xref: /spdk/lib/nvmf/subsystem.c (revision 1794c395e6be4a98b357a15c8f5d5213dc40ebfe)
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 
2000 #undef FIELD_OK
2001 #undef SET_FIELD
2002 }
2003 
2004 static void
2005 nvmf_ns_opts_copy(struct spdk_nvmf_ns_opts *opts,
2006 		  const struct spdk_nvmf_ns_opts *user_opts,
2007 		  size_t opts_size)
2008 {
2009 #define FIELD_OK(field)	\
2010 	offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= user_opts->opts_size
2011 
2012 #define SET_FIELD(field) \
2013 	if (FIELD_OK(field)) { \
2014 		opts->field = user_opts->field;	\
2015 	} \
2016 
2017 	SET_FIELD(nsid);
2018 	if (FIELD_OK(nguid)) {
2019 		memcpy(opts->nguid, user_opts->nguid, sizeof(opts->nguid));
2020 	}
2021 	if (FIELD_OK(eui64)) {
2022 		memcpy(opts->eui64, user_opts->eui64, sizeof(opts->eui64));
2023 	}
2024 	if (FIELD_OK(uuid)) {
2025 		spdk_uuid_copy(&opts->uuid, &user_opts->uuid);
2026 	}
2027 	SET_FIELD(anagrpid);
2028 	SET_FIELD(no_auto_visible);
2029 	SET_FIELD(transport_specific);
2030 
2031 	opts->opts_size = user_opts->opts_size;
2032 
2033 	/* We should not remove this statement, but need to update the assert statement
2034 	 * if we add a new field, and also add a corresponding SET_FIELD statement.
2035 	 */
2036 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ns_opts) == 72, "Incorrect size");
2037 
2038 #undef FIELD_OK
2039 #undef SET_FIELD
2040 }
2041 
2042 /* Dummy bdev module used to to claim bdevs. */
2043 static struct spdk_bdev_module ns_bdev_module = {
2044 	.name	= "NVMe-oF Target",
2045 };
2046 
2047 static int nvmf_ns_reservation_update(const struct spdk_nvmf_ns *ns,
2048 				      const struct spdk_nvmf_reservation_info *info);
2049 static int nvmf_ns_reservation_load(const struct spdk_nvmf_ns *ns,
2050 				    struct spdk_nvmf_reservation_info *info);
2051 static int nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns,
2052 				       struct spdk_nvmf_reservation_info *info);
2053 
2054 bool
2055 nvmf_subsystem_zone_append_supported(struct spdk_nvmf_subsystem *subsystem)
2056 {
2057 	struct spdk_nvmf_ns *ns;
2058 
2059 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
2060 	     ns != NULL;
2061 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2062 		if (spdk_bdev_is_zoned(ns->bdev) &&
2063 		    spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZONE_APPEND)) {
2064 			return true;
2065 		}
2066 	}
2067 
2068 	return false;
2069 }
2070 
2071 uint32_t
2072 spdk_nvmf_subsystem_add_ns_ext(struct spdk_nvmf_subsystem *subsystem, const char *bdev_name,
2073 			       const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size,
2074 			       const char *ptpl_file)
2075 {
2076 	struct spdk_nvmf_transport *transport;
2077 	struct spdk_nvmf_ns_opts opts;
2078 	struct spdk_nvmf_ns *ns, *first_ns;
2079 	struct spdk_nvmf_ctrlr *ctrlr;
2080 	struct spdk_nvmf_reservation_info info = {0};
2081 	int rc;
2082 	bool zone_append_supported;
2083 	uint64_t max_zone_append_size_kib;
2084 
2085 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
2086 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
2087 		return 0;
2088 	}
2089 
2090 	spdk_nvmf_ns_opts_get_defaults(&opts, sizeof(opts));
2091 	if (user_opts) {
2092 		nvmf_ns_opts_copy(&opts, user_opts, opts_size);
2093 	}
2094 
2095 	if (opts.nsid == SPDK_NVME_GLOBAL_NS_TAG) {
2096 		SPDK_ERRLOG("Invalid NSID %" PRIu32 "\n", opts.nsid);
2097 		return 0;
2098 	}
2099 
2100 	if (opts.nsid == 0) {
2101 		/*
2102 		 * NSID not specified - find a free index.
2103 		 *
2104 		 * If no free slots are found, return error.
2105 		 */
2106 		for (opts.nsid = 1; opts.nsid <= subsystem->max_nsid; opts.nsid++) {
2107 			if (_nvmf_subsystem_get_ns(subsystem, opts.nsid) == NULL) {
2108 				break;
2109 			}
2110 		}
2111 		if (opts.nsid > subsystem->max_nsid) {
2112 			SPDK_ERRLOG("No free namespace slot available in the subsystem\n");
2113 			return 0;
2114 		}
2115 	}
2116 
2117 	if (opts.nsid > subsystem->max_nsid) {
2118 		SPDK_ERRLOG("NSID greater than maximum not allowed\n");
2119 		return 0;
2120 	}
2121 
2122 	if (_nvmf_subsystem_get_ns(subsystem, opts.nsid)) {
2123 		SPDK_ERRLOG("Requested NSID %" PRIu32 " already in use\n", opts.nsid);
2124 		return 0;
2125 	}
2126 
2127 	if (opts.anagrpid == 0) {
2128 		opts.anagrpid = opts.nsid;
2129 	}
2130 
2131 	if (opts.anagrpid > subsystem->max_nsid) {
2132 		SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n");
2133 		return 0;
2134 	}
2135 
2136 	ns = calloc(1, sizeof(*ns));
2137 	if (ns == NULL) {
2138 		SPDK_ERRLOG("Namespace allocation failed\n");
2139 		return 0;
2140 	}
2141 
2142 	TAILQ_INIT(&ns->hosts);
2143 	ns->always_visible = !opts.no_auto_visible;
2144 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2145 		nvmf_ctrlr_ns_set_visible(ctrlr, opts.nsid, ns->always_visible);
2146 	}
2147 
2148 	rc = spdk_bdev_open_ext(bdev_name, true, nvmf_ns_event, ns, &ns->desc);
2149 	if (rc != 0) {
2150 		SPDK_ERRLOG("Subsystem %s: bdev %s cannot be opened, error=%d\n",
2151 			    subsystem->subnqn, bdev_name, rc);
2152 		free(ns);
2153 		return 0;
2154 	}
2155 
2156 	ns->bdev = spdk_bdev_desc_get_bdev(ns->desc);
2157 
2158 	if (spdk_bdev_get_md_size(ns->bdev) != 0) {
2159 		if (!spdk_bdev_is_md_interleaved(ns->bdev)) {
2160 			SPDK_ERRLOG("Can't attach bdev with separate metadata.\n");
2161 			spdk_bdev_close(ns->desc);
2162 			free(ns);
2163 			return 0;
2164 		}
2165 
2166 		if (spdk_bdev_get_md_size(ns->bdev) > SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE) {
2167 			SPDK_ERRLOG("Maximum supported interleaved md size %u, current md size %u\n",
2168 				    SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE, spdk_bdev_get_md_size(ns->bdev));
2169 			spdk_bdev_close(ns->desc);
2170 			free(ns);
2171 			return 0;
2172 		}
2173 	}
2174 
2175 	rc = spdk_bdev_module_claim_bdev(ns->bdev, ns->desc, &ns_bdev_module);
2176 	if (rc != 0) {
2177 		spdk_bdev_close(ns->desc);
2178 		free(ns);
2179 		return 0;
2180 	}
2181 
2182 	ns->passthru_nsid = spdk_bdev_get_nvme_nsid(ns->bdev);
2183 	if (subsystem->passthrough && ns->passthru_nsid == 0) {
2184 		SPDK_ERRLOG("Only bdev_nvme namespaces can be added to a passthrough subsystem.\n");
2185 		goto err;
2186 	}
2187 
2188 	/* Cache the zcopy capability of the bdev device */
2189 	ns->zcopy = spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZCOPY);
2190 
2191 	if (spdk_uuid_is_null(&opts.uuid)) {
2192 		opts.uuid = *spdk_bdev_get_uuid(ns->bdev);
2193 	}
2194 
2195 	/* if nguid descriptor is supported by bdev module (nvme) then uuid = nguid */
2196 	if (spdk_mem_all_zero(opts.nguid, sizeof(opts.nguid))) {
2197 		SPDK_STATIC_ASSERT(sizeof(opts.nguid) == sizeof(opts.uuid), "size mismatch");
2198 		memcpy(opts.nguid, spdk_bdev_get_uuid(ns->bdev), sizeof(opts.nguid));
2199 	}
2200 
2201 	if (spdk_bdev_is_zoned(ns->bdev)) {
2202 		SPDK_DEBUGLOG(nvmf, "The added namespace is backed by a zoned block device.\n");
2203 		ns->csi = SPDK_NVME_CSI_ZNS;
2204 
2205 		zone_append_supported = spdk_bdev_io_type_supported(ns->bdev,
2206 					SPDK_BDEV_IO_TYPE_ZONE_APPEND);
2207 		max_zone_append_size_kib = spdk_bdev_get_max_zone_append_size(
2208 						   ns->bdev) * spdk_bdev_get_block_size(ns->bdev);
2209 
2210 		if (_nvmf_subsystem_get_first_zoned_ns(subsystem) != NULL &&
2211 		    (nvmf_subsystem_zone_append_supported(subsystem) != zone_append_supported ||
2212 		     subsystem->max_zone_append_size_kib != max_zone_append_size_kib)) {
2213 			SPDK_ERRLOG("Namespaces with different zone append support or different zone append size are not allowed.\n");
2214 			goto err;
2215 		}
2216 
2217 		subsystem->max_zone_append_size_kib = max_zone_append_size_kib;
2218 	}
2219 
2220 	first_ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
2221 	if (!first_ns) {
2222 		if (spdk_bdev_get_nvme_ctratt(ns->bdev).bits.fdps) {
2223 			SPDK_DEBUGLOG(nvmf, "Subsystem with id: %u has FDP capability.\n",
2224 				      subsystem->id);
2225 			subsystem->fdp_supported = true;
2226 		}
2227 	} else {
2228 		if (spdk_bdev_get_nvme_ctratt(first_ns->bdev).bits.fdps !=
2229 		    spdk_bdev_get_nvme_ctratt(ns->bdev).bits.fdps) {
2230 			SPDK_ERRLOG("Subsystem with id: %u can%s FDP namespace.\n", subsystem->id,
2231 				    spdk_bdev_get_nvme_ctratt(first_ns->bdev).bits.fdps ? " only add" : "not add");
2232 			goto err;
2233 		}
2234 	}
2235 
2236 	ns->opts = opts;
2237 	ns->subsystem = subsystem;
2238 	subsystem->ns[opts.nsid - 1] = ns;
2239 	ns->nsid = opts.nsid;
2240 	ns->anagrpid = opts.anagrpid;
2241 	subsystem->ana_group[ns->anagrpid - 1]++;
2242 	TAILQ_INIT(&ns->registrants);
2243 	if (ptpl_file) {
2244 		ns->ptpl_file = strdup(ptpl_file);
2245 		if (!ns->ptpl_file) {
2246 			SPDK_ERRLOG("Namespace ns->ptpl_file allocation failed\n");
2247 			goto err;
2248 		}
2249 	}
2250 
2251 	if (nvmf_ns_is_ptpl_capable(ns)) {
2252 		rc = nvmf_ns_reservation_load(ns, &info);
2253 		if (rc) {
2254 			SPDK_ERRLOG("Subsystem load reservation failed\n");
2255 			goto err;
2256 		}
2257 
2258 		rc = nvmf_ns_reservation_restore(ns, &info);
2259 		if (rc) {
2260 			SPDK_ERRLOG("Subsystem restore reservation failed\n");
2261 			goto err;
2262 		}
2263 	}
2264 
2265 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
2266 	     transport = spdk_nvmf_transport_get_next(transport)) {
2267 		if (transport->ops->subsystem_add_ns) {
2268 			rc = transport->ops->subsystem_add_ns(transport, subsystem, ns);
2269 			if (rc) {
2270 				SPDK_ERRLOG("Namespace attachment is not allowed by %s transport\n", transport->ops->name);
2271 				nvmf_ns_reservation_clear_all_registrants(ns);
2272 				goto err;
2273 			}
2274 		}
2275 	}
2276 
2277 	/* JSON value obj is freed before sending the response. Set NULL to prevent usage of dangling pointer. */
2278 	ns->opts.transport_specific = NULL;
2279 
2280 	SPDK_DEBUGLOG(nvmf, "Subsystem %s: bdev %s assigned nsid %" PRIu32 "\n",
2281 		      spdk_nvmf_subsystem_get_nqn(subsystem),
2282 		      bdev_name,
2283 		      opts.nsid);
2284 
2285 	nvmf_subsystem_ns_changed(subsystem, opts.nsid);
2286 
2287 	SPDK_DTRACE_PROBE2(nvmf_subsystem_add_ns, subsystem->subnqn, ns->nsid);
2288 
2289 	return opts.nsid;
2290 err:
2291 	subsystem->ns[opts.nsid - 1] = NULL;
2292 	spdk_bdev_module_release_bdev(ns->bdev);
2293 	spdk_bdev_close(ns->desc);
2294 	free(ns->ptpl_file);
2295 	free(ns);
2296 
2297 	return 0;
2298 }
2299 
2300 int
2301 spdk_nvmf_subsystem_set_ns_ana_group(struct spdk_nvmf_subsystem *subsystem,
2302 				     uint32_t nsid, uint32_t anagrpid)
2303 {
2304 	struct spdk_nvmf_ns *ns;
2305 
2306 	if (anagrpid > subsystem->max_nsid) {
2307 		SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n");
2308 		return -1;
2309 	}
2310 
2311 	if (anagrpid == 0) {
2312 		SPDK_ERRLOG("Zero is not allowed to ANAGRPID\n");
2313 		return -1;
2314 	}
2315 
2316 	if (nsid == 0 || nsid > subsystem->max_nsid) {
2317 		return -1;
2318 	}
2319 
2320 	ns = subsystem->ns[nsid - 1];
2321 	if (!ns) {
2322 		return -1;
2323 	}
2324 
2325 	assert(ns->anagrpid - 1 < subsystem->max_nsid);
2326 
2327 	assert(subsystem->ana_group[ns->anagrpid - 1] > 0);
2328 
2329 	subsystem->ana_group[ns->anagrpid - 1]--;
2330 
2331 	subsystem->ana_group[anagrpid - 1]++;
2332 
2333 	ns->anagrpid = anagrpid;
2334 	ns->opts.anagrpid = anagrpid;
2335 
2336 	nvmf_subsystem_ns_changed(subsystem, nsid);
2337 
2338 	return 0;
2339 }
2340 
2341 static uint32_t
2342 nvmf_subsystem_get_next_allocated_nsid(struct spdk_nvmf_subsystem *subsystem,
2343 				       uint32_t prev_nsid)
2344 {
2345 	uint32_t nsid;
2346 
2347 	if (prev_nsid >= subsystem->max_nsid) {
2348 		return 0;
2349 	}
2350 
2351 	for (nsid = prev_nsid + 1; nsid <= subsystem->max_nsid; nsid++) {
2352 		if (subsystem->ns[nsid - 1]) {
2353 			return nsid;
2354 		}
2355 	}
2356 
2357 	return 0;
2358 }
2359 
2360 struct spdk_nvmf_ns *
2361 spdk_nvmf_subsystem_get_first_ns(struct spdk_nvmf_subsystem *subsystem)
2362 {
2363 	uint32_t first_nsid;
2364 
2365 	first_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, 0);
2366 	return _nvmf_subsystem_get_ns(subsystem, first_nsid);
2367 }
2368 
2369 struct spdk_nvmf_ns *
2370 spdk_nvmf_subsystem_get_next_ns(struct spdk_nvmf_subsystem *subsystem,
2371 				struct spdk_nvmf_ns *prev_ns)
2372 {
2373 	uint32_t next_nsid;
2374 
2375 	next_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, prev_ns->opts.nsid);
2376 	return _nvmf_subsystem_get_ns(subsystem, next_nsid);
2377 }
2378 
2379 struct spdk_nvmf_ns *
2380 spdk_nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
2381 {
2382 	return _nvmf_subsystem_get_ns(subsystem, nsid);
2383 }
2384 
2385 uint32_t
2386 spdk_nvmf_ns_get_id(const struct spdk_nvmf_ns *ns)
2387 {
2388 	return ns->opts.nsid;
2389 }
2390 
2391 struct spdk_bdev *
2392 spdk_nvmf_ns_get_bdev(struct spdk_nvmf_ns *ns)
2393 {
2394 	return ns->bdev;
2395 }
2396 
2397 void
2398 spdk_nvmf_ns_get_opts(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_ns_opts *opts,
2399 		      size_t opts_size)
2400 {
2401 	memset(opts, 0, opts_size);
2402 	memcpy(opts, &ns->opts, spdk_min(sizeof(ns->opts), opts_size));
2403 }
2404 
2405 const char *
2406 spdk_nvmf_subsystem_get_sn(const struct spdk_nvmf_subsystem *subsystem)
2407 {
2408 	return subsystem->sn;
2409 }
2410 
2411 int
2412 spdk_nvmf_subsystem_set_sn(struct spdk_nvmf_subsystem *subsystem, const char *sn)
2413 {
2414 	size_t len, max_len;
2415 
2416 	max_len = sizeof(subsystem->sn) - 1;
2417 	len = strlen(sn);
2418 	if (len > max_len) {
2419 		SPDK_DEBUGLOG(nvmf, "Invalid sn \"%s\": length %zu > max %zu\n",
2420 			      sn, len, max_len);
2421 		return -1;
2422 	}
2423 
2424 	if (!nvmf_valid_ascii_string(sn, len)) {
2425 		SPDK_DEBUGLOG(nvmf, "Non-ASCII sn\n");
2426 		SPDK_LOGDUMP(nvmf, "sn", sn, len);
2427 		return -1;
2428 	}
2429 
2430 	snprintf(subsystem->sn, sizeof(subsystem->sn), "%s", sn);
2431 
2432 	return 0;
2433 }
2434 
2435 const char *
2436 spdk_nvmf_subsystem_get_mn(const struct spdk_nvmf_subsystem *subsystem)
2437 {
2438 	return subsystem->mn;
2439 }
2440 
2441 int
2442 spdk_nvmf_subsystem_set_mn(struct spdk_nvmf_subsystem *subsystem, const char *mn)
2443 {
2444 	size_t len, max_len;
2445 
2446 	if (mn == NULL) {
2447 		mn = MODEL_NUMBER_DEFAULT;
2448 	}
2449 	max_len = sizeof(subsystem->mn) - 1;
2450 	len = strlen(mn);
2451 	if (len > max_len) {
2452 		SPDK_DEBUGLOG(nvmf, "Invalid mn \"%s\": length %zu > max %zu\n",
2453 			      mn, len, max_len);
2454 		return -1;
2455 	}
2456 
2457 	if (!nvmf_valid_ascii_string(mn, len)) {
2458 		SPDK_DEBUGLOG(nvmf, "Non-ASCII mn\n");
2459 		SPDK_LOGDUMP(nvmf, "mn", mn, len);
2460 		return -1;
2461 	}
2462 
2463 	snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", mn);
2464 
2465 	return 0;
2466 }
2467 
2468 const char *
2469 spdk_nvmf_subsystem_get_nqn(const struct spdk_nvmf_subsystem *subsystem)
2470 {
2471 	return subsystem->subnqn;
2472 }
2473 
2474 /* We have to use the typedef in the function declaration to appease astyle. */
2475 typedef enum spdk_nvmf_subtype spdk_nvmf_subtype_t;
2476 
2477 spdk_nvmf_subtype_t
2478 spdk_nvmf_subsystem_get_type(struct spdk_nvmf_subsystem *subsystem)
2479 {
2480 	return subsystem->subtype;
2481 }
2482 
2483 uint32_t
2484 spdk_nvmf_subsystem_get_max_nsid(struct spdk_nvmf_subsystem *subsystem)
2485 {
2486 	return subsystem->max_nsid;
2487 }
2488 
2489 int
2490 spdk_nvmf_subsystem_set_cntlid_range(struct spdk_nvmf_subsystem *subsystem,
2491 				     uint16_t min_cntlid, uint16_t max_cntlid)
2492 {
2493 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
2494 		return -EAGAIN;
2495 	}
2496 
2497 	if (min_cntlid > max_cntlid) {
2498 		return -EINVAL;
2499 	}
2500 	/* The spec reserves cntlid values in the range FFF0h to FFFFh. */
2501 	if (min_cntlid < NVMF_MIN_CNTLID || min_cntlid > NVMF_MAX_CNTLID ||
2502 	    max_cntlid < NVMF_MIN_CNTLID || max_cntlid > NVMF_MAX_CNTLID) {
2503 		return -EINVAL;
2504 	}
2505 	subsystem->min_cntlid = min_cntlid;
2506 	subsystem->max_cntlid = max_cntlid;
2507 	if (subsystem->next_cntlid < min_cntlid || subsystem->next_cntlid > max_cntlid) {
2508 		subsystem->next_cntlid = min_cntlid;
2509 	}
2510 
2511 	return 0;
2512 }
2513 
2514 uint16_t
2515 nvmf_subsystem_gen_cntlid(struct spdk_nvmf_subsystem *subsystem)
2516 {
2517 	int count;
2518 	uint16_t cntlid;
2519 
2520 	/*
2521 	 * In the worst case, we might have to try all CNTLID values between min_cntlid and max_cntlid
2522 	 * before we find one that is unused (or find that all values are in use).
2523 	 */
2524 	for (count = 0; count < subsystem->max_cntlid - subsystem->min_cntlid + 1; count++) {
2525 		cntlid = subsystem->next_cntlid;
2526 		subsystem->next_cntlid++;
2527 
2528 		if (subsystem->next_cntlid > subsystem->max_cntlid) {
2529 			subsystem->next_cntlid = subsystem->min_cntlid;
2530 		}
2531 
2532 		/* Check if a controller with this cntlid currently exists. */
2533 		if (nvmf_subsystem_get_ctrlr(subsystem, cntlid) == NULL) {
2534 			/* Found unused cntlid */
2535 			return cntlid;
2536 		}
2537 	}
2538 
2539 	/* All valid cntlid values are in use. */
2540 	return 0xFFFF;
2541 }
2542 
2543 int
2544 nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_ctrlr *ctrlr)
2545 {
2546 
2547 	if (ctrlr->dynamic_ctrlr) {
2548 		ctrlr->cntlid = nvmf_subsystem_gen_cntlid(subsystem);
2549 		if (ctrlr->cntlid == 0xFFFF) {
2550 			/* Unable to get a cntlid */
2551 			SPDK_ERRLOG("Reached max simultaneous ctrlrs\n");
2552 			return -EBUSY;
2553 		}
2554 	} else if (nvmf_subsystem_get_ctrlr(subsystem, ctrlr->cntlid) != NULL) {
2555 		SPDK_ERRLOG("Ctrlr with cntlid %u already exist\n", ctrlr->cntlid);
2556 		return -EEXIST;
2557 	}
2558 
2559 	TAILQ_INSERT_TAIL(&subsystem->ctrlrs, ctrlr, link);
2560 
2561 	SPDK_DTRACE_PROBE3(nvmf_subsystem_add_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn);
2562 
2563 	return 0;
2564 }
2565 
2566 void
2567 nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem,
2568 			    struct spdk_nvmf_ctrlr *ctrlr)
2569 {
2570 	SPDK_DTRACE_PROBE3(nvmf_subsystem_remove_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn);
2571 
2572 	assert(spdk_get_thread() == subsystem->thread);
2573 	assert(subsystem == ctrlr->subsys);
2574 	SPDK_DEBUGLOG(nvmf, "remove ctrlr %p id 0x%x from subsys %p %s\n", ctrlr, ctrlr->cntlid, subsystem,
2575 		      subsystem->subnqn);
2576 	TAILQ_REMOVE(&subsystem->ctrlrs, ctrlr, link);
2577 }
2578 
2579 struct spdk_nvmf_ctrlr *
2580 nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, uint16_t cntlid)
2581 {
2582 	struct spdk_nvmf_ctrlr *ctrlr;
2583 
2584 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2585 		if (ctrlr->cntlid == cntlid) {
2586 			return ctrlr;
2587 		}
2588 	}
2589 
2590 	return NULL;
2591 }
2592 
2593 uint32_t
2594 spdk_nvmf_subsystem_get_max_namespaces(const struct spdk_nvmf_subsystem *subsystem)
2595 {
2596 	return subsystem->max_nsid;
2597 }
2598 
2599 uint16_t
2600 spdk_nvmf_subsystem_get_min_cntlid(const struct spdk_nvmf_subsystem *subsystem)
2601 {
2602 	return subsystem->min_cntlid;
2603 }
2604 
2605 uint16_t
2606 spdk_nvmf_subsystem_get_max_cntlid(const struct spdk_nvmf_subsystem *subsystem)
2607 {
2608 	return subsystem->max_cntlid;
2609 }
2610 
2611 struct _nvmf_ns_registrant {
2612 	uint64_t		rkey;
2613 	char			*host_uuid;
2614 };
2615 
2616 struct _nvmf_ns_registrants {
2617 	size_t				num_regs;
2618 	struct _nvmf_ns_registrant	reg[SPDK_NVMF_MAX_NUM_REGISTRANTS];
2619 };
2620 
2621 struct _nvmf_ns_reservation {
2622 	bool					ptpl_activated;
2623 	enum spdk_nvme_reservation_type		rtype;
2624 	uint64_t				crkey;
2625 	char					*bdev_uuid;
2626 	char					*holder_uuid;
2627 	struct _nvmf_ns_registrants		regs;
2628 };
2629 
2630 static const struct spdk_json_object_decoder nvmf_ns_pr_reg_decoders[] = {
2631 	{"rkey", offsetof(struct _nvmf_ns_registrant, rkey), spdk_json_decode_uint64},
2632 	{"host_uuid", offsetof(struct _nvmf_ns_registrant, host_uuid), spdk_json_decode_string},
2633 };
2634 
2635 static int
2636 nvmf_decode_ns_pr_reg(const struct spdk_json_val *val, void *out)
2637 {
2638 	struct _nvmf_ns_registrant *reg = out;
2639 
2640 	return spdk_json_decode_object(val, nvmf_ns_pr_reg_decoders,
2641 				       SPDK_COUNTOF(nvmf_ns_pr_reg_decoders), reg);
2642 }
2643 
2644 static int
2645 nvmf_decode_ns_pr_regs(const struct spdk_json_val *val, void *out)
2646 {
2647 	struct _nvmf_ns_registrants *regs = out;
2648 
2649 	return spdk_json_decode_array(val, nvmf_decode_ns_pr_reg, regs->reg,
2650 				      SPDK_NVMF_MAX_NUM_REGISTRANTS, &regs->num_regs,
2651 				      sizeof(struct _nvmf_ns_registrant));
2652 }
2653 
2654 static const struct spdk_json_object_decoder nvmf_ns_pr_decoders[] = {
2655 	{"ptpl", offsetof(struct _nvmf_ns_reservation, ptpl_activated), spdk_json_decode_bool, true},
2656 	{"rtype", offsetof(struct _nvmf_ns_reservation, rtype), spdk_json_decode_uint32, true},
2657 	{"crkey", offsetof(struct _nvmf_ns_reservation, crkey), spdk_json_decode_uint64, true},
2658 	{"bdev_uuid", offsetof(struct _nvmf_ns_reservation, bdev_uuid), spdk_json_decode_string},
2659 	{"holder_uuid", offsetof(struct _nvmf_ns_reservation, holder_uuid), spdk_json_decode_string, true},
2660 	{"registrants", offsetof(struct _nvmf_ns_reservation, regs), nvmf_decode_ns_pr_regs},
2661 };
2662 
2663 static int
2664 nvmf_ns_reservation_load_json(const struct spdk_nvmf_ns *ns,
2665 			      struct spdk_nvmf_reservation_info *info)
2666 {
2667 	size_t json_size;
2668 	ssize_t values_cnt, rc;
2669 	void *json = NULL, *end;
2670 	struct spdk_json_val *values = NULL;
2671 	struct _nvmf_ns_reservation res = {};
2672 	const char *file = ns->ptpl_file;
2673 	uint32_t i;
2674 
2675 	/* It's not an error if the file does not exist */
2676 	if (access(file, F_OK) != 0) {
2677 		SPDK_DEBUGLOG(nvmf, "File %s does not exist\n", file);
2678 		return 0;
2679 	}
2680 
2681 	/* Load all persist file contents into a local buffer */
2682 	json = spdk_posix_file_load_from_name(file, &json_size);
2683 	if (!json) {
2684 		SPDK_ERRLOG("Load persist file %s failed\n", file);
2685 		return -ENOMEM;
2686 	}
2687 
2688 	rc = spdk_json_parse(json, json_size, NULL, 0, &end, 0);
2689 	if (rc < 0) {
2690 		SPDK_NOTICELOG("Parsing JSON configuration failed (%zd)\n", rc);
2691 		goto exit;
2692 	}
2693 
2694 	values_cnt = rc;
2695 	values = calloc(values_cnt, sizeof(struct spdk_json_val));
2696 	if (values == NULL) {
2697 		goto exit;
2698 	}
2699 
2700 	rc = spdk_json_parse(json, json_size, values, values_cnt, &end, 0);
2701 	if (rc != values_cnt) {
2702 		SPDK_ERRLOG("Parsing JSON configuration failed (%zd)\n", rc);
2703 		goto exit;
2704 	}
2705 
2706 	/* Decode json */
2707 	if (spdk_json_decode_object(values, nvmf_ns_pr_decoders,
2708 				    SPDK_COUNTOF(nvmf_ns_pr_decoders),
2709 				    &res)) {
2710 		SPDK_ERRLOG("Invalid objects in the persist file %s\n", file);
2711 		rc = -EINVAL;
2712 		goto exit;
2713 	}
2714 
2715 	if (res.regs.num_regs > SPDK_NVMF_MAX_NUM_REGISTRANTS) {
2716 		SPDK_ERRLOG("Can only support up to %u registrants\n", SPDK_NVMF_MAX_NUM_REGISTRANTS);
2717 		rc = -ERANGE;
2718 		goto exit;
2719 	}
2720 
2721 	rc = 0;
2722 	info->ptpl_activated = res.ptpl_activated;
2723 	info->rtype = res.rtype;
2724 	info->crkey = res.crkey;
2725 	snprintf(info->bdev_uuid, sizeof(info->bdev_uuid), "%s", res.bdev_uuid);
2726 	snprintf(info->holder_uuid, sizeof(info->holder_uuid), "%s", res.holder_uuid);
2727 	info->num_regs = res.regs.num_regs;
2728 	for (i = 0; i < res.regs.num_regs; i++) {
2729 		info->registrants[i].rkey = res.regs.reg[i].rkey;
2730 		snprintf(info->registrants[i].host_uuid, sizeof(info->registrants[i].host_uuid), "%s",
2731 			 res.regs.reg[i].host_uuid);
2732 	}
2733 
2734 exit:
2735 	free(json);
2736 	free(values);
2737 	free(res.bdev_uuid);
2738 	free(res.holder_uuid);
2739 	for (i = 0; i < res.regs.num_regs; i++) {
2740 		free(res.regs.reg[i].host_uuid);
2741 	}
2742 
2743 	return rc;
2744 }
2745 
2746 static bool nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns);
2747 
2748 static int
2749 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info)
2750 {
2751 	uint32_t i;
2752 	struct spdk_nvmf_registrant *reg, *holder = NULL;
2753 	struct spdk_uuid bdev_uuid, holder_uuid;
2754 	bool rkey_flag = false;
2755 
2756 	SPDK_DEBUGLOG(nvmf, "NSID %u, PTPL %u, Number of registrants %u\n",
2757 		      ns->nsid, info->ptpl_activated, info->num_regs);
2758 
2759 	/* it's not an error */
2760 	if (!info->ptpl_activated || !info->num_regs) {
2761 		return 0;
2762 	}
2763 
2764 	/* Check info->crkey exist or not in info->registrants[i].rkey */
2765 	for (i = 0; i < info->num_regs; i++) {
2766 		if (info->crkey == info->registrants[i].rkey) {
2767 			rkey_flag = true;
2768 		}
2769 	}
2770 	if (!rkey_flag && info->crkey != 0) {
2771 		return -EINVAL;
2772 	}
2773 
2774 	spdk_uuid_parse(&bdev_uuid, info->bdev_uuid);
2775 	if (spdk_uuid_compare(&bdev_uuid, spdk_bdev_get_uuid(ns->bdev))) {
2776 		SPDK_ERRLOG("Existing bdev UUID is not same with configuration file\n");
2777 		return -EINVAL;
2778 	}
2779 
2780 	ns->crkey = info->crkey;
2781 	ns->rtype = info->rtype;
2782 	ns->ptpl_activated = info->ptpl_activated;
2783 	spdk_uuid_parse(&holder_uuid, info->holder_uuid);
2784 
2785 	SPDK_DEBUGLOG(nvmf, "Bdev UUID %s\n", info->bdev_uuid);
2786 	if (info->rtype) {
2787 		SPDK_DEBUGLOG(nvmf, "Holder UUID %s, RTYPE %u, RKEY 0x%"PRIx64"\n",
2788 			      info->holder_uuid, info->rtype, info->crkey);
2789 	}
2790 
2791 	for (i = 0; i < info->num_regs; i++) {
2792 		reg = calloc(1, sizeof(*reg));
2793 		if (!reg) {
2794 			return -ENOMEM;
2795 		}
2796 		spdk_uuid_parse(&reg->hostid, info->registrants[i].host_uuid);
2797 		reg->rkey = info->registrants[i].rkey;
2798 		TAILQ_INSERT_TAIL(&ns->registrants, reg, link);
2799 		if (info->crkey != 0 && !spdk_uuid_compare(&holder_uuid, &reg->hostid)) {
2800 			holder = reg;
2801 		}
2802 		SPDK_DEBUGLOG(nvmf, "Registrant RKEY 0x%"PRIx64", Host UUID %s\n",
2803 			      info->registrants[i].rkey, info->registrants[i].host_uuid);
2804 	}
2805 
2806 	if (nvmf_ns_reservation_all_registrants_type(ns)) {
2807 		ns->holder = TAILQ_FIRST(&ns->registrants);
2808 	} else {
2809 		ns->holder = holder;
2810 	}
2811 
2812 	return 0;
2813 }
2814 
2815 static int
2816 nvmf_ns_json_write_cb(void *cb_ctx, const void *data, size_t size)
2817 {
2818 	char *file = cb_ctx;
2819 	size_t rc;
2820 	FILE *fd;
2821 
2822 	fd = fopen(file, "w");
2823 	if (!fd) {
2824 		SPDK_ERRLOG("Can't open file %s for write\n", file);
2825 		return -ENOENT;
2826 	}
2827 	rc = fwrite(data, 1, size, fd);
2828 	fclose(fd);
2829 
2830 	return rc == size ? 0 : -1;
2831 }
2832 
2833 static int
2834 nvmf_ns_reservation_update_json(const struct spdk_nvmf_ns *ns,
2835 				const struct spdk_nvmf_reservation_info *info)
2836 {
2837 	const char *file = ns->ptpl_file;
2838 	struct spdk_json_write_ctx *w;
2839 	uint32_t i;
2840 	int rc = 0;
2841 
2842 	w = spdk_json_write_begin(nvmf_ns_json_write_cb, (void *)file, 0);
2843 	if (w == NULL) {
2844 		return -ENOMEM;
2845 	}
2846 	/* clear the configuration file */
2847 	if (!info->ptpl_activated) {
2848 		goto exit;
2849 	}
2850 
2851 	spdk_json_write_object_begin(w);
2852 	spdk_json_write_named_bool(w, "ptpl", info->ptpl_activated);
2853 	spdk_json_write_named_uint32(w, "rtype", info->rtype);
2854 	spdk_json_write_named_uint64(w, "crkey", info->crkey);
2855 	spdk_json_write_named_string(w, "bdev_uuid", info->bdev_uuid);
2856 	spdk_json_write_named_string(w, "holder_uuid", info->holder_uuid);
2857 
2858 	spdk_json_write_named_array_begin(w, "registrants");
2859 	for (i = 0; i < info->num_regs; i++) {
2860 		spdk_json_write_object_begin(w);
2861 		spdk_json_write_named_uint64(w, "rkey", info->registrants[i].rkey);
2862 		spdk_json_write_named_string(w, "host_uuid", info->registrants[i].host_uuid);
2863 		spdk_json_write_object_end(w);
2864 	}
2865 	spdk_json_write_array_end(w);
2866 	spdk_json_write_object_end(w);
2867 
2868 exit:
2869 	rc = spdk_json_write_end(w);
2870 	return rc;
2871 }
2872 
2873 static int
2874 nvmf_ns_update_reservation_info(struct spdk_nvmf_ns *ns)
2875 {
2876 	struct spdk_nvmf_reservation_info info;
2877 	struct spdk_nvmf_registrant *reg, *tmp;
2878 	uint32_t i = 0;
2879 
2880 	assert(ns != NULL);
2881 
2882 	if (!ns->bdev || !nvmf_ns_is_ptpl_capable(ns)) {
2883 		return 0;
2884 	}
2885 
2886 	memset(&info, 0, sizeof(info));
2887 	spdk_uuid_fmt_lower(info.bdev_uuid, sizeof(info.bdev_uuid), spdk_bdev_get_uuid(ns->bdev));
2888 
2889 	if (ns->rtype) {
2890 		info.rtype = ns->rtype;
2891 		info.crkey = ns->crkey;
2892 		if (!nvmf_ns_reservation_all_registrants_type(ns)) {
2893 			assert(ns->holder != NULL);
2894 			spdk_uuid_fmt_lower(info.holder_uuid, sizeof(info.holder_uuid), &ns->holder->hostid);
2895 		}
2896 	}
2897 
2898 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2899 		spdk_uuid_fmt_lower(info.registrants[i].host_uuid, sizeof(info.registrants[i].host_uuid),
2900 				    &reg->hostid);
2901 		info.registrants[i++].rkey = reg->rkey;
2902 	}
2903 
2904 	info.num_regs = i;
2905 	info.ptpl_activated = ns->ptpl_activated;
2906 
2907 	return nvmf_ns_reservation_update(ns, &info);
2908 }
2909 
2910 static struct spdk_nvmf_registrant *
2911 nvmf_ns_reservation_get_registrant(struct spdk_nvmf_ns *ns,
2912 				   struct spdk_uuid *uuid)
2913 {
2914 	struct spdk_nvmf_registrant *reg, *tmp;
2915 
2916 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2917 		if (!spdk_uuid_compare(&reg->hostid, uuid)) {
2918 			return reg;
2919 		}
2920 	}
2921 
2922 	return NULL;
2923 }
2924 
2925 /* Generate reservation notice log to registered HostID controllers */
2926 static void
2927 nvmf_subsystem_gen_ctrlr_notification(struct spdk_nvmf_subsystem *subsystem,
2928 				      struct spdk_nvmf_ns *ns,
2929 				      struct spdk_uuid *hostid_list,
2930 				      uint32_t num_hostid,
2931 				      enum spdk_nvme_reservation_notification_log_page_type type)
2932 {
2933 	struct spdk_nvmf_ctrlr *ctrlr;
2934 	uint32_t i;
2935 
2936 	for (i = 0; i < num_hostid; i++) {
2937 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2938 			if (!spdk_uuid_compare(&ctrlr->hostid, &hostid_list[i])) {
2939 				nvmf_ctrlr_reservation_notice_log(ctrlr, ns, type);
2940 			}
2941 		}
2942 	}
2943 }
2944 
2945 /* Get all registrants' hostid other than the controller who issued the command */
2946 static uint32_t
2947 nvmf_ns_reservation_get_all_other_hostid(struct spdk_nvmf_ns *ns,
2948 		struct spdk_uuid *hostid_list,
2949 		uint32_t max_num_hostid,
2950 		struct spdk_uuid *current_hostid)
2951 {
2952 	struct spdk_nvmf_registrant *reg, *tmp;
2953 	uint32_t num_hostid = 0;
2954 
2955 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2956 		if (spdk_uuid_compare(&reg->hostid, current_hostid)) {
2957 			if (num_hostid == max_num_hostid) {
2958 				assert(false);
2959 				return max_num_hostid;
2960 			}
2961 			hostid_list[num_hostid++] = reg->hostid;
2962 		}
2963 	}
2964 
2965 	return num_hostid;
2966 }
2967 
2968 /* Calculate the unregistered HostID list according to list
2969  * prior to execute preempt command and list after executing
2970  * preempt command.
2971  */
2972 static uint32_t
2973 nvmf_ns_reservation_get_unregistered_hostid(struct spdk_uuid *old_hostid_list,
2974 		uint32_t old_num_hostid,
2975 		struct spdk_uuid *remaining_hostid_list,
2976 		uint32_t remaining_num_hostid)
2977 {
2978 	struct spdk_uuid temp_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
2979 	uint32_t i, j, num_hostid = 0;
2980 	bool found;
2981 
2982 	if (!remaining_num_hostid) {
2983 		return old_num_hostid;
2984 	}
2985 
2986 	for (i = 0; i < old_num_hostid; i++) {
2987 		found = false;
2988 		for (j = 0; j < remaining_num_hostid; j++) {
2989 			if (!spdk_uuid_compare(&old_hostid_list[i], &remaining_hostid_list[j])) {
2990 				found = true;
2991 				break;
2992 			}
2993 		}
2994 		if (!found) {
2995 			spdk_uuid_copy(&temp_hostid_list[num_hostid++], &old_hostid_list[i]);
2996 		}
2997 	}
2998 
2999 	if (num_hostid) {
3000 		memcpy(old_hostid_list, temp_hostid_list, sizeof(struct spdk_uuid) * num_hostid);
3001 	}
3002 
3003 	return num_hostid;
3004 }
3005 
3006 /* current reservation type is all registrants or not */
3007 static bool
3008 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns)
3009 {
3010 	return (ns->rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
3011 		ns->rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS);
3012 }
3013 
3014 /* current registrant is reservation holder or not */
3015 static bool
3016 nvmf_ns_reservation_registrant_is_holder(struct spdk_nvmf_ns *ns,
3017 		struct spdk_nvmf_registrant *reg)
3018 {
3019 	if (!reg) {
3020 		return false;
3021 	}
3022 
3023 	if (nvmf_ns_reservation_all_registrants_type(ns)) {
3024 		return true;
3025 	}
3026 
3027 	return (ns->holder == reg);
3028 }
3029 
3030 static int
3031 nvmf_ns_reservation_add_registrant(struct spdk_nvmf_ns *ns,
3032 				   struct spdk_nvmf_ctrlr *ctrlr,
3033 				   uint64_t nrkey)
3034 {
3035 	struct spdk_nvmf_registrant *reg;
3036 
3037 	reg = calloc(1, sizeof(*reg));
3038 	if (!reg) {
3039 		return -ENOMEM;
3040 	}
3041 
3042 	reg->rkey = nrkey;
3043 	/* set hostid for the registrant */
3044 	spdk_uuid_copy(&reg->hostid, &ctrlr->hostid);
3045 	TAILQ_INSERT_TAIL(&ns->registrants, reg, link);
3046 	ns->gen++;
3047 
3048 	return 0;
3049 }
3050 
3051 static void
3052 nvmf_ns_reservation_release_reservation(struct spdk_nvmf_ns *ns)
3053 {
3054 	ns->rtype = 0;
3055 	ns->crkey = 0;
3056 	ns->holder = NULL;
3057 }
3058 
3059 /* release the reservation if the last registrant was removed */
3060 static void
3061 nvmf_ns_reservation_check_release_on_remove_registrant(struct spdk_nvmf_ns *ns,
3062 		struct spdk_nvmf_registrant *reg)
3063 {
3064 	struct spdk_nvmf_registrant *next_reg;
3065 
3066 	/* no reservation holder */
3067 	if (!ns->holder) {
3068 		assert(ns->rtype == 0);
3069 		return;
3070 	}
3071 
3072 	next_reg = TAILQ_FIRST(&ns->registrants);
3073 	if (next_reg && nvmf_ns_reservation_all_registrants_type(ns)) {
3074 		/* the next valid registrant is the new holder now */
3075 		ns->holder = next_reg;
3076 	} else if (nvmf_ns_reservation_registrant_is_holder(ns, reg)) {
3077 		/* release the reservation */
3078 		nvmf_ns_reservation_release_reservation(ns);
3079 	}
3080 }
3081 
3082 static void
3083 nvmf_ns_reservation_remove_registrant(struct spdk_nvmf_ns *ns,
3084 				      struct spdk_nvmf_registrant *reg)
3085 {
3086 	TAILQ_REMOVE(&ns->registrants, reg, link);
3087 	nvmf_ns_reservation_check_release_on_remove_registrant(ns, reg);
3088 	free(reg);
3089 	ns->gen++;
3090 	return;
3091 }
3092 
3093 static uint32_t
3094 nvmf_ns_reservation_remove_registrants_by_key(struct spdk_nvmf_ns *ns,
3095 		uint64_t rkey)
3096 {
3097 	struct spdk_nvmf_registrant *reg, *tmp;
3098 	uint32_t count = 0;
3099 
3100 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3101 		if (reg->rkey == rkey) {
3102 			nvmf_ns_reservation_remove_registrant(ns, reg);
3103 			count++;
3104 		}
3105 	}
3106 	return count;
3107 }
3108 
3109 static uint32_t
3110 nvmf_ns_reservation_remove_all_other_registrants(struct spdk_nvmf_ns *ns,
3111 		struct spdk_nvmf_registrant *reg)
3112 {
3113 	struct spdk_nvmf_registrant *reg_tmp, *reg_tmp2;
3114 	uint32_t count = 0;
3115 
3116 	TAILQ_FOREACH_SAFE(reg_tmp, &ns->registrants, link, reg_tmp2) {
3117 		if (reg_tmp != reg) {
3118 			nvmf_ns_reservation_remove_registrant(ns, reg_tmp);
3119 			count++;
3120 		}
3121 	}
3122 	return count;
3123 }
3124 
3125 static uint32_t
3126 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns)
3127 {
3128 	struct spdk_nvmf_registrant *reg, *reg_tmp;
3129 	uint32_t count = 0;
3130 
3131 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, reg_tmp) {
3132 		nvmf_ns_reservation_remove_registrant(ns, reg);
3133 		count++;
3134 	}
3135 	return count;
3136 }
3137 
3138 static void
3139 nvmf_ns_reservation_acquire_reservation(struct spdk_nvmf_ns *ns, uint64_t rkey,
3140 					enum spdk_nvme_reservation_type rtype,
3141 					struct spdk_nvmf_registrant *holder)
3142 {
3143 	ns->rtype = rtype;
3144 	ns->crkey = rkey;
3145 	assert(ns->holder == NULL);
3146 	ns->holder = holder;
3147 }
3148 
3149 static bool
3150 nvmf_ns_reservation_register(struct spdk_nvmf_ns *ns,
3151 			     struct spdk_nvmf_ctrlr *ctrlr,
3152 			     struct spdk_nvmf_request *req)
3153 {
3154 	struct spdk_nvme_reservation_register_data key = { 0 };
3155 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3156 	uint8_t rrega, iekey, cptpl, rtype;
3157 	struct spdk_nvmf_registrant *reg;
3158 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3159 	bool update_sgroup = false;
3160 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3161 	uint32_t num_hostid = 0;
3162 	int rc;
3163 
3164 	rrega = cmd->cdw10_bits.resv_register.rrega;
3165 	iekey = cmd->cdw10_bits.resv_register.iekey;
3166 	cptpl = cmd->cdw10_bits.resv_register.cptpl;
3167 
3168 	if (req->iovcnt > 0 && req->length >= sizeof(key)) {
3169 		struct spdk_iov_xfer ix;
3170 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3171 		spdk_iov_xfer_to_buf(&ix, &key, sizeof(key));
3172 	} else {
3173 		SPDK_ERRLOG("No key provided. Failing request.\n");
3174 		status = SPDK_NVME_SC_INVALID_FIELD;
3175 		goto exit;
3176 	}
3177 
3178 	SPDK_DEBUGLOG(nvmf, "REGISTER: RREGA %u, IEKEY %u, CPTPL %u, "
3179 		      "NRKEY 0x%"PRIx64", NRKEY 0x%"PRIx64"\n",
3180 		      rrega, iekey, cptpl, key.crkey, key.nrkey);
3181 
3182 	if (cptpl == SPDK_NVME_RESERVE_PTPL_CLEAR_POWER_ON) {
3183 		/* True to OFF state, and need to be updated in the configuration file */
3184 		if (ns->ptpl_activated) {
3185 			ns->ptpl_activated = 0;
3186 			update_sgroup = true;
3187 		}
3188 	} else if (cptpl == SPDK_NVME_RESERVE_PTPL_PERSIST_POWER_LOSS) {
3189 		if (!nvmf_ns_is_ptpl_capable(ns)) {
3190 			status = SPDK_NVME_SC_INVALID_FIELD;
3191 			goto exit;
3192 		} else if (ns->ptpl_activated == 0) {
3193 			ns->ptpl_activated = 1;
3194 			update_sgroup = true;
3195 		}
3196 	}
3197 
3198 	/* current Host Identifier has registrant or not */
3199 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3200 
3201 	switch (rrega) {
3202 	case SPDK_NVME_RESERVE_REGISTER_KEY:
3203 		if (!reg) {
3204 			/* register new controller */
3205 			if (key.nrkey == 0) {
3206 				SPDK_ERRLOG("Can't register zeroed new key\n");
3207 				status = SPDK_NVME_SC_INVALID_FIELD;
3208 				goto exit;
3209 			}
3210 			rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey);
3211 			if (rc < 0) {
3212 				status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3213 				goto exit;
3214 			}
3215 			update_sgroup = true;
3216 		} else {
3217 			/* register with same key is not an error */
3218 			if (reg->rkey != key.nrkey) {
3219 				SPDK_ERRLOG("The same host already register a "
3220 					    "key with 0x%"PRIx64"\n",
3221 					    reg->rkey);
3222 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3223 				goto exit;
3224 			}
3225 		}
3226 		break;
3227 	case SPDK_NVME_RESERVE_UNREGISTER_KEY:
3228 		if (!reg || (!iekey && reg->rkey != key.crkey)) {
3229 			SPDK_ERRLOG("No registrant or current key doesn't match "
3230 				    "with existing registrant key\n");
3231 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3232 			goto exit;
3233 		}
3234 
3235 		rtype = ns->rtype;
3236 		num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3237 				SPDK_NVMF_MAX_NUM_REGISTRANTS,
3238 				&ctrlr->hostid);
3239 
3240 		nvmf_ns_reservation_remove_registrant(ns, reg);
3241 
3242 		if (!ns->rtype && num_hostid && (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_REG_ONLY ||
3243 						 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY)) {
3244 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3245 							      hostid_list,
3246 							      num_hostid,
3247 							      SPDK_NVME_RESERVATION_RELEASED);
3248 		}
3249 		update_sgroup = true;
3250 		break;
3251 	case SPDK_NVME_RESERVE_REPLACE_KEY:
3252 		if (key.nrkey == 0) {
3253 			SPDK_ERRLOG("Can't register zeroed new key\n");
3254 			status = SPDK_NVME_SC_INVALID_FIELD;
3255 			goto exit;
3256 		}
3257 		/* Registrant exists */
3258 		if (reg) {
3259 			if (!iekey && reg->rkey != key.crkey) {
3260 				SPDK_ERRLOG("Current key doesn't match "
3261 					    "existing registrant key\n");
3262 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3263 				goto exit;
3264 			}
3265 			if (reg->rkey == key.nrkey) {
3266 				goto exit;
3267 			}
3268 			reg->rkey = key.nrkey;
3269 		} else if (iekey) { /* No registrant but IEKEY is set */
3270 			/* new registrant */
3271 			rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey);
3272 			if (rc < 0) {
3273 				status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3274 				goto exit;
3275 			}
3276 		} else { /* No registrant */
3277 			SPDK_ERRLOG("No registrant\n");
3278 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3279 			goto exit;
3280 
3281 		}
3282 		update_sgroup = true;
3283 		break;
3284 	default:
3285 		status = SPDK_NVME_SC_INVALID_FIELD;
3286 		goto exit;
3287 	}
3288 
3289 exit:
3290 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3291 	req->rsp->nvme_cpl.status.sc = status;
3292 	return update_sgroup;
3293 }
3294 
3295 static bool
3296 nvmf_ns_reservation_acquire(struct spdk_nvmf_ns *ns,
3297 			    struct spdk_nvmf_ctrlr *ctrlr,
3298 			    struct spdk_nvmf_request *req)
3299 {
3300 	struct spdk_nvme_reservation_acquire_data key = { 0 };
3301 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3302 	uint8_t racqa, iekey, rtype;
3303 	struct spdk_nvmf_registrant *reg;
3304 	bool all_regs = false;
3305 	uint32_t count = 0;
3306 	bool update_sgroup = true;
3307 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3308 	uint32_t num_hostid = 0;
3309 	struct spdk_uuid new_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3310 	uint32_t new_num_hostid = 0;
3311 	bool reservation_released = false;
3312 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3313 
3314 	racqa = cmd->cdw10_bits.resv_acquire.racqa;
3315 	iekey = cmd->cdw10_bits.resv_acquire.iekey;
3316 	rtype = cmd->cdw10_bits.resv_acquire.rtype;
3317 
3318 	if (req->iovcnt > 0 && req->length >= sizeof(key)) {
3319 		struct spdk_iov_xfer ix;
3320 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3321 		spdk_iov_xfer_to_buf(&ix, &key, sizeof(key));
3322 	} else {
3323 		SPDK_ERRLOG("No key provided. Failing request.\n");
3324 		status = SPDK_NVME_SC_INVALID_FIELD;
3325 		goto exit;
3326 	}
3327 
3328 	SPDK_DEBUGLOG(nvmf, "ACQUIRE: RACQA %u, IEKEY %u, RTYPE %u, "
3329 		      "NRKEY 0x%"PRIx64", PRKEY 0x%"PRIx64"\n",
3330 		      racqa, iekey, rtype, key.crkey, key.prkey);
3331 
3332 	if (iekey || rtype > SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) {
3333 		SPDK_ERRLOG("Ignore existing key field set to 1\n");
3334 		status = SPDK_NVME_SC_INVALID_FIELD;
3335 		update_sgroup = false;
3336 		goto exit;
3337 	}
3338 
3339 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3340 	/* must be registrant and CRKEY must match */
3341 	if (!reg || reg->rkey != key.crkey) {
3342 		SPDK_ERRLOG("No registrant or current key doesn't match "
3343 			    "with existing registrant key\n");
3344 		status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3345 		update_sgroup = false;
3346 		goto exit;
3347 	}
3348 
3349 	all_regs = nvmf_ns_reservation_all_registrants_type(ns);
3350 
3351 	switch (racqa) {
3352 	case SPDK_NVME_RESERVE_ACQUIRE:
3353 		/* it's not an error for the holder to acquire same reservation type again */
3354 		if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && ns->rtype == rtype) {
3355 			/* do nothing */
3356 			update_sgroup = false;
3357 		} else if (ns->holder == NULL) {
3358 			/* first time to acquire the reservation */
3359 			nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg);
3360 		} else {
3361 			SPDK_ERRLOG("Invalid rtype or current registrant is not holder\n");
3362 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3363 			update_sgroup = false;
3364 			goto exit;
3365 		}
3366 		break;
3367 	case SPDK_NVME_RESERVE_PREEMPT:
3368 		/* no reservation holder */
3369 		if (!ns->holder) {
3370 			/* unregister with PRKEY */
3371 			nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3372 			break;
3373 		}
3374 		num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3375 				SPDK_NVMF_MAX_NUM_REGISTRANTS,
3376 				&ctrlr->hostid);
3377 
3378 		/* only 1 reservation holder and reservation key is valid */
3379 		if (!all_regs) {
3380 			/* preempt itself */
3381 			if (nvmf_ns_reservation_registrant_is_holder(ns, reg) &&
3382 			    ns->crkey == key.prkey) {
3383 				ns->rtype = rtype;
3384 				reservation_released = true;
3385 				break;
3386 			}
3387 
3388 			if (ns->crkey == key.prkey) {
3389 				nvmf_ns_reservation_remove_registrant(ns, ns->holder);
3390 				nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg);
3391 				reservation_released = true;
3392 			} else if (key.prkey != 0) {
3393 				nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3394 			} else {
3395 				/* PRKEY is zero */
3396 				SPDK_ERRLOG("Current PRKEY is zero\n");
3397 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3398 				update_sgroup = false;
3399 				goto exit;
3400 			}
3401 		} else {
3402 			/* release all other registrants except for the current one */
3403 			if (key.prkey == 0) {
3404 				nvmf_ns_reservation_remove_all_other_registrants(ns, reg);
3405 				assert(ns->holder == reg);
3406 			} else {
3407 				count = nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3408 				if (count == 0) {
3409 					SPDK_ERRLOG("PRKEY doesn't match any registrant\n");
3410 					status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3411 					update_sgroup = false;
3412 					goto exit;
3413 				}
3414 			}
3415 		}
3416 		break;
3417 	default:
3418 		status = SPDK_NVME_SC_INVALID_FIELD;
3419 		update_sgroup = false;
3420 		break;
3421 	}
3422 
3423 exit:
3424 	if (update_sgroup && racqa == SPDK_NVME_RESERVE_PREEMPT) {
3425 		new_num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, new_hostid_list,
3426 				 SPDK_NVMF_MAX_NUM_REGISTRANTS,
3427 				 &ctrlr->hostid);
3428 		/* Preempt notification occurs on the unregistered controllers
3429 		 * other than the controller who issued the command.
3430 		 */
3431 		num_hostid = nvmf_ns_reservation_get_unregistered_hostid(hostid_list,
3432 				num_hostid,
3433 				new_hostid_list,
3434 				new_num_hostid);
3435 		if (num_hostid) {
3436 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3437 							      hostid_list,
3438 							      num_hostid,
3439 							      SPDK_NVME_REGISTRATION_PREEMPTED);
3440 
3441 		}
3442 		/* Reservation released notification occurs on the
3443 		 * controllers which are the remaining registrants other than
3444 		 * the controller who issued the command.
3445 		 */
3446 		if (reservation_released && new_num_hostid) {
3447 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3448 							      new_hostid_list,
3449 							      new_num_hostid,
3450 							      SPDK_NVME_RESERVATION_RELEASED);
3451 
3452 		}
3453 	}
3454 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3455 	req->rsp->nvme_cpl.status.sc = status;
3456 	return update_sgroup;
3457 }
3458 
3459 static bool
3460 nvmf_ns_reservation_release(struct spdk_nvmf_ns *ns,
3461 			    struct spdk_nvmf_ctrlr *ctrlr,
3462 			    struct spdk_nvmf_request *req)
3463 {
3464 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3465 	uint8_t rrela, iekey, rtype;
3466 	struct spdk_nvmf_registrant *reg;
3467 	uint64_t crkey = 0;
3468 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3469 	bool update_sgroup = true;
3470 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3471 	uint32_t num_hostid = 0;
3472 
3473 	rrela = cmd->cdw10_bits.resv_release.rrela;
3474 	iekey = cmd->cdw10_bits.resv_release.iekey;
3475 	rtype = cmd->cdw10_bits.resv_release.rtype;
3476 
3477 	if (req->iovcnt > 0 && req->length >= sizeof(crkey)) {
3478 		struct spdk_iov_xfer ix;
3479 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3480 		spdk_iov_xfer_to_buf(&ix, &crkey, sizeof(crkey));
3481 	} else {
3482 		SPDK_ERRLOG("No key provided. Failing request.\n");
3483 		status = SPDK_NVME_SC_INVALID_FIELD;
3484 		goto exit;
3485 	}
3486 
3487 	SPDK_DEBUGLOG(nvmf, "RELEASE: RRELA %u, IEKEY %u, RTYPE %u, "
3488 		      "CRKEY 0x%"PRIx64"\n",  rrela, iekey, rtype, crkey);
3489 
3490 	if (iekey) {
3491 		SPDK_ERRLOG("Ignore existing key field set to 1\n");
3492 		status = SPDK_NVME_SC_INVALID_FIELD;
3493 		update_sgroup = false;
3494 		goto exit;
3495 	}
3496 
3497 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3498 	if (!reg || reg->rkey != crkey) {
3499 		SPDK_ERRLOG("No registrant or current key doesn't match "
3500 			    "with existing registrant key\n");
3501 		status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3502 		update_sgroup = false;
3503 		goto exit;
3504 	}
3505 
3506 	num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3507 			SPDK_NVMF_MAX_NUM_REGISTRANTS,
3508 			&ctrlr->hostid);
3509 
3510 	switch (rrela) {
3511 	case SPDK_NVME_RESERVE_RELEASE:
3512 		if (!ns->holder) {
3513 			SPDK_DEBUGLOG(nvmf, "RELEASE: no holder\n");
3514 			update_sgroup = false;
3515 			goto exit;
3516 		}
3517 		if (ns->rtype != rtype) {
3518 			SPDK_ERRLOG("Type doesn't match\n");
3519 			status = SPDK_NVME_SC_INVALID_FIELD;
3520 			update_sgroup = false;
3521 			goto exit;
3522 		}
3523 		if (!nvmf_ns_reservation_registrant_is_holder(ns, reg)) {
3524 			/* not the reservation holder, this isn't an error */
3525 			update_sgroup = false;
3526 			goto exit;
3527 		}
3528 
3529 		rtype = ns->rtype;
3530 		nvmf_ns_reservation_release_reservation(ns);
3531 
3532 		if (num_hostid && rtype != SPDK_NVME_RESERVE_WRITE_EXCLUSIVE &&
3533 		    rtype != SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3534 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3535 							      hostid_list,
3536 							      num_hostid,
3537 							      SPDK_NVME_RESERVATION_RELEASED);
3538 		}
3539 		break;
3540 	case SPDK_NVME_RESERVE_CLEAR:
3541 		nvmf_ns_reservation_clear_all_registrants(ns);
3542 		if (num_hostid) {
3543 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3544 							      hostid_list,
3545 							      num_hostid,
3546 							      SPDK_NVME_RESERVATION_PREEMPTED);
3547 		}
3548 		break;
3549 	default:
3550 		status = SPDK_NVME_SC_INVALID_FIELD;
3551 		update_sgroup = false;
3552 		goto exit;
3553 	}
3554 
3555 exit:
3556 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3557 	req->rsp->nvme_cpl.status.sc = status;
3558 	return update_sgroup;
3559 }
3560 
3561 static void
3562 nvmf_ns_reservation_report(struct spdk_nvmf_ns *ns,
3563 			   struct spdk_nvmf_ctrlr *ctrlr,
3564 			   struct spdk_nvmf_request *req)
3565 {
3566 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3567 	struct spdk_nvmf_registrant *reg, *tmp;
3568 	struct spdk_nvme_reservation_status_extended_data status_data = { 0 };
3569 	struct spdk_iov_xfer ix;
3570 	uint32_t transfer_len;
3571 	uint32_t regctl = 0;
3572 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3573 
3574 	if (req->iovcnt == 0) {
3575 		SPDK_ERRLOG("No data transfer specified for request. "
3576 			    " Unable to transfer back response.\n");
3577 		status = SPDK_NVME_SC_INVALID_FIELD;
3578 		goto exit;
3579 	}
3580 
3581 	if (!cmd->cdw11_bits.resv_report.eds) {
3582 		SPDK_ERRLOG("NVMeoF uses extended controller data structure, "
3583 			    "please set EDS bit in cdw11 and try again\n");
3584 		status = SPDK_NVME_SC_HOSTID_INCONSISTENT_FORMAT;
3585 		goto exit;
3586 	}
3587 
3588 	/* Number of Dwords of the Reservation Status data structure to transfer */
3589 	transfer_len = (cmd->cdw10 + 1) * sizeof(uint32_t);
3590 
3591 	if (transfer_len < sizeof(struct spdk_nvme_reservation_status_extended_data)) {
3592 		status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3593 		goto exit;
3594 	}
3595 
3596 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3597 
3598 	status_data.data.gen = ns->gen;
3599 	status_data.data.rtype = ns->rtype;
3600 	status_data.data.ptpls = ns->ptpl_activated;
3601 
3602 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3603 		regctl++;
3604 	}
3605 
3606 	/*
3607 	 * We report the number of registrants as per the spec here, even if
3608 	 * the iov isn't big enough to contain them all. In that case, the
3609 	 * spdk_iov_xfer_from_buf() won't actually copy any of the remaining
3610 	 * data; as it keeps track of the iov cursor itself, it's simplest to
3611 	 * just walk the entire list anyway.
3612 	 */
3613 	status_data.data.regctl = regctl;
3614 
3615 	spdk_iov_xfer_from_buf(&ix, &status_data, sizeof(status_data));
3616 
3617 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3618 		struct spdk_nvme_registered_ctrlr_extended_data ctrlr_data = { 0 };
3619 
3620 		/* Set to 0xffffh for dynamic controller */
3621 		ctrlr_data.cntlid = 0xffff;
3622 		ctrlr_data.rcsts.status = (ns->holder == reg) ? true : false;
3623 		ctrlr_data.rkey = reg->rkey;
3624 		spdk_uuid_copy((struct spdk_uuid *)ctrlr_data.hostid, &reg->hostid);
3625 
3626 		spdk_iov_xfer_from_buf(&ix, &ctrlr_data, sizeof(ctrlr_data));
3627 	}
3628 
3629 exit:
3630 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3631 	req->rsp->nvme_cpl.status.sc = status;
3632 	return;
3633 }
3634 
3635 static void
3636 nvmf_ns_reservation_complete(void *ctx)
3637 {
3638 	struct spdk_nvmf_request *req = ctx;
3639 
3640 	spdk_nvmf_request_complete(req);
3641 }
3642 
3643 static void
3644 _nvmf_ns_reservation_update_done(struct spdk_nvmf_subsystem *subsystem,
3645 				 void *cb_arg, int status)
3646 {
3647 	struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)cb_arg;
3648 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3649 
3650 	spdk_thread_send_msg(group->thread, nvmf_ns_reservation_complete, req);
3651 }
3652 
3653 void
3654 nvmf_ns_reservation_request(void *ctx)
3655 {
3656 	struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)ctx;
3657 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3658 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3659 	uint32_t nsid;
3660 	struct spdk_nvmf_ns *ns;
3661 	bool update_sgroup = false;
3662 	int status = 0;
3663 
3664 	nsid = cmd->nsid;
3665 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3666 	assert(ns != NULL);
3667 
3668 	switch (cmd->opc) {
3669 	case SPDK_NVME_OPC_RESERVATION_REGISTER:
3670 		update_sgroup = nvmf_ns_reservation_register(ns, ctrlr, req);
3671 		break;
3672 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3673 		update_sgroup = nvmf_ns_reservation_acquire(ns, ctrlr, req);
3674 		break;
3675 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3676 		update_sgroup = nvmf_ns_reservation_release(ns, ctrlr, req);
3677 		break;
3678 	case SPDK_NVME_OPC_RESERVATION_REPORT:
3679 		nvmf_ns_reservation_report(ns, ctrlr, req);
3680 		break;
3681 	default:
3682 		break;
3683 	}
3684 
3685 	/* update reservation information to subsystem's poll group */
3686 	if (update_sgroup) {
3687 		if (ns->ptpl_activated || cmd->opc == SPDK_NVME_OPC_RESERVATION_REGISTER) {
3688 			if (nvmf_ns_update_reservation_info(ns) != 0) {
3689 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3690 			}
3691 		}
3692 		status = nvmf_subsystem_update_ns(ctrlr->subsys, _nvmf_ns_reservation_update_done, req);
3693 		if (status == 0) {
3694 			return;
3695 		}
3696 	}
3697 
3698 	_nvmf_ns_reservation_update_done(ctrlr->subsys, req, status);
3699 }
3700 
3701 static bool
3702 nvmf_ns_is_ptpl_capable_json(const struct spdk_nvmf_ns *ns)
3703 {
3704 	return ns->ptpl_file != NULL;
3705 }
3706 
3707 static struct spdk_nvmf_ns_reservation_ops g_reservation_ops = {
3708 	.is_ptpl_capable = nvmf_ns_is_ptpl_capable_json,
3709 	.update = nvmf_ns_reservation_update_json,
3710 	.load = nvmf_ns_reservation_load_json,
3711 };
3712 
3713 bool
3714 nvmf_ns_is_ptpl_capable(const struct spdk_nvmf_ns *ns)
3715 {
3716 	return g_reservation_ops.is_ptpl_capable(ns);
3717 }
3718 
3719 static int
3720 nvmf_ns_reservation_update(const struct spdk_nvmf_ns *ns,
3721 			   const struct spdk_nvmf_reservation_info *info)
3722 {
3723 	return g_reservation_ops.update(ns, info);
3724 }
3725 
3726 static int
3727 nvmf_ns_reservation_load(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info)
3728 {
3729 	return g_reservation_ops.load(ns, info);
3730 }
3731 
3732 void
3733 spdk_nvmf_set_custom_ns_reservation_ops(const struct spdk_nvmf_ns_reservation_ops *ops)
3734 {
3735 	g_reservation_ops = *ops;
3736 }
3737 
3738 int
3739 spdk_nvmf_subsystem_set_ana_reporting(struct spdk_nvmf_subsystem *subsystem,
3740 				      bool ana_reporting)
3741 {
3742 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
3743 		return -EAGAIN;
3744 	}
3745 
3746 	subsystem->flags.ana_reporting = ana_reporting;
3747 
3748 	return 0;
3749 }
3750 
3751 bool
3752 spdk_nvmf_subsystem_get_ana_reporting(struct spdk_nvmf_subsystem *subsystem)
3753 {
3754 	return subsystem->flags.ana_reporting;
3755 }
3756 
3757 struct subsystem_listener_update_ctx {
3758 	struct spdk_nvmf_subsystem_listener *listener;
3759 
3760 	spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn;
3761 	void *cb_arg;
3762 };
3763 
3764 static void
3765 subsystem_listener_update_done(struct spdk_io_channel_iter *i, int status)
3766 {
3767 	struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
3768 
3769 	if (ctx->cb_fn) {
3770 		ctx->cb_fn(ctx->cb_arg, status);
3771 	}
3772 	free(ctx);
3773 }
3774 
3775 static void
3776 subsystem_listener_update_on_pg(struct spdk_io_channel_iter *i)
3777 {
3778 	struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
3779 	struct spdk_nvmf_subsystem_listener *listener;
3780 	struct spdk_nvmf_poll_group *group;
3781 	struct spdk_nvmf_ctrlr *ctrlr;
3782 
3783 	listener = ctx->listener;
3784 	group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i));
3785 
3786 	TAILQ_FOREACH(ctrlr, &listener->subsystem->ctrlrs, link) {
3787 		if (ctrlr->thread != spdk_get_thread()) {
3788 			continue;
3789 		}
3790 
3791 		if (ctrlr->admin_qpair && ctrlr->admin_qpair->group == group && ctrlr->listener == listener) {
3792 			nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
3793 		}
3794 	}
3795 
3796 	spdk_for_each_channel_continue(i, 0);
3797 }
3798 
3799 void
3800 spdk_nvmf_subsystem_set_ana_state(struct spdk_nvmf_subsystem *subsystem,
3801 				  const struct spdk_nvme_transport_id *trid,
3802 				  enum spdk_nvme_ana_state ana_state, uint32_t anagrpid,
3803 				  spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, void *cb_arg)
3804 {
3805 	struct spdk_nvmf_subsystem_listener *listener;
3806 	struct subsystem_listener_update_ctx *ctx;
3807 	uint32_t i;
3808 
3809 	assert(cb_fn != NULL);
3810 	assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
3811 	       subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED);
3812 
3813 	if (!subsystem->flags.ana_reporting) {
3814 		SPDK_ERRLOG("ANA reporting is disabled\n");
3815 		cb_fn(cb_arg, -EINVAL);
3816 		return;
3817 	}
3818 
3819 	/* ANA Change state is not used, ANA Persistent Loss state
3820 	 * is not supported yet.
3821 	 */
3822 	if (!(ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE ||
3823 	      ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE ||
3824 	      ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE)) {
3825 		SPDK_ERRLOG("ANA state %d is not supported\n", ana_state);
3826 		cb_fn(cb_arg, -ENOTSUP);
3827 		return;
3828 	}
3829 
3830 	if (anagrpid > subsystem->max_nsid) {
3831 		SPDK_ERRLOG("ANA group ID %" PRIu32 " is more than maximum\n", anagrpid);
3832 		cb_fn(cb_arg, -EINVAL);
3833 		return;
3834 	}
3835 
3836 	listener = nvmf_subsystem_find_listener(subsystem, trid);
3837 	if (!listener) {
3838 		SPDK_ERRLOG("Unable to find listener.\n");
3839 		cb_fn(cb_arg, -EINVAL);
3840 		return;
3841 	}
3842 
3843 	if (anagrpid != 0 && listener->ana_state[anagrpid - 1] == ana_state) {
3844 		cb_fn(cb_arg, 0);
3845 		return;
3846 	}
3847 
3848 	ctx = calloc(1, sizeof(*ctx));
3849 	if (!ctx) {
3850 		SPDK_ERRLOG("Unable to allocate context\n");
3851 		cb_fn(cb_arg, -ENOMEM);
3852 		return;
3853 	}
3854 
3855 	for (i = 1; i <= subsystem->max_nsid; i++) {
3856 		if (anagrpid == 0 || i == anagrpid) {
3857 			listener->ana_state[i - 1] = ana_state;
3858 		}
3859 	}
3860 	listener->ana_state_change_count++;
3861 
3862 	ctx->listener = listener;
3863 	ctx->cb_fn = cb_fn;
3864 	ctx->cb_arg = cb_arg;
3865 
3866 	spdk_for_each_channel(subsystem->tgt,
3867 			      subsystem_listener_update_on_pg,
3868 			      ctx,
3869 			      subsystem_listener_update_done);
3870 }
3871 
3872 int
3873 spdk_nvmf_subsystem_get_ana_state(struct spdk_nvmf_subsystem *subsystem,
3874 				  const struct spdk_nvme_transport_id *trid,
3875 				  uint32_t anagrpid,
3876 				  enum spdk_nvme_ana_state *ana_state)
3877 {
3878 	assert(ana_state != NULL);
3879 
3880 	struct spdk_nvmf_subsystem_listener *listener;
3881 
3882 	if (!subsystem->flags.ana_reporting) {
3883 		SPDK_ERRLOG("ANA reporting is disabled\n");
3884 		return -EINVAL;
3885 	}
3886 
3887 	if (anagrpid <= 0 || anagrpid > subsystem->max_nsid) {
3888 		SPDK_ERRLOG("ANA group ID %" PRIu32 " is invalid\n", anagrpid);
3889 		return -EINVAL;
3890 	}
3891 
3892 	listener = nvmf_subsystem_find_listener(subsystem, trid);
3893 	if (!listener) {
3894 		SPDK_ERRLOG("Unable to find listener.\n");
3895 		return -EINVAL;
3896 	}
3897 
3898 	*ana_state = listener->ana_state[anagrpid - 1];
3899 	return 0;
3900 }
3901 
3902 bool
3903 spdk_nvmf_subsystem_is_discovery(struct spdk_nvmf_subsystem *subsystem)
3904 {
3905 	return subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
3906 	       subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY;
3907 }
3908 
3909 bool
3910 nvmf_nqn_is_discovery(const char *nqn)
3911 {
3912 	return strcmp(nqn, SPDK_NVMF_DISCOVERY_NQN) == 0;
3913 }
3914