		nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT);
2418 	}
2419 	return;
2420 }
2421 
2422 static int
2423 nvmf_ctrlr_get_log_page(struct spdk_nvmf_request *req)
2424 {
2425 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2426 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2427 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2428 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
2429 	struct spdk_nvme_transport_id cmd_source_trid;
2430 	uint64_t offset, len;
2431 	uint32_t rae, numdl, numdu;
2432 	uint8_t lid;
2433 
2434 	if (req->data == NULL) {
2435 		SPDK_DEBUGLOG(nvmf, "get log command with no buffer\n");
2436 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2437 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2438 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2439 	}
2440 
2441 	offset = (uint64_t)cmd->cdw12 | ((uint64_t)cmd->cdw13 << 32);
2442 	if (offset & 3) {
2443 		SPDK_ERRLOG("Invalid log page offset 0x%" PRIx64 "\n", offset);
2444 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2445 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2446 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2447 	}
2448 
2449 	rae = cmd->cdw10_bits.get_log_page.rae;
2450 	numdl = cmd->cdw10_bits.get_log_page.numdl;
2451 	numdu = cmd->cdw11_bits.get_log_page.numdu;
2452 	len = ((numdu << 16) + numdl + (uint64_t)1) * 4;
2453 	if (len > req->length) {
2454 		SPDK_ERRLOG("Get log page: len (%" PRIu64 ") > buf size (%u)\n",
2455 			    len, req->length);
2456 		response->status.sct = SPDK_NVME_SCT_GENERIC;
2457 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2458 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2459 	}
2460 
2461 	lid = cmd->cdw10_bits.get_log_page.lid;
2462 	SPDK_DEBUGLOG(nvmf, "Get log page: LID=0x%02X offset=0x%" PRIx64 " len=0x%" PRIx64 " rae=%u\n",
2463 		      lid, offset, len, rae);
2464 
2465 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2466 		switch (lid) {
2467 		case SPDK_NVME_LOG_DISCOVERY:
2468 			if (spdk_nvmf_qpair_get_listen_trid(req->qpair, &cmd_source_trid)) {
2469 				SPDK_ERRLOG("Failed to get LOG_DISCOVERY source trid\n");
2470 				response->status.sct = SPDK_NVME_SCT_GENERIC;
2471 				response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
2472 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2473 			}
2474 			nvmf_get_discovery_log_page(subsystem->tgt, ctrlr->hostnqn, req->iov, req->iovcnt,
2475 						    offset, len, &cmd_source_trid);
2476 			if (!rae) {
2477 				nvmf_ctrlr_unmask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT);
2478 			}
2479 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2480 		default:
2481 			goto invalid_log_page;
2482 		}
2483 	} else {
2484 		if (offset > len) {
2485 			SPDK_ERRLOG("Get log page: offset (%" PRIu64 ") > len (%" PRIu64 ")\n",
2486 				    offset, len);
2487 			response->status.sct = SPDK_NVME_SCT_GENERIC;
2488 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2489 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2490 		}
2491 
2492 		switch (lid) {
2493 		case SPDK_NVME_LOG_ERROR:
2494 			nvmf_get_error_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2495 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2496 		case SPDK_NVME_LOG_HEALTH_INFORMATION:
2497 			/* TODO: actually fill out log page data */
2498 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2499 		case SPDK_NVME_LOG_FIRMWARE_SLOT:
2500 			nvmf_get_firmware_slot_log_page(req->iov, req->iovcnt, offset, len);
2501 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2502 		case SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS:
2503 			if (subsystem->flags.ana_reporting) {
2504 				nvmf_get_ana_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2505 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2506 			} else {
2507 				goto invalid_log_page;
2508 			}
2509 		case SPDK_NVME_LOG_COMMAND_EFFECTS_LOG:
2510 			nvmf_get_cmds_and_effects_log_page(req->iov, req->iovcnt, offset, len);
2511 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2512 		case SPDK_NVME_LOG_CHANGED_NS_LIST:
2513 			nvmf_get_changed_ns_list_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2514 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2515 		case SPDK_NVME_LOG_RESERVATION_NOTIFICATION:
2516 			nvmf_get_reservation_notification_log_page(ctrlr, req->iov, req->iovcnt, offset, len, rae);
2517 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2518 		default:
2519 			goto invalid_log_page;
2520 		}
2521 	}
2522 
2523 invalid_log_page:
2524 	SPDK_INFOLOG(nvmf, "Unsupported Get Log Page 0x%02X\n", lid);
2525 	response->status.sct = SPDK_NVME_SCT_GENERIC;
2526 	response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2527 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2528 }
2529 
2530 int
2531 spdk_nvmf_ctrlr_identify_ns(struct spdk_nvmf_ctrlr *ctrlr,
2532 			    struct spdk_nvme_cmd *cmd,
2533 			    struct spdk_nvme_cpl *rsp,
2534 			    struct spdk_nvme_ns_data *nsdata)
2535 {
2536 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2537 	struct spdk_nvmf_ns *ns;
2538 	uint32_t max_num_blocks;
2539 	enum spdk_nvme_ana_state ana_state;
2540 
2541 	if (cmd->nsid == 0 || cmd->nsid > subsystem->max_nsid) {
2542 		SPDK_ERRLOG("Identify Namespace for invalid NSID %u\n", cmd->nsid);
2543 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2544 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2545 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2546 	}
2547 
2548 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2549 	if (ns == NULL || ns->bdev == NULL) {
2550 		/*
2551 		 * Inactive namespaces should return a zero filled data structure.
2552 		 * The data buffer is already zeroed by nvmf_ctrlr_process_admin_cmd(),
2553 		 * so we can just return early here.
2554 		 */
2555 		SPDK_DEBUGLOG(nvmf, "Identify Namespace for inactive NSID %u\n", cmd->nsid);
2556 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2557 		rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2558 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2559 	}
2560 
2561 	nvmf_bdev_ctrlr_identify_ns(ns, nsdata, ctrlr->dif_insert_or_strip);
2562 
2563 	assert(ctrlr->admin_qpair);
2564 	/* Due to bug in the Linux kernel NVMe driver we have to set noiob no larger than mdts */
2565 	max_num_blocks = ctrlr->admin_qpair->transport->opts.max_io_size /
2566 			 (1U << nsdata->lbaf[nsdata->flbas.format].lbads);
2567 	if (nsdata->noiob > max_num_blocks) {
2568 		nsdata->noiob = max_num_blocks;
2569 	}
2570 
2571 	/* Set NOWS equal to Controller MDTS */
2572 	if (nsdata->nsfeat.optperf) {
2573 		nsdata->nows = max_num_blocks - 1;
2574 	}
2575 
2576 	if (subsystem->flags.ana_reporting) {
2577 		assert(ns->anagrpid - 1 < subsystem->max_nsid);
2578 		nsdata->anagrpid = ns->anagrpid;
2579 
2580 		ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
2581 		if (ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE ||
2582 		    ana_state == SPDK_NVME_ANA_PERSISTENT_LOSS_STATE) {
2583 			nsdata->nuse = 0;
2584 		}
2585 	}
2586 
2587 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2588 }
2589 
2590 static void
2591 nvmf_ctrlr_populate_oacs(struct spdk_nvmf_ctrlr *ctrlr,
2592 			 struct spdk_nvme_ctrlr_data *cdata)
2593 {
2594 	cdata->oacs = ctrlr->cdata.oacs;
2595 
2596 	cdata->oacs.virtualization_management =
2597 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_VIRTUALIZATION_MANAGEMENT].hdlr != NULL;
2598 	cdata->oacs.nvme_mi = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_SEND].hdlr != NULL
2599 			      && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NVME_MI_RECEIVE].hdlr != NULL;
2600 	cdata->oacs.directives = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_SEND].hdlr != NULL
2601 				 && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DIRECTIVE_RECEIVE].hdlr != NULL;
2602 	cdata->oacs.device_self_test =
2603 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_DEVICE_SELF_TEST].hdlr != NULL;
2604 	cdata->oacs.ns_manage = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_MANAGEMENT].hdlr != NULL
2605 				&& g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_NS_ATTACHMENT].hdlr != NULL;
2606 	cdata->oacs.firmware = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD].hdlr !=
2607 			       NULL
2608 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FIRMWARE_COMMIT].hdlr != NULL;
2609 	cdata->oacs.format =
2610 		g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_FORMAT_NVM].hdlr != NULL;
2611 	cdata->oacs.security = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_SEND].hdlr != NULL
2612 			       && g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_SECURITY_RECEIVE].hdlr != NULL;
2613 	cdata->oacs.get_lba_status = g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_GET_LBA_STATUS].hdlr !=
2614 				     NULL;
2615 }
2616 
2617 int
2618 spdk_nvmf_ctrlr_identify_ctrlr(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_ctrlr_data *cdata)
2619 {
2620 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2621 	struct spdk_nvmf_transport *transport;
2622 
2623 	/*
2624 	 * Common fields for discovery and NVM subsystems
2625 	 */
2626 	assert(ctrlr->admin_qpair);
2627 	transport = ctrlr->admin_qpair->transport;
2628 	spdk_strcpy_pad(cdata->fr, FW_VERSION, sizeof(cdata->fr), ' ');
2629 	assert((transport->opts.max_io_size % 4096) == 0);
2630 	cdata->mdts = spdk_u32log2(transport->opts.max_io_size / 4096);
2631 	cdata->cntlid = ctrlr->cntlid;
2632 	cdata->ver = ctrlr->vcprop.vs;
2633 	cdata->aerl = ctrlr->cdata.aerl;
2634 	cdata->lpa.edlp = 1;
2635 	cdata->elpe = 127;
2636 	cdata->maxcmd = transport->opts.max_queue_depth;
2637 	cdata->sgls = ctrlr->cdata.sgls;
2638 	cdata->fuses.compare_and_write = 1;
2639 	cdata->acwu = 0; /* ACWU is 0-based. */
2640 	if (subsystem->flags.ana_reporting) {
2641 		cdata->mnan = subsystem->max_nsid;
2642 	}
2643 	spdk_strcpy_pad(cdata->subnqn, subsystem->subnqn, sizeof(cdata->subnqn), '\0');
2644 
2645 	SPDK_DEBUGLOG(nvmf, "ctrlr data: maxcmd 0x%x\n", cdata->maxcmd);
2646 	SPDK_DEBUGLOG(nvmf, "sgls data: 0x%x\n", from_le32(&cdata->sgls));
2647 
2648 
2649 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
2650 		/*
2651 		 * NVM Discovery subsystem fields
2652 		 */
2653 		cdata->oaes.discovery_log_change_notices = 1;
2654 	} else {
2655 		cdata->vid = ctrlr->cdata.vid;
2656 		cdata->ssvid = ctrlr->cdata.ssvid;
2657 		cdata->ieee[0] = ctrlr->cdata.ieee[0];
2658 		cdata->ieee[1] = ctrlr->cdata.ieee[1];
2659 		cdata->ieee[2] = ctrlr->cdata.ieee[2];
2660 
2661 		/*
2662 		 * NVM subsystem fields (reserved for discovery subsystems)
2663 		 */
2664 		spdk_strcpy_pad(cdata->mn, spdk_nvmf_subsystem_get_mn(subsystem), sizeof(cdata->mn), ' ');
2665 		spdk_strcpy_pad(cdata->sn, spdk_nvmf_subsystem_get_sn(subsystem), sizeof(cdata->sn), ' ');
2666 		cdata->kas = ctrlr->cdata.kas;
2667 
2668 		cdata->rab = 6;
2669 		cdata->cmic.multi_port = 1;
2670 		cdata->cmic.multi_ctrlr = 1;
2671 		cdata->oaes.ns_attribute_notices = 1;
2672 		cdata->ctratt.host_id_exhid_supported = 1;
2673 		/* We do not have any actual limitation to the number of abort commands.
2674 		 * We follow the recommendation by the NVMe specification.
2675 		 */
2676 		cdata->acl = NVMF_ABORT_COMMAND_LIMIT;
2677 		cdata->frmw.slot1_ro = 1;
2678 		cdata->frmw.num_slots = 1;
2679 
2680 		cdata->lpa.celp = 1; /* Command Effects log page supported */
2681 
2682 		cdata->sqes.min = 6;
2683 		cdata->sqes.max = 6;
2684 		cdata->cqes.min = 4;
2685 		cdata->cqes.max = 4;
2686 		cdata->nn = subsystem->max_nsid;
2687 		cdata->vwc.present = 1;
2688 		cdata->vwc.flush_broadcast = SPDK_NVME_FLUSH_BROADCAST_NOT_SUPPORTED;
2689 
2690 		cdata->nvmf_specific = ctrlr->cdata.nvmf_specific;
2691 
2692 		cdata->oncs.dsm = nvmf_ctrlr_dsm_supported(ctrlr);
2693 		cdata->oncs.write_zeroes = nvmf_ctrlr_write_zeroes_supported(ctrlr);
2694 		cdata->oncs.reservations = ctrlr->cdata.oncs.reservations;
2695 		if (subsystem->flags.ana_reporting) {
2696 			/* Asymmetric Namespace Access Reporting is supported. */
2697 			cdata->cmic.ana_reporting = 1;
2698 			cdata->oaes.ana_change_notices = 1;
2699 
2700 			cdata->anatt = ANA_TRANSITION_TIME_IN_SEC;
2701 			/* ANA Change state is not used, and ANA Persistent Loss state
2702 			 * is not supported for now.
2703 			 */
2704 			cdata->anacap.ana_optimized_state = 1;
2705 			cdata->anacap.ana_non_optimized_state = 1;
2706 			cdata->anacap.ana_inaccessible_state = 1;
2707 			/* ANAGRPID does not change while namespace is attached to controller */
2708 			cdata->anacap.no_change_anagrpid = 1;
2709 			cdata->anagrpmax = subsystem->max_nsid;
2710 			cdata->nanagrpid = subsystem->max_nsid;
2711 		}
2712 
2713 		nvmf_ctrlr_populate_oacs(ctrlr, cdata);
2714 
2715 		assert(subsystem->tgt != NULL);
2716 		cdata->crdt[0] = subsystem->tgt->crdt[0];
2717 		cdata->crdt[1] = subsystem->tgt->crdt[1];
2718 		cdata->crdt[2] = subsystem->tgt->crdt[2];
2719 
2720 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ioccsz 0x%x\n",
2721 			      cdata->nvmf_specific.ioccsz);
2722 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: iorcsz 0x%x\n",
2723 			      cdata->nvmf_specific.iorcsz);
2724 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: icdoff 0x%x\n",
2725 			      cdata->nvmf_specific.icdoff);
2726 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: ctrattr 0x%x\n",
2727 			      *(uint8_t *)&cdata->nvmf_specific.ctrattr);
2728 		SPDK_DEBUGLOG(nvmf, "ext ctrlr data: msdbd 0x%x\n",
2729 			      cdata->nvmf_specific.msdbd);
2730 	}
2731 
2732 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2733 }
2734 
2735 static int
2736 nvmf_ctrlr_identify_active_ns_list(struct spdk_nvmf_subsystem *subsystem,
2737 				   struct spdk_nvme_cmd *cmd,
2738 				   struct spdk_nvme_cpl *rsp,
2739 				   struct spdk_nvme_ns_list *ns_list)
2740 {
2741 	struct spdk_nvmf_ns *ns;
2742 	uint32_t count = 0;
2743 
2744 	if (cmd->nsid >= 0xfffffffeUL) {
2745 		SPDK_ERRLOG("Identify Active Namespace List with invalid NSID %u\n", cmd->nsid);
2746 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2747 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2748 	}
2749 
2750 	memset(ns_list, 0, sizeof(*ns_list));
2751 
2752 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
2753 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2754 		if (ns->opts.nsid <= cmd->nsid) {
2755 			continue;
2756 		}
2757 
2758 		ns_list->ns_list[count++] = ns->opts.nsid;
2759 		if (count == SPDK_COUNTOF(ns_list->ns_list)) {
2760 			break;
2761 		}
2762 	}
2763 
2764 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2765 }
2766 
2767 static void
2768 _add_ns_id_desc(void **buf_ptr, size_t *buf_remain,
2769 		enum spdk_nvme_nidt type,
2770 		const void *data, size_t data_size)
2771 {
2772 	struct spdk_nvme_ns_id_desc *desc;
2773 	size_t desc_size = sizeof(*desc) + data_size;
2774 
2775 	/*
2776 	 * These should never fail in practice, since all valid NS ID descriptors
2777 	 * should be defined so that they fit in the available 4096-byte buffer.
2778 	 */
2779 	assert(data_size > 0);
2780 	assert(data_size <= UINT8_MAX);
2781 	assert(desc_size < *buf_remain);
2782 	if (data_size == 0 || data_size > UINT8_MAX || desc_size > *buf_remain) {
2783 		return;
2784 	}
2785 
2786 	desc = *buf_ptr;
2787 	desc->nidt = type;
2788 	desc->nidl = data_size;
2789 	memcpy(desc->nid, data, data_size);
2790 
2791 	*buf_ptr += desc_size;
2792 	*buf_remain -= desc_size;
2793 }
2794 
2795 static int
2796 nvmf_ctrlr_identify_ns_id_descriptor_list(
2797 	struct spdk_nvmf_subsystem *subsystem,
2798 	struct spdk_nvme_cmd *cmd,
2799 	struct spdk_nvme_cpl *rsp,
2800 	void *id_desc_list, size_t id_desc_list_size)
2801 {
2802 	struct spdk_nvmf_ns *ns;
2803 	size_t buf_remain = id_desc_list_size;
2804 	void *buf_ptr = id_desc_list;
2805 
2806 	ns = _nvmf_subsystem_get_ns(subsystem, cmd->nsid);
2807 	if (ns == NULL || ns->bdev == NULL) {
2808 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2809 		rsp->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
2810 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2811 	}
2812 
2813 #define ADD_ID_DESC(type, data, size) \
2814 	do { \
2815 		if (!spdk_mem_all_zero(data, size)) { \
2816 			_add_ns_id_desc(&buf_ptr, &buf_remain, type, data, size); \
2817 		} \
2818 	} while (0)
2819 
2820 	ADD_ID_DESC(SPDK_NVME_NIDT_EUI64, ns->opts.eui64, sizeof(ns->opts.eui64));
2821 	ADD_ID_DESC(SPDK_NVME_NIDT_NGUID, ns->opts.nguid, sizeof(ns->opts.nguid));
2822 	ADD_ID_DESC(SPDK_NVME_NIDT_UUID, &ns->opts.uuid, sizeof(ns->opts.uuid));
2823 
2824 	/*
2825 	 * The list is automatically 0-terminated because controller to host buffers in
2826 	 * admin commands always get zeroed in nvmf_ctrlr_process_admin_cmd().
2827 	 */
2828 
2829 #undef ADD_ID_DESC
2830 
2831 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2832 }
2833 
2834 static int
2835 nvmf_ctrlr_identify(struct spdk_nvmf_request *req)
2836 {
2837 	uint8_t cns;
2838 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
2839 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
2840 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2841 	struct spdk_nvmf_subsystem *subsystem = ctrlr->subsys;
2842 
2843 	if (req->data == NULL || req->length < 4096) {
2844 		SPDK_DEBUGLOG(nvmf, "identify command with invalid buffer\n");
2845 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2846 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2847 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2848 	}
2849 
2850 	cns = cmd->cdw10_bits.identify.cns;
2851 
2852 	if (subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY &&
2853 	    cns != SPDK_NVME_IDENTIFY_CTRLR) {
2854 		/* Discovery controllers only support Identify Controller */
2855 		goto invalid_cns;
2856 	}
2857 
2858 	switch (cns) {
2859 	case SPDK_NVME_IDENTIFY_NS:
2860 		return spdk_nvmf_ctrlr_identify_ns(ctrlr, cmd, rsp, req->data);
2861 	case SPDK_NVME_IDENTIFY_CTRLR:
2862 		return spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, req->data);
2863 	case SPDK_NVME_IDENTIFY_ACTIVE_NS_LIST:
2864 		return nvmf_ctrlr_identify_active_ns_list(subsystem, cmd, rsp, req->data);
2865 	case SPDK_NVME_IDENTIFY_NS_ID_DESCRIPTOR_LIST:
2866 		return nvmf_ctrlr_identify_ns_id_descriptor_list(subsystem, cmd, rsp, req->data, req->length);
2867 	default:
2868 		goto invalid_cns;
2869 	}
2870 
2871 invalid_cns:
2872 	SPDK_INFOLOG(nvmf, "Identify command with unsupported CNS 0x%02x\n", cns);
2873 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2874 	rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
2875 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
2876 }
2877 
2878 static bool
2879 nvmf_qpair_abort_aer(struct spdk_nvmf_qpair *qpair, uint16_t cid)
2880 {
2881 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
2882 	struct spdk_nvmf_request *req;
2883 	int i;
2884 
2885 	if (!nvmf_qpair_is_admin_queue(qpair)) {
2886 		return false;
2887 	}
2888 
2889 	assert(spdk_get_thread() == ctrlr->thread);
2890 
2891 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
2892 		if (ctrlr->aer_req[i]->cmd->nvme_cmd.cid == cid) {
2893 			SPDK_DEBUGLOG(nvmf, "Aborting AER request\n");
2894 			req = ctrlr->aer_req[i];
2895 			ctrlr->aer_req[i] = NULL;
2896 			ctrlr->nr_aer_reqs--;
2897 
2898 			/* Move the last req to the aborting position for making aer_reqs
2899 			 * in continuous
2900 			 */
2901 			if (i < ctrlr->nr_aer_reqs) {
2902 				ctrlr->aer_req[i] = ctrlr->aer_req[ctrlr->nr_aer_reqs];
2903 				ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
2904 			}
2905 
2906 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2907 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2908 			_nvmf_request_complete(req);
2909 			return true;
2910 		}
2911 	}
2912 
2913 	return false;
2914 }
2915 
2916 void
2917 nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair)
2918 {
2919 	struct spdk_nvmf_request *req, *tmp;
2920 
2921 	TAILQ_FOREACH_SAFE(req, &qpair->outstanding, link, tmp) {
2922 		if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) {
2923 			/* Zero-copy requests are kept on the outstanding queue from the moment
2924 			 * zcopy_start is sent until a zcopy_end callback is received.  Therefore,
2925 			 * we can't remove them from the outstanding queue here, but need to rely on
2926 			 * the transport to do a zcopy_end to release their buffers and, in turn,
2927 			 * remove them from the queue.
2928 			 */
2929 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
2930 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
2931 			nvmf_transport_req_free(req);
2932 		}
2933 	}
2934 }
2935 
2936 static void
2937 nvmf_qpair_abort_request(struct spdk_nvmf_qpair *qpair, struct spdk_nvmf_request *req)
2938 {
2939 	uint16_t cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid;
2940 
2941 	if (nvmf_qpair_abort_aer(qpair, cid)) {
2942 		SPDK_DEBUGLOG(nvmf, "abort ctrlr=%p sqid=%u cid=%u successful\n",
2943 			      qpair->ctrlr, qpair->qid, cid);
2944 		req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command successfully aborted */
2945 
2946 		spdk_nvmf_request_complete(req);
2947 		return;
2948 	}
2949 
2950 	nvmf_transport_qpair_abort_request(qpair, req);
2951 }
2952 
2953 static void
2954 nvmf_ctrlr_abort_done(struct spdk_io_channel_iter *i, int status)
2955 {
2956 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2957 
2958 	if (status == 0) {
2959 		/* There was no qpair whose ID matches SQID of the abort command.
2960 		 * Hence call _nvmf_request_complete() here.
2961 		 */
2962 		_nvmf_request_complete(req);
2963 	}
2964 }
2965 
2966 static void
2967 nvmf_ctrlr_abort_on_pg(struct spdk_io_channel_iter *i)
2968 {
2969 	struct spdk_nvmf_request *req = spdk_io_channel_iter_get_ctx(i);
2970 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
2971 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
2972 	uint16_t sqid = req->cmd->nvme_cmd.cdw10_bits.abort.sqid;
2973 	struct spdk_nvmf_qpair *qpair;
2974 
2975 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
2976 		if (qpair->ctrlr == req->qpair->ctrlr && qpair->qid == sqid) {
2977 			/* Found the qpair */
2978 
2979 			nvmf_qpair_abort_request(qpair, req);
2980 
2981 			/* Return -1 for the status so the iteration across threads stops. */
2982 			spdk_for_each_channel_continue(i, -1);
2983 			return;
2984 		}
2985 	}
2986 
2987 	spdk_for_each_channel_continue(i, 0);
2988 }
2989 
2990 static int
2991 nvmf_ctrlr_abort(struct spdk_nvmf_request *req)
2992 {
2993 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
2994 
2995 	rsp->cdw0 = 1U; /* Command not aborted */
2996 	rsp->status.sct = SPDK_NVME_SCT_GENERIC;
2997 	rsp->status.sc = SPDK_NVME_SC_SUCCESS;
2998 
2999 	/* Send a message to each poll group, searching for this ctrlr, sqid, and command. */
3000 	spdk_for_each_channel(req->qpair->ctrlr->subsys->tgt,
3001 			      nvmf_ctrlr_abort_on_pg,
3002 			      req,
3003 			      nvmf_ctrlr_abort_done
3004 			     );
3005 
3006 	return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3007 }
3008 
3009 int
3010 nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req)
3011 {
3012 	struct spdk_nvmf_request *req_to_abort = req->req_to_abort;
3013 	struct spdk_bdev *bdev;
3014 	struct spdk_bdev_desc *desc;
3015 	struct spdk_io_channel *ch;
3016 	int rc;
3017 
3018 	assert(req_to_abort != NULL);
3019 
3020 	if (g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr &&
3021 	    nvmf_qpair_is_admin_queue(req_to_abort->qpair)) {
3022 		return g_nvmf_custom_admin_cmd_hdlrs[SPDK_NVME_OPC_ABORT].hdlr(req);
3023 	}
3024 
3025 	rc = spdk_nvmf_request_get_bdev(req_to_abort->cmd->nvme_cmd.nsid, req_to_abort,
3026 					&bdev, &desc, &ch);
3027 	if (rc != 0) {
3028 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3029 	}
3030 
3031 	return spdk_nvmf_bdev_ctrlr_abort_cmd(bdev, desc, ch, req, req_to_abort);
3032 }
3033 
3034 static int
3035 get_features_generic(struct spdk_nvmf_request *req, uint32_t cdw0)
3036 {
3037 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3038 
3039 	rsp->cdw0 = cdw0;
3040 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3041 }
3042 
3043 /* we have to use the typedef in the function declaration to appease astyle. */
3044 typedef enum spdk_nvme_path_status_code spdk_nvme_path_status_code_t;
3045 
3046 static spdk_nvme_path_status_code_t
3047 _nvme_ana_state_to_path_status(enum spdk_nvme_ana_state ana_state)
3048 {
3049 	switch (ana_state) {
3050 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3051 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE;
3052 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3053 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3054 	case SPDK_NVME_ANA_CHANGE_STATE:
3055 		return SPDK_NVME_SC_ASYMMETRIC_ACCESS_TRANSITION;
3056 	default:
3057 		return SPDK_NVME_SC_INTERNAL_PATH_ERROR;
3058 	}
3059 }
3060 
3061 static int
3062 nvmf_ctrlr_get_features(struct spdk_nvmf_request *req)
3063 {
3064 	uint8_t feature;
3065 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3066 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3067 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3068 	enum spdk_nvme_ana_state ana_state;
3069 
3070 	feature = cmd->cdw10_bits.get_features.fid;
3071 
3072 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
3073 		/*
3074 		 * Features supported by Discovery controller
3075 		 */
3076 		switch (feature) {
3077 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3078 			return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3079 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3080 			return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3081 		default:
3082 			SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3083 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3084 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3085 		}
3086 	}
3087 	/*
3088 	 * Process Get Features command for non-discovery controller
3089 	 */
3090 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3091 	switch (ana_state) {
3092 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3093 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3094 	case SPDK_NVME_ANA_CHANGE_STATE:
3095 		switch (feature) {
3096 		case SPDK_NVME_FEAT_ERROR_RECOVERY:
3097 		case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3098 		case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3099 		case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3100 			response->status.sct = SPDK_NVME_SCT_PATH;
3101 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3102 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3103 		default:
3104 			break;
3105 		}
3106 		break;
3107 	default:
3108 		break;
3109 	}
3110 
3111 	switch (feature) {
3112 	case SPDK_NVME_FEAT_ARBITRATION:
3113 		return get_features_generic(req, ctrlr->feat.arbitration.raw);
3114 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3115 		return get_features_generic(req, ctrlr->feat.power_management.raw);
3116 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3117 		return nvmf_ctrlr_get_features_temperature_threshold(req);
3118 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3119 		return get_features_generic(req, ctrlr->feat.error_recovery.raw);
3120 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3121 		return get_features_generic(req, ctrlr->feat.volatile_write_cache.raw);
3122 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3123 		return get_features_generic(req, ctrlr->feat.number_of_queues.raw);
3124 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3125 		return get_features_generic(req, ctrlr->feat.interrupt_coalescing.raw);
3126 	case SPDK_NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION:
3127 		return nvmf_ctrlr_get_features_interrupt_vector_configuration(req);
3128 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3129 		return get_features_generic(req, ctrlr->feat.write_atomicity.raw);
3130 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3131 		return get_features_generic(req, ctrlr->feat.async_event_configuration.raw);
3132 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3133 		return get_features_generic(req, ctrlr->feat.keep_alive_timer.raw);
3134 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3135 		return nvmf_ctrlr_get_features_host_identifier(req);
3136 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3137 		return nvmf_ctrlr_get_features_reservation_notification_mask(req);
3138 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3139 		return nvmf_ctrlr_get_features_reservation_persistence(req);
3140 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3141 		return nvmf_ctrlr_get_features_host_behavior_support(req);
3142 	default:
3143 		SPDK_INFOLOG(nvmf, "Get Features command with unsupported feature ID 0x%02x\n", feature);
3144 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3145 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3146 	}
3147 }
3148 
3149 static int
3150 nvmf_ctrlr_set_features(struct spdk_nvmf_request *req)
3151 {
3152 	uint8_t feature, save;
3153 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3154 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3155 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3156 	enum spdk_nvme_ana_state ana_state;
3157 	/*
3158 	 * Features are not saveable by the controller as indicated by
3159 	 * ONCS field of the Identify Controller data.
3160 	 * */
3161 	save = cmd->cdw10_bits.set_features.sv;
3162 	if (save) {
3163 		response->status.sc = SPDK_NVME_SC_FEATURE_ID_NOT_SAVEABLE;
3164 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3165 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3166 	}
3167 
3168 	feature = cmd->cdw10_bits.set_features.fid;
3169 
3170 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
3171 		/*
3172 		 * Features supported by Discovery controller
3173 		 */
3174 		switch (feature) {
3175 		case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3176 			return nvmf_ctrlr_set_features_keep_alive_timer(req);
3177 		case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3178 			return nvmf_ctrlr_set_features_async_event_configuration(req);
3179 		default:
3180 			SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3181 			response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3182 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3183 		}
3184 	}
3185 	/*
3186 	 * Process Set Features command for non-discovery controller
3187 	 */
3188 	ana_state = nvmf_ctrlr_get_ana_state_from_nsid(ctrlr, cmd->nsid);
3189 	switch (ana_state) {
3190 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3191 	case SPDK_NVME_ANA_CHANGE_STATE:
3192 		if (cmd->nsid == SPDK_NVME_GLOBAL_NS_TAG) {
3193 			response->status.sct = SPDK_NVME_SCT_PATH;
3194 			response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3195 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3196 		} else {
3197 			switch (feature) {
3198 			case SPDK_NVME_FEAT_ERROR_RECOVERY:
3199 			case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3200 			case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3201 			case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3202 				response->status.sct = SPDK_NVME_SCT_PATH;
3203 				response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3204 				return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3205 			default:
3206 				break;
3207 			}
3208 		}
3209 		break;
3210 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3211 		response->status.sct = SPDK_NVME_SCT_PATH;
3212 		response->status.sc = SPDK_NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS;
3213 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3214 	default:
3215 		break;
3216 	}
3217 
3218 	switch (feature) {
3219 	case SPDK_NVME_FEAT_ARBITRATION:
3220 		return nvmf_ctrlr_set_features_arbitration(req);
3221 	case SPDK_NVME_FEAT_POWER_MANAGEMENT:
3222 		return nvmf_ctrlr_set_features_power_management(req);
3223 	case SPDK_NVME_FEAT_TEMPERATURE_THRESHOLD:
3224 		return nvmf_ctrlr_set_features_temperature_threshold(req);
3225 	case SPDK_NVME_FEAT_ERROR_RECOVERY:
3226 		return nvmf_ctrlr_set_features_error_recovery(req);
3227 	case SPDK_NVME_FEAT_VOLATILE_WRITE_CACHE:
3228 		return nvmf_ctrlr_set_features_volatile_write_cache(req);
3229 	case SPDK_NVME_FEAT_NUMBER_OF_QUEUES:
3230 		return nvmf_ctrlr_set_features_number_of_queues(req);
3231 	case SPDK_NVME_FEAT_INTERRUPT_COALESCING:
3232 		response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC;
3233 		response->status.sc = SPDK_NVME_SC_FEATURE_NOT_CHANGEABLE;
3234 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3235 	case SPDK_NVME_FEAT_WRITE_ATOMICITY:
3236 		return nvmf_ctrlr_set_features_write_atomicity(req);
3237 	case SPDK_NVME_FEAT_ASYNC_EVENT_CONFIGURATION:
3238 		return nvmf_ctrlr_set_features_async_event_configuration(req);
3239 	case SPDK_NVME_FEAT_KEEP_ALIVE_TIMER:
3240 		return nvmf_ctrlr_set_features_keep_alive_timer(req);
3241 	case SPDK_NVME_FEAT_HOST_IDENTIFIER:
3242 		return nvmf_ctrlr_set_features_host_identifier(req);
3243 	case SPDK_NVME_FEAT_HOST_RESERVE_MASK:
3244 		return nvmf_ctrlr_set_features_reservation_notification_mask(req);
3245 	case SPDK_NVME_FEAT_HOST_RESERVE_PERSIST:
3246 		return nvmf_ctrlr_set_features_reservation_persistence(req);
3247 	case SPDK_NVME_FEAT_HOST_BEHAVIOR_SUPPORT:
3248 		return nvmf_ctrlr_set_features_host_behavior_support(req);
3249 	default:
3250 		SPDK_INFOLOG(nvmf, "Set Features command with unsupported feature ID 0x%02x\n", feature);
3251 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3252 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3253 	}
3254 }
3255 
3256 static int
3257 nvmf_ctrlr_keep_alive(struct spdk_nvmf_request *req)
3258 {
3259 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3260 
3261 	SPDK_DEBUGLOG(nvmf, "Keep Alive\n");
3262 	/*
3263 	 * To handle keep alive just clear or reset the
3264 	 * ctrlr based keep alive duration counter.
3265 	 * When added, a separate timer based process
3266 	 * will monitor if the time since last recorded
3267 	 * keep alive has exceeded the max duration and
3268 	 * take appropriate action.
3269 	 */
3270 	ctrlr->last_keep_alive_tick = spdk_get_ticks();
3271 
3272 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3273 }
3274 
3275 int
3276 nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req)
3277 {
3278 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3279 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3280 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3281 	int rc;
3282 
3283 	if (ctrlr == NULL) {
3284 		SPDK_ERRLOG("Admin command sent before CONNECT\n");
3285 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3286 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3287 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3288 	}
3289 
3290 	assert(spdk_get_thread() == ctrlr->thread);
3291 
3292 	if (cmd->fuse != 0) {
3293 		/* Fused admin commands are not supported. */
3294 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3295 		response->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3296 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3297 	}
3298 
3299 	if (ctrlr->vcprop.cc.bits.en != 1) {
3300 		SPDK_ERRLOG("Admin command sent to disabled controller\n");
3301 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3302 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3303 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3304 	}
3305 
3306 	if (req->data && spdk_nvme_opc_get_data_transfer(cmd->opc) == SPDK_NVME_DATA_CONTROLLER_TO_HOST) {
3307 		_clear_iovs(req->iov, req->iovcnt);
3308 	}
3309 
3310 	if (ctrlr->subsys->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
3311 		/* Discovery controllers only support these admin OPS. */
3312 		switch (cmd->opc) {
3313 		case SPDK_NVME_OPC_IDENTIFY:
3314 		case SPDK_NVME_OPC_GET_LOG_PAGE:
3315 		case SPDK_NVME_OPC_KEEP_ALIVE:
3316 		case SPDK_NVME_OPC_SET_FEATURES:
3317 		case SPDK_NVME_OPC_GET_FEATURES:
3318 		case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3319 			break;
3320 		default:
3321 			goto invalid_opcode;
3322 		}
3323 	}
3324 
3325 	/* Call a custom adm cmd handler if set. Aborts are handled in a different path (see nvmf_passthru_admin_cmd) */
3326 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr && cmd->opc != SPDK_NVME_OPC_ABORT) {
3327 		rc = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].hdlr(req);
3328 		if (rc >= SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3329 			/* The handler took care of this command */
3330 			return rc;
3331 		}
3332 	}
3333 
3334 	switch (cmd->opc) {
3335 	case SPDK_NVME_OPC_GET_LOG_PAGE:
3336 		return nvmf_ctrlr_get_log_page(req);
3337 	case SPDK_NVME_OPC_IDENTIFY:
3338 		return nvmf_ctrlr_identify(req);
3339 	case SPDK_NVME_OPC_ABORT:
3340 		return nvmf_ctrlr_abort(req);
3341 	case SPDK_NVME_OPC_GET_FEATURES:
3342 		return nvmf_ctrlr_get_features(req);
3343 	case SPDK_NVME_OPC_SET_FEATURES:
3344 		return nvmf_ctrlr_set_features(req);
3345 	case SPDK_NVME_OPC_ASYNC_EVENT_REQUEST:
3346 		return nvmf_ctrlr_async_event_request(req);
3347 	case SPDK_NVME_OPC_KEEP_ALIVE:
3348 		return nvmf_ctrlr_keep_alive(req);
3349 
3350 	case SPDK_NVME_OPC_CREATE_IO_SQ:
3351 	case SPDK_NVME_OPC_CREATE_IO_CQ:
3352 	case SPDK_NVME_OPC_DELETE_IO_SQ:
3353 	case SPDK_NVME_OPC_DELETE_IO_CQ:
3354 		/* Create and Delete I/O CQ/SQ not allowed in NVMe-oF */
3355 		goto invalid_opcode;
3356 
3357 	default:
3358 		goto invalid_opcode;
3359 	}
3360 
3361 invalid_opcode:
3362 	SPDK_INFOLOG(nvmf, "Unsupported admin opcode 0x%x\n", cmd->opc);
3363 	response->status.sct = SPDK_NVME_SCT_GENERIC;
3364 	response->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3365 	return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3366 }
3367 
3368 static int
3369 nvmf_ctrlr_process_fabrics_cmd(struct spdk_nvmf_request *req)
3370 {
3371 	struct spdk_nvmf_qpair *qpair = req->qpair;
3372 	struct spdk_nvmf_capsule_cmd *cap_hdr;
3373 
3374 	cap_hdr = &req->cmd->nvmf_cmd;
3375 
3376 	if (qpair->ctrlr == NULL) {
3377 		/* No ctrlr established yet; the only valid command is Connect */
3378 		if (cap_hdr->fctype == SPDK_NVMF_FABRIC_COMMAND_CONNECT) {
3379 			return nvmf_ctrlr_cmd_connect(req);
3380 		} else {
3381 			SPDK_DEBUGLOG(nvmf, "Got fctype 0x%x, expected Connect\n",
3382 				      cap_hdr->fctype);
3383 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3384 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3385 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3386 		}
3387 	} else if (nvmf_qpair_is_admin_queue(qpair)) {
3388 		/*
3389 		 * Controller session is established, and this is an admin queue.
3390 		 * Disallow Connect and allow other fabrics commands.
3391 		 */
3392 		switch (cap_hdr->fctype) {
3393 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_SET:
3394 			return nvmf_property_set(req);
3395 		case SPDK_NVMF_FABRIC_COMMAND_PROPERTY_GET:
3396 			return nvmf_property_get(req);
3397 		default:
3398 			SPDK_DEBUGLOG(nvmf, "unknown fctype 0x%02x\n",
3399 				      cap_hdr->fctype);
3400 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3401 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3402 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3403 		}
3404 	} else {
3405 		/* Controller session is established, and this is an I/O queue */
3406 		/* For now, no I/O-specific Fabrics commands are implemented (other than Connect) */
3407 		SPDK_DEBUGLOG(nvmf, "Unexpected I/O fctype 0x%x\n", cap_hdr->fctype);
3408 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3409 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3410 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3411 	}
3412 }
3413 
3414 static inline void
3415 nvmf_ctrlr_queue_pending_async_event(struct spdk_nvmf_ctrlr *ctrlr,
3416 				     union spdk_nvme_async_event_completion *event)
3417 {
3418 	struct spdk_nvmf_async_event_completion *nvmf_event;
3419 
3420 	nvmf_event = calloc(1, sizeof(*nvmf_event));
3421 	if (!nvmf_event) {
3422 		SPDK_ERRLOG("Alloc nvmf event failed, ignore the event\n");
3423 		return;
3424 	}
3425 	nvmf_event->event.raw = event->raw;
3426 	STAILQ_INSERT_TAIL(&ctrlr->async_events, nvmf_event, link);
3427 }
3428 
3429 static inline int
3430 nvmf_ctrlr_async_event_notification(struct spdk_nvmf_ctrlr *ctrlr,
3431 				    union spdk_nvme_async_event_completion *event)
3432 {
3433 	struct spdk_nvmf_request *req;
3434 	struct spdk_nvme_cpl *rsp;
3435 
3436 	assert(spdk_get_thread() == ctrlr->thread);
3437 
3438 	/* If there is no outstanding AER request, queue the event.  Then
3439 	 * if an AER is later submitted, this event can be sent as a
3440 	 * response.
3441 	 */
3442 	if (ctrlr->nr_aer_reqs == 0) {
3443 		nvmf_ctrlr_queue_pending_async_event(ctrlr, event);
3444 		return 0;
3445 	}
3446 
3447 	req = ctrlr->aer_req[--ctrlr->nr_aer_reqs];
3448 	rsp = &req->rsp->nvme_cpl;
3449 
3450 	rsp->cdw0 = event->raw;
3451 
3452 	_nvmf_request_complete(req);
3453 	ctrlr->aer_req[ctrlr->nr_aer_reqs] = NULL;
3454 
3455 	return 0;
3456 }
3457 
3458 int
3459 nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
3460 {
3461 	union spdk_nvme_async_event_completion event = {0};
3462 
3463 	/* Users may disable the event notification */
3464 	if (!ctrlr->feat.async_event_configuration.bits.ns_attr_notice) {
3465 		return 0;
3466 	}
3467 
3468 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGE_MASK_BIT)) {
3469 		return 0;
3470 	}
3471 
3472 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3473 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED;
3474 	event.bits.log_page_identifier = SPDK_NVME_LOG_CHANGED_NS_LIST;
3475 
3476 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3477 }
3478 
3479 int
3480 nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr)
3481 {
3482 	union spdk_nvme_async_event_completion event = {0};
3483 
3484 	/* Users may disable the event notification */
3485 	if (!ctrlr->feat.async_event_configuration.bits.ana_change_notice) {
3486 		return 0;
3487 	}
3488 
3489 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ANA_CHANGE_MASK_BIT)) {
3490 		return 0;
3491 	}
3492 
3493 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3494 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_ANA_CHANGE;
3495 	event.bits.log_page_identifier = SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS;
3496 
3497 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3498 }
3499 
3500 void
3501 nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr)
3502 {
3503 	union spdk_nvme_async_event_completion event = {0};
3504 
3505 	if (!ctrlr->num_avail_log_pages) {
3506 		return;
3507 	}
3508 
3509 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL_MASK_BIT)) {
3510 		return;
3511 	}
3512 
3513 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_IO;
3514 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_RESERVATION_LOG_AVAIL;
3515 	event.bits.log_page_identifier = SPDK_NVME_LOG_RESERVATION_NOTIFICATION;
3516 
3517 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3518 }
3519 
3520 void
3521 nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx)
3522 {
3523 	union spdk_nvme_async_event_completion event = {0};
3524 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
3525 
3526 	/* Users may disable the event notification manually or
3527 	 * it may not be enabled due to keep alive timeout
3528 	 * not being set in connect command to discovery controller.
3529 	 */
3530 	if (!ctrlr->feat.async_event_configuration.bits.discovery_log_change_notice) {
3531 		return;
3532 	}
3533 
3534 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE_MASK_BIT)) {
3535 		return;
3536 	}
3537 
3538 	event.bits.async_event_type = SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE;
3539 	event.bits.async_event_info = SPDK_NVME_ASYNC_EVENT_DISCOVERY_LOG_CHANGE;
3540 	event.bits.log_page_identifier = SPDK_NVME_LOG_DISCOVERY;
3541 
3542 	nvmf_ctrlr_async_event_notification(ctrlr, &event);
3543 }
3544 
3545 int
3546 nvmf_ctrlr_async_event_error_event(struct spdk_nvmf_ctrlr *ctrlr,
3547 				   union spdk_nvme_async_event_completion event)
3548 {
3549 	if (!nvmf_ctrlr_mask_aen(ctrlr, SPDK_NVME_ASYNC_EVENT_ERROR_MASK_BIT)) {
3550 		return 0;
3551 	}
3552 
3553 	if (event.bits.async_event_type != SPDK_NVME_ASYNC_EVENT_TYPE_ERROR ||
3554 	    event.bits.async_event_info > SPDK_NVME_ASYNC_EVENT_FW_IMAGE_LOAD) {
3555 		return 0;
3556 	}
3557 
3558 	return nvmf_ctrlr_async_event_notification(ctrlr, &event);
3559 }
3560 
3561 void
3562 nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair)
3563 {
3564 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
3565 	int i;
3566 
3567 	if (!nvmf_qpair_is_admin_queue(qpair)) {
3568 		return;
3569 	}
3570 
3571 	assert(spdk_get_thread() == ctrlr->thread);
3572 
3573 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3574 		spdk_nvmf_request_free(ctrlr->aer_req[i]);
3575 		ctrlr->aer_req[i] = NULL;
3576 	}
3577 
3578 	ctrlr->nr_aer_reqs = 0;
3579 }
3580 
3581 void
3582 nvmf_ctrlr_abort_aer(struct spdk_nvmf_ctrlr *ctrlr)
3583 {
3584 	struct spdk_nvmf_request *req;
3585 	int i;
3586 
3587 	assert(spdk_get_thread() == ctrlr->thread);
3588 
3589 	if (!ctrlr->nr_aer_reqs) {
3590 		return;
3591 	}
3592 
3593 	for (i = 0; i < ctrlr->nr_aer_reqs; i++) {
3594 		req = ctrlr->aer_req[i];
3595 
3596 		req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3597 		req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
3598 		_nvmf_request_complete(req);
3599 
3600 		ctrlr->aer_req[i] = NULL;
3601 	}
3602 
3603 	ctrlr->nr_aer_reqs = 0;
3604 }
3605 
3606 static void
3607 _nvmf_ctrlr_add_reservation_log(void *ctx)
3608 {
3609 	struct spdk_nvmf_reservation_log *log = (struct spdk_nvmf_reservation_log *)ctx;
3610 	struct spdk_nvmf_ctrlr *ctrlr = log->ctrlr;
3611 
3612 	ctrlr->log_page_count++;
3613 
3614 	/* Maximum number of queued log pages is 255 */
3615 	if (ctrlr->num_avail_log_pages == 0xff) {
3616 		struct spdk_nvmf_reservation_log *entry;
3617 		entry = TAILQ_LAST(&ctrlr->log_head, log_page_head);
3618 		entry->log.log_page_count = ctrlr->log_page_count;
3619 		free(log);
3620 		return;
3621 	}
3622 
3623 	log->log.log_page_count = ctrlr->log_page_count;
3624 	log->log.num_avail_log_pages = ctrlr->num_avail_log_pages++;
3625 	TAILQ_INSERT_TAIL(&ctrlr->log_head, log, link);
3626 
3627 	nvmf_ctrlr_async_event_reservation_notification(ctrlr);
3628 }
3629 
3630 void
3631 nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr,
3632 				  struct spdk_nvmf_ns *ns,
3633 				  enum spdk_nvme_reservation_notification_log_page_type type)
3634 {
3635 	struct spdk_nvmf_reservation_log *log;
3636 
3637 	switch (type) {
3638 	case SPDK_NVME_RESERVATION_LOG_PAGE_EMPTY:
3639 		return;
3640 	case SPDK_NVME_REGISTRATION_PREEMPTED:
3641 		if (ns->mask & SPDK_NVME_REGISTRATION_PREEMPTED_MASK) {
3642 			return;
3643 		}
3644 		break;
3645 	case SPDK_NVME_RESERVATION_RELEASED:
3646 		if (ns->mask & SPDK_NVME_RESERVATION_RELEASED_MASK) {
3647 			return;
3648 		}
3649 		break;
3650 	case SPDK_NVME_RESERVATION_PREEMPTED:
3651 		if (ns->mask & SPDK_NVME_RESERVATION_PREEMPTED_MASK) {
3652 			return;
3653 		}
3654 		break;
3655 	default:
3656 		return;
3657 	}
3658 
3659 	log = calloc(1, sizeof(*log));
3660 	if (!log) {
3661 		SPDK_ERRLOG("Alloc log page failed, ignore the log\n");
3662 		return;
3663 	}
3664 	log->ctrlr = ctrlr;
3665 	log->log.type = type;
3666 	log->log.nsid = ns->nsid;
3667 
3668 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_add_reservation_log, log);
3669 }
3670 
3671 /* Check from subsystem poll group's namespace information data structure */
3672 static bool
3673 nvmf_ns_info_ctrlr_is_registrant(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3674 				 struct spdk_nvmf_ctrlr *ctrlr)
3675 {
3676 	uint32_t i;
3677 
3678 	for (i = 0; i < SPDK_NVMF_MAX_NUM_REGISTRANTS; i++) {
3679 		if (!spdk_uuid_compare(&ns_info->reg_hostid[i], &ctrlr->hostid)) {
3680 			return true;
3681 		}
3682 	}
3683 
3684 	return false;
3685 }
3686 
3687 /*
3688  * Check the NVMe command is permitted or not for current controller(Host).
3689  */
3690 static int
3691 nvmf_ns_reservation_request_check(struct spdk_nvmf_subsystem_pg_ns_info *ns_info,
3692 				  struct spdk_nvmf_ctrlr *ctrlr,
3693 				  struct spdk_nvmf_request *req)
3694 {
3695 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3696 	enum spdk_nvme_reservation_type rtype = ns_info->rtype;
3697 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3698 	uint8_t racqa;
3699 	bool is_registrant;
3700 
3701 	/* No valid reservation */
3702 	if (!rtype) {
3703 		return 0;
3704 	}
3705 
3706 	is_registrant = nvmf_ns_info_ctrlr_is_registrant(ns_info, ctrlr);
3707 	/* All registrants type and current ctrlr is a valid registrant */
3708 	if ((rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3709 	     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && is_registrant) {
3710 		return 0;
3711 	} else if (!spdk_uuid_compare(&ns_info->holder_id, &ctrlr->hostid)) {
3712 		return 0;
3713 	}
3714 
3715 	/* Non-holder for current controller */
3716 	switch (cmd->opc) {
3717 	case SPDK_NVME_OPC_READ:
3718 	case SPDK_NVME_OPC_COMPARE:
3719 		if (rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3720 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3721 			goto exit;
3722 		}
3723 		if ((rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY ||
3724 		     rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) && !is_registrant) {
3725 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3726 		}
3727 		break;
3728 	case SPDK_NVME_OPC_FLUSH:
3729 	case SPDK_NVME_OPC_WRITE:
3730 	case SPDK_NVME_OPC_WRITE_UNCORRECTABLE:
3731 	case SPDK_NVME_OPC_WRITE_ZEROES:
3732 	case SPDK_NVME_OPC_DATASET_MANAGEMENT:
3733 		if (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE ||
3734 		    rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3735 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3736 			goto exit;
3737 		}
3738 		if (!is_registrant) {
3739 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3740 		}
3741 		break;
3742 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3743 		racqa = cmd->cdw10_bits.resv_acquire.racqa;
3744 		if (racqa == SPDK_NVME_RESERVE_ACQUIRE) {
3745 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3746 			goto exit;
3747 		}
3748 		if (!is_registrant) {
3749 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3750 		}
3751 		break;
3752 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3753 		if (!is_registrant) {
3754 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3755 		}
3756 		break;
3757 	default:
3758 		break;
3759 	}
3760 
3761 exit:
3762 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3763 	req->rsp->nvme_cpl.status.sc = status;
3764 	if (status == SPDK_NVME_SC_RESERVATION_CONFLICT) {
3765 		return -EPERM;
3766 	}
3767 
3768 	return 0;
3769 }
3770 
3771 static int
3772 nvmf_ctrlr_process_io_fused_cmd(struct spdk_nvmf_request *req, struct spdk_bdev *bdev,
3773 				struct spdk_bdev_desc *desc, struct spdk_io_channel *ch)
3774 {
3775 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3776 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
3777 	struct spdk_nvmf_request *first_fused_req = req->qpair->first_fused_req;
3778 	int rc;
3779 
3780 	if (cmd->fuse == SPDK_NVME_CMD_FUSE_FIRST) {
3781 		/* first fused operation (should be compare) */
3782 		if (first_fused_req != NULL) {
3783 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3784 
3785 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3786 
3787 			/* abort req->qpair->first_fused_request and continue with new fused command */
3788 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3789 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3790 			_nvmf_request_complete(first_fused_req);
3791 		} else if (cmd->opc != SPDK_NVME_OPC_COMPARE) {
3792 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3793 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3794 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3795 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3796 		}
3797 
3798 		req->qpair->first_fused_req = req;
3799 		return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
3800 	} else if (cmd->fuse == SPDK_NVME_CMD_FUSE_SECOND) {
3801 		/* second fused operation (should be write) */
3802 		if (first_fused_req == NULL) {
3803 			SPDK_ERRLOG("Wrong sequence of fused operations\n");
3804 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3805 			rsp->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3806 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3807 		} else if (cmd->opc != SPDK_NVME_OPC_WRITE) {
3808 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3809 
3810 			SPDK_ERRLOG("Wrong op code of fused operations\n");
3811 
3812 			/* abort req->qpair->first_fused_request and fail current command */
3813 			fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3814 			fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3815 			_nvmf_request_complete(first_fused_req);
3816 
3817 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3818 			rsp->status.sc = SPDK_NVME_SC_INVALID_OPCODE;
3819 			req->qpair->first_fused_req = NULL;
3820 			return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3821 		}
3822 
3823 		/* save request of first command to generate response later */
3824 		req->first_fused_req = first_fused_req;
3825 		req->qpair->first_fused_req = NULL;
3826 	} else {
3827 		SPDK_ERRLOG("Invalid fused command fuse field.\n");
3828 		rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3829 		rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD;
3830 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3831 	}
3832 
3833 	rc = nvmf_bdev_ctrlr_compare_and_write_cmd(bdev, desc, ch, req->first_fused_req, req);
3834 
3835 	if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
3836 		if (spdk_nvme_cpl_is_error(rsp)) {
3837 			struct spdk_nvme_cpl *fused_response = &first_fused_req->rsp->nvme_cpl;
3838 
3839 			fused_response->status = rsp->status;
3840 			rsp->status.sct = SPDK_NVME_SCT_GENERIC;
3841 			rsp->status.sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
3842 			/* Complete first of fused commands. Second will be completed by upper layer */
3843 			_nvmf_request_complete(first_fused_req);
3844 			req->first_fused_req = NULL;
3845 		}
3846 	}
3847 
3848 	return rc;
3849 }
3850 
3851 bool
3852 nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req)
3853 {
3854 	struct spdk_nvmf_transport *transport = req->qpair->transport;
3855 	struct spdk_nvmf_ns *ns;
3856 
3857 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE);
3858 
3859 	if (!transport->opts.zcopy) {
3860 		return false;
3861 	}
3862 
3863 	if (nvmf_qpair_is_admin_queue(req->qpair)) {
3864 		/* Admin queue */
3865 		return false;
3866 	}
3867 
3868 	if ((req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_WRITE) &&
3869 	    (req->cmd->nvme_cmd.opc != SPDK_NVME_OPC_READ)) {
3870 		/* Not a READ or WRITE command */
3871 		return false;
3872 	}
3873 
3874 	if (req->cmd->nvme_cmd.fuse != SPDK_NVME_CMD_FUSE_NONE) {
3875 		/* Fused commands dont use zcopy buffers */
3876 		return false;
3877 	}
3878 
3879 	ns = _nvmf_subsystem_get_ns(req->qpair->ctrlr->subsys, req->cmd->nvme_cmd.nsid);
3880 	if (ns == NULL || ns->bdev == NULL || !ns->zcopy) {
3881 		return false;
3882 	}
3883 
3884 	req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT;
3885 	return true;
3886 }
3887 
3888 void
3889 spdk_nvmf_request_zcopy_start(struct spdk_nvmf_request *req)
3890 {
3891 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
3892 
3893 	/* Set iovcnt to be the maximum number of iovs that the ZCOPY can use */
3894 	req->iovcnt = NVMF_REQ_MAX_BUFFERS;
3895 
3896 	spdk_nvmf_request_exec(req);
3897 }
3898 
3899 void
3900 spdk_nvmf_request_zcopy_end(struct spdk_nvmf_request *req, bool commit)
3901 {
3902 	assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE);
3903 	req->zcopy_phase = NVMF_ZCOPY_PHASE_END_PENDING;
3904 
3905 	nvmf_bdev_ctrlr_zcopy_end(req, commit);
3906 }
3907 
3908 int
3909 nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req)
3910 {
3911 	uint32_t nsid;
3912 	struct spdk_nvmf_ns *ns;
3913 	struct spdk_bdev *bdev;
3914 	struct spdk_bdev_desc *desc;
3915 	struct spdk_io_channel *ch;
3916 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3917 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3918 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3919 	struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl;
3920 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
3921 	enum spdk_nvme_ana_state ana_state;
3922 
3923 	/* pre-set response details for this command */
3924 	response->status.sc = SPDK_NVME_SC_SUCCESS;
3925 	nsid = cmd->nsid;
3926 
3927 	if (spdk_unlikely(ctrlr == NULL)) {
3928 		SPDK_ERRLOG("I/O command sent before CONNECT\n");
3929 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3930 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3931 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3932 	}
3933 
3934 	if (spdk_unlikely(ctrlr->vcprop.cc.bits.en != 1)) {
3935 		SPDK_ERRLOG("I/O command sent to disabled controller\n");
3936 		response->status.sct = SPDK_NVME_SCT_GENERIC;
3937 		response->status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
3938 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3939 	}
3940 
3941 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3942 	if (ns == NULL || ns->bdev == NULL) {
3943 		SPDK_DEBUGLOG(nvmf, "Unsuccessful query for nsid %u\n", cmd->nsid);
3944 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
3945 		response->status.dnr = 1;
3946 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3947 	}
3948 
3949 	ana_state = nvmf_ctrlr_get_ana_state(ctrlr, ns->anagrpid);
3950 	if (spdk_unlikely(ana_state != SPDK_NVME_ANA_OPTIMIZED_STATE &&
3951 			  ana_state != SPDK_NVME_ANA_NON_OPTIMIZED_STATE)) {
3952 		SPDK_DEBUGLOG(nvmf, "Fail I/O command due to ANA state %d\n",
3953 			      ana_state);
3954 		response->status.sct = SPDK_NVME_SCT_PATH;
3955 		response->status.sc = _nvme_ana_state_to_path_status(ana_state);
3956 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3957 	}
3958 
3959 	if (spdk_likely(ctrlr->listener != NULL)) {
3960 		SPDK_DTRACE_PROBE3(nvmf_request_io_exec_path, req,
3961 				   ctrlr->listener->trid->traddr,
3962 				   ctrlr->listener->trid->trsvcid);
3963 	}
3964 
3965 	/* scan-build falsely reporting dereference of null pointer */
3966 	assert(group != NULL && group->sgroups != NULL);
3967 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
3968 	if (nvmf_ns_reservation_request_check(ns_info, ctrlr, req)) {
3969 		SPDK_DEBUGLOG(nvmf, "Reservation Conflict for nsid %u, opcode %u\n",
3970 			      cmd->nsid, cmd->opc);
3971 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
3972 	}
3973 
3974 	bdev = ns->bdev;
3975 	desc = ns->desc;
3976 	ch = ns_info->channel;
3977 
3978 	if (spdk_unlikely(cmd->fuse & SPDK_NVME_CMD_FUSE_MASK)) {
3979 		return nvmf_ctrlr_process_io_fused_cmd(req, bdev, desc, ch);
3980 	} else if (spdk_unlikely(req->qpair->first_fused_req != NULL)) {
3981 		struct spdk_nvme_cpl *fused_response = &req->qpair->first_fused_req->rsp->nvme_cpl;
3982 
3983 		SPDK_ERRLOG("Expected second of fused commands - failing first of fused commands\n");
3984 
3985 		/* abort req->qpair->first_fused_request and continue with new command */
3986 		fused_response->status.sc = SPDK_NVME_SC_ABORTED_MISSING_FUSED;
3987 		fused_response->status.sct = SPDK_NVME_SCT_GENERIC;
3988 		_nvmf_request_complete(req->qpair->first_fused_req);
3989 		req->qpair->first_fused_req = NULL;
3990 	}
3991 
3992 	if (spdk_nvmf_request_using_zcopy(req)) {
3993 		assert(req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT);
3994 		return nvmf_bdev_ctrlr_zcopy_start(bdev, desc, ch, req);
3995 	} else {
3996 		switch (cmd->opc) {
3997 		case SPDK_NVME_OPC_READ:
3998 			return nvmf_bdev_ctrlr_read_cmd(bdev, desc, ch, req);
3999 		case SPDK_NVME_OPC_WRITE:
4000 			return nvmf_bdev_ctrlr_write_cmd(bdev, desc, ch, req);
4001 		case SPDK_NVME_OPC_COMPARE:
4002 			return nvmf_bdev_ctrlr_compare_cmd(bdev, desc, ch, req);
4003 		case SPDK_NVME_OPC_WRITE_ZEROES:
4004 			return nvmf_bdev_ctrlr_write_zeroes_cmd(bdev, desc, ch, req);
4005 		case SPDK_NVME_OPC_FLUSH:
4006 			return nvmf_bdev_ctrlr_flush_cmd(bdev, desc, ch, req);
4007 		case SPDK_NVME_OPC_DATASET_MANAGEMENT:
4008 			return nvmf_bdev_ctrlr_dsm_cmd(bdev, desc, ch, req);
4009 		case SPDK_NVME_OPC_RESERVATION_REGISTER:
4010 		case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
4011 		case SPDK_NVME_OPC_RESERVATION_RELEASE:
4012 		case SPDK_NVME_OPC_RESERVATION_REPORT:
4013 			spdk_thread_send_msg(ctrlr->subsys->thread, nvmf_ns_reservation_request, req);
4014 			return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS;
4015 		default:
4016 			return nvmf_bdev_ctrlr_nvme_passthru_io(bdev, desc, ch, req);
4017 		}
4018 	}
4019 }
4020 
4021 static void
4022 nvmf_qpair_request_cleanup(struct spdk_nvmf_qpair *qpair)
4023 {
4024 	if (qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING) {
4025 		assert(qpair->state_cb != NULL);
4026 
4027 		if (TAILQ_EMPTY(&qpair->outstanding)) {
4028 			qpair->state_cb(qpair->state_cb_arg, 0);
4029 		}
4030 	}
4031 }
4032 
4033 int
4034 spdk_nvmf_request_free(struct spdk_nvmf_request *req)
4035 {
4036 	struct spdk_nvmf_qpair *qpair = req->qpair;
4037 
4038 	TAILQ_REMOVE(&qpair->outstanding, req, link);
4039 	if (nvmf_transport_req_free(req)) {
4040 		SPDK_ERRLOG("Unable to free transport level request resources.\n");
4041 	}
4042 
4043 	nvmf_qpair_request_cleanup(qpair);
4044 
4045 	return 0;
4046 }
4047 
4048 static void
4049 _nvmf_request_complete(void *ctx)
4050 {
4051 	struct spdk_nvmf_request *req = ctx;
4052 	struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl;
4053 	struct spdk_nvmf_qpair *qpair;
4054 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4055 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4056 	bool is_aer = false;
4057 	uint32_t nsid;
4058 	bool paused;
4059 	uint8_t opcode;
4060 
4061 	rsp->sqid = 0;
4062 	rsp->status.p = 0;
4063 	rsp->cid = req->cmd->nvme_cmd.cid;
4064 	nsid = req->cmd->nvme_cmd.nsid;
4065 	opcode = req->cmd->nvmf_cmd.opcode;
4066 
4067 	qpair = req->qpair;
4068 	if (qpair->ctrlr) {
4069 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4070 		assert(sgroup != NULL);
4071 		is_aer = req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_ASYNC_EVENT_REQUEST;
4072 
4073 		/*
4074 		 * Set the crd value.
4075 		 * If the the IO has any error, and dnr (DoNotRetry) is not 1,
4076 		 * and ACRE is enabled, we will set the crd to 1 to select the first CRDT.
4077 		 */
4078 		if (spdk_nvme_cpl_is_error(rsp) &&
4079 		    rsp->status.dnr == 0 &&
4080 		    qpair->ctrlr->acre_enabled) {
4081 			rsp->status.crd = 1;
4082 		}
4083 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4084 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4085 	}
4086 
4087 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4088 		spdk_nvme_print_completion(qpair->qid, rsp);
4089 	}
4090 
4091 	switch (req->zcopy_phase) {
4092 	case NVMF_ZCOPY_PHASE_NONE:
4093 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4094 		break;
4095 	case NVMF_ZCOPY_PHASE_INIT:
4096 		if (spdk_unlikely(spdk_nvme_cpl_is_error(rsp))) {
4097 			req->zcopy_phase = NVMF_ZCOPY_PHASE_INIT_FAILED;
4098 			TAILQ_REMOVE(&qpair->outstanding, req, link);
4099 		} else {
4100 			req->zcopy_phase = NVMF_ZCOPY_PHASE_EXECUTE;
4101 		}
4102 		break;
4103 	case NVMF_ZCOPY_PHASE_EXECUTE:
4104 		break;
4105 	case NVMF_ZCOPY_PHASE_END_PENDING:
4106 		TAILQ_REMOVE(&qpair->outstanding, req, link);
4107 		req->zcopy_phase = NVMF_ZCOPY_PHASE_COMPLETE;
4108 		break;
4109 	default:
4110 		SPDK_ERRLOG("Invalid ZCOPY phase %u\n", req->zcopy_phase);
4111 		break;
4112 	}
4113 
4114 	if (nvmf_transport_req_complete(req)) {
4115 		SPDK_ERRLOG("Transport request completion error!\n");
4116 	}
4117 
4118 	/* AER cmd is an exception */
4119 	if (sgroup && !is_aer) {
4120 		if (spdk_unlikely(opcode == SPDK_NVME_OPC_FABRIC ||
4121 				  nvmf_qpair_is_admin_queue(qpair))) {
4122 			assert(sgroup->mgmt_io_outstanding > 0);
4123 			sgroup->mgmt_io_outstanding--;
4124 		} else {
4125 			if (req->zcopy_phase == NVMF_ZCOPY_PHASE_NONE ||
4126 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_COMPLETE ||
4127 			    req->zcopy_phase == NVMF_ZCOPY_PHASE_INIT_FAILED) {
4128 				/* End of request */
4129 
4130 				/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4131 				if (spdk_likely(nsid - 1 < sgroup->num_ns)) {
4132 					sgroup->ns_info[nsid - 1].io_outstanding--;
4133 				}
4134 			}
4135 		}
4136 
4137 		if (spdk_unlikely(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4138 				  sgroup->mgmt_io_outstanding == 0)) {
4139 			paused = true;
4140 			for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
4141 				ns_info = &sgroup->ns_info[nsid];
4142 
4143 				if (ns_info->state == SPDK_NVMF_SUBSYSTEM_PAUSING &&
4144 				    ns_info->io_outstanding > 0) {
4145 					paused = false;
4146 					break;
4147 				}
4148 			}
4149 
4150 			if (paused) {
4151 				sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
4152 				sgroup->cb_fn(sgroup->cb_arg, 0);
4153 				sgroup->cb_fn = NULL;
4154 				sgroup->cb_arg = NULL;
4155 			}
4156 		}
4157 
4158 	}
4159 
4160 	nvmf_qpair_request_cleanup(qpair);
4161 }
4162 
4163 int
4164 spdk_nvmf_request_complete(struct spdk_nvmf_request *req)
4165 {
4166 	struct spdk_nvmf_qpair *qpair = req->qpair;
4167 
4168 	spdk_thread_exec_msg(qpair->group->thread, _nvmf_request_complete, req);
4169 
4170 	return 0;
4171 }
4172 
4173 void
4174 spdk_nvmf_request_exec_fabrics(struct spdk_nvmf_request *req)
4175 {
4176 	struct spdk_nvmf_qpair *qpair = req->qpair;
4177 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4178 	enum spdk_nvmf_request_exec_status status;
4179 
4180 	if (qpair->ctrlr) {
4181 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4182 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4183 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4184 	}
4185 
4186 	assert(sgroup != NULL);
4187 	sgroup->mgmt_io_outstanding++;
4188 
4189 	/* Place the request on the outstanding list so we can keep track of it */
4190 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4191 
4192 	assert(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC);
4193 	status = nvmf_ctrlr_process_fabrics_cmd(req);
4194 
4195 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4196 		_nvmf_request_complete(req);
4197 	}
4198 }
4199 
4200 static bool nvmf_check_subsystem_active(struct spdk_nvmf_request *req)
4201 {
4202 	struct spdk_nvmf_qpair *qpair = req->qpair;
4203 	struct spdk_nvmf_subsystem_poll_group *sgroup = NULL;
4204 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4205 	uint32_t nsid;
4206 
4207 	if (qpair->ctrlr) {
4208 		sgroup = &qpair->group->sgroups[qpair->ctrlr->subsys->id];
4209 		assert(sgroup != NULL);
4210 	} else if (spdk_unlikely(nvmf_request_is_fabric_connect(req))) {
4211 		sgroup = nvmf_subsystem_pg_from_connect_cmd(req);
4212 	}
4213 
4214 	/* Check if the subsystem is paused (if there is a subsystem) */
4215 	if (sgroup != NULL) {
4216 		if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC ||
4217 				  nvmf_qpair_is_admin_queue(qpair))) {
4218 			if (sgroup->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
4219 				/* The subsystem is not currently active. Queue this request. */
4220 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4221 				return false;
4222 			}
4223 			sgroup->mgmt_io_outstanding++;
4224 		} else {
4225 			nsid = req->cmd->nvme_cmd.nsid;
4226 
4227 			/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
4228 			if (spdk_unlikely(nsid - 1 >= sgroup->num_ns)) {
4229 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4230 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4231 				req->rsp->nvme_cpl.status.dnr = 1;
4232 				TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4233 				_nvmf_request_complete(req);
4234 				return false;
4235 			}
4236 
4237 			ns_info = &sgroup->ns_info[nsid - 1];
4238 			if (ns_info->channel == NULL) {
4239 				/* This can can happen if host sends I/O to a namespace that is
4240 				 * in the process of being added, but before the full addition
4241 				 * process is complete.  Report invalid namespace in that case.
4242 				 */
4243 				req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4244 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4245 				req->rsp->nvme_cpl.status.dnr = 1;
4246 				TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4247 				ns_info->io_outstanding++;
4248 				_nvmf_request_complete(req);
4249 				return false;
4250 			}
4251 
4252 			if (ns_info->state != SPDK_NVMF_SUBSYSTEM_ACTIVE) {
4253 				/* The namespace is not currently active. Queue this request. */
4254 				TAILQ_INSERT_TAIL(&sgroup->queued, req, link);
4255 				return false;
4256 			}
4257 
4258 			ns_info->io_outstanding++;
4259 		}
4260 
4261 		if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
4262 			req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
4263 			req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_COMMAND_SEQUENCE_ERROR;
4264 			TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4265 			_nvmf_request_complete(req);
4266 			return false;
4267 		}
4268 	}
4269 
4270 	return true;
4271 }
4272 
4273 void
4274 spdk_nvmf_request_exec(struct spdk_nvmf_request *req)
4275 {
4276 	struct spdk_nvmf_qpair *qpair = req->qpair;
4277 	struct spdk_nvmf_transport *transport = qpair->transport;
4278 	enum spdk_nvmf_request_exec_status status;
4279 
4280 	if (!nvmf_check_subsystem_active(req)) {
4281 		return;
4282 	}
4283 
4284 	if (SPDK_DEBUGLOG_FLAG_ENABLED("nvmf")) {
4285 		spdk_nvme_print_command(qpair->qid, &req->cmd->nvme_cmd);
4286 	}
4287 
4288 	/* Place the request on the outstanding list so we can keep track of it */
4289 	TAILQ_INSERT_TAIL(&qpair->outstanding, req, link);
4290 
4291 	if (spdk_unlikely((req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC) &&
4292 			  spdk_nvme_trtype_is_fabrics(transport->ops->type))) {
4293 		status = nvmf_ctrlr_process_fabrics_cmd(req);
4294 	} else if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4295 		status = nvmf_ctrlr_process_admin_cmd(req);
4296 	} else {
4297 		status = nvmf_ctrlr_process_io_cmd(req);
4298 	}
4299 
4300 	if (status == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
4301 		_nvmf_request_complete(req);
4302 	}
4303 }
4304 
4305 static bool
4306 nvmf_ctrlr_get_dif_ctx(struct spdk_nvmf_ctrlr *ctrlr, struct spdk_nvme_cmd *cmd,
4307 		       struct spdk_dif_ctx *dif_ctx)
4308 {
4309 	struct spdk_nvmf_ns *ns;
4310 	struct spdk_bdev *bdev;
4311 
4312 	if (ctrlr == NULL || cmd == NULL) {
4313 		return false;
4314 	}
4315 
4316 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, cmd->nsid);
4317 	if (ns == NULL || ns->bdev == NULL) {
4318 		return false;
4319 	}
4320 
4321 	bdev = ns->bdev;
4322 
4323 	switch (cmd->opc) {
4324 	case SPDK_NVME_OPC_READ:
4325 	case SPDK_NVME_OPC_WRITE:
4326 	case SPDK_NVME_OPC_COMPARE:
4327 		return nvmf_bdev_ctrlr_get_dif_ctx(bdev, cmd, dif_ctx);
4328 	default:
4329 		break;
4330 	}
4331 
4332 	return false;
4333 }
4334 
4335 bool
4336 spdk_nvmf_request_get_dif_ctx(struct spdk_nvmf_request *req, struct spdk_dif_ctx *dif_ctx)
4337 {
4338 	struct spdk_nvmf_qpair *qpair = req->qpair;
4339 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
4340 
4341 	if (spdk_likely(ctrlr == NULL || !ctrlr->dif_insert_or_strip)) {
4342 		return false;
4343 	}
4344 
4345 	if (spdk_unlikely(qpair->state != SPDK_NVMF_QPAIR_ACTIVE)) {
4346 		return false;
4347 	}
4348 
4349 	if (spdk_unlikely(req->cmd->nvmf_cmd.opcode == SPDK_NVME_OPC_FABRIC)) {
4350 		return false;
4351 	}
4352 
4353 	if (spdk_unlikely(nvmf_qpair_is_admin_queue(qpair))) {
4354 		return false;
4355 	}
4356 
4357 	return nvmf_ctrlr_get_dif_ctx(ctrlr, &req->cmd->nvme_cmd, dif_ctx);
4358 }
4359 
4360 void
4361 spdk_nvmf_set_custom_admin_cmd_hdlr(uint8_t opc, spdk_nvmf_custom_cmd_hdlr hdlr)
4362 {
4363 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = hdlr;
4364 }
4365 
4366 static int
4367 nvmf_passthru_admin_cmd(struct spdk_nvmf_request *req)
4368 {
4369 	struct spdk_bdev *bdev;
4370 	struct spdk_bdev_desc *desc;
4371 	struct spdk_io_channel *ch;
4372 	struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
4373 	struct spdk_nvme_cpl *response = spdk_nvmf_request_get_response(req);
4374 	uint32_t bdev_nsid;
4375 	int rc;
4376 
4377 	if (g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid == 0) {
4378 		bdev_nsid = cmd->nsid;
4379 	} else {
4380 		bdev_nsid = g_nvmf_custom_admin_cmd_hdlrs[cmd->opc].nsid;
4381 	}
4382 
4383 	rc = spdk_nvmf_request_get_bdev(bdev_nsid, req, &bdev, &desc, &ch);
4384 	if (rc) {
4385 		response->status.sct = SPDK_NVME_SCT_GENERIC;
4386 		response->status.sc = SPDK_NVME_SC_INVALID_NAMESPACE_OR_FORMAT;
4387 		return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE;
4388 	}
4389 	return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, NULL);
4390 }
4391 
4392 void
4393 spdk_nvmf_set_passthru_admin_cmd(uint8_t opc, uint32_t forward_nsid)
4394 {
4395 	g_nvmf_custom_admin_cmd_hdlrs[opc].hdlr = nvmf_passthru_admin_cmd;
4396 	g_nvmf_custom_admin_cmd_hdlrs[opc].nsid = forward_nsid;
4397 }
4398 
4399 int
4400 spdk_nvmf_request_get_bdev(uint32_t nsid, struct spdk_nvmf_request *req,
4401 			   struct spdk_bdev **bdev, struct spdk_bdev_desc **desc, struct spdk_io_channel **ch)
4402 {
4403 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
4404 	struct spdk_nvmf_ns *ns;
4405 	struct spdk_nvmf_poll_group *group = req->qpair->group;
4406 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
4407 
4408 	*bdev = NULL;
4409 	*desc = NULL;
4410 	*ch = NULL;
4411 
4412 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
4413 	if (ns == NULL || ns->bdev == NULL) {
4414 		return -EINVAL;
4415 	}
4416 
4417 	assert(group != NULL && group->sgroups != NULL);
4418 	ns_info = &group->sgroups[ctrlr->subsys->id].ns_info[nsid - 1];
4419 	*bdev = ns->bdev;
4420 	*desc = ns->desc;
4421 	*ch = ns_info->channel;
4422 
4423 	return 0;
4424 }
4425 
4426 struct spdk_nvmf_ctrlr *spdk_nvmf_request_get_ctrlr(struct spdk_nvmf_request *req)
4427 {
4428 	return req->qpair->ctrlr;
4429 }
4430 
4431 struct spdk_nvme_cmd *spdk_nvmf_request_get_cmd(struct spdk_nvmf_request *req)
4432 {
4433 	return &req->cmd->nvme_cmd;
4434 }
4435 
4436 struct spdk_nvme_cpl *spdk_nvmf_request_get_response(struct spdk_nvmf_request *req)
4437 {
4438 	return &req->rsp->nvme_cpl;
4439 }
4440 
4441 struct spdk_nvmf_subsystem *spdk_nvmf_request_get_subsystem(struct spdk_nvmf_request *req)
4442 {
4443 	return req->qpair->ctrlr->subsys;
4444 }
4445 
4446 void spdk_nvmf_request_get_data(struct spdk_nvmf_request *req, void **data, uint32_t *length)
4447 {
4448 	*data = req->data;
4449 	*length = req->length;
4450 }
4451 
4452 struct spdk_nvmf_subsystem *spdk_nvmf_ctrlr_get_subsystem(struct spdk_nvmf_ctrlr *ctrlr)
4453 {
4454 	return ctrlr->subsys;
4455 }
4456 
4457 uint16_t spdk_nvmf_ctrlr_get_id(struct spdk_nvmf_ctrlr *ctrlr)
4458 {
4459 	return ctrlr->cntlid;
4460 }
4461 
4462 struct spdk_nvmf_request *spdk_nvmf_request_get_req_to_abort(struct spdk_nvmf_request *req)
4463 {
4464 	return req->req_to_abort;
4465 }
4466