xref: /spdk/lib/nvmf/subsystem.c (revision b37db06935181fd0e8f5592a96d860040abaa201)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (C) 2016 Intel Corporation. All rights reserved.
3  *   Copyright (c) 2019 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/assert.h"
13 #include "spdk/likely.h"
14 #include "spdk/string.h"
15 #include "spdk/trace.h"
16 #include "spdk/nvmf_spec.h"
17 #include "spdk/uuid.h"
18 #include "spdk/json.h"
19 #include "spdk/file.h"
20 #include "spdk/bit_array.h"
21 #include "spdk/bdev.h"
22 
23 #define __SPDK_BDEV_MODULE_ONLY
24 #include "spdk/bdev_module.h"
25 #include "spdk/log.h"
26 #include "spdk_internal/utf.h"
27 #include "spdk_internal/usdt.h"
28 
29 #define MODEL_NUMBER_DEFAULT "SPDK bdev Controller"
30 #define NVMF_SUBSYSTEM_DEFAULT_NAMESPACES 32
31 
32 /*
33  * States for parsing valid domains in NQNs according to RFC 1034
34  */
35 enum spdk_nvmf_nqn_domain_states {
36 	/* First character of a domain must be a letter */
37 	SPDK_NVMF_DOMAIN_ACCEPT_LETTER = 0,
38 
39 	/* Subsequent characters can be any of letter, digit, or hyphen */
40 	SPDK_NVMF_DOMAIN_ACCEPT_LDH = 1,
41 
42 	/* A domain label must end with either a letter or digit */
43 	SPDK_NVMF_DOMAIN_ACCEPT_ANY = 2
44 };
45 
46 static int _nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem);
47 
48 /* Returns true if is a valid ASCII string as defined by the NVMe spec */
49 static bool
50 nvmf_valid_ascii_string(const void *buf, size_t size)
51 {
52 	const uint8_t *str = buf;
53 	size_t i;
54 
55 	for (i = 0; i < size; i++) {
56 		if (str[i] < 0x20 || str[i] > 0x7E) {
57 			return false;
58 		}
59 	}
60 
61 	return true;
62 }
63 
64 bool
65 nvmf_nqn_is_valid(const char *nqn)
66 {
67 	size_t len;
68 	struct spdk_uuid uuid_value;
69 	uint32_t i;
70 	int bytes_consumed;
71 	uint32_t domain_label_length;
72 	char *reverse_domain_end;
73 	uint32_t reverse_domain_end_index;
74 	enum spdk_nvmf_nqn_domain_states domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LETTER;
75 
76 	/* Check for length requirements */
77 	len = strlen(nqn);
78 	if (len > SPDK_NVMF_NQN_MAX_LEN) {
79 		SPDK_ERRLOG("Invalid NQN \"%s\": length %zu > max %d\n", nqn, len, SPDK_NVMF_NQN_MAX_LEN);
80 		return false;
81 	}
82 
83 	/* The nqn must be at least as long as SPDK_NVMF_NQN_MIN_LEN to contain the necessary prefix. */
84 	if (len < SPDK_NVMF_NQN_MIN_LEN) {
85 		SPDK_ERRLOG("Invalid NQN \"%s\": length %zu < min %d\n", nqn, len, SPDK_NVMF_NQN_MIN_LEN);
86 		return false;
87 	}
88 
89 	/* Check for discovery controller nqn */
90 	if (!strcmp(nqn, SPDK_NVMF_DISCOVERY_NQN)) {
91 		return true;
92 	}
93 
94 	/* Check for equality with the generic nqn structure of the form "nqn.2014-08.org.nvmexpress:uuid:11111111-2222-3333-4444-555555555555" */
95 	if (!strncmp(nqn, SPDK_NVMF_NQN_UUID_PRE, SPDK_NVMF_NQN_UUID_PRE_LEN)) {
96 		if (len != SPDK_NVMF_NQN_UUID_PRE_LEN + SPDK_NVMF_UUID_STRING_LEN) {
97 			SPDK_ERRLOG("Invalid NQN \"%s\": uuid is not the correct length\n", nqn);
98 			return false;
99 		}
100 
101 		if (spdk_uuid_parse(&uuid_value, &nqn[SPDK_NVMF_NQN_UUID_PRE_LEN])) {
102 			SPDK_ERRLOG("Invalid NQN \"%s\": uuid is not formatted correctly\n", nqn);
103 			return false;
104 		}
105 		return true;
106 	}
107 
108 	/* If the nqn does not match the uuid structure, the next several checks validate the form "nqn.yyyy-mm.reverse.domain:user-string" */
109 
110 	if (strncmp(nqn, "nqn.", 4) != 0) {
111 		SPDK_ERRLOG("Invalid NQN \"%s\": NQN must begin with \"nqn.\".\n", nqn);
112 		return false;
113 	}
114 
115 	/* Check for yyyy-mm. */
116 	if (!(isdigit(nqn[4]) && isdigit(nqn[5]) && isdigit(nqn[6]) && isdigit(nqn[7]) &&
117 	      nqn[8] == '-' && isdigit(nqn[9]) && isdigit(nqn[10]) && nqn[11] == '.')) {
118 		SPDK_ERRLOG("Invalid date code in NQN \"%s\"\n", nqn);
119 		return false;
120 	}
121 
122 	reverse_domain_end = strchr(nqn, ':');
123 	if (reverse_domain_end != NULL && (reverse_domain_end_index = reverse_domain_end - nqn) < len - 1) {
124 	} else {
125 		SPDK_ERRLOG("Invalid NQN \"%s\". NQN must contain user specified name with a ':' as a prefix.\n",
126 			    nqn);
127 		return false;
128 	}
129 
130 	/* Check for valid reverse domain */
131 	domain_label_length = 0;
132 	for (i = 12; i < reverse_domain_end_index; i++) {
133 		if (domain_label_length > SPDK_DOMAIN_LABEL_MAX_LEN) {
134 			SPDK_ERRLOG("Invalid domain name in NQN \"%s\". At least one Label is too long.\n", nqn);
135 			return false;
136 		}
137 
138 		switch (domain_state) {
139 
140 		case SPDK_NVMF_DOMAIN_ACCEPT_LETTER: {
141 			if (isalpha(nqn[i])) {
142 				domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY;
143 				domain_label_length++;
144 				break;
145 			} else {
146 				SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must start with a letter.\n", nqn);
147 				return false;
148 			}
149 		}
150 
151 		case SPDK_NVMF_DOMAIN_ACCEPT_LDH: {
152 			if (isalpha(nqn[i]) || isdigit(nqn[i])) {
153 				domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY;
154 				domain_label_length++;
155 				break;
156 			} else if (nqn[i] == '-') {
157 				if (i == reverse_domain_end_index - 1) {
158 					SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n",
159 						    nqn);
160 					return false;
161 				}
162 				domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LDH;
163 				domain_label_length++;
164 				break;
165 			} else if (nqn[i] == '.') {
166 				SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n",
167 					    nqn);
168 				return false;
169 			} else {
170 				SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only [a-z,A-Z,0-9,'-','.'].\n",
171 					    nqn);
172 				return false;
173 			}
174 		}
175 
176 		case SPDK_NVMF_DOMAIN_ACCEPT_ANY: {
177 			if (isalpha(nqn[i]) || isdigit(nqn[i])) {
178 				domain_state = SPDK_NVMF_DOMAIN_ACCEPT_ANY;
179 				domain_label_length++;
180 				break;
181 			} else if (nqn[i] == '-') {
182 				if (i == reverse_domain_end_index - 1) {
183 					SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must end with an alphanumeric symbol.\n",
184 						    nqn);
185 					return false;
186 				}
187 				domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LDH;
188 				domain_label_length++;
189 				break;
190 			} else if (nqn[i] == '.') {
191 				domain_state = SPDK_NVMF_DOMAIN_ACCEPT_LETTER;
192 				domain_label_length = 0;
193 				break;
194 			} else {
195 				SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only [a-z,A-Z,0-9,'-','.'].\n",
196 					    nqn);
197 				return false;
198 			}
199 		}
200 		}
201 	}
202 
203 	i = reverse_domain_end_index + 1;
204 	while (i < len) {
205 		bytes_consumed = utf8_valid(&nqn[i], &nqn[len]);
206 		if (bytes_consumed <= 0) {
207 			SPDK_ERRLOG("Invalid domain name in NQN \"%s\". Label names must contain only valid utf-8.\n", nqn);
208 			return false;
209 		}
210 
211 		i += bytes_consumed;
212 	}
213 	return true;
214 }
215 
216 static void subsystem_state_change_on_pg(struct spdk_io_channel_iter *i);
217 
218 struct spdk_nvmf_subsystem *
219 spdk_nvmf_subsystem_create(struct spdk_nvmf_tgt *tgt,
220 			   const char *nqn,
221 			   enum spdk_nvmf_subtype type,
222 			   uint32_t num_ns)
223 {
224 	struct spdk_nvmf_subsystem	*subsystem;
225 	uint32_t			sid;
226 
227 	if (spdk_nvmf_tgt_find_subsystem(tgt, nqn)) {
228 		SPDK_ERRLOG("Subsystem NQN '%s' already exists\n", nqn);
229 		return NULL;
230 	}
231 
232 	if (!nvmf_nqn_is_valid(nqn)) {
233 		SPDK_ERRLOG("Subsystem NQN '%s' is invalid\n", nqn);
234 		return NULL;
235 	}
236 
237 	if (type == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
238 	    type == SPDK_NVMF_SUBTYPE_DISCOVERY) {
239 		if (num_ns != 0) {
240 			SPDK_ERRLOG("Discovery subsystem cannot have namespaces.\n");
241 			return NULL;
242 		}
243 	} else if (num_ns == 0) {
244 		num_ns = NVMF_SUBSYSTEM_DEFAULT_NAMESPACES;
245 	}
246 
247 	/* Find a free subsystem id (sid) */
248 	sid = spdk_bit_array_find_first_clear(tgt->subsystem_ids, 0);
249 	if (sid == UINT32_MAX) {
250 		SPDK_ERRLOG("No free subsystem IDs are available for subsystem creation\n");
251 		return NULL;
252 	}
253 	subsystem = calloc(1, sizeof(struct spdk_nvmf_subsystem));
254 	if (subsystem == NULL) {
255 		SPDK_ERRLOG("Subsystem memory allocation failed\n");
256 		return NULL;
257 	}
258 
259 	subsystem->thread = spdk_get_thread();
260 	subsystem->state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
261 	subsystem->tgt = tgt;
262 	subsystem->id = sid;
263 	subsystem->subtype = type;
264 	subsystem->max_nsid = num_ns;
265 	subsystem->next_cntlid = 1;
266 	subsystem->min_cntlid = NVMF_MIN_CNTLID;
267 	subsystem->max_cntlid = NVMF_MAX_CNTLID;
268 	snprintf(subsystem->subnqn, sizeof(subsystem->subnqn), "%s", nqn);
269 	pthread_mutex_init(&subsystem->mutex, NULL);
270 	TAILQ_INIT(&subsystem->listeners);
271 	TAILQ_INIT(&subsystem->hosts);
272 	TAILQ_INIT(&subsystem->ctrlrs);
273 	TAILQ_INIT(&subsystem->state_changes);
274 	subsystem->used_listener_ids = spdk_bit_array_create(NVMF_MAX_LISTENERS_PER_SUBSYSTEM);
275 	if (subsystem->used_listener_ids == NULL) {
276 		pthread_mutex_destroy(&subsystem->mutex);
277 		free(subsystem);
278 		SPDK_ERRLOG("Listener id array memory allocation failed\n");
279 		return NULL;
280 	}
281 
282 	if (num_ns != 0) {
283 		subsystem->ns = calloc(num_ns, sizeof(struct spdk_nvmf_ns *));
284 		if (subsystem->ns == NULL) {
285 			SPDK_ERRLOG("Namespace memory allocation failed\n");
286 			pthread_mutex_destroy(&subsystem->mutex);
287 			spdk_bit_array_free(&subsystem->used_listener_ids);
288 			free(subsystem);
289 			return NULL;
290 		}
291 		subsystem->ana_group = calloc(num_ns, sizeof(uint32_t));
292 		if (subsystem->ana_group == NULL) {
293 			SPDK_ERRLOG("ANA group memory allocation failed\n");
294 			pthread_mutex_destroy(&subsystem->mutex);
295 			free(subsystem->ns);
296 			spdk_bit_array_free(&subsystem->used_listener_ids);
297 			free(subsystem);
298 			return NULL;
299 		}
300 	}
301 
302 	memset(subsystem->sn, '0', sizeof(subsystem->sn) - 1);
303 	subsystem->sn[sizeof(subsystem->sn) - 1] = '\0';
304 
305 	snprintf(subsystem->mn, sizeof(subsystem->mn), "%s",
306 		 MODEL_NUMBER_DEFAULT);
307 
308 	spdk_bit_array_set(tgt->subsystem_ids, sid);
309 	RB_INSERT(subsystem_tree, &tgt->subsystems, subsystem);
310 
311 	SPDK_DTRACE_PROBE1(nvmf_subsystem_create, subsystem->subnqn);
312 
313 	return subsystem;
314 }
315 
316 static void
317 nvmf_host_free(struct spdk_nvmf_host *host)
318 {
319 	spdk_keyring_put_key(host->dhchap_key);
320 	spdk_keyring_put_key(host->dhchap_ctrlr_key);
321 	free(host);
322 }
323 
324 /* Must hold subsystem->mutex while calling this function */
325 static void
326 nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_host *host)
327 {
328 	TAILQ_REMOVE(&subsystem->hosts, host, link);
329 	nvmf_host_free(host);
330 }
331 
332 static void
333 _nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem,
334 				struct spdk_nvmf_subsystem_listener *listener,
335 				bool stop)
336 {
337 	struct spdk_nvmf_transport *transport;
338 	struct spdk_nvmf_ctrlr *ctrlr;
339 
340 	if (stop) {
341 		transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, listener->trid->trstring);
342 		if (transport != NULL) {
343 			spdk_nvmf_transport_stop_listen(transport, listener->trid);
344 		}
345 	}
346 
347 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
348 		if (ctrlr->listener == listener) {
349 			ctrlr->listener = NULL;
350 		}
351 	}
352 
353 	TAILQ_REMOVE(&subsystem->listeners, listener, link);
354 	if (spdk_nvmf_subsystem_is_discovery(listener->subsystem)) {
355 		nvmf_tgt_update_mdns_prr(listener->subsystem->tgt);
356 	}
357 	spdk_nvmf_send_discovery_log_notice(listener->subsystem->tgt, NULL);
358 	free(listener->ana_state);
359 	spdk_bit_array_clear(subsystem->used_listener_ids, listener->id);
360 	free(listener->opts.sock_impl);
361 	free(listener);
362 }
363 
364 static void
365 _nvmf_subsystem_destroy_msg(void *cb_arg)
366 {
367 	struct spdk_nvmf_subsystem *subsystem = cb_arg;
368 
369 	_nvmf_subsystem_destroy(subsystem);
370 }
371 
372 static int
373 _nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem)
374 {
375 	struct nvmf_subsystem_state_change_ctx *ctx;
376 	struct spdk_nvmf_ns		*ns;
377 	nvmf_subsystem_destroy_cb	async_destroy_cb = NULL;
378 	void				*async_destroy_cb_arg = NULL;
379 	int				rc;
380 
381 	if (!TAILQ_EMPTY(&subsystem->ctrlrs)) {
382 		SPDK_DEBUGLOG(nvmf, "subsystem %p %s has active controllers\n", subsystem, subsystem->subnqn);
383 		subsystem->async_destroy = true;
384 		rc = spdk_thread_send_msg(subsystem->thread, _nvmf_subsystem_destroy_msg, subsystem);
385 		if (rc) {
386 			SPDK_ERRLOG("Failed to send thread msg, rc %d\n", rc);
387 			assert(0);
388 			return rc;
389 		}
390 		return -EINPROGRESS;
391 	}
392 
393 	ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
394 	while (ns != NULL) {
395 		struct spdk_nvmf_ns *next_ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns);
396 
397 		spdk_nvmf_subsystem_remove_ns(subsystem, ns->opts.nsid);
398 		ns = next_ns;
399 	}
400 
401 	while ((ctx = TAILQ_FIRST(&subsystem->state_changes))) {
402 		SPDK_WARNLOG("subsystem %s has pending state change requests\n", subsystem->subnqn);
403 		TAILQ_REMOVE(&subsystem->state_changes, ctx, link);
404 		if (ctx->cb_fn != NULL) {
405 			ctx->cb_fn(subsystem, ctx->cb_arg, -ECANCELED);
406 		}
407 		free(ctx);
408 	}
409 
410 	free(subsystem->ns);
411 	free(subsystem->ana_group);
412 
413 	RB_REMOVE(subsystem_tree, &subsystem->tgt->subsystems, subsystem);
414 	assert(spdk_bit_array_get(subsystem->tgt->subsystem_ids, subsystem->id) == true);
415 	spdk_bit_array_clear(subsystem->tgt->subsystem_ids, subsystem->id);
416 
417 	pthread_mutex_destroy(&subsystem->mutex);
418 
419 	spdk_bit_array_free(&subsystem->used_listener_ids);
420 
421 	if (subsystem->async_destroy) {
422 		async_destroy_cb = subsystem->async_destroy_cb;
423 		async_destroy_cb_arg = subsystem->async_destroy_cb_arg;
424 	}
425 
426 	free(subsystem);
427 
428 	if (async_destroy_cb) {
429 		async_destroy_cb(async_destroy_cb_arg);
430 	}
431 
432 	return 0;
433 }
434 
435 static struct spdk_nvmf_ns *
436 _nvmf_subsystem_get_first_zoned_ns(struct spdk_nvmf_subsystem *subsystem)
437 {
438 	struct spdk_nvmf_ns *ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
439 	while (ns != NULL) {
440 		if (ns->csi == SPDK_NVME_CSI_ZNS) {
441 			return ns;
442 		}
443 		ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns);
444 	}
445 	return NULL;
446 }
447 
448 int
449 spdk_nvmf_subsystem_destroy(struct spdk_nvmf_subsystem *subsystem, nvmf_subsystem_destroy_cb cpl_cb,
450 			    void *cpl_cb_arg)
451 {
452 	struct spdk_nvmf_host *host, *host_tmp;
453 	struct spdk_nvmf_transport *transport;
454 
455 	if (!subsystem) {
456 		return -EINVAL;
457 	}
458 
459 	SPDK_DTRACE_PROBE1(nvmf_subsystem_destroy, subsystem->subnqn);
460 
461 	assert(spdk_get_thread() == subsystem->thread);
462 
463 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
464 		SPDK_ERRLOG("Subsystem can only be destroyed in inactive state, %s state %d\n",
465 			    subsystem->subnqn, subsystem->state);
466 		return -EAGAIN;
467 	}
468 	if (subsystem->destroying) {
469 		SPDK_ERRLOG("Subsystem destruction is already started\n");
470 		assert(0);
471 		return -EALREADY;
472 	}
473 
474 	subsystem->destroying = true;
475 
476 	SPDK_DEBUGLOG(nvmf, "subsystem is %p %s\n", subsystem, subsystem->subnqn);
477 
478 	nvmf_subsystem_remove_all_listeners(subsystem, false);
479 
480 	pthread_mutex_lock(&subsystem->mutex);
481 
482 	TAILQ_FOREACH_SAFE(host, &subsystem->hosts, link, host_tmp) {
483 		for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
484 		     transport = spdk_nvmf_transport_get_next(transport)) {
485 			if (transport->ops->subsystem_remove_host) {
486 				transport->ops->subsystem_remove_host(transport, subsystem, host->nqn);
487 			}
488 		}
489 		nvmf_subsystem_remove_host(subsystem, host);
490 	}
491 
492 	pthread_mutex_unlock(&subsystem->mutex);
493 
494 	subsystem->async_destroy_cb = cpl_cb;
495 	subsystem->async_destroy_cb_arg = cpl_cb_arg;
496 
497 	return _nvmf_subsystem_destroy(subsystem);
498 }
499 
500 /* we have to use the typedef in the function declaration to appease astyle. */
501 typedef enum spdk_nvmf_subsystem_state spdk_nvmf_subsystem_state_t;
502 
503 static spdk_nvmf_subsystem_state_t
504 nvmf_subsystem_get_intermediate_state(enum spdk_nvmf_subsystem_state current_state,
505 				      enum spdk_nvmf_subsystem_state requested_state)
506 {
507 	switch (requested_state) {
508 	case SPDK_NVMF_SUBSYSTEM_INACTIVE:
509 		return SPDK_NVMF_SUBSYSTEM_DEACTIVATING;
510 	case SPDK_NVMF_SUBSYSTEM_ACTIVE:
511 		if (current_state == SPDK_NVMF_SUBSYSTEM_PAUSED) {
512 			return SPDK_NVMF_SUBSYSTEM_RESUMING;
513 		} else {
514 			return SPDK_NVMF_SUBSYSTEM_ACTIVATING;
515 		}
516 	case SPDK_NVMF_SUBSYSTEM_PAUSED:
517 		return SPDK_NVMF_SUBSYSTEM_PAUSING;
518 	default:
519 		assert(false);
520 		return SPDK_NVMF_SUBSYSTEM_NUM_STATES;
521 	}
522 }
523 
524 static int
525 nvmf_subsystem_set_state(struct spdk_nvmf_subsystem *subsystem,
526 			 enum spdk_nvmf_subsystem_state state)
527 {
528 	enum spdk_nvmf_subsystem_state actual_old_state, expected_old_state;
529 	bool exchanged;
530 
531 	switch (state) {
532 	case SPDK_NVMF_SUBSYSTEM_INACTIVE:
533 		expected_old_state = SPDK_NVMF_SUBSYSTEM_DEACTIVATING;
534 		break;
535 	case SPDK_NVMF_SUBSYSTEM_ACTIVATING:
536 		expected_old_state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
537 		break;
538 	case SPDK_NVMF_SUBSYSTEM_ACTIVE:
539 		expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING;
540 		break;
541 	case SPDK_NVMF_SUBSYSTEM_PAUSING:
542 		expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
543 		break;
544 	case SPDK_NVMF_SUBSYSTEM_PAUSED:
545 		expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSING;
546 		break;
547 	case SPDK_NVMF_SUBSYSTEM_RESUMING:
548 		expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED;
549 		break;
550 	case SPDK_NVMF_SUBSYSTEM_DEACTIVATING:
551 		expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
552 		break;
553 	default:
554 		assert(false);
555 		return -1;
556 	}
557 
558 	actual_old_state = expected_old_state;
559 	exchanged = __atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false,
560 						__ATOMIC_RELAXED, __ATOMIC_RELAXED);
561 	if (spdk_unlikely(exchanged == false)) {
562 		if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING &&
563 		    state == SPDK_NVMF_SUBSYSTEM_ACTIVE) {
564 			expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING;
565 		}
566 		/* This is for the case when activating the subsystem fails. */
567 		if (actual_old_state == SPDK_NVMF_SUBSYSTEM_ACTIVATING &&
568 		    state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) {
569 			expected_old_state = SPDK_NVMF_SUBSYSTEM_ACTIVATING;
570 		}
571 		/* This is for the case when resuming the subsystem fails. */
572 		if (actual_old_state == SPDK_NVMF_SUBSYSTEM_RESUMING &&
573 		    state == SPDK_NVMF_SUBSYSTEM_PAUSING) {
574 			expected_old_state = SPDK_NVMF_SUBSYSTEM_RESUMING;
575 		}
576 		/* This is for the case when stopping paused subsystem */
577 		if (actual_old_state == SPDK_NVMF_SUBSYSTEM_PAUSED &&
578 		    state == SPDK_NVMF_SUBSYSTEM_DEACTIVATING) {
579 			expected_old_state = SPDK_NVMF_SUBSYSTEM_PAUSED;
580 		}
581 		actual_old_state = expected_old_state;
582 		__atomic_compare_exchange_n(&subsystem->state, &actual_old_state, state, false,
583 					    __ATOMIC_RELAXED, __ATOMIC_RELAXED);
584 	}
585 	assert(actual_old_state == expected_old_state);
586 	return actual_old_state - expected_old_state;
587 }
588 
589 static void nvmf_subsystem_do_state_change(struct nvmf_subsystem_state_change_ctx *ctx);
590 
591 static void
592 _nvmf_subsystem_state_change_complete(void *_ctx)
593 {
594 	struct nvmf_subsystem_state_change_ctx *next, *ctx = _ctx;
595 	struct spdk_nvmf_subsystem *subsystem = ctx->subsystem;
596 
597 	pthread_mutex_lock(&subsystem->mutex);
598 	assert(TAILQ_FIRST(&subsystem->state_changes) == ctx);
599 	TAILQ_REMOVE(&subsystem->state_changes, ctx, link);
600 	next = TAILQ_FIRST(&subsystem->state_changes);
601 	pthread_mutex_unlock(&subsystem->mutex);
602 
603 	if (ctx->cb_fn != NULL) {
604 		ctx->cb_fn(subsystem, ctx->cb_arg, ctx->status);
605 	}
606 	free(ctx);
607 
608 	if (next != NULL) {
609 		nvmf_subsystem_do_state_change(next);
610 	}
611 }
612 
613 static void
614 nvmf_subsystem_state_change_complete(struct nvmf_subsystem_state_change_ctx *ctx, int status)
615 {
616 	ctx->status = status;
617 	spdk_thread_exec_msg(ctx->thread, _nvmf_subsystem_state_change_complete, ctx);
618 }
619 
620 static void
621 subsystem_state_change_revert_done(struct spdk_io_channel_iter *i, int status)
622 {
623 	struct nvmf_subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
624 
625 	/* Nothing to be done here if the state setting fails, we are just screwed. */
626 	if (nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state)) {
627 		SPDK_ERRLOG("Unable to revert the subsystem state after operation failure.\n");
628 	}
629 
630 	/* return a failure here. This function only exists in an error path. */
631 	nvmf_subsystem_state_change_complete(ctx, -1);
632 }
633 
634 static void
635 subsystem_state_change_done(struct spdk_io_channel_iter *i, int status)
636 {
637 	struct nvmf_subsystem_state_change_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
638 	enum spdk_nvmf_subsystem_state intermediate_state;
639 
640 	SPDK_DTRACE_PROBE4(nvmf_subsystem_change_state_done, ctx->subsystem->subnqn,
641 			   ctx->requested_state, ctx->original_state, status);
642 
643 	if (status == 0) {
644 		status = nvmf_subsystem_set_state(ctx->subsystem, ctx->requested_state);
645 		if (status) {
646 			status = -1;
647 		}
648 	}
649 
650 	if (status) {
651 		intermediate_state = nvmf_subsystem_get_intermediate_state(ctx->requested_state,
652 				     ctx->original_state);
653 		assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES);
654 
655 		if (nvmf_subsystem_set_state(ctx->subsystem, intermediate_state)) {
656 			goto out;
657 		}
658 		ctx->requested_state = ctx->original_state;
659 		spdk_for_each_channel(ctx->subsystem->tgt,
660 				      subsystem_state_change_on_pg,
661 				      ctx,
662 				      subsystem_state_change_revert_done);
663 		return;
664 	}
665 
666 out:
667 	nvmf_subsystem_state_change_complete(ctx, status);
668 }
669 
670 static void
671 subsystem_state_change_continue(void *ctx, int status)
672 {
673 	struct spdk_io_channel_iter *i = ctx;
674 	struct nvmf_subsystem_state_change_ctx *_ctx __attribute__((unused));
675 
676 	_ctx = spdk_io_channel_iter_get_ctx(i);
677 	SPDK_DTRACE_PROBE3(nvmf_pg_change_state_done, _ctx->subsystem->subnqn,
678 			   _ctx->requested_state, spdk_thread_get_id(spdk_get_thread()));
679 
680 	spdk_for_each_channel_continue(i, status);
681 }
682 
683 static void
684 subsystem_state_change_on_pg(struct spdk_io_channel_iter *i)
685 {
686 	struct nvmf_subsystem_state_change_ctx *ctx;
687 	struct spdk_io_channel *ch;
688 	struct spdk_nvmf_poll_group *group;
689 
690 	ctx = spdk_io_channel_iter_get_ctx(i);
691 	ch = spdk_io_channel_iter_get_channel(i);
692 	group = spdk_io_channel_get_ctx(ch);
693 
694 	SPDK_DTRACE_PROBE3(nvmf_pg_change_state, ctx->subsystem->subnqn,
695 			   ctx->requested_state, spdk_thread_get_id(spdk_get_thread()));
696 	switch (ctx->requested_state) {
697 	case SPDK_NVMF_SUBSYSTEM_INACTIVE:
698 		nvmf_poll_group_remove_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i);
699 		break;
700 	case SPDK_NVMF_SUBSYSTEM_ACTIVE:
701 		if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_ACTIVATING) {
702 			nvmf_poll_group_add_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i);
703 		} else if (ctx->subsystem->state == SPDK_NVMF_SUBSYSTEM_RESUMING) {
704 			nvmf_poll_group_resume_subsystem(group, ctx->subsystem, subsystem_state_change_continue, i);
705 		}
706 		break;
707 	case SPDK_NVMF_SUBSYSTEM_PAUSED:
708 		nvmf_poll_group_pause_subsystem(group, ctx->subsystem, ctx->nsid, subsystem_state_change_continue,
709 						i);
710 		break;
711 	default:
712 		assert(false);
713 		break;
714 	}
715 }
716 
717 static void
718 nvmf_subsystem_do_state_change(struct nvmf_subsystem_state_change_ctx *ctx)
719 {
720 	struct spdk_nvmf_subsystem *subsystem = ctx->subsystem;
721 	enum spdk_nvmf_subsystem_state intermediate_state;
722 	int rc;
723 
724 	SPDK_DTRACE_PROBE3(nvmf_subsystem_change_state, subsystem->subnqn,
725 			   ctx->requested_state, subsystem->state);
726 
727 	/* If we are already in the requested state, just call the callback immediately. */
728 	if (subsystem->state == ctx->requested_state) {
729 		nvmf_subsystem_state_change_complete(ctx, 0);
730 		return;
731 	}
732 
733 	intermediate_state = nvmf_subsystem_get_intermediate_state(subsystem->state,
734 			     ctx->requested_state);
735 	assert(intermediate_state != SPDK_NVMF_SUBSYSTEM_NUM_STATES);
736 
737 	ctx->original_state = subsystem->state;
738 	rc = nvmf_subsystem_set_state(subsystem, intermediate_state);
739 	if (rc) {
740 		nvmf_subsystem_state_change_complete(ctx, -1);
741 		return;
742 	}
743 
744 	spdk_for_each_channel(subsystem->tgt,
745 			      subsystem_state_change_on_pg,
746 			      ctx,
747 			      subsystem_state_change_done);
748 }
749 
750 
751 static int
752 nvmf_subsystem_state_change(struct spdk_nvmf_subsystem *subsystem,
753 			    uint32_t nsid,
754 			    enum spdk_nvmf_subsystem_state requested_state,
755 			    spdk_nvmf_subsystem_state_change_done cb_fn,
756 			    void *cb_arg)
757 {
758 	struct nvmf_subsystem_state_change_ctx *ctx;
759 	struct spdk_thread *thread;
760 
761 	thread = spdk_get_thread();
762 	if (thread == NULL) {
763 		return -EINVAL;
764 	}
765 
766 	ctx = calloc(1, sizeof(*ctx));
767 	if (!ctx) {
768 		return -ENOMEM;
769 	}
770 
771 	ctx->subsystem = subsystem;
772 	ctx->nsid = nsid;
773 	ctx->requested_state = requested_state;
774 	ctx->cb_fn = cb_fn;
775 	ctx->cb_arg = cb_arg;
776 	ctx->thread = thread;
777 
778 	pthread_mutex_lock(&subsystem->mutex);
779 	TAILQ_INSERT_TAIL(&subsystem->state_changes, ctx, link);
780 	if (ctx != TAILQ_FIRST(&subsystem->state_changes)) {
781 		pthread_mutex_unlock(&subsystem->mutex);
782 		return 0;
783 	}
784 	pthread_mutex_unlock(&subsystem->mutex);
785 
786 	nvmf_subsystem_do_state_change(ctx);
787 
788 	return 0;
789 }
790 
791 int
792 spdk_nvmf_subsystem_start(struct spdk_nvmf_subsystem *subsystem,
793 			  spdk_nvmf_subsystem_state_change_done cb_fn,
794 			  void *cb_arg)
795 {
796 	return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg);
797 }
798 
799 int
800 spdk_nvmf_subsystem_stop(struct spdk_nvmf_subsystem *subsystem,
801 			 spdk_nvmf_subsystem_state_change_done cb_fn,
802 			 void *cb_arg)
803 {
804 	return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_INACTIVE, cb_fn, cb_arg);
805 }
806 
807 int
808 spdk_nvmf_subsystem_pause(struct spdk_nvmf_subsystem *subsystem,
809 			  uint32_t nsid,
810 			  spdk_nvmf_subsystem_state_change_done cb_fn,
811 			  void *cb_arg)
812 {
813 	return nvmf_subsystem_state_change(subsystem, nsid, SPDK_NVMF_SUBSYSTEM_PAUSED, cb_fn, cb_arg);
814 }
815 
816 int
817 spdk_nvmf_subsystem_resume(struct spdk_nvmf_subsystem *subsystem,
818 			   spdk_nvmf_subsystem_state_change_done cb_fn,
819 			   void *cb_arg)
820 {
821 	return nvmf_subsystem_state_change(subsystem, 0, SPDK_NVMF_SUBSYSTEM_ACTIVE, cb_fn, cb_arg);
822 }
823 
824 struct spdk_nvmf_subsystem *
825 spdk_nvmf_subsystem_get_first(struct spdk_nvmf_tgt *tgt)
826 {
827 	return RB_MIN(subsystem_tree, &tgt->subsystems);
828 }
829 
830 struct spdk_nvmf_subsystem *
831 spdk_nvmf_subsystem_get_next(struct spdk_nvmf_subsystem *subsystem)
832 {
833 	if (!subsystem) {
834 		return NULL;
835 	}
836 
837 	return RB_NEXT(subsystem_tree, &tgt->subsystems, subsystem);
838 }
839 
840 static int
841 nvmf_ns_add_host(struct spdk_nvmf_ns *ns, const char *hostnqn)
842 {
843 	struct spdk_nvmf_host *host;
844 
845 	host = calloc(1, sizeof(*host));
846 	if (!host) {
847 		return -ENOMEM;
848 	}
849 	snprintf(host->nqn, sizeof(host->nqn), "%s", hostnqn);
850 	TAILQ_INSERT_HEAD(&ns->hosts, host, link);
851 	return 0;
852 }
853 
854 static void
855 nvmf_ns_remove_host(struct spdk_nvmf_ns *ns, struct spdk_nvmf_host *host)
856 {
857 	TAILQ_REMOVE(&ns->hosts, host, link);
858 	free(host);
859 }
860 
861 static void
862 _async_event_ns_notice(void *_ctrlr)
863 {
864 	struct spdk_nvmf_ctrlr *ctrlr = _ctrlr;
865 
866 	nvmf_ctrlr_async_event_ns_notice(ctrlr);
867 }
868 
869 static void
870 send_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr)
871 {
872 	spdk_thread_send_msg(ctrlr->thread, _async_event_ns_notice, ctrlr);
873 }
874 
875 static int
876 nvmf_ns_visible(struct spdk_nvmf_subsystem *subsystem,
877 		uint32_t nsid,
878 		const char *hostnqn,
879 		bool visible)
880 {
881 	struct spdk_nvmf_ns *ns;
882 	struct spdk_nvmf_ctrlr *ctrlr;
883 	struct spdk_nvmf_host *host;
884 	int rc;
885 
886 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
887 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
888 		assert(false);
889 		return -1;
890 	}
891 
892 	if (hostnqn == NULL || !nvmf_nqn_is_valid(hostnqn)) {
893 		return -EINVAL;
894 	}
895 
896 	if (nsid == 0 || nsid > subsystem->max_nsid) {
897 		return -EINVAL;
898 	}
899 
900 	ns = subsystem->ns[nsid - 1];
901 	if (!ns) {
902 		return -ENOENT;
903 	}
904 
905 	if (ns->always_visible) {
906 		/* No individual host control */
907 		return -EPERM;
908 	}
909 
910 	/* Save host info to use for any future controllers. */
911 	host = nvmf_ns_find_host(ns, hostnqn);
912 	if (visible && host == NULL) {
913 		rc = nvmf_ns_add_host(ns, hostnqn);
914 		if (rc) {
915 			return rc;
916 		}
917 	} else if (!visible && host != NULL) {
918 		nvmf_ns_remove_host(ns, host);
919 	}
920 
921 	/* Also apply to existing controllers. */
922 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
923 		if (strcmp(hostnqn, ctrlr->hostnqn) ||
924 		    nvmf_ctrlr_ns_is_visible(ctrlr, nsid) == visible) {
925 			continue;
926 		}
927 		nvmf_ctrlr_ns_set_visible(ctrlr, nsid, visible);
928 		send_async_event_ns_notice(ctrlr);
929 		nvmf_ctrlr_ns_changed(ctrlr, nsid);
930 	}
931 
932 	return 0;
933 }
934 
935 int
936 spdk_nvmf_ns_add_host(struct spdk_nvmf_subsystem *subsystem,
937 		      uint32_t nsid,
938 		      const char *hostnqn,
939 		      uint32_t flags)
940 {
941 	SPDK_DTRACE_PROBE4(spdk_nvmf_ns_add_host,
942 			   subsystem->subnqn,
943 			   nsid,
944 			   hostnqn,
945 			   flags);
946 	return nvmf_ns_visible(subsystem, nsid, hostnqn, true);
947 }
948 
949 int
950 spdk_nvmf_ns_remove_host(struct spdk_nvmf_subsystem *subsystem,
951 			 uint32_t nsid,
952 			 const char *hostnqn,
953 			 uint32_t flags)
954 {
955 	SPDK_DTRACE_PROBE4(spdk_nvmf_ns_remove_host,
956 			   subsystem->subnqn,
957 			   nsid,
958 			   hostnqn,
959 			   flags);
960 	return nvmf_ns_visible(subsystem, nsid, hostnqn, false);
961 }
962 
963 /* Must hold subsystem->mutex while calling this function */
964 static struct spdk_nvmf_host *
965 nvmf_subsystem_find_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
966 {
967 	struct spdk_nvmf_host *host = NULL;
968 
969 	TAILQ_FOREACH(host, &subsystem->hosts, link) {
970 		if (strcmp(hostnqn, host->nqn) == 0) {
971 			return host;
972 		}
973 	}
974 
975 	return NULL;
976 }
977 
978 int
979 spdk_nvmf_subsystem_add_host_ext(struct spdk_nvmf_subsystem *subsystem,
980 				 const char *hostnqn, struct spdk_nvmf_host_opts *opts)
981 {
982 	struct spdk_nvmf_host *host;
983 	struct spdk_nvmf_transport *transport;
984 	struct spdk_key *key;
985 	int rc;
986 
987 	if (!nvmf_nqn_is_valid(hostnqn)) {
988 		return -EINVAL;
989 	}
990 
991 	pthread_mutex_lock(&subsystem->mutex);
992 
993 	if (nvmf_subsystem_find_host(subsystem, hostnqn)) {
994 		/* This subsystem already allows the specified host. */
995 		pthread_mutex_unlock(&subsystem->mutex);
996 		return -EINVAL;
997 	}
998 
999 	host = calloc(1, sizeof(*host));
1000 	if (!host) {
1001 		pthread_mutex_unlock(&subsystem->mutex);
1002 		return -ENOMEM;
1003 	}
1004 
1005 	key = SPDK_GET_FIELD(opts, dhchap_key, NULL);
1006 	if (key != NULL) {
1007 		if (!nvmf_auth_is_supported()) {
1008 			SPDK_ERRLOG("NVMe in-band authentication is unsupported\n");
1009 			pthread_mutex_unlock(&subsystem->mutex);
1010 			nvmf_host_free(host);
1011 			return -EINVAL;
1012 		}
1013 		host->dhchap_key = spdk_key_dup(key);
1014 		if (host->dhchap_key == NULL) {
1015 			pthread_mutex_unlock(&subsystem->mutex);
1016 			nvmf_host_free(host);
1017 			return -EINVAL;
1018 		}
1019 		key = SPDK_GET_FIELD(opts, dhchap_ctrlr_key, NULL);
1020 		if (key != NULL) {
1021 			host->dhchap_ctrlr_key = spdk_key_dup(key);
1022 			if (host->dhchap_ctrlr_key == NULL) {
1023 				pthread_mutex_unlock(&subsystem->mutex);
1024 				nvmf_host_free(host);
1025 				return -EINVAL;
1026 			}
1027 		}
1028 	} else if (SPDK_GET_FIELD(opts, dhchap_ctrlr_key, NULL) != NULL) {
1029 		SPDK_ERRLOG("DH-HMAC-CHAP controller key requires host key to be set\n");
1030 		pthread_mutex_unlock(&subsystem->mutex);
1031 		nvmf_host_free(host);
1032 		return -EINVAL;
1033 	}
1034 
1035 	snprintf(host->nqn, sizeof(host->nqn), "%s", hostnqn);
1036 
1037 	SPDK_DTRACE_PROBE2(nvmf_subsystem_add_host, subsystem->subnqn, host->nqn);
1038 
1039 	TAILQ_INSERT_HEAD(&subsystem->hosts, host, link);
1040 
1041 	if (!TAILQ_EMPTY(&subsystem->listeners)) {
1042 		spdk_nvmf_send_discovery_log_notice(subsystem->tgt, hostnqn);
1043 	}
1044 
1045 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
1046 	     transport = spdk_nvmf_transport_get_next(transport)) {
1047 		if (transport->ops->subsystem_add_host) {
1048 			rc = transport->ops->subsystem_add_host(transport, subsystem, hostnqn,
1049 								SPDK_GET_FIELD(opts, params, NULL));
1050 			if (rc) {
1051 				SPDK_ERRLOG("Unable to add host to %s transport\n", transport->ops->name);
1052 				/* Remove this host from all transports we've managed to add it to. */
1053 				pthread_mutex_unlock(&subsystem->mutex);
1054 				spdk_nvmf_subsystem_remove_host(subsystem, hostnqn);
1055 				return rc;
1056 			}
1057 		}
1058 	}
1059 
1060 	pthread_mutex_unlock(&subsystem->mutex);
1061 
1062 	return 0;
1063 }
1064 
1065 int
1066 spdk_nvmf_subsystem_add_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
1067 			     const struct spdk_json_val *params)
1068 {
1069 	struct spdk_nvmf_host_opts opts = {};
1070 
1071 	opts.size = SPDK_SIZEOF(&opts, params);
1072 	opts.params = params;
1073 
1074 	return spdk_nvmf_subsystem_add_host_ext(subsystem, hostnqn, &opts);
1075 }
1076 
1077 int
1078 spdk_nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
1079 {
1080 	struct spdk_nvmf_host *host;
1081 	struct spdk_nvmf_transport *transport;
1082 
1083 	pthread_mutex_lock(&subsystem->mutex);
1084 
1085 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1086 	if (host == NULL) {
1087 		pthread_mutex_unlock(&subsystem->mutex);
1088 		return -ENOENT;
1089 	}
1090 
1091 	SPDK_DTRACE_PROBE2(nvmf_subsystem_remove_host, subsystem->subnqn, host->nqn);
1092 
1093 	nvmf_subsystem_remove_host(subsystem, host);
1094 
1095 	if (!TAILQ_EMPTY(&subsystem->listeners)) {
1096 		spdk_nvmf_send_discovery_log_notice(subsystem->tgt, hostnqn);
1097 	}
1098 
1099 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
1100 	     transport = spdk_nvmf_transport_get_next(transport)) {
1101 		if (transport->ops->subsystem_remove_host) {
1102 			transport->ops->subsystem_remove_host(transport, subsystem, hostnqn);
1103 		}
1104 	}
1105 
1106 	pthread_mutex_unlock(&subsystem->mutex);
1107 
1108 	return 0;
1109 }
1110 
1111 int
1112 spdk_nvmf_subsystem_set_keys(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
1113 			     struct spdk_nvmf_subsystem_key_opts *opts)
1114 {
1115 	struct spdk_nvmf_host *host;
1116 	struct spdk_key *key, *ckey;
1117 
1118 	if (!nvmf_auth_is_supported()) {
1119 		SPDK_ERRLOG("NVMe in-band authentication is unsupported\n");
1120 		return -EINVAL;
1121 	}
1122 
1123 	pthread_mutex_lock(&subsystem->mutex);
1124 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1125 	if (host == NULL) {
1126 		pthread_mutex_unlock(&subsystem->mutex);
1127 		return -EINVAL;
1128 	}
1129 
1130 	if (SPDK_GET_FIELD(opts, dhchap_key, host->dhchap_key) == NULL &&
1131 	    SPDK_GET_FIELD(opts, dhchap_ctrlr_key, host->dhchap_ctrlr_key) != NULL) {
1132 		SPDK_ERRLOG("DH-HMAC-CHAP controller key requires host key to be set\n");
1133 		pthread_mutex_unlock(&subsystem->mutex);
1134 		return -EINVAL;
1135 	}
1136 	key = SPDK_GET_FIELD(opts, dhchap_key, NULL);
1137 	if (key != NULL) {
1138 		key = spdk_key_dup(key);
1139 		if (key == NULL) {
1140 			pthread_mutex_unlock(&subsystem->mutex);
1141 			return -EINVAL;
1142 		}
1143 	}
1144 	ckey = SPDK_GET_FIELD(opts, dhchap_ctrlr_key, NULL);
1145 	if (ckey != NULL) {
1146 		ckey = spdk_key_dup(ckey);
1147 		if (ckey == NULL) {
1148 			pthread_mutex_unlock(&subsystem->mutex);
1149 			spdk_keyring_put_key(key);
1150 			return -EINVAL;
1151 		}
1152 	}
1153 	if (SPDK_FIELD_VALID(opts, dhchap_key)) {
1154 		spdk_keyring_put_key(host->dhchap_key);
1155 		host->dhchap_key = key;
1156 	}
1157 	if (SPDK_FIELD_VALID(opts, dhchap_ctrlr_key)) {
1158 		spdk_keyring_put_key(host->dhchap_ctrlr_key);
1159 		host->dhchap_ctrlr_key = ckey;
1160 	}
1161 	pthread_mutex_unlock(&subsystem->mutex);
1162 
1163 	return 0;
1164 }
1165 
1166 struct nvmf_subsystem_disconnect_host_ctx {
1167 	struct spdk_nvmf_subsystem		*subsystem;
1168 	char					*hostnqn;
1169 	spdk_nvmf_tgt_subsystem_listen_done_fn	cb_fn;
1170 	void					*cb_arg;
1171 };
1172 
1173 static void
1174 nvmf_subsystem_disconnect_host_fini(struct spdk_io_channel_iter *i, int status)
1175 {
1176 	struct nvmf_subsystem_disconnect_host_ctx *ctx;
1177 
1178 	ctx = spdk_io_channel_iter_get_ctx(i);
1179 
1180 	if (ctx->cb_fn) {
1181 		ctx->cb_fn(ctx->cb_arg, status);
1182 	}
1183 	free(ctx->hostnqn);
1184 	free(ctx);
1185 }
1186 
1187 static void
1188 nvmf_subsystem_disconnect_qpairs_by_host(struct spdk_io_channel_iter *i)
1189 {
1190 	struct nvmf_subsystem_disconnect_host_ctx *ctx;
1191 	struct spdk_nvmf_poll_group *group;
1192 	struct spdk_io_channel *ch;
1193 	struct spdk_nvmf_qpair *qpair, *tmp_qpair;
1194 	struct spdk_nvmf_ctrlr *ctrlr;
1195 
1196 	ctx = spdk_io_channel_iter_get_ctx(i);
1197 	ch = spdk_io_channel_iter_get_channel(i);
1198 	group = spdk_io_channel_get_ctx(ch);
1199 
1200 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, tmp_qpair) {
1201 		ctrlr = qpair->ctrlr;
1202 
1203 		if (ctrlr == NULL || ctrlr->subsys != ctx->subsystem) {
1204 			continue;
1205 		}
1206 
1207 		if (strncmp(ctrlr->hostnqn, ctx->hostnqn, sizeof(ctrlr->hostnqn)) == 0) {
1208 			/* Right now this does not wait for the queue pairs to actually disconnect. */
1209 			spdk_nvmf_qpair_disconnect(qpair);
1210 		}
1211 	}
1212 	spdk_for_each_channel_continue(i, 0);
1213 }
1214 
1215 int
1216 spdk_nvmf_subsystem_disconnect_host(struct spdk_nvmf_subsystem *subsystem,
1217 				    const char *hostnqn,
1218 				    spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1219 				    void *cb_arg)
1220 {
1221 	struct nvmf_subsystem_disconnect_host_ctx *ctx;
1222 
1223 	ctx = calloc(1, sizeof(struct nvmf_subsystem_disconnect_host_ctx));
1224 	if (ctx == NULL) {
1225 		return -ENOMEM;
1226 	}
1227 
1228 	ctx->hostnqn = strdup(hostnqn);
1229 	if (ctx->hostnqn == NULL) {
1230 		free(ctx);
1231 		return -ENOMEM;
1232 	}
1233 
1234 	ctx->subsystem = subsystem;
1235 	ctx->cb_fn = cb_fn;
1236 	ctx->cb_arg = cb_arg;
1237 
1238 	spdk_for_each_channel(subsystem->tgt, nvmf_subsystem_disconnect_qpairs_by_host, ctx,
1239 			      nvmf_subsystem_disconnect_host_fini);
1240 
1241 	return 0;
1242 }
1243 
1244 int
1245 spdk_nvmf_subsystem_set_allow_any_host(struct spdk_nvmf_subsystem *subsystem, bool allow_any_host)
1246 {
1247 	pthread_mutex_lock(&subsystem->mutex);
1248 	subsystem->allow_any_host = allow_any_host;
1249 	if (!TAILQ_EMPTY(&subsystem->listeners)) {
1250 		spdk_nvmf_send_discovery_log_notice(subsystem->tgt, NULL);
1251 	}
1252 	pthread_mutex_unlock(&subsystem->mutex);
1253 
1254 	return 0;
1255 }
1256 
1257 bool
1258 spdk_nvmf_subsystem_get_allow_any_host(const struct spdk_nvmf_subsystem *subsystem)
1259 {
1260 	bool allow_any_host;
1261 	struct spdk_nvmf_subsystem *sub;
1262 
1263 	/* Technically, taking the mutex modifies data in the subsystem. But the const
1264 	 * is still important to convey that this doesn't mutate any other data. Cast
1265 	 * it away to work around this. */
1266 	sub = (struct spdk_nvmf_subsystem *)subsystem;
1267 
1268 	pthread_mutex_lock(&sub->mutex);
1269 	allow_any_host = sub->allow_any_host;
1270 	pthread_mutex_unlock(&sub->mutex);
1271 
1272 	return allow_any_host;
1273 }
1274 
1275 bool
1276 spdk_nvmf_subsystem_host_allowed(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
1277 {
1278 	bool allowed;
1279 
1280 	if (!hostnqn) {
1281 		return false;
1282 	}
1283 
1284 	pthread_mutex_lock(&subsystem->mutex);
1285 
1286 	if (subsystem->allow_any_host) {
1287 		pthread_mutex_unlock(&subsystem->mutex);
1288 		return true;
1289 	}
1290 
1291 	allowed =  nvmf_subsystem_find_host(subsystem, hostnqn) != NULL;
1292 	pthread_mutex_unlock(&subsystem->mutex);
1293 
1294 	return allowed;
1295 }
1296 
1297 bool
1298 nvmf_subsystem_host_auth_required(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
1299 {
1300 	struct spdk_nvmf_host *host;
1301 	bool status;
1302 
1303 	pthread_mutex_lock(&subsystem->mutex);
1304 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1305 	status = host != NULL && host->dhchap_key != NULL;
1306 	pthread_mutex_unlock(&subsystem->mutex);
1307 
1308 	return status;
1309 }
1310 
1311 struct spdk_key *
1312 nvmf_subsystem_get_dhchap_key(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
1313 			      enum nvmf_auth_key_type type)
1314 {
1315 	struct spdk_nvmf_host *host;
1316 	struct spdk_key *key = NULL;
1317 
1318 	pthread_mutex_lock(&subsystem->mutex);
1319 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1320 	if (host != NULL) {
1321 		switch (type) {
1322 		case NVMF_AUTH_KEY_HOST:
1323 			key = host->dhchap_key;
1324 			break;
1325 		case NVMF_AUTH_KEY_CTRLR:
1326 			key = host->dhchap_ctrlr_key;
1327 			break;
1328 		}
1329 		if (key != NULL) {
1330 			key = spdk_key_dup(key);
1331 		}
1332 	}
1333 	pthread_mutex_unlock(&subsystem->mutex);
1334 
1335 	return key;
1336 }
1337 
1338 struct spdk_nvmf_host *
1339 spdk_nvmf_subsystem_get_first_host(struct spdk_nvmf_subsystem *subsystem)
1340 {
1341 	return TAILQ_FIRST(&subsystem->hosts);
1342 }
1343 
1344 
1345 struct spdk_nvmf_host *
1346 spdk_nvmf_subsystem_get_next_host(struct spdk_nvmf_subsystem *subsystem,
1347 				  struct spdk_nvmf_host *prev_host)
1348 {
1349 	return TAILQ_NEXT(prev_host, link);
1350 }
1351 
1352 const char *
1353 spdk_nvmf_host_get_nqn(const struct spdk_nvmf_host *host)
1354 {
1355 	return host->nqn;
1356 }
1357 
1358 struct spdk_nvmf_subsystem_listener *
1359 nvmf_subsystem_find_listener(struct spdk_nvmf_subsystem *subsystem,
1360 			     const struct spdk_nvme_transport_id *trid)
1361 {
1362 	struct spdk_nvmf_subsystem_listener *listener;
1363 
1364 	TAILQ_FOREACH(listener, &subsystem->listeners, link) {
1365 		if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) {
1366 			return listener;
1367 		}
1368 	}
1369 
1370 	return NULL;
1371 }
1372 
1373 /**
1374  * Function to be called once the target is listening.
1375  *
1376  * \param ctx Context argument passed to this function.
1377  * \param status 0 if it completed successfully, or negative errno if it failed.
1378  */
1379 static void
1380 _nvmf_subsystem_add_listener_done(void *ctx, int status)
1381 {
1382 	struct spdk_nvmf_subsystem_listener *listener = ctx;
1383 
1384 	if (status) {
1385 		listener->cb_fn(listener->cb_arg, status);
1386 		free(listener);
1387 		return;
1388 	}
1389 
1390 	TAILQ_INSERT_HEAD(&listener->subsystem->listeners, listener, link);
1391 
1392 	if (spdk_nvmf_subsystem_is_discovery(listener->subsystem)) {
1393 		status = nvmf_tgt_update_mdns_prr(listener->subsystem->tgt);
1394 		if (status) {
1395 			TAILQ_REMOVE(&listener->subsystem->listeners, listener, link);
1396 			listener->cb_fn(listener->cb_arg, status);
1397 			free(listener);
1398 			return;
1399 		}
1400 	}
1401 
1402 	spdk_nvmf_send_discovery_log_notice(listener->subsystem->tgt, NULL);
1403 	listener->cb_fn(listener->cb_arg, status);
1404 }
1405 
1406 void
1407 spdk_nvmf_subsystem_listener_opts_init(struct spdk_nvmf_listener_opts *opts, size_t size)
1408 {
1409 	if (opts == NULL) {
1410 		SPDK_ERRLOG("opts should not be NULL\n");
1411 		assert(false);
1412 		return;
1413 	}
1414 	if (size == 0) {
1415 		SPDK_ERRLOG("size should not be zero\n");
1416 		assert(false);
1417 		return;
1418 	}
1419 
1420 	memset(opts, 0, size);
1421 	opts->opts_size = size;
1422 
1423 #define FIELD_OK(field) \
1424 	offsetof(struct spdk_nvmf_listener_opts, field) + sizeof(opts->field) <= size
1425 
1426 #define SET_FIELD(field, value) \
1427 	if (FIELD_OK(field)) { \
1428 		opts->field = value; \
1429 	} \
1430 
1431 	SET_FIELD(secure_channel, false);
1432 	SET_FIELD(ana_state, SPDK_NVME_ANA_OPTIMIZED_STATE);
1433 	SET_FIELD(sock_impl, NULL);
1434 
1435 #undef FIELD_OK
1436 #undef SET_FIELD
1437 }
1438 
1439 static int
1440 listener_opts_copy(struct spdk_nvmf_listener_opts *src, struct spdk_nvmf_listener_opts *dst)
1441 {
1442 	if (src->opts_size == 0) {
1443 		SPDK_ERRLOG("source structure size should not be zero\n");
1444 		assert(false);
1445 		return -EINVAL;
1446 	}
1447 
1448 	memset(dst, 0, sizeof(*dst));
1449 	dst->opts_size = src->opts_size;
1450 
1451 #define FIELD_OK(field) \
1452 	offsetof(struct spdk_nvmf_listener_opts, field) + sizeof(src->field) <= src->opts_size
1453 
1454 #define SET_FIELD(field) \
1455 	if (FIELD_OK(field)) { \
1456 		dst->field = src->field; \
1457 	} \
1458 
1459 	SET_FIELD(secure_channel);
1460 	SET_FIELD(ana_state);
1461 	SET_FIELD(sock_impl);
1462 	/* We should not remove this statement, but need to update the assert statement
1463 	 * if we add a new field, and also add a corresponding SET_FIELD statement. */
1464 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listener_opts) == 24, "Incorrect size");
1465 
1466 #undef SET_FIELD
1467 #undef FIELD_OK
1468 
1469 	return 0;
1470 }
1471 
1472 static void
1473 _nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem,
1474 			     struct spdk_nvme_transport_id *trid,
1475 			     spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1476 			     void *cb_arg, struct spdk_nvmf_listener_opts *opts)
1477 {
1478 	struct spdk_nvmf_transport *transport;
1479 	struct spdk_nvmf_subsystem_listener *listener;
1480 	struct spdk_nvmf_listener *tr_listener;
1481 	uint32_t i;
1482 	uint32_t id;
1483 	int rc = 0;
1484 
1485 	assert(cb_fn != NULL);
1486 
1487 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
1488 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
1489 		cb_fn(cb_arg, -EAGAIN);
1490 		return;
1491 	}
1492 
1493 	if (nvmf_subsystem_find_listener(subsystem, trid)) {
1494 		/* Listener already exists in this subsystem */
1495 		cb_fn(cb_arg, 0);
1496 		return;
1497 	}
1498 
1499 	transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, trid->trstring);
1500 	if (!transport) {
1501 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
1502 			    trid->trstring);
1503 		cb_fn(cb_arg, -EINVAL);
1504 		return;
1505 	}
1506 
1507 	tr_listener = nvmf_transport_find_listener(transport, trid);
1508 	if (!tr_listener) {
1509 		SPDK_ERRLOG("Cannot find transport listener for %s\n", trid->traddr);
1510 		cb_fn(cb_arg, -EINVAL);
1511 		return;
1512 	}
1513 
1514 	listener = calloc(1, sizeof(*listener));
1515 	if (!listener) {
1516 		cb_fn(cb_arg, -ENOMEM);
1517 		return;
1518 	}
1519 
1520 	listener->trid = &tr_listener->trid;
1521 	listener->transport = transport;
1522 	listener->cb_fn = cb_fn;
1523 	listener->cb_arg = cb_arg;
1524 	listener->subsystem = subsystem;
1525 	listener->ana_state = calloc(subsystem->max_nsid, sizeof(enum spdk_nvme_ana_state));
1526 	if (!listener->ana_state) {
1527 		free(listener);
1528 		cb_fn(cb_arg, -ENOMEM);
1529 		return;
1530 	}
1531 
1532 	spdk_nvmf_subsystem_listener_opts_init(&listener->opts, sizeof(listener->opts));
1533 	if (opts != NULL) {
1534 		rc = listener_opts_copy(opts, &listener->opts);
1535 		if (rc) {
1536 			SPDK_ERRLOG("Unable to copy listener options\n");
1537 			free(listener->ana_state);
1538 			free(listener);
1539 			cb_fn(cb_arg, -EINVAL);
1540 			return;
1541 		}
1542 	}
1543 
1544 	id = spdk_bit_array_find_first_clear(subsystem->used_listener_ids, 0);
1545 	if (id == UINT32_MAX) {
1546 		SPDK_ERRLOG("Cannot add any more listeners\n");
1547 		free(listener->ana_state);
1548 		free(listener->opts.sock_impl);
1549 		free(listener);
1550 		cb_fn(cb_arg, -EINVAL);
1551 		return;
1552 	}
1553 
1554 	spdk_bit_array_set(subsystem->used_listener_ids, id);
1555 	listener->id = id;
1556 
1557 	for (i = 0; i < subsystem->max_nsid; i++) {
1558 		listener->ana_state[i] = listener->opts.ana_state;
1559 	}
1560 
1561 	if (transport->ops->listen_associate != NULL) {
1562 		rc = transport->ops->listen_associate(transport, subsystem, trid);
1563 	}
1564 
1565 	SPDK_DTRACE_PROBE4(nvmf_subsystem_add_listener, subsystem->subnqn, listener->trid->trtype,
1566 			   listener->trid->traddr, listener->trid->trsvcid);
1567 
1568 	_nvmf_subsystem_add_listener_done(listener, rc);
1569 }
1570 
1571 void
1572 spdk_nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem,
1573 				 struct spdk_nvme_transport_id *trid,
1574 				 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1575 				 void *cb_arg)
1576 {
1577 	_nvmf_subsystem_add_listener(subsystem, trid, cb_fn, cb_arg, NULL);
1578 }
1579 
1580 void
1581 spdk_nvmf_subsystem_add_listener_ext(struct spdk_nvmf_subsystem *subsystem,
1582 				     struct spdk_nvme_transport_id *trid,
1583 				     spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1584 				     void *cb_arg, struct spdk_nvmf_listener_opts *opts)
1585 {
1586 	_nvmf_subsystem_add_listener(subsystem, trid, cb_fn, cb_arg, opts);
1587 }
1588 
1589 int
1590 spdk_nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem,
1591 				    const struct spdk_nvme_transport_id *trid)
1592 {
1593 	struct spdk_nvmf_subsystem_listener *listener;
1594 
1595 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
1596 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
1597 		return -EAGAIN;
1598 	}
1599 
1600 	listener = nvmf_subsystem_find_listener(subsystem, trid);
1601 	if (listener == NULL) {
1602 		return -ENOENT;
1603 	}
1604 
1605 	SPDK_DTRACE_PROBE4(nvmf_subsystem_remove_listener, subsystem->subnqn, listener->trid->trtype,
1606 			   listener->trid->traddr, listener->trid->trsvcid);
1607 
1608 	_nvmf_subsystem_remove_listener(subsystem, listener, false);
1609 
1610 	return 0;
1611 }
1612 
1613 void
1614 nvmf_subsystem_remove_all_listeners(struct spdk_nvmf_subsystem *subsystem,
1615 				    bool stop)
1616 {
1617 	struct spdk_nvmf_subsystem_listener *listener, *listener_tmp;
1618 
1619 	TAILQ_FOREACH_SAFE(listener, &subsystem->listeners, link, listener_tmp) {
1620 		_nvmf_subsystem_remove_listener(subsystem, listener, stop);
1621 	}
1622 }
1623 
1624 bool
1625 spdk_nvmf_subsystem_listener_allowed(struct spdk_nvmf_subsystem *subsystem,
1626 				     const struct spdk_nvme_transport_id *trid)
1627 {
1628 	struct spdk_nvmf_subsystem_listener *listener;
1629 
1630 	TAILQ_FOREACH(listener, &subsystem->listeners, link) {
1631 		if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) {
1632 			return true;
1633 		}
1634 	}
1635 
1636 	if (!strcmp(subsystem->subnqn, SPDK_NVMF_DISCOVERY_NQN)) {
1637 		SPDK_WARNLOG("Allowing connection to discovery subsystem on %s/%s/%s, "
1638 			     "even though this listener was not added to the discovery "
1639 			     "subsystem.  This behavior is deprecated and will be removed "
1640 			     "in a future release.\n",
1641 			     spdk_nvme_transport_id_trtype_str(trid->trtype), trid->traddr, trid->trsvcid);
1642 		return true;
1643 	}
1644 
1645 	return false;
1646 }
1647 
1648 struct spdk_nvmf_subsystem_listener *
1649 spdk_nvmf_subsystem_get_first_listener(struct spdk_nvmf_subsystem *subsystem)
1650 {
1651 	return TAILQ_FIRST(&subsystem->listeners);
1652 }
1653 
1654 struct spdk_nvmf_subsystem_listener *
1655 spdk_nvmf_subsystem_get_next_listener(struct spdk_nvmf_subsystem *subsystem,
1656 				      struct spdk_nvmf_subsystem_listener *prev_listener)
1657 {
1658 	return TAILQ_NEXT(prev_listener, link);
1659 }
1660 
1661 const struct spdk_nvme_transport_id *
1662 spdk_nvmf_subsystem_listener_get_trid(struct spdk_nvmf_subsystem_listener *listener)
1663 {
1664 	return listener->trid;
1665 }
1666 
1667 void
1668 spdk_nvmf_subsystem_allow_any_listener(struct spdk_nvmf_subsystem *subsystem,
1669 				       bool allow_any_listener)
1670 {
1671 	subsystem->flags.allow_any_listener = allow_any_listener;
1672 }
1673 
1674 bool
1675 spdk_nvmf_subsystem_any_listener_allowed(struct spdk_nvmf_subsystem *subsystem)
1676 {
1677 	return subsystem->flags.allow_any_listener;
1678 }
1679 
1680 struct subsystem_update_ns_ctx {
1681 	struct spdk_nvmf_subsystem *subsystem;
1682 
1683 	spdk_nvmf_subsystem_state_change_done cb_fn;
1684 	void *cb_arg;
1685 };
1686 
1687 static void
1688 subsystem_update_ns_done(struct spdk_io_channel_iter *i, int status)
1689 {
1690 	struct subsystem_update_ns_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
1691 
1692 	if (ctx->cb_fn) {
1693 		ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status);
1694 	}
1695 	free(ctx);
1696 }
1697 
1698 static void
1699 subsystem_update_ns_on_pg(struct spdk_io_channel_iter *i)
1700 {
1701 	int rc;
1702 	struct subsystem_update_ns_ctx *ctx;
1703 	struct spdk_nvmf_poll_group *group;
1704 	struct spdk_nvmf_subsystem *subsystem;
1705 
1706 	ctx = spdk_io_channel_iter_get_ctx(i);
1707 	group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i));
1708 	subsystem = ctx->subsystem;
1709 
1710 	rc = nvmf_poll_group_update_subsystem(group, subsystem);
1711 	spdk_for_each_channel_continue(i, rc);
1712 }
1713 
1714 static int
1715 nvmf_subsystem_update_ns(struct spdk_nvmf_subsystem *subsystem,
1716 			 spdk_nvmf_subsystem_state_change_done cb_fn, void *cb_arg)
1717 {
1718 	struct subsystem_update_ns_ctx *ctx;
1719 
1720 	ctx = calloc(1, sizeof(*ctx));
1721 	if (ctx == NULL) {
1722 		SPDK_ERRLOG("Can't alloc subsystem poll group update context\n");
1723 		return -ENOMEM;
1724 	}
1725 	ctx->subsystem = subsystem;
1726 	ctx->cb_fn = cb_fn;
1727 	ctx->cb_arg = cb_arg;
1728 
1729 	spdk_for_each_channel(subsystem->tgt,
1730 			      subsystem_update_ns_on_pg,
1731 			      ctx,
1732 			      subsystem_update_ns_done);
1733 	return 0;
1734 }
1735 
1736 static void
1737 nvmf_subsystem_ns_changed(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
1738 {
1739 	struct spdk_nvmf_ctrlr *ctrlr;
1740 
1741 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1742 		if (nvmf_ctrlr_ns_is_visible(ctrlr, nsid)) {
1743 			nvmf_ctrlr_ns_changed(ctrlr, nsid);
1744 		}
1745 	}
1746 }
1747 
1748 static uint32_t nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns);
1749 
1750 int
1751 spdk_nvmf_subsystem_remove_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
1752 {
1753 	struct spdk_nvmf_transport *transport;
1754 	struct spdk_nvmf_ns *ns;
1755 	struct spdk_nvmf_host *host, *tmp;
1756 	struct spdk_nvmf_ctrlr *ctrlr;
1757 
1758 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
1759 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
1760 		assert(false);
1761 		return -1;
1762 	}
1763 
1764 	if (nsid == 0 || nsid > subsystem->max_nsid) {
1765 		return -1;
1766 	}
1767 
1768 	ns = subsystem->ns[nsid - 1];
1769 	if (!ns) {
1770 		return -1;
1771 	}
1772 
1773 	subsystem->ns[nsid - 1] = NULL;
1774 
1775 	assert(ns->anagrpid - 1 < subsystem->max_nsid);
1776 	assert(subsystem->ana_group[ns->anagrpid - 1] > 0);
1777 
1778 	subsystem->ana_group[ns->anagrpid - 1]--;
1779 
1780 	TAILQ_FOREACH_SAFE(host, &ns->hosts, link, tmp) {
1781 		nvmf_ns_remove_host(ns, host);
1782 	}
1783 
1784 	free(ns->ptpl_file);
1785 	nvmf_ns_reservation_clear_all_registrants(ns);
1786 	spdk_bdev_module_release_bdev(ns->bdev);
1787 	spdk_bdev_close(ns->desc);
1788 	free(ns);
1789 
1790 	if (subsystem->fdp_supported && !spdk_nvmf_subsystem_get_first_ns(subsystem)) {
1791 		subsystem->fdp_supported = false;
1792 		SPDK_DEBUGLOG(nvmf, "Subsystem with id: %u doesn't have FDP capability.\n",
1793 			      subsystem->id);
1794 	}
1795 
1796 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
1797 	     transport = spdk_nvmf_transport_get_next(transport)) {
1798 		if (transport->ops->subsystem_remove_ns) {
1799 			transport->ops->subsystem_remove_ns(transport, subsystem, nsid);
1800 		}
1801 	}
1802 
1803 	nvmf_subsystem_ns_changed(subsystem, nsid);
1804 
1805 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1806 		nvmf_ctrlr_ns_set_visible(ctrlr, nsid, false);
1807 	}
1808 
1809 	return 0;
1810 }
1811 
1812 struct subsystem_ns_change_ctx {
1813 	struct spdk_nvmf_subsystem		*subsystem;
1814 	spdk_nvmf_subsystem_state_change_done	cb_fn;
1815 	uint32_t				nsid;
1816 };
1817 
1818 static void
1819 _nvmf_ns_hot_remove(struct spdk_nvmf_subsystem *subsystem,
1820 		    void *cb_arg, int status)
1821 {
1822 	struct subsystem_ns_change_ctx *ctx = cb_arg;
1823 	int rc;
1824 
1825 	rc = spdk_nvmf_subsystem_remove_ns(subsystem, ctx->nsid);
1826 	if (rc != 0) {
1827 		SPDK_ERRLOG("Failed to make changes to NVME-oF subsystem with id: %u\n", subsystem->id);
1828 	}
1829 
1830 	rc = spdk_nvmf_subsystem_resume(subsystem, NULL, NULL);
1831 	if (rc != 0) {
1832 		SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id);
1833 	}
1834 
1835 	free(ctx);
1836 }
1837 
1838 static void
1839 nvmf_ns_change_msg(void *ns_ctx)
1840 {
1841 	struct subsystem_ns_change_ctx *ctx = ns_ctx;
1842 	int rc;
1843 
1844 	SPDK_DTRACE_PROBE2(nvmf_ns_change, ctx->nsid, ctx->subsystem->subnqn);
1845 
1846 	rc = spdk_nvmf_subsystem_pause(ctx->subsystem, ctx->nsid, ctx->cb_fn, ctx);
1847 	if (rc) {
1848 		if (rc == -EBUSY) {
1849 			/* Try again, this is not a permanent situation. */
1850 			spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ctx);
1851 		} else {
1852 			free(ctx);
1853 			SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n");
1854 		}
1855 	}
1856 }
1857 
1858 static void
1859 nvmf_ns_hot_remove(void *remove_ctx)
1860 {
1861 	struct spdk_nvmf_ns *ns = remove_ctx;
1862 	struct subsystem_ns_change_ctx *ns_ctx;
1863 	int rc;
1864 
1865 	/* We have to allocate a new context because this op
1866 	 * is asynchronous and we could lose the ns in the middle.
1867 	 */
1868 	ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx));
1869 	if (!ns_ctx) {
1870 		SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n");
1871 		return;
1872 	}
1873 
1874 	ns_ctx->subsystem = ns->subsystem;
1875 	ns_ctx->nsid = ns->opts.nsid;
1876 	ns_ctx->cb_fn = _nvmf_ns_hot_remove;
1877 
1878 	rc = spdk_nvmf_subsystem_pause(ns->subsystem, ns_ctx->nsid, _nvmf_ns_hot_remove, ns_ctx);
1879 	if (rc) {
1880 		if (rc == -EBUSY) {
1881 			/* Try again, this is not a permanent situation. */
1882 			spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx);
1883 		} else {
1884 			SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n");
1885 			free(ns_ctx);
1886 		}
1887 	}
1888 }
1889 
1890 static void
1891 _nvmf_ns_resize(struct spdk_nvmf_subsystem *subsystem, void *cb_arg, int status)
1892 {
1893 	struct subsystem_ns_change_ctx *ctx = cb_arg;
1894 
1895 	nvmf_subsystem_ns_changed(subsystem, ctx->nsid);
1896 	if (spdk_nvmf_subsystem_resume(subsystem, NULL, NULL) != 0) {
1897 		SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id);
1898 	}
1899 
1900 	free(ctx);
1901 }
1902 
1903 static void
1904 nvmf_ns_resize(void *event_ctx)
1905 {
1906 	struct spdk_nvmf_ns *ns = event_ctx;
1907 	struct subsystem_ns_change_ctx *ns_ctx;
1908 	int rc;
1909 
1910 	/* We have to allocate a new context because this op
1911 	 * is asynchronous and we could lose the ns in the middle.
1912 	 */
1913 	ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx));
1914 	if (!ns_ctx) {
1915 		SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n");
1916 		return;
1917 	}
1918 
1919 	ns_ctx->subsystem = ns->subsystem;
1920 	ns_ctx->nsid = ns->opts.nsid;
1921 	ns_ctx->cb_fn = _nvmf_ns_resize;
1922 
1923 	/* Specify 0 for the nsid here, because we do not need to pause the namespace.
1924 	 * Namespaces can only be resized bigger, so there is no need to quiesce I/O.
1925 	 */
1926 	rc = spdk_nvmf_subsystem_pause(ns->subsystem, 0, _nvmf_ns_resize, ns_ctx);
1927 	if (rc) {
1928 		if (rc == -EBUSY) {
1929 			/* Try again, this is not a permanent situation. */
1930 			spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx);
1931 		} else {
1932 			SPDK_ERRLOG("Unable to pause subsystem to process namespace resize!\n");
1933 			free(ns_ctx);
1934 		}
1935 	}
1936 }
1937 
1938 static void
1939 nvmf_ns_event(enum spdk_bdev_event_type type,
1940 	      struct spdk_bdev *bdev,
1941 	      void *event_ctx)
1942 {
1943 	SPDK_DEBUGLOG(nvmf, "Bdev event: type %d, name %s, subsystem_id %d, ns_id %d\n",
1944 		      type,
1945 		      spdk_bdev_get_name(bdev),
1946 		      ((struct spdk_nvmf_ns *)event_ctx)->subsystem->id,
1947 		      ((struct spdk_nvmf_ns *)event_ctx)->nsid);
1948 
1949 	switch (type) {
1950 	case SPDK_BDEV_EVENT_REMOVE:
1951 		nvmf_ns_hot_remove(event_ctx);
1952 		break;
1953 	case SPDK_BDEV_EVENT_RESIZE:
1954 		nvmf_ns_resize(event_ctx);
1955 		break;
1956 	default:
1957 		SPDK_NOTICELOG("Unsupported bdev event: type %d\n", type);
1958 		break;
1959 	}
1960 }
1961 
1962 void
1963 spdk_nvmf_ns_opts_get_defaults(struct spdk_nvmf_ns_opts *opts, size_t opts_size)
1964 {
1965 	if (!opts) {
1966 		SPDK_ERRLOG("opts should not be NULL.\n");
1967 		return;
1968 	}
1969 
1970 	if (!opts_size) {
1971 		SPDK_ERRLOG("opts_size should not be zero.\n");
1972 		return;
1973 	}
1974 
1975 	memset(opts, 0, opts_size);
1976 	opts->opts_size = opts_size;
1977 
1978 #define FIELD_OK(field) \
1979 	offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= opts_size
1980 
1981 #define SET_FIELD(field, value) \
1982 	if (FIELD_OK(field)) { \
1983 		opts->field = value; \
1984 	} \
1985 
1986 	/* All current fields are set to 0 by default. */
1987 	SET_FIELD(nsid, 0);
1988 	if (FIELD_OK(nguid)) {
1989 		memset(opts->nguid, 0, sizeof(opts->nguid));
1990 	}
1991 	if (FIELD_OK(eui64)) {
1992 		memset(opts->eui64, 0, sizeof(opts->eui64));
1993 	}
1994 	if (FIELD_OK(uuid)) {
1995 		spdk_uuid_set_null(&opts->uuid);
1996 	}
1997 	SET_FIELD(anagrpid, 0);
1998 	SET_FIELD(transport_specific, NULL);
1999 	SET_FIELD(hide_metadata, false);
2000 
2001 #undef FIELD_OK
2002 #undef SET_FIELD
2003 }
2004 
2005 static void
2006 nvmf_ns_opts_copy(struct spdk_nvmf_ns_opts *opts,
2007 		  const struct spdk_nvmf_ns_opts *user_opts,
2008 		  size_t opts_size)
2009 {
2010 #define FIELD_OK(field)	\
2011 	offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= user_opts->opts_size
2012 
2013 #define SET_FIELD(field) \
2014 	if (FIELD_OK(field)) { \
2015 		opts->field = user_opts->field;	\
2016 	} \
2017 
2018 	SET_FIELD(nsid);
2019 	if (FIELD_OK(nguid)) {
2020 		memcpy(opts->nguid, user_opts->nguid, sizeof(opts->nguid));
2021 	}
2022 	if (FIELD_OK(eui64)) {
2023 		memcpy(opts->eui64, user_opts->eui64, sizeof(opts->eui64));
2024 	}
2025 	if (FIELD_OK(uuid)) {
2026 		spdk_uuid_copy(&opts->uuid, &user_opts->uuid);
2027 	}
2028 	SET_FIELD(anagrpid);
2029 	SET_FIELD(no_auto_visible);
2030 	SET_FIELD(transport_specific);
2031 	SET_FIELD(hide_metadata);
2032 
2033 	opts->opts_size = user_opts->opts_size;
2034 
2035 	/* We should not remove this statement, but need to update the assert statement
2036 	 * if we add a new field, and also add a corresponding SET_FIELD statement.
2037 	 */
2038 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ns_opts) == 73, "Incorrect size");
2039 
2040 #undef FIELD_OK
2041 #undef SET_FIELD
2042 }
2043 
2044 /* Dummy bdev module used to to claim bdevs. */
2045 static struct spdk_bdev_module ns_bdev_module = {
2046 	.name	= "NVMe-oF Target",
2047 };
2048 
2049 static int nvmf_ns_reservation_update(const struct spdk_nvmf_ns *ns,
2050 				      const struct spdk_nvmf_reservation_info *info);
2051 static int nvmf_ns_reservation_load(const struct spdk_nvmf_ns *ns,
2052 				    struct spdk_nvmf_reservation_info *info);
2053 static int nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns,
2054 				       struct spdk_nvmf_reservation_info *info);
2055 
2056 bool
2057 nvmf_subsystem_zone_append_supported(struct spdk_nvmf_subsystem *subsystem)
2058 {
2059 	struct spdk_nvmf_ns *ns;
2060 
2061 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
2062 	     ns != NULL;
2063 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2064 		if (spdk_bdev_is_zoned(ns->bdev) &&
2065 		    spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZONE_APPEND)) {
2066 			return true;
2067 		}
2068 	}
2069 
2070 	return false;
2071 }
2072 
2073 uint32_t
2074 spdk_nvmf_subsystem_add_ns_ext(struct spdk_nvmf_subsystem *subsystem, const char *bdev_name,
2075 			       const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size,
2076 			       const char *ptpl_file)
2077 {
2078 	struct spdk_nvmf_transport *transport;
2079 	struct spdk_nvmf_ns_opts opts;
2080 	struct spdk_bdev_open_opts open_opts = {};
2081 	struct spdk_nvmf_ns *ns, *first_ns;
2082 	struct spdk_nvmf_ctrlr *ctrlr;
2083 	struct spdk_nvmf_reservation_info info = {0};
2084 	int rc;
2085 	bool zone_append_supported;
2086 	uint64_t max_zone_append_size_kib;
2087 
2088 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
2089 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
2090 		return 0;
2091 	}
2092 
2093 	spdk_nvmf_ns_opts_get_defaults(&opts, sizeof(opts));
2094 	if (user_opts) {
2095 		nvmf_ns_opts_copy(&opts, user_opts, opts_size);
2096 	}
2097 
2098 	if (opts.nsid == SPDK_NVME_GLOBAL_NS_TAG) {
2099 		SPDK_ERRLOG("Invalid NSID %" PRIu32 "\n", opts.nsid);
2100 		return 0;
2101 	}
2102 
2103 	if (opts.nsid == 0) {
2104 		/*
2105 		 * NSID not specified - find a free index.
2106 		 *
2107 		 * If no free slots are found, return error.
2108 		 */
2109 		for (opts.nsid = 1; opts.nsid <= subsystem->max_nsid; opts.nsid++) {
2110 			if (_nvmf_subsystem_get_ns(subsystem, opts.nsid) == NULL) {
2111 				break;
2112 			}
2113 		}
2114 		if (opts.nsid > subsystem->max_nsid) {
2115 			SPDK_ERRLOG("No free namespace slot available in the subsystem\n");
2116 			return 0;
2117 		}
2118 	}
2119 
2120 	if (opts.nsid > subsystem->max_nsid) {
2121 		SPDK_ERRLOG("NSID greater than maximum not allowed\n");
2122 		return 0;
2123 	}
2124 
2125 	if (_nvmf_subsystem_get_ns(subsystem, opts.nsid)) {
2126 		SPDK_ERRLOG("Requested NSID %" PRIu32 " already in use\n", opts.nsid);
2127 		return 0;
2128 	}
2129 
2130 	if (opts.anagrpid == 0) {
2131 		opts.anagrpid = opts.nsid;
2132 	}
2133 
2134 	if (opts.anagrpid > subsystem->max_nsid) {
2135 		SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n");
2136 		return 0;
2137 	}
2138 
2139 	ns = calloc(1, sizeof(*ns));
2140 	if (ns == NULL) {
2141 		SPDK_ERRLOG("Namespace allocation failed\n");
2142 		return 0;
2143 	}
2144 
2145 	TAILQ_INIT(&ns->hosts);
2146 	ns->always_visible = !opts.no_auto_visible;
2147 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2148 		nvmf_ctrlr_ns_set_visible(ctrlr, opts.nsid, ns->always_visible);
2149 	}
2150 
2151 	spdk_bdev_open_opts_init(&open_opts, sizeof(open_opts));
2152 	open_opts.hide_metadata = opts.hide_metadata;
2153 
2154 	rc = spdk_bdev_open_ext_v2(bdev_name, true, nvmf_ns_event, ns, &open_opts, &ns->desc);
2155 	if (rc != 0) {
2156 		SPDK_ERRLOG("Subsystem %s: bdev %s cannot be opened, error=%d\n",
2157 			    subsystem->subnqn, bdev_name, rc);
2158 		free(ns);
2159 		return 0;
2160 	}
2161 
2162 	ns->bdev = spdk_bdev_desc_get_bdev(ns->desc);
2163 
2164 	if (spdk_bdev_desc_get_md_size(ns->desc) != 0) {
2165 		if (!spdk_bdev_desc_is_md_interleaved(ns->desc)) {
2166 			SPDK_ERRLOG("Can't attach bdev with separate metadata.\n");
2167 			spdk_bdev_close(ns->desc);
2168 			free(ns);
2169 			return 0;
2170 		}
2171 
2172 		if (spdk_bdev_desc_get_md_size(ns->desc) > SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE) {
2173 			SPDK_ERRLOG("Maximum supported interleaved md size %u, current md size %u\n",
2174 				    SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE,
2175 				    spdk_bdev_desc_get_md_size(ns->desc));
2176 			spdk_bdev_close(ns->desc);
2177 			free(ns);
2178 			return 0;
2179 		}
2180 	}
2181 
2182 	rc = spdk_bdev_module_claim_bdev(ns->bdev, ns->desc, &ns_bdev_module);
2183 	if (rc != 0) {
2184 		spdk_bdev_close(ns->desc);
2185 		free(ns);
2186 		return 0;
2187 	}
2188 
2189 	ns->passthru_nsid = spdk_bdev_get_nvme_nsid(ns->bdev);
2190 	if (subsystem->passthrough && ns->passthru_nsid == 0) {
2191 		SPDK_ERRLOG("Only bdev_nvme namespaces can be added to a passthrough subsystem.\n");
2192 		goto err;
2193 	}
2194 
2195 	/* Cache the zcopy capability of the bdev device */
2196 	ns->zcopy = spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZCOPY);
2197 
2198 	if (spdk_uuid_is_null(&opts.uuid)) {
2199 		opts.uuid = *spdk_bdev_get_uuid(ns->bdev);
2200 	}
2201 
2202 	/* if nguid descriptor is supported by bdev module (nvme) then uuid = nguid */
2203 	if (spdk_mem_all_zero(opts.nguid, sizeof(opts.nguid))) {
2204 		SPDK_STATIC_ASSERT(sizeof(opts.nguid) == sizeof(opts.uuid), "size mismatch");
2205 		memcpy(opts.nguid, spdk_bdev_get_uuid(ns->bdev), sizeof(opts.nguid));
2206 	}
2207 
2208 	if (spdk_bdev_is_zoned(ns->bdev)) {
2209 		SPDK_DEBUGLOG(nvmf, "The added namespace is backed by a zoned block device.\n");
2210 		ns->csi = SPDK_NVME_CSI_ZNS;
2211 
2212 		zone_append_supported = spdk_bdev_io_type_supported(ns->bdev,
2213 					SPDK_BDEV_IO_TYPE_ZONE_APPEND);
2214 		max_zone_append_size_kib = spdk_bdev_get_max_zone_append_size(ns->bdev) *
2215 					   spdk_bdev_desc_get_block_size(ns->desc);
2216 
2217 		if (_nvmf_subsystem_get_first_zoned_ns(subsystem) != NULL &&
2218 		    (nvmf_subsystem_zone_append_supported(subsystem) != zone_append_supported ||
2219 		     subsystem->max_zone_append_size_kib != max_zone_append_size_kib)) {
2220 			SPDK_ERRLOG("Namespaces with different zone append support or different zone append size are not allowed.\n");
2221 			goto err;
2222 		}
2223 
2224 		subsystem->max_zone_append_size_kib = max_zone_append_size_kib;
2225 	}
2226 
2227 	first_ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
2228 	if (!first_ns) {
2229 		if (spdk_bdev_get_nvme_ctratt(ns->bdev).bits.fdps) {
2230 			SPDK_DEBUGLOG(nvmf, "Subsystem with id: %u has FDP capability.\n",
2231 				      subsystem->id);
2232 			subsystem->fdp_supported = true;
2233 		}
2234 	} else {
2235 		if (spdk_bdev_get_nvme_ctratt(first_ns->bdev).bits.fdps !=
2236 		    spdk_bdev_get_nvme_ctratt(ns->bdev).bits.fdps) {
2237 			SPDK_ERRLOG("Subsystem with id: %u can%s FDP namespace.\n", subsystem->id,
2238 				    spdk_bdev_get_nvme_ctratt(first_ns->bdev).bits.fdps ? " only add" : "not add");
2239 			goto err;
2240 		}
2241 	}
2242 
2243 	ns->opts = opts;
2244 	ns->subsystem = subsystem;
2245 	subsystem->ns[opts.nsid - 1] = ns;
2246 	ns->nsid = opts.nsid;
2247 	ns->anagrpid = opts.anagrpid;
2248 	subsystem->ana_group[ns->anagrpid - 1]++;
2249 	TAILQ_INIT(&ns->registrants);
2250 	if (ptpl_file) {
2251 		ns->ptpl_file = strdup(ptpl_file);
2252 		if (!ns->ptpl_file) {
2253 			SPDK_ERRLOG("Namespace ns->ptpl_file allocation failed\n");
2254 			goto err;
2255 		}
2256 	}
2257 
2258 	if (nvmf_ns_is_ptpl_capable(ns)) {
2259 		rc = nvmf_ns_reservation_load(ns, &info);
2260 		if (rc) {
2261 			SPDK_ERRLOG("Subsystem load reservation failed\n");
2262 			goto err;
2263 		}
2264 
2265 		rc = nvmf_ns_reservation_restore(ns, &info);
2266 		if (rc) {
2267 			SPDK_ERRLOG("Subsystem restore reservation failed\n");
2268 			goto err;
2269 		}
2270 	}
2271 
2272 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
2273 	     transport = spdk_nvmf_transport_get_next(transport)) {
2274 		if (transport->ops->subsystem_add_ns) {
2275 			rc = transport->ops->subsystem_add_ns(transport, subsystem, ns);
2276 			if (rc) {
2277 				SPDK_ERRLOG("Namespace attachment is not allowed by %s transport\n", transport->ops->name);
2278 				nvmf_ns_reservation_clear_all_registrants(ns);
2279 				goto err;
2280 			}
2281 		}
2282 	}
2283 
2284 	/* JSON value obj is freed before sending the response. Set NULL to prevent usage of dangling pointer. */
2285 	ns->opts.transport_specific = NULL;
2286 
2287 	SPDK_DEBUGLOG(nvmf, "Subsystem %s: bdev %s assigned nsid %" PRIu32 "\n",
2288 		      spdk_nvmf_subsystem_get_nqn(subsystem),
2289 		      bdev_name,
2290 		      opts.nsid);
2291 
2292 	nvmf_subsystem_ns_changed(subsystem, opts.nsid);
2293 
2294 	SPDK_DTRACE_PROBE2(nvmf_subsystem_add_ns, subsystem->subnqn, ns->nsid);
2295 
2296 	return opts.nsid;
2297 err:
2298 	subsystem->ns[opts.nsid - 1] = NULL;
2299 	spdk_bdev_module_release_bdev(ns->bdev);
2300 	spdk_bdev_close(ns->desc);
2301 	free(ns->ptpl_file);
2302 	free(ns);
2303 
2304 	return 0;
2305 }
2306 
2307 int
2308 spdk_nvmf_subsystem_set_ns_ana_group(struct spdk_nvmf_subsystem *subsystem,
2309 				     uint32_t nsid, uint32_t anagrpid)
2310 {
2311 	struct spdk_nvmf_ns *ns;
2312 
2313 	if (anagrpid > subsystem->max_nsid) {
2314 		SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n");
2315 		return -1;
2316 	}
2317 
2318 	if (anagrpid == 0) {
2319 		SPDK_ERRLOG("Zero is not allowed to ANAGRPID\n");
2320 		return -1;
2321 	}
2322 
2323 	if (nsid == 0 || nsid > subsystem->max_nsid) {
2324 		return -1;
2325 	}
2326 
2327 	ns = subsystem->ns[nsid - 1];
2328 	if (!ns) {
2329 		return -1;
2330 	}
2331 
2332 	assert(ns->anagrpid - 1 < subsystem->max_nsid);
2333 
2334 	assert(subsystem->ana_group[ns->anagrpid - 1] > 0);
2335 
2336 	subsystem->ana_group[ns->anagrpid - 1]--;
2337 
2338 	subsystem->ana_group[anagrpid - 1]++;
2339 
2340 	ns->anagrpid = anagrpid;
2341 	ns->opts.anagrpid = anagrpid;
2342 
2343 	nvmf_subsystem_ns_changed(subsystem, nsid);
2344 
2345 	return 0;
2346 }
2347 
2348 static uint32_t
2349 nvmf_subsystem_get_next_allocated_nsid(struct spdk_nvmf_subsystem *subsystem,
2350 				       uint32_t prev_nsid)
2351 {
2352 	uint32_t nsid;
2353 
2354 	if (prev_nsid >= subsystem->max_nsid) {
2355 		return 0;
2356 	}
2357 
2358 	for (nsid = prev_nsid + 1; nsid <= subsystem->max_nsid; nsid++) {
2359 		if (subsystem->ns[nsid - 1]) {
2360 			return nsid;
2361 		}
2362 	}
2363 
2364 	return 0;
2365 }
2366 
2367 struct spdk_nvmf_ns *
2368 spdk_nvmf_subsystem_get_first_ns(struct spdk_nvmf_subsystem *subsystem)
2369 {
2370 	uint32_t first_nsid;
2371 
2372 	first_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, 0);
2373 	return _nvmf_subsystem_get_ns(subsystem, first_nsid);
2374 }
2375 
2376 struct spdk_nvmf_ns *
2377 spdk_nvmf_subsystem_get_next_ns(struct spdk_nvmf_subsystem *subsystem,
2378 				struct spdk_nvmf_ns *prev_ns)
2379 {
2380 	uint32_t next_nsid;
2381 
2382 	next_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, prev_ns->opts.nsid);
2383 	return _nvmf_subsystem_get_ns(subsystem, next_nsid);
2384 }
2385 
2386 struct spdk_nvmf_ns *
2387 spdk_nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
2388 {
2389 	return _nvmf_subsystem_get_ns(subsystem, nsid);
2390 }
2391 
2392 uint32_t
2393 spdk_nvmf_ns_get_id(const struct spdk_nvmf_ns *ns)
2394 {
2395 	return ns->opts.nsid;
2396 }
2397 
2398 struct spdk_bdev *
2399 spdk_nvmf_ns_get_bdev(struct spdk_nvmf_ns *ns)
2400 {
2401 	return ns->bdev;
2402 }
2403 
2404 void
2405 spdk_nvmf_ns_get_opts(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_ns_opts *opts,
2406 		      size_t opts_size)
2407 {
2408 	memset(opts, 0, opts_size);
2409 	memcpy(opts, &ns->opts, spdk_min(sizeof(ns->opts), opts_size));
2410 }
2411 
2412 const char *
2413 spdk_nvmf_subsystem_get_sn(const struct spdk_nvmf_subsystem *subsystem)
2414 {
2415 	return subsystem->sn;
2416 }
2417 
2418 int
2419 spdk_nvmf_subsystem_set_sn(struct spdk_nvmf_subsystem *subsystem, const char *sn)
2420 {
2421 	size_t len, max_len;
2422 
2423 	max_len = sizeof(subsystem->sn) - 1;
2424 	len = strlen(sn);
2425 	if (len > max_len) {
2426 		SPDK_DEBUGLOG(nvmf, "Invalid sn \"%s\": length %zu > max %zu\n",
2427 			      sn, len, max_len);
2428 		return -1;
2429 	}
2430 
2431 	if (!nvmf_valid_ascii_string(sn, len)) {
2432 		SPDK_DEBUGLOG(nvmf, "Non-ASCII sn\n");
2433 		SPDK_LOGDUMP(nvmf, "sn", sn, len);
2434 		return -1;
2435 	}
2436 
2437 	snprintf(subsystem->sn, sizeof(subsystem->sn), "%s", sn);
2438 
2439 	return 0;
2440 }
2441 
2442 const char *
2443 spdk_nvmf_subsystem_get_mn(const struct spdk_nvmf_subsystem *subsystem)
2444 {
2445 	return subsystem->mn;
2446 }
2447 
2448 int
2449 spdk_nvmf_subsystem_set_mn(struct spdk_nvmf_subsystem *subsystem, const char *mn)
2450 {
2451 	size_t len, max_len;
2452 
2453 	if (mn == NULL) {
2454 		mn = MODEL_NUMBER_DEFAULT;
2455 	}
2456 	max_len = sizeof(subsystem->mn) - 1;
2457 	len = strlen(mn);
2458 	if (len > max_len) {
2459 		SPDK_DEBUGLOG(nvmf, "Invalid mn \"%s\": length %zu > max %zu\n",
2460 			      mn, len, max_len);
2461 		return -1;
2462 	}
2463 
2464 	if (!nvmf_valid_ascii_string(mn, len)) {
2465 		SPDK_DEBUGLOG(nvmf, "Non-ASCII mn\n");
2466 		SPDK_LOGDUMP(nvmf, "mn", mn, len);
2467 		return -1;
2468 	}
2469 
2470 	snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", mn);
2471 
2472 	return 0;
2473 }
2474 
2475 const char *
2476 spdk_nvmf_subsystem_get_nqn(const struct spdk_nvmf_subsystem *subsystem)
2477 {
2478 	return subsystem->subnqn;
2479 }
2480 
2481 /* We have to use the typedef in the function declaration to appease astyle. */
2482 typedef enum spdk_nvmf_subtype spdk_nvmf_subtype_t;
2483 
2484 spdk_nvmf_subtype_t
2485 spdk_nvmf_subsystem_get_type(struct spdk_nvmf_subsystem *subsystem)
2486 {
2487 	return subsystem->subtype;
2488 }
2489 
2490 uint32_t
2491 spdk_nvmf_subsystem_get_max_nsid(struct spdk_nvmf_subsystem *subsystem)
2492 {
2493 	return subsystem->max_nsid;
2494 }
2495 
2496 int
2497 spdk_nvmf_subsystem_set_cntlid_range(struct spdk_nvmf_subsystem *subsystem,
2498 				     uint16_t min_cntlid, uint16_t max_cntlid)
2499 {
2500 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
2501 		return -EAGAIN;
2502 	}
2503 
2504 	if (min_cntlid > max_cntlid) {
2505 		return -EINVAL;
2506 	}
2507 	/* The spec reserves cntlid values in the range FFF0h to FFFFh. */
2508 	if (min_cntlid < NVMF_MIN_CNTLID || min_cntlid > NVMF_MAX_CNTLID ||
2509 	    max_cntlid < NVMF_MIN_CNTLID || max_cntlid > NVMF_MAX_CNTLID) {
2510 		return -EINVAL;
2511 	}
2512 	subsystem->min_cntlid = min_cntlid;
2513 	subsystem->max_cntlid = max_cntlid;
2514 	if (subsystem->next_cntlid < min_cntlid || subsystem->next_cntlid > max_cntlid) {
2515 		subsystem->next_cntlid = min_cntlid;
2516 	}
2517 
2518 	return 0;
2519 }
2520 
2521 uint16_t
2522 nvmf_subsystem_gen_cntlid(struct spdk_nvmf_subsystem *subsystem)
2523 {
2524 	int count;
2525 	uint16_t cntlid;
2526 
2527 	/*
2528 	 * In the worst case, we might have to try all CNTLID values between min_cntlid and max_cntlid
2529 	 * before we find one that is unused (or find that all values are in use).
2530 	 */
2531 	for (count = 0; count < subsystem->max_cntlid - subsystem->min_cntlid + 1; count++) {
2532 		cntlid = subsystem->next_cntlid;
2533 		subsystem->next_cntlid++;
2534 
2535 		if (subsystem->next_cntlid > subsystem->max_cntlid) {
2536 			subsystem->next_cntlid = subsystem->min_cntlid;
2537 		}
2538 
2539 		/* Check if a controller with this cntlid currently exists. */
2540 		if (nvmf_subsystem_get_ctrlr(subsystem, cntlid) == NULL) {
2541 			/* Found unused cntlid */
2542 			return cntlid;
2543 		}
2544 	}
2545 
2546 	/* All valid cntlid values are in use. */
2547 	return 0xFFFF;
2548 }
2549 
2550 int
2551 nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_ctrlr *ctrlr)
2552 {
2553 
2554 	if (ctrlr->dynamic_ctrlr) {
2555 		ctrlr->cntlid = nvmf_subsystem_gen_cntlid(subsystem);
2556 		if (ctrlr->cntlid == 0xFFFF) {
2557 			/* Unable to get a cntlid */
2558 			SPDK_ERRLOG("Reached max simultaneous ctrlrs\n");
2559 			return -EBUSY;
2560 		}
2561 	} else if (nvmf_subsystem_get_ctrlr(subsystem, ctrlr->cntlid) != NULL) {
2562 		SPDK_ERRLOG("Ctrlr with cntlid %u already exist\n", ctrlr->cntlid);
2563 		return -EEXIST;
2564 	}
2565 
2566 	TAILQ_INSERT_TAIL(&subsystem->ctrlrs, ctrlr, link);
2567 
2568 	SPDK_DTRACE_PROBE3(nvmf_subsystem_add_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn);
2569 
2570 	return 0;
2571 }
2572 
2573 void
2574 nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem,
2575 			    struct spdk_nvmf_ctrlr *ctrlr)
2576 {
2577 	SPDK_DTRACE_PROBE3(nvmf_subsystem_remove_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn);
2578 
2579 	assert(spdk_get_thread() == subsystem->thread);
2580 	assert(subsystem == ctrlr->subsys);
2581 	SPDK_DEBUGLOG(nvmf, "remove ctrlr %p id 0x%x from subsys %p %s\n", ctrlr, ctrlr->cntlid, subsystem,
2582 		      subsystem->subnqn);
2583 	TAILQ_REMOVE(&subsystem->ctrlrs, ctrlr, link);
2584 }
2585 
2586 struct spdk_nvmf_ctrlr *
2587 nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, uint16_t cntlid)
2588 {
2589 	struct spdk_nvmf_ctrlr *ctrlr;
2590 
2591 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2592 		if (ctrlr->cntlid == cntlid) {
2593 			return ctrlr;
2594 		}
2595 	}
2596 
2597 	return NULL;
2598 }
2599 
2600 uint32_t
2601 spdk_nvmf_subsystem_get_max_namespaces(const struct spdk_nvmf_subsystem *subsystem)
2602 {
2603 	return subsystem->max_nsid;
2604 }
2605 
2606 uint16_t
2607 spdk_nvmf_subsystem_get_min_cntlid(const struct spdk_nvmf_subsystem *subsystem)
2608 {
2609 	return subsystem->min_cntlid;
2610 }
2611 
2612 uint16_t
2613 spdk_nvmf_subsystem_get_max_cntlid(const struct spdk_nvmf_subsystem *subsystem)
2614 {
2615 	return subsystem->max_cntlid;
2616 }
2617 
2618 struct _nvmf_ns_registrant {
2619 	uint64_t		rkey;
2620 	char			*host_uuid;
2621 };
2622 
2623 struct _nvmf_ns_registrants {
2624 	size_t				num_regs;
2625 	struct _nvmf_ns_registrant	reg[SPDK_NVMF_MAX_NUM_REGISTRANTS];
2626 };
2627 
2628 struct _nvmf_ns_reservation {
2629 	bool					ptpl_activated;
2630 	enum spdk_nvme_reservation_type		rtype;
2631 	uint64_t				crkey;
2632 	char					*bdev_uuid;
2633 	char					*holder_uuid;
2634 	struct _nvmf_ns_registrants		regs;
2635 };
2636 
2637 static const struct spdk_json_object_decoder nvmf_ns_pr_reg_decoders[] = {
2638 	{"rkey", offsetof(struct _nvmf_ns_registrant, rkey), spdk_json_decode_uint64},
2639 	{"host_uuid", offsetof(struct _nvmf_ns_registrant, host_uuid), spdk_json_decode_string},
2640 };
2641 
2642 static int
2643 nvmf_decode_ns_pr_reg(const struct spdk_json_val *val, void *out)
2644 {
2645 	struct _nvmf_ns_registrant *reg = out;
2646 
2647 	return spdk_json_decode_object(val, nvmf_ns_pr_reg_decoders,
2648 				       SPDK_COUNTOF(nvmf_ns_pr_reg_decoders), reg);
2649 }
2650 
2651 static int
2652 nvmf_decode_ns_pr_regs(const struct spdk_json_val *val, void *out)
2653 {
2654 	struct _nvmf_ns_registrants *regs = out;
2655 
2656 	return spdk_json_decode_array(val, nvmf_decode_ns_pr_reg, regs->reg,
2657 				      SPDK_NVMF_MAX_NUM_REGISTRANTS, &regs->num_regs,
2658 				      sizeof(struct _nvmf_ns_registrant));
2659 }
2660 
2661 static const struct spdk_json_object_decoder nvmf_ns_pr_decoders[] = {
2662 	{"ptpl", offsetof(struct _nvmf_ns_reservation, ptpl_activated), spdk_json_decode_bool, true},
2663 	{"rtype", offsetof(struct _nvmf_ns_reservation, rtype), spdk_json_decode_uint32, true},
2664 	{"crkey", offsetof(struct _nvmf_ns_reservation, crkey), spdk_json_decode_uint64, true},
2665 	{"bdev_uuid", offsetof(struct _nvmf_ns_reservation, bdev_uuid), spdk_json_decode_string},
2666 	{"holder_uuid", offsetof(struct _nvmf_ns_reservation, holder_uuid), spdk_json_decode_string, true},
2667 	{"registrants", offsetof(struct _nvmf_ns_reservation, regs), nvmf_decode_ns_pr_regs},
2668 };
2669 
2670 static int
2671 nvmf_ns_reservation_load_json(const struct spdk_nvmf_ns *ns,
2672 			      struct spdk_nvmf_reservation_info *info)
2673 {
2674 	size_t json_size;
2675 	ssize_t values_cnt, rc;
2676 	void *json = NULL, *end;
2677 	struct spdk_json_val *values = NULL;
2678 	struct _nvmf_ns_reservation res = {};
2679 	const char *file = ns->ptpl_file;
2680 	uint32_t i;
2681 
2682 	/* It's not an error if the file does not exist */
2683 	if (access(file, F_OK) != 0) {
2684 		SPDK_DEBUGLOG(nvmf, "File %s does not exist\n", file);
2685 		return 0;
2686 	}
2687 
2688 	/* Load all persist file contents into a local buffer */
2689 	json = spdk_posix_file_load_from_name(file, &json_size);
2690 	if (!json) {
2691 		SPDK_ERRLOG("Load persist file %s failed\n", file);
2692 		return -ENOMEM;
2693 	}
2694 
2695 	rc = spdk_json_parse(json, json_size, NULL, 0, &end, 0);
2696 	if (rc < 0) {
2697 		SPDK_NOTICELOG("Parsing JSON configuration failed (%zd)\n", rc);
2698 		goto exit;
2699 	}
2700 
2701 	values_cnt = rc;
2702 	values = calloc(values_cnt, sizeof(struct spdk_json_val));
2703 	if (values == NULL) {
2704 		goto exit;
2705 	}
2706 
2707 	rc = spdk_json_parse(json, json_size, values, values_cnt, &end, 0);
2708 	if (rc != values_cnt) {
2709 		SPDK_ERRLOG("Parsing JSON configuration failed (%zd)\n", rc);
2710 		goto exit;
2711 	}
2712 
2713 	/* Decode json */
2714 	if (spdk_json_decode_object(values, nvmf_ns_pr_decoders,
2715 				    SPDK_COUNTOF(nvmf_ns_pr_decoders),
2716 				    &res)) {
2717 		SPDK_ERRLOG("Invalid objects in the persist file %s\n", file);
2718 		rc = -EINVAL;
2719 		goto exit;
2720 	}
2721 
2722 	if (res.regs.num_regs > SPDK_NVMF_MAX_NUM_REGISTRANTS) {
2723 		SPDK_ERRLOG("Can only support up to %u registrants\n", SPDK_NVMF_MAX_NUM_REGISTRANTS);
2724 		rc = -ERANGE;
2725 		goto exit;
2726 	}
2727 
2728 	rc = 0;
2729 	info->ptpl_activated = res.ptpl_activated;
2730 	info->rtype = res.rtype;
2731 	info->crkey = res.crkey;
2732 	snprintf(info->bdev_uuid, sizeof(info->bdev_uuid), "%s", res.bdev_uuid);
2733 	snprintf(info->holder_uuid, sizeof(info->holder_uuid), "%s", res.holder_uuid);
2734 	info->num_regs = res.regs.num_regs;
2735 	for (i = 0; i < res.regs.num_regs; i++) {
2736 		info->registrants[i].rkey = res.regs.reg[i].rkey;
2737 		snprintf(info->registrants[i].host_uuid, sizeof(info->registrants[i].host_uuid), "%s",
2738 			 res.regs.reg[i].host_uuid);
2739 	}
2740 
2741 exit:
2742 	free(json);
2743 	free(values);
2744 	free(res.bdev_uuid);
2745 	free(res.holder_uuid);
2746 	for (i = 0; i < res.regs.num_regs; i++) {
2747 		free(res.regs.reg[i].host_uuid);
2748 	}
2749 
2750 	return rc;
2751 }
2752 
2753 static bool nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns);
2754 
2755 static int
2756 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info)
2757 {
2758 	uint32_t i;
2759 	struct spdk_nvmf_registrant *reg, *holder = NULL;
2760 	struct spdk_uuid bdev_uuid, holder_uuid;
2761 	bool rkey_flag = false;
2762 
2763 	SPDK_DEBUGLOG(nvmf, "NSID %u, PTPL %u, Number of registrants %u\n",
2764 		      ns->nsid, info->ptpl_activated, info->num_regs);
2765 
2766 	/* it's not an error */
2767 	if (!info->ptpl_activated || !info->num_regs) {
2768 		return 0;
2769 	}
2770 
2771 	/* Check info->crkey exist or not in info->registrants[i].rkey */
2772 	for (i = 0; i < info->num_regs; i++) {
2773 		if (info->crkey == info->registrants[i].rkey) {
2774 			rkey_flag = true;
2775 		}
2776 	}
2777 	if (!rkey_flag && info->crkey != 0) {
2778 		return -EINVAL;
2779 	}
2780 
2781 	spdk_uuid_parse(&bdev_uuid, info->bdev_uuid);
2782 	if (spdk_uuid_compare(&bdev_uuid, spdk_bdev_get_uuid(ns->bdev))) {
2783 		SPDK_ERRLOG("Existing bdev UUID is not same with configuration file\n");
2784 		return -EINVAL;
2785 	}
2786 
2787 	ns->crkey = info->crkey;
2788 	ns->rtype = info->rtype;
2789 	ns->ptpl_activated = info->ptpl_activated;
2790 	spdk_uuid_parse(&holder_uuid, info->holder_uuid);
2791 
2792 	SPDK_DEBUGLOG(nvmf, "Bdev UUID %s\n", info->bdev_uuid);
2793 	if (info->rtype) {
2794 		SPDK_DEBUGLOG(nvmf, "Holder UUID %s, RTYPE %u, RKEY 0x%"PRIx64"\n",
2795 			      info->holder_uuid, info->rtype, info->crkey);
2796 	}
2797 
2798 	for (i = 0; i < info->num_regs; i++) {
2799 		reg = calloc(1, sizeof(*reg));
2800 		if (!reg) {
2801 			return -ENOMEM;
2802 		}
2803 		spdk_uuid_parse(&reg->hostid, info->registrants[i].host_uuid);
2804 		reg->rkey = info->registrants[i].rkey;
2805 		TAILQ_INSERT_TAIL(&ns->registrants, reg, link);
2806 		if (info->crkey != 0 && !spdk_uuid_compare(&holder_uuid, &reg->hostid)) {
2807 			holder = reg;
2808 		}
2809 		SPDK_DEBUGLOG(nvmf, "Registrant RKEY 0x%"PRIx64", Host UUID %s\n",
2810 			      info->registrants[i].rkey, info->registrants[i].host_uuid);
2811 	}
2812 
2813 	if (nvmf_ns_reservation_all_registrants_type(ns)) {
2814 		ns->holder = TAILQ_FIRST(&ns->registrants);
2815 	} else {
2816 		ns->holder = holder;
2817 	}
2818 
2819 	return 0;
2820 }
2821 
2822 static int
2823 nvmf_ns_json_write_cb(void *cb_ctx, const void *data, size_t size)
2824 {
2825 	char *file = cb_ctx;
2826 	size_t rc;
2827 	FILE *fd;
2828 
2829 	fd = fopen(file, "w");
2830 	if (!fd) {
2831 		SPDK_ERRLOG("Can't open file %s for write\n", file);
2832 		return -ENOENT;
2833 	}
2834 	rc = fwrite(data, 1, size, fd);
2835 	fclose(fd);
2836 
2837 	return rc == size ? 0 : -1;
2838 }
2839 
2840 static int
2841 nvmf_ns_reservation_update_json(const struct spdk_nvmf_ns *ns,
2842 				const struct spdk_nvmf_reservation_info *info)
2843 {
2844 	const char *file = ns->ptpl_file;
2845 	struct spdk_json_write_ctx *w;
2846 	uint32_t i;
2847 	int rc = 0;
2848 
2849 	w = spdk_json_write_begin(nvmf_ns_json_write_cb, (void *)file, 0);
2850 	if (w == NULL) {
2851 		return -ENOMEM;
2852 	}
2853 	/* clear the configuration file */
2854 	if (!info->ptpl_activated) {
2855 		goto exit;
2856 	}
2857 
2858 	spdk_json_write_object_begin(w);
2859 	spdk_json_write_named_bool(w, "ptpl", info->ptpl_activated);
2860 	spdk_json_write_named_uint32(w, "rtype", info->rtype);
2861 	spdk_json_write_named_uint64(w, "crkey", info->crkey);
2862 	spdk_json_write_named_string(w, "bdev_uuid", info->bdev_uuid);
2863 	spdk_json_write_named_string(w, "holder_uuid", info->holder_uuid);
2864 
2865 	spdk_json_write_named_array_begin(w, "registrants");
2866 	for (i = 0; i < info->num_regs; i++) {
2867 		spdk_json_write_object_begin(w);
2868 		spdk_json_write_named_uint64(w, "rkey", info->registrants[i].rkey);
2869 		spdk_json_write_named_string(w, "host_uuid", info->registrants[i].host_uuid);
2870 		spdk_json_write_object_end(w);
2871 	}
2872 	spdk_json_write_array_end(w);
2873 	spdk_json_write_object_end(w);
2874 
2875 exit:
2876 	rc = spdk_json_write_end(w);
2877 	return rc;
2878 }
2879 
2880 static int
2881 nvmf_ns_update_reservation_info(struct spdk_nvmf_ns *ns)
2882 {
2883 	struct spdk_nvmf_reservation_info info;
2884 	struct spdk_nvmf_registrant *reg, *tmp;
2885 	uint32_t i = 0;
2886 
2887 	assert(ns != NULL);
2888 
2889 	if (!ns->bdev || !nvmf_ns_is_ptpl_capable(ns)) {
2890 		return 0;
2891 	}
2892 
2893 	memset(&info, 0, sizeof(info));
2894 	spdk_uuid_fmt_lower(info.bdev_uuid, sizeof(info.bdev_uuid), spdk_bdev_get_uuid(ns->bdev));
2895 
2896 	if (ns->rtype) {
2897 		info.rtype = ns->rtype;
2898 		info.crkey = ns->crkey;
2899 		if (!nvmf_ns_reservation_all_registrants_type(ns)) {
2900 			assert(ns->holder != NULL);
2901 			spdk_uuid_fmt_lower(info.holder_uuid, sizeof(info.holder_uuid), &ns->holder->hostid);
2902 		}
2903 	}
2904 
2905 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2906 		spdk_uuid_fmt_lower(info.registrants[i].host_uuid, sizeof(info.registrants[i].host_uuid),
2907 				    &reg->hostid);
2908 		info.registrants[i++].rkey = reg->rkey;
2909 	}
2910 
2911 	info.num_regs = i;
2912 	info.ptpl_activated = ns->ptpl_activated;
2913 
2914 	return nvmf_ns_reservation_update(ns, &info);
2915 }
2916 
2917 static struct spdk_nvmf_registrant *
2918 nvmf_ns_reservation_get_registrant(struct spdk_nvmf_ns *ns,
2919 				   struct spdk_uuid *uuid)
2920 {
2921 	struct spdk_nvmf_registrant *reg, *tmp;
2922 
2923 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2924 		if (!spdk_uuid_compare(&reg->hostid, uuid)) {
2925 			return reg;
2926 		}
2927 	}
2928 
2929 	return NULL;
2930 }
2931 
2932 /* Generate reservation notice log to registered HostID controllers */
2933 static void
2934 nvmf_subsystem_gen_ctrlr_notification(struct spdk_nvmf_subsystem *subsystem,
2935 				      struct spdk_nvmf_ns *ns,
2936 				      struct spdk_uuid *hostid_list,
2937 				      uint32_t num_hostid,
2938 				      enum spdk_nvme_reservation_notification_log_page_type type)
2939 {
2940 	struct spdk_nvmf_ctrlr *ctrlr;
2941 	uint32_t i;
2942 
2943 	for (i = 0; i < num_hostid; i++) {
2944 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2945 			if (!spdk_uuid_compare(&ctrlr->hostid, &hostid_list[i])) {
2946 				nvmf_ctrlr_reservation_notice_log(ctrlr, ns, type);
2947 			}
2948 		}
2949 	}
2950 }
2951 
2952 /* Get all registrants' hostid other than the controller who issued the command */
2953 static uint32_t
2954 nvmf_ns_reservation_get_all_other_hostid(struct spdk_nvmf_ns *ns,
2955 		struct spdk_uuid *hostid_list,
2956 		uint32_t max_num_hostid,
2957 		struct spdk_uuid *current_hostid)
2958 {
2959 	struct spdk_nvmf_registrant *reg, *tmp;
2960 	uint32_t num_hostid = 0;
2961 
2962 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2963 		if (spdk_uuid_compare(&reg->hostid, current_hostid)) {
2964 			if (num_hostid == max_num_hostid) {
2965 				assert(false);
2966 				return max_num_hostid;
2967 			}
2968 			hostid_list[num_hostid++] = reg->hostid;
2969 		}
2970 	}
2971 
2972 	return num_hostid;
2973 }
2974 
2975 /* Calculate the unregistered HostID list according to list
2976  * prior to execute preempt command and list after executing
2977  * preempt command.
2978  */
2979 static uint32_t
2980 nvmf_ns_reservation_get_unregistered_hostid(struct spdk_uuid *old_hostid_list,
2981 		uint32_t old_num_hostid,
2982 		struct spdk_uuid *remaining_hostid_list,
2983 		uint32_t remaining_num_hostid)
2984 {
2985 	struct spdk_uuid temp_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
2986 	uint32_t i, j, num_hostid = 0;
2987 	bool found;
2988 
2989 	if (!remaining_num_hostid) {
2990 		return old_num_hostid;
2991 	}
2992 
2993 	for (i = 0; i < old_num_hostid; i++) {
2994 		found = false;
2995 		for (j = 0; j < remaining_num_hostid; j++) {
2996 			if (!spdk_uuid_compare(&old_hostid_list[i], &remaining_hostid_list[j])) {
2997 				found = true;
2998 				break;
2999 			}
3000 		}
3001 		if (!found) {
3002 			spdk_uuid_copy(&temp_hostid_list[num_hostid++], &old_hostid_list[i]);
3003 		}
3004 	}
3005 
3006 	if (num_hostid) {
3007 		memcpy(old_hostid_list, temp_hostid_list, sizeof(struct spdk_uuid) * num_hostid);
3008 	}
3009 
3010 	return num_hostid;
3011 }
3012 
3013 /* current reservation type is all registrants or not */
3014 static bool
3015 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns)
3016 {
3017 	return (ns->rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3018 		ns->rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS);
3019 }
3020 
3021 /* current registrant is reservation holder or not */
3022 static bool
3023 nvmf_ns_reservation_registrant_is_holder(struct spdk_nvmf_ns *ns,
3024 		struct spdk_nvmf_registrant *reg)
3025 {
3026 	if (!reg) {
3027 		return false;
3028 	}
3029 
3030 	if (nvmf_ns_reservation_all_registrants_type(ns)) {
3031 		return true;
3032 	}
3033 
3034 	return (ns->holder == reg);
3035 }
3036 
3037 static int
3038 nvmf_ns_reservation_add_registrant(struct spdk_nvmf_ns *ns,
3039 				   struct spdk_nvmf_ctrlr *ctrlr,
3040 				   uint64_t nrkey)
3041 {
3042 	struct spdk_nvmf_registrant *reg;
3043 
3044 	reg = calloc(1, sizeof(*reg));
3045 	if (!reg) {
3046 		return -ENOMEM;
3047 	}
3048 
3049 	reg->rkey = nrkey;
3050 	/* set hostid for the registrant */
3051 	spdk_uuid_copy(&reg->hostid, &ctrlr->hostid);
3052 	TAILQ_INSERT_TAIL(&ns->registrants, reg, link);
3053 	ns->gen++;
3054 
3055 	return 0;
3056 }
3057 
3058 static void
3059 nvmf_ns_reservation_release_reservation(struct spdk_nvmf_ns *ns)
3060 {
3061 	ns->rtype = 0;
3062 	ns->crkey = 0;
3063 	ns->holder = NULL;
3064 }
3065 
3066 /* release the reservation if the last registrant was removed */
3067 static void
3068 nvmf_ns_reservation_check_release_on_remove_registrant(struct spdk_nvmf_ns *ns,
3069 		struct spdk_nvmf_registrant *reg)
3070 {
3071 	struct spdk_nvmf_registrant *next_reg;
3072 
3073 	/* no reservation holder */
3074 	if (!ns->holder) {
3075 		assert(ns->rtype == 0);
3076 		return;
3077 	}
3078 
3079 	next_reg = TAILQ_FIRST(&ns->registrants);
3080 	if (next_reg && nvmf_ns_reservation_all_registrants_type(ns)) {
3081 		/* the next valid registrant is the new holder now */
3082 		ns->holder = next_reg;
3083 	} else if (nvmf_ns_reservation_registrant_is_holder(ns, reg)) {
3084 		/* release the reservation */
3085 		nvmf_ns_reservation_release_reservation(ns);
3086 	}
3087 }
3088 
3089 static void
3090 nvmf_ns_reservation_remove_registrant(struct spdk_nvmf_ns *ns,
3091 				      struct spdk_nvmf_registrant *reg)
3092 {
3093 	TAILQ_REMOVE(&ns->registrants, reg, link);
3094 	nvmf_ns_reservation_check_release_on_remove_registrant(ns, reg);
3095 	free(reg);
3096 	ns->gen++;
3097 	return;
3098 }
3099 
3100 static uint32_t
3101 nvmf_ns_reservation_remove_registrants_by_key(struct spdk_nvmf_ns *ns,
3102 		uint64_t rkey)
3103 {
3104 	struct spdk_nvmf_registrant *reg, *tmp;
3105 	uint32_t count = 0;
3106 
3107 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3108 		if (reg->rkey == rkey) {
3109 			nvmf_ns_reservation_remove_registrant(ns, reg);
3110 			count++;
3111 		}
3112 	}
3113 	return count;
3114 }
3115 
3116 static uint32_t
3117 nvmf_ns_reservation_remove_all_other_registrants(struct spdk_nvmf_ns *ns,
3118 		struct spdk_nvmf_registrant *reg)
3119 {
3120 	struct spdk_nvmf_registrant *reg_tmp, *reg_tmp2;
3121 	uint32_t count = 0;
3122 
3123 	TAILQ_FOREACH_SAFE(reg_tmp, &ns->registrants, link, reg_tmp2) {
3124 		if (reg_tmp != reg) {
3125 			nvmf_ns_reservation_remove_registrant(ns, reg_tmp);
3126 			count++;
3127 		}
3128 	}
3129 	return count;
3130 }
3131 
3132 static uint32_t
3133 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns)
3134 {
3135 	struct spdk_nvmf_registrant *reg, *reg_tmp;
3136 	uint32_t count = 0;
3137 
3138 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, reg_tmp) {
3139 		nvmf_ns_reservation_remove_registrant(ns, reg);
3140 		count++;
3141 	}
3142 	return count;
3143 }
3144 
3145 static void
3146 nvmf_ns_reservation_acquire_reservation(struct spdk_nvmf_ns *ns, uint64_t rkey,
3147 					enum spdk_nvme_reservation_type rtype,
3148 					struct spdk_nvmf_registrant *holder)
3149 {
3150 	ns->rtype = rtype;
3151 	ns->crkey = rkey;
3152 	assert(ns->holder == NULL);
3153 	ns->holder = holder;
3154 }
3155 
3156 static bool
3157 nvmf_ns_reservation_register(struct spdk_nvmf_ns *ns,
3158 			     struct spdk_nvmf_ctrlr *ctrlr,
3159 			     struct spdk_nvmf_request *req)
3160 {
3161 	struct spdk_nvme_reservation_register_data key = { 0 };
3162 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3163 	uint8_t rrega, iekey, cptpl, rtype;
3164 	struct spdk_nvmf_registrant *reg;
3165 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3166 	bool update_sgroup = false;
3167 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3168 	uint32_t num_hostid = 0;
3169 	int rc;
3170 
3171 	rrega = cmd->cdw10_bits.resv_register.rrega;
3172 	iekey = cmd->cdw10_bits.resv_register.iekey;
3173 	cptpl = cmd->cdw10_bits.resv_register.cptpl;
3174 
3175 	if (req->iovcnt > 0 && req->length >= sizeof(key)) {
3176 		struct spdk_iov_xfer ix;
3177 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3178 		spdk_iov_xfer_to_buf(&ix, &key, sizeof(key));
3179 	} else {
3180 		SPDK_ERRLOG("No key provided. Failing request.\n");
3181 		status = SPDK_NVME_SC_INVALID_FIELD;
3182 		goto exit;
3183 	}
3184 
3185 	SPDK_DEBUGLOG(nvmf, "REGISTER: RREGA %u, IEKEY %u, CPTPL %u, "
3186 		      "NRKEY 0x%"PRIx64", NRKEY 0x%"PRIx64"\n",
3187 		      rrega, iekey, cptpl, key.crkey, key.nrkey);
3188 
3189 	if (cptpl == SPDK_NVME_RESERVE_PTPL_CLEAR_POWER_ON) {
3190 		/* True to OFF state, and need to be updated in the configuration file */
3191 		if (ns->ptpl_activated) {
3192 			ns->ptpl_activated = 0;
3193 			update_sgroup = true;
3194 		}
3195 	} else if (cptpl == SPDK_NVME_RESERVE_PTPL_PERSIST_POWER_LOSS) {
3196 		if (!nvmf_ns_is_ptpl_capable(ns)) {
3197 			status = SPDK_NVME_SC_INVALID_FIELD;
3198 			goto exit;
3199 		} else if (ns->ptpl_activated == 0) {
3200 			ns->ptpl_activated = 1;
3201 			update_sgroup = true;
3202 		}
3203 	}
3204 
3205 	/* current Host Identifier has registrant or not */
3206 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3207 
3208 	switch (rrega) {
3209 	case SPDK_NVME_RESERVE_REGISTER_KEY:
3210 		if (!reg) {
3211 			/* register new controller */
3212 			if (key.nrkey == 0) {
3213 				SPDK_ERRLOG("Can't register zeroed new key\n");
3214 				status = SPDK_NVME_SC_INVALID_FIELD;
3215 				goto exit;
3216 			}
3217 			rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey);
3218 			if (rc < 0) {
3219 				status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3220 				goto exit;
3221 			}
3222 			update_sgroup = true;
3223 		} else {
3224 			/* register with same key is not an error */
3225 			if (reg->rkey != key.nrkey) {
3226 				SPDK_ERRLOG("The same host already register a "
3227 					    "key with 0x%"PRIx64"\n",
3228 					    reg->rkey);
3229 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3230 				goto exit;
3231 			}
3232 		}
3233 		break;
3234 	case SPDK_NVME_RESERVE_UNREGISTER_KEY:
3235 		if (!reg || (!iekey && reg->rkey != key.crkey)) {
3236 			SPDK_ERRLOG("No registrant or current key doesn't match "
3237 				    "with existing registrant key\n");
3238 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3239 			goto exit;
3240 		}
3241 
3242 		rtype = ns->rtype;
3243 		num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3244 				SPDK_NVMF_MAX_NUM_REGISTRANTS,
3245 				&ctrlr->hostid);
3246 
3247 		nvmf_ns_reservation_remove_registrant(ns, reg);
3248 
3249 		if (!ns->rtype && num_hostid && (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_REG_ONLY ||
3250 						 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY)) {
3251 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3252 							      hostid_list,
3253 							      num_hostid,
3254 							      SPDK_NVME_RESERVATION_RELEASED);
3255 		}
3256 		update_sgroup = true;
3257 		break;
3258 	case SPDK_NVME_RESERVE_REPLACE_KEY:
3259 		if (key.nrkey == 0) {
3260 			SPDK_ERRLOG("Can't register zeroed new key\n");
3261 			status = SPDK_NVME_SC_INVALID_FIELD;
3262 			goto exit;
3263 		}
3264 		/* Registrant exists */
3265 		if (reg) {
3266 			if (!iekey && reg->rkey != key.crkey) {
3267 				SPDK_ERRLOG("Current key doesn't match "
3268 					    "existing registrant key\n");
3269 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3270 				goto exit;
3271 			}
3272 			if (reg->rkey == key.nrkey) {
3273 				goto exit;
3274 			}
3275 			reg->rkey = key.nrkey;
3276 		} else if (iekey) { /* No registrant but IEKEY is set */
3277 			/* new registrant */
3278 			rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey);
3279 			if (rc < 0) {
3280 				status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3281 				goto exit;
3282 			}
3283 		} else { /* No registrant */
3284 			SPDK_ERRLOG("No registrant\n");
3285 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3286 			goto exit;
3287 
3288 		}
3289 		update_sgroup = true;
3290 		break;
3291 	default:
3292 		status = SPDK_NVME_SC_INVALID_FIELD;
3293 		goto exit;
3294 	}
3295 
3296 exit:
3297 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3298 	req->rsp->nvme_cpl.status.sc = status;
3299 	return update_sgroup;
3300 }
3301 
3302 static bool
3303 nvmf_ns_reservation_acquire(struct spdk_nvmf_ns *ns,
3304 			    struct spdk_nvmf_ctrlr *ctrlr,
3305 			    struct spdk_nvmf_request *req)
3306 {
3307 	struct spdk_nvme_reservation_acquire_data key = { 0 };
3308 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3309 	uint8_t racqa, iekey, rtype;
3310 	struct spdk_nvmf_registrant *reg;
3311 	bool all_regs = false;
3312 	uint32_t count = 0;
3313 	bool update_sgroup = true;
3314 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3315 	uint32_t num_hostid = 0;
3316 	struct spdk_uuid new_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3317 	uint32_t new_num_hostid = 0;
3318 	bool reservation_released = false;
3319 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3320 
3321 	racqa = cmd->cdw10_bits.resv_acquire.racqa;
3322 	iekey = cmd->cdw10_bits.resv_acquire.iekey;
3323 	rtype = cmd->cdw10_bits.resv_acquire.rtype;
3324 
3325 	if (req->iovcnt > 0 && req->length >= sizeof(key)) {
3326 		struct spdk_iov_xfer ix;
3327 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3328 		spdk_iov_xfer_to_buf(&ix, &key, sizeof(key));
3329 	} else {
3330 		SPDK_ERRLOG("No key provided. Failing request.\n");
3331 		status = SPDK_NVME_SC_INVALID_FIELD;
3332 		goto exit;
3333 	}
3334 
3335 	SPDK_DEBUGLOG(nvmf, "ACQUIRE: RACQA %u, IEKEY %u, RTYPE %u, "
3336 		      "NRKEY 0x%"PRIx64", PRKEY 0x%"PRIx64"\n",
3337 		      racqa, iekey, rtype, key.crkey, key.prkey);
3338 
3339 	if (iekey || rtype > SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) {
3340 		SPDK_ERRLOG("Ignore existing key field set to 1\n");
3341 		status = SPDK_NVME_SC_INVALID_FIELD;
3342 		update_sgroup = false;
3343 		goto exit;
3344 	}
3345 
3346 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3347 	/* must be registrant and CRKEY must match */
3348 	if (!reg || reg->rkey != key.crkey) {
3349 		SPDK_ERRLOG("No registrant or current key doesn't match "
3350 			    "with existing registrant key\n");
3351 		status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3352 		update_sgroup = false;
3353 		goto exit;
3354 	}
3355 
3356 	all_regs = nvmf_ns_reservation_all_registrants_type(ns);
3357 
3358 	switch (racqa) {
3359 	case SPDK_NVME_RESERVE_ACQUIRE:
3360 		/* it's not an error for the holder to acquire same reservation type again */
3361 		if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && ns->rtype == rtype) {
3362 			/* do nothing */
3363 			update_sgroup = false;
3364 		} else if (ns->holder == NULL) {
3365 			/* first time to acquire the reservation */
3366 			nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg);
3367 		} else {
3368 			SPDK_ERRLOG("Invalid rtype or current registrant is not holder\n");
3369 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3370 			update_sgroup = false;
3371 			goto exit;
3372 		}
3373 		break;
3374 	case SPDK_NVME_RESERVE_PREEMPT:
3375 		/* no reservation holder */
3376 		if (!ns->holder) {
3377 			/* unregister with PRKEY */
3378 			nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3379 			break;
3380 		}
3381 		num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3382 				SPDK_NVMF_MAX_NUM_REGISTRANTS,
3383 				&ctrlr->hostid);
3384 
3385 		/* only 1 reservation holder and reservation key is valid */
3386 		if (!all_regs) {
3387 			/* preempt itself */
3388 			if (nvmf_ns_reservation_registrant_is_holder(ns, reg) &&
3389 			    ns->crkey == key.prkey) {
3390 				ns->rtype = rtype;
3391 				reservation_released = true;
3392 				break;
3393 			}
3394 
3395 			if (ns->crkey == key.prkey) {
3396 				nvmf_ns_reservation_remove_registrant(ns, ns->holder);
3397 				nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg);
3398 				reservation_released = true;
3399 			} else if (key.prkey != 0) {
3400 				nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3401 			} else {
3402 				/* PRKEY is zero */
3403 				SPDK_ERRLOG("Current PRKEY is zero\n");
3404 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3405 				update_sgroup = false;
3406 				goto exit;
3407 			}
3408 		} else {
3409 			/* release all other registrants except for the current one */
3410 			if (key.prkey == 0) {
3411 				nvmf_ns_reservation_remove_all_other_registrants(ns, reg);
3412 				assert(ns->holder == reg);
3413 			} else {
3414 				count = nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3415 				if (count == 0) {
3416 					SPDK_ERRLOG("PRKEY doesn't match any registrant\n");
3417 					status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3418 					update_sgroup = false;
3419 					goto exit;
3420 				}
3421 			}
3422 		}
3423 		break;
3424 	default:
3425 		status = SPDK_NVME_SC_INVALID_FIELD;
3426 		update_sgroup = false;
3427 		break;
3428 	}
3429 
3430 exit:
3431 	if (update_sgroup && racqa == SPDK_NVME_RESERVE_PREEMPT) {
3432 		new_num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, new_hostid_list,
3433 				 SPDK_NVMF_MAX_NUM_REGISTRANTS,
3434 				 &ctrlr->hostid);
3435 		/* Preempt notification occurs on the unregistered controllers
3436 		 * other than the controller who issued the command.
3437 		 */
3438 		num_hostid = nvmf_ns_reservation_get_unregistered_hostid(hostid_list,
3439 				num_hostid,
3440 				new_hostid_list,
3441 				new_num_hostid);
3442 		if (num_hostid) {
3443 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3444 							      hostid_list,
3445 							      num_hostid,
3446 							      SPDK_NVME_REGISTRATION_PREEMPTED);
3447 
3448 		}
3449 		/* Reservation released notification occurs on the
3450 		 * controllers which are the remaining registrants other than
3451 		 * the controller who issued the command.
3452 		 */
3453 		if (reservation_released && new_num_hostid) {
3454 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3455 							      new_hostid_list,
3456 							      new_num_hostid,
3457 							      SPDK_NVME_RESERVATION_RELEASED);
3458 
3459 		}
3460 	}
3461 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3462 	req->rsp->nvme_cpl.status.sc = status;
3463 	return update_sgroup;
3464 }
3465 
3466 static bool
3467 nvmf_ns_reservation_release(struct spdk_nvmf_ns *ns,
3468 			    struct spdk_nvmf_ctrlr *ctrlr,
3469 			    struct spdk_nvmf_request *req)
3470 {
3471 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3472 	uint8_t rrela, iekey, rtype;
3473 	struct spdk_nvmf_registrant *reg;
3474 	uint64_t crkey = 0;
3475 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3476 	bool update_sgroup = true;
3477 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3478 	uint32_t num_hostid = 0;
3479 
3480 	rrela = cmd->cdw10_bits.resv_release.rrela;
3481 	iekey = cmd->cdw10_bits.resv_release.iekey;
3482 	rtype = cmd->cdw10_bits.resv_release.rtype;
3483 
3484 	if (req->iovcnt > 0 && req->length >= sizeof(crkey)) {
3485 		struct spdk_iov_xfer ix;
3486 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3487 		spdk_iov_xfer_to_buf(&ix, &crkey, sizeof(crkey));
3488 	} else {
3489 		SPDK_ERRLOG("No key provided. Failing request.\n");
3490 		status = SPDK_NVME_SC_INVALID_FIELD;
3491 		goto exit;
3492 	}
3493 
3494 	SPDK_DEBUGLOG(nvmf, "RELEASE: RRELA %u, IEKEY %u, RTYPE %u, "
3495 		      "CRKEY 0x%"PRIx64"\n",  rrela, iekey, rtype, crkey);
3496 
3497 	if (iekey) {
3498 		SPDK_ERRLOG("Ignore existing key field set to 1\n");
3499 		status = SPDK_NVME_SC_INVALID_FIELD;
3500 		update_sgroup = false;
3501 		goto exit;
3502 	}
3503 
3504 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3505 	if (!reg || reg->rkey != crkey) {
3506 		SPDK_ERRLOG("No registrant or current key doesn't match "
3507 			    "with existing registrant key\n");
3508 		status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3509 		update_sgroup = false;
3510 		goto exit;
3511 	}
3512 
3513 	num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3514 			SPDK_NVMF_MAX_NUM_REGISTRANTS,
3515 			&ctrlr->hostid);
3516 
3517 	switch (rrela) {
3518 	case SPDK_NVME_RESERVE_RELEASE:
3519 		if (!ns->holder) {
3520 			SPDK_DEBUGLOG(nvmf, "RELEASE: no holder\n");
3521 			update_sgroup = false;
3522 			goto exit;
3523 		}
3524 		if (ns->rtype != rtype) {
3525 			SPDK_ERRLOG("Type doesn't match\n");
3526 			status = SPDK_NVME_SC_INVALID_FIELD;
3527 			update_sgroup = false;
3528 			goto exit;
3529 		}
3530 		if (!nvmf_ns_reservation_registrant_is_holder(ns, reg)) {
3531 			/* not the reservation holder, this isn't an error */
3532 			update_sgroup = false;
3533 			goto exit;
3534 		}
3535 
3536 		rtype = ns->rtype;
3537 		nvmf_ns_reservation_release_reservation(ns);
3538 
3539 		if (num_hostid && rtype != SPDK_NVME_RESERVE_WRITE_EXCLUSIVE &&
3540 		    rtype != SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3541 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3542 							      hostid_list,
3543 							      num_hostid,
3544 							      SPDK_NVME_RESERVATION_RELEASED);
3545 		}
3546 		break;
3547 	case SPDK_NVME_RESERVE_CLEAR:
3548 		nvmf_ns_reservation_clear_all_registrants(ns);
3549 		if (num_hostid) {
3550 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3551 							      hostid_list,
3552 							      num_hostid,
3553 							      SPDK_NVME_RESERVATION_PREEMPTED);
3554 		}
3555 		break;
3556 	default:
3557 		status = SPDK_NVME_SC_INVALID_FIELD;
3558 		update_sgroup = false;
3559 		goto exit;
3560 	}
3561 
3562 exit:
3563 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3564 	req->rsp->nvme_cpl.status.sc = status;
3565 	return update_sgroup;
3566 }
3567 
3568 static void
3569 nvmf_ns_reservation_report(struct spdk_nvmf_ns *ns,
3570 			   struct spdk_nvmf_ctrlr *ctrlr,
3571 			   struct spdk_nvmf_request *req)
3572 {
3573 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3574 	struct spdk_nvmf_registrant *reg, *tmp;
3575 	struct spdk_nvme_reservation_status_extended_data status_data = { 0 };
3576 	struct spdk_iov_xfer ix;
3577 	uint32_t transfer_len;
3578 	uint32_t regctl = 0;
3579 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3580 
3581 	if (req->iovcnt == 0) {
3582 		SPDK_ERRLOG("No data transfer specified for request. "
3583 			    " Unable to transfer back response.\n");
3584 		status = SPDK_NVME_SC_INVALID_FIELD;
3585 		goto exit;
3586 	}
3587 
3588 	if (!cmd->cdw11_bits.resv_report.eds) {
3589 		SPDK_ERRLOG("NVMeoF uses extended controller data structure, "
3590 			    "please set EDS bit in cdw11 and try again\n");
3591 		status = SPDK_NVME_SC_HOSTID_INCONSISTENT_FORMAT;
3592 		goto exit;
3593 	}
3594 
3595 	/* Number of Dwords of the Reservation Status data structure to transfer */
3596 	transfer_len = (cmd->cdw10 + 1) * sizeof(uint32_t);
3597 
3598 	if (transfer_len < sizeof(struct spdk_nvme_reservation_status_extended_data)) {
3599 		status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3600 		goto exit;
3601 	}
3602 
3603 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3604 
3605 	status_data.data.gen = ns->gen;
3606 	status_data.data.rtype = ns->rtype;
3607 	status_data.data.ptpls = ns->ptpl_activated;
3608 
3609 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3610 		regctl++;
3611 	}
3612 
3613 	/*
3614 	 * We report the number of registrants as per the spec here, even if
3615 	 * the iov isn't big enough to contain them all. In that case, the
3616 	 * spdk_iov_xfer_from_buf() won't actually copy any of the remaining
3617 	 * data; as it keeps track of the iov cursor itself, it's simplest to
3618 	 * just walk the entire list anyway.
3619 	 */
3620 	status_data.data.regctl = regctl;
3621 
3622 	spdk_iov_xfer_from_buf(&ix, &status_data, sizeof(status_data));
3623 
3624 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3625 		struct spdk_nvme_registered_ctrlr_extended_data ctrlr_data = { 0 };
3626 
3627 		/* Set to 0xffffh for dynamic controller */
3628 		ctrlr_data.cntlid = 0xffff;
3629 		ctrlr_data.rcsts.status = (ns->holder == reg) ? true : false;
3630 		ctrlr_data.rkey = reg->rkey;
3631 		spdk_uuid_copy((struct spdk_uuid *)ctrlr_data.hostid, &reg->hostid);
3632 
3633 		spdk_iov_xfer_from_buf(&ix, &ctrlr_data, sizeof(ctrlr_data));
3634 	}
3635 
3636 exit:
3637 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3638 	req->rsp->nvme_cpl.status.sc = status;
3639 	return;
3640 }
3641 
3642 static void
3643 nvmf_ns_reservation_complete(void *ctx)
3644 {
3645 	struct spdk_nvmf_request *req = ctx;
3646 
3647 	spdk_nvmf_request_complete(req);
3648 }
3649 
3650 static void
3651 _nvmf_ns_reservation_update_done(struct spdk_nvmf_subsystem *subsystem,
3652 				 void *cb_arg, int status)
3653 {
3654 	struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)cb_arg;
3655 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3656 
3657 	spdk_thread_send_msg(group->thread, nvmf_ns_reservation_complete, req);
3658 }
3659 
3660 void
3661 nvmf_ns_reservation_request(void *ctx)
3662 {
3663 	struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)ctx;
3664 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3665 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3666 	uint32_t nsid;
3667 	struct spdk_nvmf_ns *ns;
3668 	bool update_sgroup = false;
3669 	int status = 0;
3670 
3671 	nsid = cmd->nsid;
3672 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3673 	assert(ns != NULL);
3674 
3675 	switch (cmd->opc) {
3676 	case SPDK_NVME_OPC_RESERVATION_REGISTER:
3677 		update_sgroup = nvmf_ns_reservation_register(ns, ctrlr, req);
3678 		break;
3679 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3680 		update_sgroup = nvmf_ns_reservation_acquire(ns, ctrlr, req);
3681 		break;
3682 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3683 		update_sgroup = nvmf_ns_reservation_release(ns, ctrlr, req);
3684 		break;
3685 	case SPDK_NVME_OPC_RESERVATION_REPORT:
3686 		nvmf_ns_reservation_report(ns, ctrlr, req);
3687 		break;
3688 	default:
3689 		break;
3690 	}
3691 
3692 	/* update reservation information to subsystem's poll group */
3693 	if (update_sgroup) {
3694 		if (ns->ptpl_activated || cmd->opc == SPDK_NVME_OPC_RESERVATION_REGISTER) {
3695 			if (nvmf_ns_update_reservation_info(ns) != 0) {
3696 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3697 			}
3698 		}
3699 		status = nvmf_subsystem_update_ns(ctrlr->subsys, _nvmf_ns_reservation_update_done, req);
3700 		if (status == 0) {
3701 			return;
3702 		}
3703 	}
3704 
3705 	_nvmf_ns_reservation_update_done(ctrlr->subsys, req, status);
3706 }
3707 
3708 static bool
3709 nvmf_ns_is_ptpl_capable_json(const struct spdk_nvmf_ns *ns)
3710 {
3711 	return ns->ptpl_file != NULL;
3712 }
3713 
3714 static struct spdk_nvmf_ns_reservation_ops g_reservation_ops = {
3715 	.is_ptpl_capable = nvmf_ns_is_ptpl_capable_json,
3716 	.update = nvmf_ns_reservation_update_json,
3717 	.load = nvmf_ns_reservation_load_json,
3718 };
3719 
3720 bool
3721 nvmf_ns_is_ptpl_capable(const struct spdk_nvmf_ns *ns)
3722 {
3723 	return g_reservation_ops.is_ptpl_capable(ns);
3724 }
3725 
3726 static int
3727 nvmf_ns_reservation_update(const struct spdk_nvmf_ns *ns,
3728 			   const struct spdk_nvmf_reservation_info *info)
3729 {
3730 	return g_reservation_ops.update(ns, info);
3731 }
3732 
3733 static int
3734 nvmf_ns_reservation_load(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info)
3735 {
3736 	return g_reservation_ops.load(ns, info);
3737 }
3738 
3739 void
3740 spdk_nvmf_set_custom_ns_reservation_ops(const struct spdk_nvmf_ns_reservation_ops *ops)
3741 {
3742 	g_reservation_ops = *ops;
3743 }
3744 
3745 int
3746 spdk_nvmf_subsystem_set_ana_reporting(struct spdk_nvmf_subsystem *subsystem,
3747 				      bool ana_reporting)
3748 {
3749 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
3750 		return -EAGAIN;
3751 	}
3752 
3753 	subsystem->flags.ana_reporting = ana_reporting;
3754 
3755 	return 0;
3756 }
3757 
3758 bool
3759 spdk_nvmf_subsystem_get_ana_reporting(struct spdk_nvmf_subsystem *subsystem)
3760 {
3761 	return subsystem->flags.ana_reporting;
3762 }
3763 
3764 struct subsystem_listener_update_ctx {
3765 	struct spdk_nvmf_subsystem_listener *listener;
3766 
3767 	spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn;
3768 	void *cb_arg;
3769 };
3770 
3771 static void
3772 subsystem_listener_update_done(struct spdk_io_channel_iter *i, int status)
3773 {
3774 	struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
3775 
3776 	if (ctx->cb_fn) {
3777 		ctx->cb_fn(ctx->cb_arg, status);
3778 	}
3779 	free(ctx);
3780 }
3781 
3782 static void
3783 subsystem_listener_update_on_pg(struct spdk_io_channel_iter *i)
3784 {
3785 	struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
3786 	struct spdk_nvmf_subsystem_listener *listener;
3787 	struct spdk_nvmf_poll_group *group;
3788 	struct spdk_nvmf_ctrlr *ctrlr;
3789 
3790 	listener = ctx->listener;
3791 	group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i));
3792 
3793 	TAILQ_FOREACH(ctrlr, &listener->subsystem->ctrlrs, link) {
3794 		if (ctrlr->thread != spdk_get_thread()) {
3795 			continue;
3796 		}
3797 
3798 		if (ctrlr->admin_qpair && ctrlr->admin_qpair->group == group && ctrlr->listener == listener) {
3799 			nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
3800 		}
3801 	}
3802 
3803 	spdk_for_each_channel_continue(i, 0);
3804 }
3805 
3806 void
3807 spdk_nvmf_subsystem_set_ana_state(struct spdk_nvmf_subsystem *subsystem,
3808 				  const struct spdk_nvme_transport_id *trid,
3809 				  enum spdk_nvme_ana_state ana_state, uint32_t anagrpid,
3810 				  spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, void *cb_arg)
3811 {
3812 	struct spdk_nvmf_subsystem_listener *listener;
3813 	struct subsystem_listener_update_ctx *ctx;
3814 	uint32_t i;
3815 
3816 	assert(cb_fn != NULL);
3817 	assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
3818 	       subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED);
3819 
3820 	if (!subsystem->flags.ana_reporting) {
3821 		SPDK_ERRLOG("ANA reporting is disabled\n");
3822 		cb_fn(cb_arg, -EINVAL);
3823 		return;
3824 	}
3825 
3826 	/* ANA Change state is not used, ANA Persistent Loss state
3827 	 * is not supported yet.
3828 	 */
3829 	if (!(ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE ||
3830 	      ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE ||
3831 	      ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE)) {
3832 		SPDK_ERRLOG("ANA state %d is not supported\n", ana_state);
3833 		cb_fn(cb_arg, -ENOTSUP);
3834 		return;
3835 	}
3836 
3837 	if (anagrpid > subsystem->max_nsid) {
3838 		SPDK_ERRLOG("ANA group ID %" PRIu32 " is more than maximum\n", anagrpid);
3839 		cb_fn(cb_arg, -EINVAL);
3840 		return;
3841 	}
3842 
3843 	listener = nvmf_subsystem_find_listener(subsystem, trid);
3844 	if (!listener) {
3845 		SPDK_ERRLOG("Unable to find listener.\n");
3846 		cb_fn(cb_arg, -EINVAL);
3847 		return;
3848 	}
3849 
3850 	if (anagrpid != 0 && listener->ana_state[anagrpid - 1] == ana_state) {
3851 		cb_fn(cb_arg, 0);
3852 		return;
3853 	}
3854 
3855 	ctx = calloc(1, sizeof(*ctx));
3856 	if (!ctx) {
3857 		SPDK_ERRLOG("Unable to allocate context\n");
3858 		cb_fn(cb_arg, -ENOMEM);
3859 		return;
3860 	}
3861 
3862 	for (i = 1; i <= subsystem->max_nsid; i++) {
3863 		if (anagrpid == 0 || i == anagrpid) {
3864 			listener->ana_state[i - 1] = ana_state;
3865 		}
3866 	}
3867 	listener->ana_state_change_count++;
3868 
3869 	ctx->listener = listener;
3870 	ctx->cb_fn = cb_fn;
3871 	ctx->cb_arg = cb_arg;
3872 
3873 	spdk_for_each_channel(subsystem->tgt,
3874 			      subsystem_listener_update_on_pg,
3875 			      ctx,
3876 			      subsystem_listener_update_done);
3877 }
3878 
3879 int
3880 spdk_nvmf_subsystem_get_ana_state(struct spdk_nvmf_subsystem *subsystem,
3881 				  const struct spdk_nvme_transport_id *trid,
3882 				  uint32_t anagrpid,
3883 				  enum spdk_nvme_ana_state *ana_state)
3884 {
3885 	assert(ana_state != NULL);
3886 
3887 	struct spdk_nvmf_subsystem_listener *listener;
3888 
3889 	if (!subsystem->flags.ana_reporting) {
3890 		SPDK_ERRLOG("ANA reporting is disabled\n");
3891 		return -EINVAL;
3892 	}
3893 
3894 	if (anagrpid <= 0 || anagrpid > subsystem->max_nsid) {
3895 		SPDK_ERRLOG("ANA group ID %" PRIu32 " is invalid\n", anagrpid);
3896 		return -EINVAL;
3897 	}
3898 
3899 	listener = nvmf_subsystem_find_listener(subsystem, trid);
3900 	if (!listener) {
3901 		SPDK_ERRLOG("Unable to find listener.\n");
3902 		return -EINVAL;
3903 	}
3904 
3905 	*ana_state = listener->ana_state[anagrpid - 1];
3906 	return 0;
3907 }
3908 
3909 bool
3910 spdk_nvmf_subsystem_is_discovery(struct spdk_nvmf_subsystem *subsystem)
3911 {
3912 	return subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
3913 	       subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY;
3914 }
3915 
3916 bool
3917 nvmf_nqn_is_discovery(const char *nqn)
3918 {
3919 	return strcmp(nqn, SPDK_NVMF_DISCOVERY_NQN) == 0;
3920 }
3921