xref: /spdk/lib/nvmf/subsystem.c (revision 2186fc0389db7611703e15b2131e64846c1e65a0)
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 = 0;
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 	nvmf_update_discovery_log(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 		    spdk_bit_array_get(ctrlr->visible_ns, nsid - 1) == visible) {
925 			continue;
926 		}
927 		if (visible) {
928 			spdk_bit_array_set(ctrlr->visible_ns, nsid - 1);
929 		} else {
930 			spdk_bit_array_clear(ctrlr->visible_ns, nsid - 1);
931 		}
932 		send_async_event_ns_notice(ctrlr);
933 		nvmf_ctrlr_ns_changed(ctrlr, nsid);
934 	}
935 
936 	return 0;
937 }
938 
939 int
940 spdk_nvmf_ns_add_host(struct spdk_nvmf_subsystem *subsystem,
941 		      uint32_t nsid,
942 		      const char *hostnqn,
943 		      uint32_t flags)
944 {
945 	SPDK_DTRACE_PROBE4(spdk_nvmf_ns_add_host,
946 			   subsystem->subnqn,
947 			   nsid,
948 			   hostnqn,
949 			   flags);
950 	return nvmf_ns_visible(subsystem, nsid, hostnqn, true);
951 }
952 
953 int
954 spdk_nvmf_ns_remove_host(struct spdk_nvmf_subsystem *subsystem,
955 			 uint32_t nsid,
956 			 const char *hostnqn,
957 			 uint32_t flags)
958 {
959 	SPDK_DTRACE_PROBE4(spdk_nvmf_ns_remove_host,
960 			   subsystem->subnqn,
961 			   nsid,
962 			   hostnqn,
963 			   flags);
964 	return nvmf_ns_visible(subsystem, nsid, hostnqn, false);
965 }
966 
967 /* Must hold subsystem->mutex while calling this function */
968 static struct spdk_nvmf_host *
969 nvmf_subsystem_find_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
970 {
971 	struct spdk_nvmf_host *host = NULL;
972 
973 	TAILQ_FOREACH(host, &subsystem->hosts, link) {
974 		if (strcmp(hostnqn, host->nqn) == 0) {
975 			return host;
976 		}
977 	}
978 
979 	return NULL;
980 }
981 
982 int
983 spdk_nvmf_subsystem_add_host_ext(struct spdk_nvmf_subsystem *subsystem,
984 				 const char *hostnqn, struct spdk_nvmf_host_opts *opts)
985 {
986 	struct spdk_nvmf_host *host;
987 	struct spdk_nvmf_transport *transport;
988 	struct spdk_key *key;
989 	int rc;
990 
991 	if (!nvmf_nqn_is_valid(hostnqn)) {
992 		return -EINVAL;
993 	}
994 
995 	pthread_mutex_lock(&subsystem->mutex);
996 
997 	if (nvmf_subsystem_find_host(subsystem, hostnqn)) {
998 		/* This subsystem already allows the specified host. */
999 		pthread_mutex_unlock(&subsystem->mutex);
1000 		return -EINVAL;
1001 	}
1002 
1003 	host = calloc(1, sizeof(*host));
1004 	if (!host) {
1005 		pthread_mutex_unlock(&subsystem->mutex);
1006 		return -ENOMEM;
1007 	}
1008 
1009 	key = SPDK_GET_FIELD(opts, dhchap_key, NULL);
1010 	if (key != NULL) {
1011 		if (!nvmf_auth_is_supported()) {
1012 			SPDK_ERRLOG("NVMe in-band authentication is unsupported\n");
1013 			pthread_mutex_unlock(&subsystem->mutex);
1014 			nvmf_host_free(host);
1015 			return -EINVAL;
1016 		}
1017 		host->dhchap_key = spdk_key_dup(key);
1018 		if (host->dhchap_key == NULL) {
1019 			pthread_mutex_unlock(&subsystem->mutex);
1020 			nvmf_host_free(host);
1021 			return -EINVAL;
1022 		}
1023 		key = SPDK_GET_FIELD(opts, dhchap_ctrlr_key, NULL);
1024 		if (key != NULL) {
1025 			host->dhchap_ctrlr_key = spdk_key_dup(key);
1026 			if (host->dhchap_ctrlr_key == NULL) {
1027 				pthread_mutex_unlock(&subsystem->mutex);
1028 				nvmf_host_free(host);
1029 				return -EINVAL;
1030 			}
1031 		}
1032 	} else if (SPDK_GET_FIELD(opts, dhchap_ctrlr_key, NULL) != NULL) {
1033 		SPDK_ERRLOG("DH-HMAC-CHAP controller key requires host key to be set\n");
1034 		pthread_mutex_unlock(&subsystem->mutex);
1035 		nvmf_host_free(host);
1036 		return -EINVAL;
1037 	}
1038 
1039 	snprintf(host->nqn, sizeof(host->nqn), "%s", hostnqn);
1040 
1041 	SPDK_DTRACE_PROBE2(nvmf_subsystem_add_host, subsystem->subnqn, host->nqn);
1042 
1043 	TAILQ_INSERT_HEAD(&subsystem->hosts, host, link);
1044 
1045 	if (!TAILQ_EMPTY(&subsystem->listeners)) {
1046 		nvmf_update_discovery_log(subsystem->tgt, hostnqn);
1047 	}
1048 
1049 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
1050 	     transport = spdk_nvmf_transport_get_next(transport)) {
1051 		if (transport->ops->subsystem_add_host) {
1052 			rc = transport->ops->subsystem_add_host(transport, subsystem, hostnqn,
1053 								SPDK_GET_FIELD(opts, params, NULL));
1054 			if (rc) {
1055 				SPDK_ERRLOG("Unable to add host to %s transport\n", transport->ops->name);
1056 				/* Remove this host from all transports we've managed to add it to. */
1057 				pthread_mutex_unlock(&subsystem->mutex);
1058 				spdk_nvmf_subsystem_remove_host(subsystem, hostnqn);
1059 				return rc;
1060 			}
1061 		}
1062 	}
1063 
1064 	pthread_mutex_unlock(&subsystem->mutex);
1065 
1066 	return 0;
1067 }
1068 
1069 int
1070 spdk_nvmf_subsystem_add_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
1071 			     const struct spdk_json_val *params)
1072 {
1073 	struct spdk_nvmf_host_opts opts = {};
1074 
1075 	opts.size = SPDK_SIZEOF(&opts, params);
1076 	opts.params = params;
1077 
1078 	return spdk_nvmf_subsystem_add_host_ext(subsystem, hostnqn, &opts);
1079 }
1080 
1081 int
1082 spdk_nvmf_subsystem_remove_host(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
1083 {
1084 	struct spdk_nvmf_host *host;
1085 	struct spdk_nvmf_transport *transport;
1086 
1087 	pthread_mutex_lock(&subsystem->mutex);
1088 
1089 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1090 	if (host == NULL) {
1091 		pthread_mutex_unlock(&subsystem->mutex);
1092 		return -ENOENT;
1093 	}
1094 
1095 	SPDK_DTRACE_PROBE2(nvmf_subsystem_remove_host, subsystem->subnqn, host->nqn);
1096 
1097 	nvmf_subsystem_remove_host(subsystem, host);
1098 
1099 	if (!TAILQ_EMPTY(&subsystem->listeners)) {
1100 		nvmf_update_discovery_log(subsystem->tgt, hostnqn);
1101 	}
1102 
1103 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
1104 	     transport = spdk_nvmf_transport_get_next(transport)) {
1105 		if (transport->ops->subsystem_remove_host) {
1106 			transport->ops->subsystem_remove_host(transport, subsystem, hostnqn);
1107 		}
1108 	}
1109 
1110 	pthread_mutex_unlock(&subsystem->mutex);
1111 
1112 	return 0;
1113 }
1114 
1115 int
1116 spdk_nvmf_subsystem_set_keys(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
1117 			     struct spdk_nvmf_subsystem_key_opts *opts)
1118 {
1119 	struct spdk_nvmf_host *host;
1120 	struct spdk_key *key, *ckey;
1121 
1122 	if (!nvmf_auth_is_supported()) {
1123 		SPDK_ERRLOG("NVMe in-band authentication is unsupported\n");
1124 		return -EINVAL;
1125 	}
1126 
1127 	pthread_mutex_lock(&subsystem->mutex);
1128 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1129 	if (host == NULL) {
1130 		pthread_mutex_unlock(&subsystem->mutex);
1131 		return -EINVAL;
1132 	}
1133 
1134 	if (SPDK_GET_FIELD(opts, dhchap_key, host->dhchap_key) == NULL &&
1135 	    SPDK_GET_FIELD(opts, dhchap_ctrlr_key, host->dhchap_ctrlr_key) != NULL) {
1136 		SPDK_ERRLOG("DH-HMAC-CHAP controller key requires host key to be set\n");
1137 		pthread_mutex_unlock(&subsystem->mutex);
1138 		return -EINVAL;
1139 	}
1140 	key = SPDK_GET_FIELD(opts, dhchap_key, NULL);
1141 	if (key != NULL) {
1142 		key = spdk_key_dup(key);
1143 		if (key == NULL) {
1144 			pthread_mutex_unlock(&subsystem->mutex);
1145 			return -EINVAL;
1146 		}
1147 	}
1148 	ckey = SPDK_GET_FIELD(opts, dhchap_ctrlr_key, NULL);
1149 	if (ckey != NULL) {
1150 		ckey = spdk_key_dup(ckey);
1151 		if (ckey == NULL) {
1152 			pthread_mutex_unlock(&subsystem->mutex);
1153 			spdk_keyring_put_key(key);
1154 			return -EINVAL;
1155 		}
1156 	}
1157 	if (SPDK_FIELD_VALID(opts, dhchap_key)) {
1158 		spdk_keyring_put_key(host->dhchap_key);
1159 		host->dhchap_key = key;
1160 	}
1161 	if (SPDK_FIELD_VALID(opts, dhchap_ctrlr_key)) {
1162 		spdk_keyring_put_key(host->dhchap_ctrlr_key);
1163 		host->dhchap_ctrlr_key = ckey;
1164 	}
1165 	pthread_mutex_unlock(&subsystem->mutex);
1166 
1167 	return 0;
1168 }
1169 
1170 struct nvmf_subsystem_disconnect_host_ctx {
1171 	struct spdk_nvmf_subsystem		*subsystem;
1172 	char					*hostnqn;
1173 	spdk_nvmf_tgt_subsystem_listen_done_fn	cb_fn;
1174 	void					*cb_arg;
1175 };
1176 
1177 static void
1178 nvmf_subsystem_disconnect_host_fini(struct spdk_io_channel_iter *i, int status)
1179 {
1180 	struct nvmf_subsystem_disconnect_host_ctx *ctx;
1181 
1182 	ctx = spdk_io_channel_iter_get_ctx(i);
1183 
1184 	if (ctx->cb_fn) {
1185 		ctx->cb_fn(ctx->cb_arg, status);
1186 	}
1187 	free(ctx->hostnqn);
1188 	free(ctx);
1189 }
1190 
1191 static void
1192 nvmf_subsystem_disconnect_qpairs_by_host(struct spdk_io_channel_iter *i)
1193 {
1194 	struct nvmf_subsystem_disconnect_host_ctx *ctx;
1195 	struct spdk_nvmf_poll_group *group;
1196 	struct spdk_io_channel *ch;
1197 	struct spdk_nvmf_qpair *qpair, *tmp_qpair;
1198 	struct spdk_nvmf_ctrlr *ctrlr;
1199 
1200 	ctx = spdk_io_channel_iter_get_ctx(i);
1201 	ch = spdk_io_channel_iter_get_channel(i);
1202 	group = spdk_io_channel_get_ctx(ch);
1203 
1204 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, tmp_qpair) {
1205 		ctrlr = qpair->ctrlr;
1206 
1207 		if (ctrlr == NULL || ctrlr->subsys != ctx->subsystem) {
1208 			continue;
1209 		}
1210 
1211 		if (strncmp(ctrlr->hostnqn, ctx->hostnqn, sizeof(ctrlr->hostnqn)) == 0) {
1212 			/* Right now this does not wait for the queue pairs to actually disconnect. */
1213 			spdk_nvmf_qpair_disconnect(qpair);
1214 		}
1215 	}
1216 	spdk_for_each_channel_continue(i, 0);
1217 }
1218 
1219 int
1220 spdk_nvmf_subsystem_disconnect_host(struct spdk_nvmf_subsystem *subsystem,
1221 				    const char *hostnqn,
1222 				    spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1223 				    void *cb_arg)
1224 {
1225 	struct nvmf_subsystem_disconnect_host_ctx *ctx;
1226 
1227 	ctx = calloc(1, sizeof(struct nvmf_subsystem_disconnect_host_ctx));
1228 	if (ctx == NULL) {
1229 		return -ENOMEM;
1230 	}
1231 
1232 	ctx->hostnqn = strdup(hostnqn);
1233 	if (ctx->hostnqn == NULL) {
1234 		free(ctx);
1235 		return -ENOMEM;
1236 	}
1237 
1238 	ctx->subsystem = subsystem;
1239 	ctx->cb_fn = cb_fn;
1240 	ctx->cb_arg = cb_arg;
1241 
1242 	spdk_for_each_channel(subsystem->tgt, nvmf_subsystem_disconnect_qpairs_by_host, ctx,
1243 			      nvmf_subsystem_disconnect_host_fini);
1244 
1245 	return 0;
1246 }
1247 
1248 int
1249 spdk_nvmf_subsystem_set_allow_any_host(struct spdk_nvmf_subsystem *subsystem, bool allow_any_host)
1250 {
1251 	pthread_mutex_lock(&subsystem->mutex);
1252 	subsystem->allow_any_host = allow_any_host;
1253 	if (!TAILQ_EMPTY(&subsystem->listeners)) {
1254 		nvmf_update_discovery_log(subsystem->tgt, NULL);
1255 	}
1256 	pthread_mutex_unlock(&subsystem->mutex);
1257 
1258 	return 0;
1259 }
1260 
1261 bool
1262 spdk_nvmf_subsystem_get_allow_any_host(const struct spdk_nvmf_subsystem *subsystem)
1263 {
1264 	bool allow_any_host;
1265 	struct spdk_nvmf_subsystem *sub;
1266 
1267 	/* Technically, taking the mutex modifies data in the subsystem. But the const
1268 	 * is still important to convey that this doesn't mutate any other data. Cast
1269 	 * it away to work around this. */
1270 	sub = (struct spdk_nvmf_subsystem *)subsystem;
1271 
1272 	pthread_mutex_lock(&sub->mutex);
1273 	allow_any_host = sub->allow_any_host;
1274 	pthread_mutex_unlock(&sub->mutex);
1275 
1276 	return allow_any_host;
1277 }
1278 
1279 bool
1280 spdk_nvmf_subsystem_host_allowed(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
1281 {
1282 	bool allowed;
1283 
1284 	if (!hostnqn) {
1285 		return false;
1286 	}
1287 
1288 	pthread_mutex_lock(&subsystem->mutex);
1289 
1290 	if (subsystem->allow_any_host) {
1291 		pthread_mutex_unlock(&subsystem->mutex);
1292 		return true;
1293 	}
1294 
1295 	allowed =  nvmf_subsystem_find_host(subsystem, hostnqn) != NULL;
1296 	pthread_mutex_unlock(&subsystem->mutex);
1297 
1298 	return allowed;
1299 }
1300 
1301 bool
1302 nvmf_subsystem_host_auth_required(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn)
1303 {
1304 	struct spdk_nvmf_host *host;
1305 	bool status;
1306 
1307 	pthread_mutex_lock(&subsystem->mutex);
1308 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1309 	status = host != NULL && host->dhchap_key != NULL;
1310 	pthread_mutex_unlock(&subsystem->mutex);
1311 
1312 	return status;
1313 }
1314 
1315 struct spdk_key *
1316 nvmf_subsystem_get_dhchap_key(struct spdk_nvmf_subsystem *subsystem, const char *hostnqn,
1317 			      enum nvmf_auth_key_type type)
1318 {
1319 	struct spdk_nvmf_host *host;
1320 	struct spdk_key *key = NULL;
1321 
1322 	pthread_mutex_lock(&subsystem->mutex);
1323 	host = nvmf_subsystem_find_host(subsystem, hostnqn);
1324 	if (host != NULL) {
1325 		switch (type) {
1326 		case NVMF_AUTH_KEY_HOST:
1327 			key = host->dhchap_key;
1328 			break;
1329 		case NVMF_AUTH_KEY_CTRLR:
1330 			key = host->dhchap_ctrlr_key;
1331 			break;
1332 		}
1333 		if (key != NULL) {
1334 			key = spdk_key_dup(key);
1335 		}
1336 	}
1337 	pthread_mutex_unlock(&subsystem->mutex);
1338 
1339 	return key;
1340 }
1341 
1342 struct spdk_nvmf_host *
1343 spdk_nvmf_subsystem_get_first_host(struct spdk_nvmf_subsystem *subsystem)
1344 {
1345 	return TAILQ_FIRST(&subsystem->hosts);
1346 }
1347 
1348 
1349 struct spdk_nvmf_host *
1350 spdk_nvmf_subsystem_get_next_host(struct spdk_nvmf_subsystem *subsystem,
1351 				  struct spdk_nvmf_host *prev_host)
1352 {
1353 	return TAILQ_NEXT(prev_host, link);
1354 }
1355 
1356 const char *
1357 spdk_nvmf_host_get_nqn(const struct spdk_nvmf_host *host)
1358 {
1359 	return host->nqn;
1360 }
1361 
1362 struct spdk_nvmf_subsystem_listener *
1363 nvmf_subsystem_find_listener(struct spdk_nvmf_subsystem *subsystem,
1364 			     const struct spdk_nvme_transport_id *trid)
1365 {
1366 	struct spdk_nvmf_subsystem_listener *listener;
1367 
1368 	TAILQ_FOREACH(listener, &subsystem->listeners, link) {
1369 		if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) {
1370 			return listener;
1371 		}
1372 	}
1373 
1374 	return NULL;
1375 }
1376 
1377 /**
1378  * Function to be called once the target is listening.
1379  *
1380  * \param ctx Context argument passed to this function.
1381  * \param status 0 if it completed successfully, or negative errno if it failed.
1382  */
1383 static void
1384 _nvmf_subsystem_add_listener_done(void *ctx, int status)
1385 {
1386 	struct spdk_nvmf_subsystem_listener *listener = ctx;
1387 
1388 	if (status) {
1389 		listener->cb_fn(listener->cb_arg, status);
1390 		free(listener);
1391 		return;
1392 	}
1393 
1394 	TAILQ_INSERT_HEAD(&listener->subsystem->listeners, listener, link);
1395 
1396 	if (spdk_nvmf_subsystem_is_discovery(listener->subsystem)) {
1397 		status = nvmf_tgt_update_mdns_prr(listener->subsystem->tgt);
1398 		if (status) {
1399 			TAILQ_REMOVE(&listener->subsystem->listeners, listener, link);
1400 			listener->cb_fn(listener->cb_arg, status);
1401 			free(listener);
1402 			return;
1403 		}
1404 	}
1405 
1406 	nvmf_update_discovery_log(listener->subsystem->tgt, NULL);
1407 	listener->cb_fn(listener->cb_arg, status);
1408 }
1409 
1410 void
1411 spdk_nvmf_subsystem_listener_opts_init(struct spdk_nvmf_listener_opts *opts, size_t size)
1412 {
1413 	if (opts == NULL) {
1414 		SPDK_ERRLOG("opts should not be NULL\n");
1415 		assert(false);
1416 		return;
1417 	}
1418 	if (size == 0) {
1419 		SPDK_ERRLOG("size should not be zero\n");
1420 		assert(false);
1421 		return;
1422 	}
1423 
1424 	memset(opts, 0, size);
1425 	opts->opts_size = size;
1426 
1427 #define FIELD_OK(field) \
1428 	offsetof(struct spdk_nvmf_listener_opts, field) + sizeof(opts->field) <= size
1429 
1430 #define SET_FIELD(field, value) \
1431 	if (FIELD_OK(field)) { \
1432 		opts->field = value; \
1433 	} \
1434 
1435 	SET_FIELD(secure_channel, false);
1436 	SET_FIELD(ana_state, SPDK_NVME_ANA_OPTIMIZED_STATE);
1437 	SET_FIELD(sock_impl, NULL);
1438 
1439 #undef FIELD_OK
1440 #undef SET_FIELD
1441 }
1442 
1443 static int
1444 listener_opts_copy(struct spdk_nvmf_listener_opts *src, struct spdk_nvmf_listener_opts *dst)
1445 {
1446 	if (src->opts_size == 0) {
1447 		SPDK_ERRLOG("source structure size should not be zero\n");
1448 		assert(false);
1449 		return -EINVAL;
1450 	}
1451 
1452 	memset(dst, 0, sizeof(*dst));
1453 	dst->opts_size = src->opts_size;
1454 
1455 #define FIELD_OK(field) \
1456 	offsetof(struct spdk_nvmf_listener_opts, field) + sizeof(src->field) <= src->opts_size
1457 
1458 #define SET_FIELD(field) \
1459 	if (FIELD_OK(field)) { \
1460 		dst->field = src->field; \
1461 	} \
1462 
1463 	SET_FIELD(secure_channel);
1464 	SET_FIELD(ana_state);
1465 	SET_FIELD(sock_impl);
1466 	/* We should not remove this statement, but need to update the assert statement
1467 	 * if we add a new field, and also add a corresponding SET_FIELD statement. */
1468 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listener_opts) == 24, "Incorrect size");
1469 
1470 #undef SET_FIELD
1471 #undef FIELD_OK
1472 
1473 	return 0;
1474 }
1475 
1476 static void
1477 _nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem,
1478 			     struct spdk_nvme_transport_id *trid,
1479 			     spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1480 			     void *cb_arg, struct spdk_nvmf_listener_opts *opts)
1481 {
1482 	struct spdk_nvmf_transport *transport;
1483 	struct spdk_nvmf_subsystem_listener *listener;
1484 	struct spdk_nvmf_listener *tr_listener;
1485 	uint32_t i;
1486 	uint32_t id;
1487 	int rc = 0;
1488 
1489 	assert(cb_fn != NULL);
1490 
1491 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
1492 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
1493 		cb_fn(cb_arg, -EAGAIN);
1494 		return;
1495 	}
1496 
1497 	if (nvmf_subsystem_find_listener(subsystem, trid)) {
1498 		/* Listener already exists in this subsystem */
1499 		cb_fn(cb_arg, 0);
1500 		return;
1501 	}
1502 
1503 	transport = spdk_nvmf_tgt_get_transport(subsystem->tgt, trid->trstring);
1504 	if (!transport) {
1505 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
1506 			    trid->trstring);
1507 		cb_fn(cb_arg, -EINVAL);
1508 		return;
1509 	}
1510 
1511 	tr_listener = nvmf_transport_find_listener(transport, trid);
1512 	if (!tr_listener) {
1513 		SPDK_ERRLOG("Cannot find transport listener for %s\n", trid->traddr);
1514 		cb_fn(cb_arg, -EINVAL);
1515 		return;
1516 	}
1517 
1518 	listener = calloc(1, sizeof(*listener));
1519 	if (!listener) {
1520 		cb_fn(cb_arg, -ENOMEM);
1521 		return;
1522 	}
1523 
1524 	listener->trid = &tr_listener->trid;
1525 	listener->transport = transport;
1526 	listener->cb_fn = cb_fn;
1527 	listener->cb_arg = cb_arg;
1528 	listener->subsystem = subsystem;
1529 	listener->ana_state = calloc(subsystem->max_nsid, sizeof(enum spdk_nvme_ana_state));
1530 	if (!listener->ana_state) {
1531 		free(listener);
1532 		cb_fn(cb_arg, -ENOMEM);
1533 		return;
1534 	}
1535 
1536 	spdk_nvmf_subsystem_listener_opts_init(&listener->opts, sizeof(listener->opts));
1537 	if (opts != NULL) {
1538 		rc = listener_opts_copy(opts, &listener->opts);
1539 		if (rc) {
1540 			SPDK_ERRLOG("Unable to copy listener options\n");
1541 			free(listener->ana_state);
1542 			free(listener);
1543 			cb_fn(cb_arg, -EINVAL);
1544 			return;
1545 		}
1546 	}
1547 
1548 	id = spdk_bit_array_find_first_clear(subsystem->used_listener_ids, 0);
1549 	if (id == UINT32_MAX) {
1550 		SPDK_ERRLOG("Cannot add any more listeners\n");
1551 		free(listener->ana_state);
1552 		free(listener->opts.sock_impl);
1553 		free(listener);
1554 		cb_fn(cb_arg, -EINVAL);
1555 		return;
1556 	}
1557 
1558 	spdk_bit_array_set(subsystem->used_listener_ids, id);
1559 	listener->id = id;
1560 
1561 	for (i = 0; i < subsystem->max_nsid; i++) {
1562 		listener->ana_state[i] = listener->opts.ana_state;
1563 	}
1564 
1565 	if (transport->ops->listen_associate != NULL) {
1566 		rc = transport->ops->listen_associate(transport, subsystem, trid);
1567 	}
1568 
1569 	SPDK_DTRACE_PROBE4(nvmf_subsystem_add_listener, subsystem->subnqn, listener->trid->trtype,
1570 			   listener->trid->traddr, listener->trid->trsvcid);
1571 
1572 	_nvmf_subsystem_add_listener_done(listener, rc);
1573 }
1574 
1575 void
1576 spdk_nvmf_subsystem_add_listener(struct spdk_nvmf_subsystem *subsystem,
1577 				 struct spdk_nvme_transport_id *trid,
1578 				 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1579 				 void *cb_arg)
1580 {
1581 	_nvmf_subsystem_add_listener(subsystem, trid, cb_fn, cb_arg, NULL);
1582 }
1583 
1584 void
1585 spdk_nvmf_subsystem_add_listener_ext(struct spdk_nvmf_subsystem *subsystem,
1586 				     struct spdk_nvme_transport_id *trid,
1587 				     spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn,
1588 				     void *cb_arg, struct spdk_nvmf_listener_opts *opts)
1589 {
1590 	_nvmf_subsystem_add_listener(subsystem, trid, cb_fn, cb_arg, opts);
1591 }
1592 
1593 int
1594 spdk_nvmf_subsystem_remove_listener(struct spdk_nvmf_subsystem *subsystem,
1595 				    const struct spdk_nvme_transport_id *trid)
1596 {
1597 	struct spdk_nvmf_subsystem_listener *listener;
1598 
1599 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
1600 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
1601 		return -EAGAIN;
1602 	}
1603 
1604 	listener = nvmf_subsystem_find_listener(subsystem, trid);
1605 	if (listener == NULL) {
1606 		return -ENOENT;
1607 	}
1608 
1609 	SPDK_DTRACE_PROBE4(nvmf_subsystem_remove_listener, subsystem->subnqn, listener->trid->trtype,
1610 			   listener->trid->traddr, listener->trid->trsvcid);
1611 
1612 	_nvmf_subsystem_remove_listener(subsystem, listener, false);
1613 
1614 	return 0;
1615 }
1616 
1617 void
1618 nvmf_subsystem_remove_all_listeners(struct spdk_nvmf_subsystem *subsystem,
1619 				    bool stop)
1620 {
1621 	struct spdk_nvmf_subsystem_listener *listener, *listener_tmp;
1622 
1623 	TAILQ_FOREACH_SAFE(listener, &subsystem->listeners, link, listener_tmp) {
1624 		_nvmf_subsystem_remove_listener(subsystem, listener, stop);
1625 	}
1626 }
1627 
1628 bool
1629 spdk_nvmf_subsystem_listener_allowed(struct spdk_nvmf_subsystem *subsystem,
1630 				     const struct spdk_nvme_transport_id *trid)
1631 {
1632 	struct spdk_nvmf_subsystem_listener *listener;
1633 
1634 	TAILQ_FOREACH(listener, &subsystem->listeners, link) {
1635 		if (spdk_nvme_transport_id_compare(listener->trid, trid) == 0) {
1636 			return true;
1637 		}
1638 	}
1639 
1640 	if (!strcmp(subsystem->subnqn, SPDK_NVMF_DISCOVERY_NQN)) {
1641 		SPDK_WARNLOG("Allowing connection to discovery subsystem on %s/%s/%s, "
1642 			     "even though this listener was not added to the discovery "
1643 			     "subsystem.  This behavior is deprecated and will be removed "
1644 			     "in a future release.\n",
1645 			     spdk_nvme_transport_id_trtype_str(trid->trtype), trid->traddr, trid->trsvcid);
1646 		return true;
1647 	}
1648 
1649 	return false;
1650 }
1651 
1652 struct spdk_nvmf_subsystem_listener *
1653 spdk_nvmf_subsystem_get_first_listener(struct spdk_nvmf_subsystem *subsystem)
1654 {
1655 	return TAILQ_FIRST(&subsystem->listeners);
1656 }
1657 
1658 struct spdk_nvmf_subsystem_listener *
1659 spdk_nvmf_subsystem_get_next_listener(struct spdk_nvmf_subsystem *subsystem,
1660 				      struct spdk_nvmf_subsystem_listener *prev_listener)
1661 {
1662 	return TAILQ_NEXT(prev_listener, link);
1663 }
1664 
1665 const struct spdk_nvme_transport_id *
1666 spdk_nvmf_subsystem_listener_get_trid(struct spdk_nvmf_subsystem_listener *listener)
1667 {
1668 	return listener->trid;
1669 }
1670 
1671 void
1672 spdk_nvmf_subsystem_allow_any_listener(struct spdk_nvmf_subsystem *subsystem,
1673 				       bool allow_any_listener)
1674 {
1675 	subsystem->flags.allow_any_listener = allow_any_listener;
1676 }
1677 
1678 bool
1679 spdk_nvmf_subsystem_any_listener_allowed(struct spdk_nvmf_subsystem *subsystem)
1680 {
1681 	return subsystem->flags.allow_any_listener;
1682 }
1683 
1684 struct subsystem_update_ns_ctx {
1685 	struct spdk_nvmf_subsystem *subsystem;
1686 
1687 	spdk_nvmf_subsystem_state_change_done cb_fn;
1688 	void *cb_arg;
1689 };
1690 
1691 static void
1692 subsystem_update_ns_done(struct spdk_io_channel_iter *i, int status)
1693 {
1694 	struct subsystem_update_ns_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
1695 
1696 	if (ctx->cb_fn) {
1697 		ctx->cb_fn(ctx->subsystem, ctx->cb_arg, status);
1698 	}
1699 	free(ctx);
1700 }
1701 
1702 static void
1703 subsystem_update_ns_on_pg(struct spdk_io_channel_iter *i)
1704 {
1705 	int rc;
1706 	struct subsystem_update_ns_ctx *ctx;
1707 	struct spdk_nvmf_poll_group *group;
1708 	struct spdk_nvmf_subsystem *subsystem;
1709 
1710 	ctx = spdk_io_channel_iter_get_ctx(i);
1711 	group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i));
1712 	subsystem = ctx->subsystem;
1713 
1714 	rc = nvmf_poll_group_update_subsystem(group, subsystem);
1715 	spdk_for_each_channel_continue(i, rc);
1716 }
1717 
1718 static int
1719 nvmf_subsystem_update_ns(struct spdk_nvmf_subsystem *subsystem,
1720 			 spdk_nvmf_subsystem_state_change_done cb_fn, void *cb_arg)
1721 {
1722 	struct subsystem_update_ns_ctx *ctx;
1723 
1724 	ctx = calloc(1, sizeof(*ctx));
1725 	if (ctx == NULL) {
1726 		SPDK_ERRLOG("Can't alloc subsystem poll group update context\n");
1727 		return -ENOMEM;
1728 	}
1729 	ctx->subsystem = subsystem;
1730 	ctx->cb_fn = cb_fn;
1731 	ctx->cb_arg = cb_arg;
1732 
1733 	spdk_for_each_channel(subsystem->tgt,
1734 			      subsystem_update_ns_on_pg,
1735 			      ctx,
1736 			      subsystem_update_ns_done);
1737 	return 0;
1738 }
1739 
1740 static void
1741 nvmf_subsystem_ns_changed(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
1742 {
1743 	struct spdk_nvmf_ctrlr *ctrlr;
1744 
1745 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1746 		if (nvmf_ctrlr_ns_is_visible(ctrlr, nsid)) {
1747 			nvmf_ctrlr_ns_changed(ctrlr, nsid);
1748 		}
1749 	}
1750 }
1751 
1752 static uint32_t nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns);
1753 
1754 int
1755 spdk_nvmf_subsystem_remove_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
1756 {
1757 	struct spdk_nvmf_transport *transport;
1758 	struct spdk_nvmf_ns *ns;
1759 	struct spdk_nvmf_host *host, *tmp;
1760 	struct spdk_nvmf_ctrlr *ctrlr;
1761 
1762 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
1763 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
1764 		assert(false);
1765 		return -1;
1766 	}
1767 
1768 	if (nsid == 0 || nsid > subsystem->max_nsid) {
1769 		return -1;
1770 	}
1771 
1772 	ns = subsystem->ns[nsid - 1];
1773 	if (!ns) {
1774 		return -1;
1775 	}
1776 
1777 	subsystem->ns[nsid - 1] = NULL;
1778 
1779 	assert(ns->anagrpid - 1 < subsystem->max_nsid);
1780 	assert(subsystem->ana_group[ns->anagrpid - 1] > 0);
1781 
1782 	subsystem->ana_group[ns->anagrpid - 1]--;
1783 
1784 	TAILQ_FOREACH_SAFE(host, &ns->hosts, link, tmp) {
1785 		nvmf_ns_remove_host(ns, host);
1786 	}
1787 
1788 	free(ns->ptpl_file);
1789 	nvmf_ns_reservation_clear_all_registrants(ns);
1790 	spdk_bdev_module_release_bdev(ns->bdev);
1791 	spdk_bdev_close(ns->desc);
1792 	free(ns);
1793 
1794 	if (subsystem->fdp_supported && !spdk_nvmf_subsystem_get_first_ns(subsystem)) {
1795 		subsystem->fdp_supported = false;
1796 		SPDK_DEBUGLOG(nvmf, "Subsystem with id: %u doesn't have FDP capability.\n",
1797 			      subsystem->id);
1798 	}
1799 
1800 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
1801 	     transport = spdk_nvmf_transport_get_next(transport)) {
1802 		if (transport->ops->subsystem_remove_ns) {
1803 			transport->ops->subsystem_remove_ns(transport, subsystem, nsid);
1804 		}
1805 	}
1806 
1807 	nvmf_subsystem_ns_changed(subsystem, nsid);
1808 
1809 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1810 		spdk_bit_array_clear(ctrlr->visible_ns, nsid - 1);
1811 	}
1812 
1813 	return 0;
1814 }
1815 
1816 struct subsystem_ns_change_ctx {
1817 	struct spdk_nvmf_subsystem		*subsystem;
1818 	spdk_nvmf_subsystem_state_change_done	cb_fn;
1819 	uint32_t				nsid;
1820 };
1821 
1822 static void
1823 _nvmf_ns_hot_remove(struct spdk_nvmf_subsystem *subsystem,
1824 		    void *cb_arg, int status)
1825 {
1826 	struct subsystem_ns_change_ctx *ctx = cb_arg;
1827 	int rc;
1828 
1829 	rc = spdk_nvmf_subsystem_remove_ns(subsystem, ctx->nsid);
1830 	if (rc != 0) {
1831 		SPDK_ERRLOG("Failed to make changes to NVME-oF subsystem with id: %u\n", subsystem->id);
1832 	}
1833 
1834 	rc = spdk_nvmf_subsystem_resume(subsystem, NULL, NULL);
1835 	if (rc != 0) {
1836 		SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id);
1837 	}
1838 
1839 	free(ctx);
1840 }
1841 
1842 static void
1843 nvmf_ns_change_msg(void *ns_ctx)
1844 {
1845 	struct subsystem_ns_change_ctx *ctx = ns_ctx;
1846 	int rc;
1847 
1848 	SPDK_DTRACE_PROBE2(nvmf_ns_change, ctx->nsid, ctx->subsystem->subnqn);
1849 
1850 	rc = spdk_nvmf_subsystem_pause(ctx->subsystem, ctx->nsid, ctx->cb_fn, ctx);
1851 	if (rc) {
1852 		if (rc == -EBUSY) {
1853 			/* Try again, this is not a permanent situation. */
1854 			spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ctx);
1855 		} else {
1856 			free(ctx);
1857 			SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n");
1858 		}
1859 	}
1860 }
1861 
1862 static void
1863 nvmf_ns_hot_remove(void *remove_ctx)
1864 {
1865 	struct spdk_nvmf_ns *ns = remove_ctx;
1866 	struct subsystem_ns_change_ctx *ns_ctx;
1867 	int rc;
1868 
1869 	/* We have to allocate a new context because this op
1870 	 * is asynchronous and we could lose the ns in the middle.
1871 	 */
1872 	ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx));
1873 	if (!ns_ctx) {
1874 		SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n");
1875 		return;
1876 	}
1877 
1878 	ns_ctx->subsystem = ns->subsystem;
1879 	ns_ctx->nsid = ns->opts.nsid;
1880 	ns_ctx->cb_fn = _nvmf_ns_hot_remove;
1881 
1882 	rc = spdk_nvmf_subsystem_pause(ns->subsystem, ns_ctx->nsid, _nvmf_ns_hot_remove, ns_ctx);
1883 	if (rc) {
1884 		if (rc == -EBUSY) {
1885 			/* Try again, this is not a permanent situation. */
1886 			spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx);
1887 		} else {
1888 			SPDK_ERRLOG("Unable to pause subsystem to process namespace removal!\n");
1889 			free(ns_ctx);
1890 		}
1891 	}
1892 }
1893 
1894 static void
1895 _nvmf_ns_resize(struct spdk_nvmf_subsystem *subsystem, void *cb_arg, int status)
1896 {
1897 	struct subsystem_ns_change_ctx *ctx = cb_arg;
1898 
1899 	nvmf_subsystem_ns_changed(subsystem, ctx->nsid);
1900 	if (spdk_nvmf_subsystem_resume(subsystem, NULL, NULL) != 0) {
1901 		SPDK_ERRLOG("Failed to resume NVME-oF subsystem with id: %u\n", subsystem->id);
1902 	}
1903 
1904 	free(ctx);
1905 }
1906 
1907 static void
1908 nvmf_ns_resize(void *event_ctx)
1909 {
1910 	struct spdk_nvmf_ns *ns = event_ctx;
1911 	struct subsystem_ns_change_ctx *ns_ctx;
1912 	int rc;
1913 
1914 	/* We have to allocate a new context because this op
1915 	 * is asynchronous and we could lose the ns in the middle.
1916 	 */
1917 	ns_ctx = calloc(1, sizeof(struct subsystem_ns_change_ctx));
1918 	if (!ns_ctx) {
1919 		SPDK_ERRLOG("Unable to allocate context to process namespace removal!\n");
1920 		return;
1921 	}
1922 
1923 	ns_ctx->subsystem = ns->subsystem;
1924 	ns_ctx->nsid = ns->opts.nsid;
1925 	ns_ctx->cb_fn = _nvmf_ns_resize;
1926 
1927 	/* Specify 0 for the nsid here, because we do not need to pause the namespace.
1928 	 * Namespaces can only be resized bigger, so there is no need to quiesce I/O.
1929 	 */
1930 	rc = spdk_nvmf_subsystem_pause(ns->subsystem, 0, _nvmf_ns_resize, ns_ctx);
1931 	if (rc) {
1932 		if (rc == -EBUSY) {
1933 			/* Try again, this is not a permanent situation. */
1934 			spdk_thread_send_msg(spdk_get_thread(), nvmf_ns_change_msg, ns_ctx);
1935 		} else {
1936 			SPDK_ERRLOG("Unable to pause subsystem to process namespace resize!\n");
1937 			free(ns_ctx);
1938 		}
1939 	}
1940 }
1941 
1942 static void
1943 nvmf_ns_event(enum spdk_bdev_event_type type,
1944 	      struct spdk_bdev *bdev,
1945 	      void *event_ctx)
1946 {
1947 	SPDK_DEBUGLOG(nvmf, "Bdev event: type %d, name %s, subsystem_id %d, ns_id %d\n",
1948 		      type,
1949 		      spdk_bdev_get_name(bdev),
1950 		      ((struct spdk_nvmf_ns *)event_ctx)->subsystem->id,
1951 		      ((struct spdk_nvmf_ns *)event_ctx)->nsid);
1952 
1953 	switch (type) {
1954 	case SPDK_BDEV_EVENT_REMOVE:
1955 		nvmf_ns_hot_remove(event_ctx);
1956 		break;
1957 	case SPDK_BDEV_EVENT_RESIZE:
1958 		nvmf_ns_resize(event_ctx);
1959 		break;
1960 	default:
1961 		SPDK_NOTICELOG("Unsupported bdev event: type %d\n", type);
1962 		break;
1963 	}
1964 }
1965 
1966 void
1967 spdk_nvmf_ns_opts_get_defaults(struct spdk_nvmf_ns_opts *opts, size_t opts_size)
1968 {
1969 	if (!opts) {
1970 		SPDK_ERRLOG("opts should not be NULL.\n");
1971 		return;
1972 	}
1973 
1974 	if (!opts_size) {
1975 		SPDK_ERRLOG("opts_size should not be zero.\n");
1976 		return;
1977 	}
1978 
1979 	memset(opts, 0, opts_size);
1980 	opts->opts_size = opts_size;
1981 
1982 #define FIELD_OK(field) \
1983 	offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= opts_size
1984 
1985 #define SET_FIELD(field, value) \
1986 	if (FIELD_OK(field)) { \
1987 		opts->field = value; \
1988 	} \
1989 
1990 	/* All current fields are set to 0 by default. */
1991 	SET_FIELD(nsid, 0);
1992 	if (FIELD_OK(nguid)) {
1993 		memset(opts->nguid, 0, sizeof(opts->nguid));
1994 	}
1995 	if (FIELD_OK(eui64)) {
1996 		memset(opts->eui64, 0, sizeof(opts->eui64));
1997 	}
1998 	if (FIELD_OK(uuid)) {
1999 		spdk_uuid_set_null(&opts->uuid);
2000 	}
2001 	SET_FIELD(anagrpid, 0);
2002 	SET_FIELD(transport_specific, NULL);
2003 
2004 #undef FIELD_OK
2005 #undef SET_FIELD
2006 }
2007 
2008 static void
2009 nvmf_ns_opts_copy(struct spdk_nvmf_ns_opts *opts,
2010 		  const struct spdk_nvmf_ns_opts *user_opts,
2011 		  size_t opts_size)
2012 {
2013 #define FIELD_OK(field)	\
2014 	offsetof(struct spdk_nvmf_ns_opts, field) + sizeof(opts->field) <= user_opts->opts_size
2015 
2016 #define SET_FIELD(field) \
2017 	if (FIELD_OK(field)) { \
2018 		opts->field = user_opts->field;	\
2019 	} \
2020 
2021 	SET_FIELD(nsid);
2022 	if (FIELD_OK(nguid)) {
2023 		memcpy(opts->nguid, user_opts->nguid, sizeof(opts->nguid));
2024 	}
2025 	if (FIELD_OK(eui64)) {
2026 		memcpy(opts->eui64, user_opts->eui64, sizeof(opts->eui64));
2027 	}
2028 	if (FIELD_OK(uuid)) {
2029 		spdk_uuid_copy(&opts->uuid, &user_opts->uuid);
2030 	}
2031 	SET_FIELD(anagrpid);
2032 	SET_FIELD(no_auto_visible);
2033 	SET_FIELD(transport_specific);
2034 
2035 	opts->opts_size = user_opts->opts_size;
2036 
2037 	/* We should not remove this statement, but need to update the assert statement
2038 	 * if we add a new field, and also add a corresponding SET_FIELD statement.
2039 	 */
2040 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_ns_opts) == 72, "Incorrect size");
2041 
2042 #undef FIELD_OK
2043 #undef SET_FIELD
2044 }
2045 
2046 /* Dummy bdev module used to to claim bdevs. */
2047 static struct spdk_bdev_module ns_bdev_module = {
2048 	.name	= "NVMe-oF Target",
2049 };
2050 
2051 static int nvmf_ns_reservation_update(const struct spdk_nvmf_ns *ns,
2052 				      const struct spdk_nvmf_reservation_info *info);
2053 static int nvmf_ns_reservation_load(const struct spdk_nvmf_ns *ns,
2054 				    struct spdk_nvmf_reservation_info *info);
2055 static int nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns,
2056 				       struct spdk_nvmf_reservation_info *info);
2057 
2058 bool
2059 nvmf_subsystem_zone_append_supported(struct spdk_nvmf_subsystem *subsystem)
2060 {
2061 	struct spdk_nvmf_ns *ns;
2062 
2063 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
2064 	     ns != NULL;
2065 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
2066 		if (spdk_bdev_is_zoned(ns->bdev) &&
2067 		    spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZONE_APPEND)) {
2068 			return true;
2069 		}
2070 	}
2071 
2072 	return false;
2073 }
2074 
2075 uint32_t
2076 spdk_nvmf_subsystem_add_ns_ext(struct spdk_nvmf_subsystem *subsystem, const char *bdev_name,
2077 			       const struct spdk_nvmf_ns_opts *user_opts, size_t opts_size,
2078 			       const char *ptpl_file)
2079 {
2080 	struct spdk_nvmf_transport *transport;
2081 	struct spdk_nvmf_ns_opts opts;
2082 	struct spdk_nvmf_ns *ns, *first_ns;
2083 	struct spdk_nvmf_ctrlr *ctrlr;
2084 	struct spdk_nvmf_reservation_info info = {0};
2085 	struct spdk_nvme_ns *nvme_ns;
2086 	int rc;
2087 	bool zone_append_supported;
2088 	uint64_t max_zone_append_size_kib;
2089 
2090 	if (!(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
2091 	      subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED)) {
2092 		return 0;
2093 	}
2094 
2095 	spdk_nvmf_ns_opts_get_defaults(&opts, sizeof(opts));
2096 	if (user_opts) {
2097 		nvmf_ns_opts_copy(&opts, user_opts, opts_size);
2098 	}
2099 
2100 	if (opts.nsid == SPDK_NVME_GLOBAL_NS_TAG) {
2101 		SPDK_ERRLOG("Invalid NSID %" PRIu32 "\n", opts.nsid);
2102 		return 0;
2103 	}
2104 
2105 	if (opts.nsid == 0) {
2106 		/*
2107 		 * NSID not specified - find a free index.
2108 		 *
2109 		 * If no free slots are found, return error.
2110 		 */
2111 		for (opts.nsid = 1; opts.nsid <= subsystem->max_nsid; opts.nsid++) {
2112 			if (_nvmf_subsystem_get_ns(subsystem, opts.nsid) == NULL) {
2113 				break;
2114 			}
2115 		}
2116 		if (opts.nsid > subsystem->max_nsid) {
2117 			SPDK_ERRLOG("No free namespace slot available in the subsystem\n");
2118 			return 0;
2119 		}
2120 	}
2121 
2122 	if (opts.nsid > subsystem->max_nsid) {
2123 		SPDK_ERRLOG("NSID greater than maximum not allowed\n");
2124 		return 0;
2125 	}
2126 
2127 	if (_nvmf_subsystem_get_ns(subsystem, opts.nsid)) {
2128 		SPDK_ERRLOG("Requested NSID %" PRIu32 " already in use\n", opts.nsid);
2129 		return 0;
2130 	}
2131 
2132 	if (opts.anagrpid == 0) {
2133 		opts.anagrpid = opts.nsid;
2134 	}
2135 
2136 	if (opts.anagrpid > subsystem->max_nsid) {
2137 		SPDK_ERRLOG("ANAGRPID greater than maximum NSID not allowed\n");
2138 		return 0;
2139 	}
2140 
2141 	ns = calloc(1, sizeof(*ns));
2142 	if (ns == NULL) {
2143 		SPDK_ERRLOG("Namespace allocation failed\n");
2144 		return 0;
2145 	}
2146 
2147 	TAILQ_INIT(&ns->hosts);
2148 	ns->always_visible = !opts.no_auto_visible;
2149 	if (ns->always_visible) {
2150 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2151 			spdk_bit_array_set(ctrlr->visible_ns, opts.nsid - 1);
2152 		}
2153 	}
2154 
2155 	rc = spdk_bdev_open_ext(bdev_name, true, nvmf_ns_event, ns, &ns->desc);
2156 	if (rc != 0) {
2157 		SPDK_ERRLOG("Subsystem %s: bdev %s cannot be opened, error=%d\n",
2158 			    subsystem->subnqn, bdev_name, rc);
2159 		free(ns);
2160 		return 0;
2161 	}
2162 
2163 	ns->bdev = spdk_bdev_desc_get_bdev(ns->desc);
2164 
2165 	if (spdk_bdev_get_md_size(ns->bdev) != 0) {
2166 		if (!spdk_bdev_is_md_interleaved(ns->bdev)) {
2167 			SPDK_ERRLOG("Can't attach bdev with separate metadata.\n");
2168 			spdk_bdev_close(ns->desc);
2169 			free(ns);
2170 			return 0;
2171 		}
2172 
2173 		if (spdk_bdev_get_md_size(ns->bdev) > SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE) {
2174 			SPDK_ERRLOG("Maximum supported interleaved md size %u, current md size %u\n",
2175 				    SPDK_BDEV_MAX_INTERLEAVED_MD_SIZE, spdk_bdev_get_md_size(ns->bdev));
2176 			spdk_bdev_close(ns->desc);
2177 			free(ns);
2178 			return 0;
2179 		}
2180 	}
2181 
2182 	rc = spdk_bdev_module_claim_bdev(ns->bdev, ns->desc, &ns_bdev_module);
2183 	if (rc != 0) {
2184 		spdk_bdev_close(ns->desc);
2185 		free(ns);
2186 		return 0;
2187 	}
2188 
2189 	if (!strncmp(spdk_bdev_get_module_name(ns->bdev), "nvme",
2190 		     strlen(spdk_bdev_get_module_name(ns->bdev)))) {
2191 		nvme_ns = (struct spdk_nvme_ns *) spdk_bdev_get_module_ctx(ns->desc);
2192 		ns->passthrough_nsid = spdk_nvme_ns_get_id(nvme_ns);
2193 	} else if (subsystem->passthrough) {
2194 		SPDK_ERRLOG("Only bdev_nvme namespaces can be added to a passthrough subsystem.\n");
2195 		goto err;
2196 	}
2197 
2198 	/* Cache the zcopy capability of the bdev device */
2199 	ns->zcopy = spdk_bdev_io_type_supported(ns->bdev, SPDK_BDEV_IO_TYPE_ZCOPY);
2200 
2201 	if (spdk_uuid_is_null(&opts.uuid)) {
2202 		opts.uuid = *spdk_bdev_get_uuid(ns->bdev);
2203 	}
2204 
2205 	/* if nguid descriptor is supported by bdev module (nvme) then uuid = nguid */
2206 	if (spdk_mem_all_zero(opts.nguid, sizeof(opts.nguid))) {
2207 		SPDK_STATIC_ASSERT(sizeof(opts.nguid) == sizeof(opts.uuid), "size mismatch");
2208 		memcpy(opts.nguid, spdk_bdev_get_uuid(ns->bdev), sizeof(opts.nguid));
2209 	}
2210 
2211 	if (spdk_bdev_is_zoned(ns->bdev)) {
2212 		SPDK_DEBUGLOG(nvmf, "The added namespace is backed by a zoned block device.\n");
2213 		ns->csi = SPDK_NVME_CSI_ZNS;
2214 
2215 		zone_append_supported = spdk_bdev_io_type_supported(ns->bdev,
2216 					SPDK_BDEV_IO_TYPE_ZONE_APPEND);
2217 		max_zone_append_size_kib = spdk_bdev_get_max_zone_append_size(
2218 						   ns->bdev) * spdk_bdev_get_block_size(ns->bdev);
2219 
2220 		if (_nvmf_subsystem_get_first_zoned_ns(subsystem) != NULL &&
2221 		    (nvmf_subsystem_zone_append_supported(subsystem) != zone_append_supported ||
2222 		     subsystem->max_zone_append_size_kib != max_zone_append_size_kib)) {
2223 			SPDK_ERRLOG("Namespaces with different zone append support or different zone append size are not allowed.\n");
2224 			goto err;
2225 		}
2226 
2227 		subsystem->max_zone_append_size_kib = max_zone_append_size_kib;
2228 	}
2229 
2230 	first_ns = spdk_nvmf_subsystem_get_first_ns(subsystem);
2231 	if (!first_ns) {
2232 		if (spdk_bdev_get_nvme_ctratt(ns->bdev).bits.fdps) {
2233 			SPDK_DEBUGLOG(nvmf, "Subsystem with id: %u has FDP capability.\n",
2234 				      subsystem->id);
2235 			subsystem->fdp_supported = true;
2236 		}
2237 	} else {
2238 		if (spdk_bdev_get_nvme_ctratt(first_ns->bdev).bits.fdps !=
2239 		    spdk_bdev_get_nvme_ctratt(ns->bdev).bits.fdps) {
2240 			SPDK_ERRLOG("Subsystem with id: %u can%s FDP namespace.\n", subsystem->id,
2241 				    spdk_bdev_get_nvme_ctratt(first_ns->bdev).bits.fdps ? " only add" : "not add");
2242 			goto err;
2243 		}
2244 	}
2245 
2246 	ns->opts = opts;
2247 	ns->subsystem = subsystem;
2248 	subsystem->ns[opts.nsid - 1] = ns;
2249 	ns->nsid = opts.nsid;
2250 	ns->anagrpid = opts.anagrpid;
2251 	subsystem->ana_group[ns->anagrpid - 1]++;
2252 	TAILQ_INIT(&ns->registrants);
2253 	if (ptpl_file) {
2254 		ns->ptpl_file = strdup(ptpl_file);
2255 		if (!ns->ptpl_file) {
2256 			SPDK_ERRLOG("Namespace ns->ptpl_file allocation failed\n");
2257 			goto err;
2258 		}
2259 	}
2260 
2261 	if (nvmf_ns_is_ptpl_capable(ns)) {
2262 		rc = nvmf_ns_reservation_load(ns, &info);
2263 		if (rc) {
2264 			SPDK_ERRLOG("Subsystem load reservation failed\n");
2265 			goto err;
2266 		}
2267 
2268 		rc = nvmf_ns_reservation_restore(ns, &info);
2269 		if (rc) {
2270 			SPDK_ERRLOG("Subsystem restore reservation failed\n");
2271 			goto err;
2272 		}
2273 	}
2274 
2275 	for (transport = spdk_nvmf_transport_get_first(subsystem->tgt); transport;
2276 	     transport = spdk_nvmf_transport_get_next(transport)) {
2277 		if (transport->ops->subsystem_add_ns) {
2278 			rc = transport->ops->subsystem_add_ns(transport, subsystem, ns);
2279 			if (rc) {
2280 				SPDK_ERRLOG("Namespace attachment is not allowed by %s transport\n", transport->ops->name);
2281 				nvmf_ns_reservation_clear_all_registrants(ns);
2282 				goto err;
2283 			}
2284 		}
2285 	}
2286 
2287 	/* JSON value obj is freed before sending the response. Set NULL to prevent usage of dangling pointer. */
2288 	ns->opts.transport_specific = NULL;
2289 
2290 	SPDK_DEBUGLOG(nvmf, "Subsystem %s: bdev %s assigned nsid %" PRIu32 "\n",
2291 		      spdk_nvmf_subsystem_get_nqn(subsystem),
2292 		      bdev_name,
2293 		      opts.nsid);
2294 
2295 	nvmf_subsystem_ns_changed(subsystem, opts.nsid);
2296 
2297 	SPDK_DTRACE_PROBE2(nvmf_subsystem_add_ns, subsystem->subnqn, ns->nsid);
2298 
2299 	return opts.nsid;
2300 err:
2301 	subsystem->ns[opts.nsid - 1] = NULL;
2302 	spdk_bdev_module_release_bdev(ns->bdev);
2303 	spdk_bdev_close(ns->desc);
2304 	free(ns->ptpl_file);
2305 	free(ns);
2306 
2307 	return 0;
2308 }
2309 
2310 static uint32_t
2311 nvmf_subsystem_get_next_allocated_nsid(struct spdk_nvmf_subsystem *subsystem,
2312 				       uint32_t prev_nsid)
2313 {
2314 	uint32_t nsid;
2315 
2316 	if (prev_nsid >= subsystem->max_nsid) {
2317 		return 0;
2318 	}
2319 
2320 	for (nsid = prev_nsid + 1; nsid <= subsystem->max_nsid; nsid++) {
2321 		if (subsystem->ns[nsid - 1]) {
2322 			return nsid;
2323 		}
2324 	}
2325 
2326 	return 0;
2327 }
2328 
2329 struct spdk_nvmf_ns *
2330 spdk_nvmf_subsystem_get_first_ns(struct spdk_nvmf_subsystem *subsystem)
2331 {
2332 	uint32_t first_nsid;
2333 
2334 	first_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, 0);
2335 	return _nvmf_subsystem_get_ns(subsystem, first_nsid);
2336 }
2337 
2338 struct spdk_nvmf_ns *
2339 spdk_nvmf_subsystem_get_next_ns(struct spdk_nvmf_subsystem *subsystem,
2340 				struct spdk_nvmf_ns *prev_ns)
2341 {
2342 	uint32_t next_nsid;
2343 
2344 	next_nsid = nvmf_subsystem_get_next_allocated_nsid(subsystem, prev_ns->opts.nsid);
2345 	return _nvmf_subsystem_get_ns(subsystem, next_nsid);
2346 }
2347 
2348 struct spdk_nvmf_ns *
2349 spdk_nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid)
2350 {
2351 	return _nvmf_subsystem_get_ns(subsystem, nsid);
2352 }
2353 
2354 uint32_t
2355 spdk_nvmf_ns_get_id(const struct spdk_nvmf_ns *ns)
2356 {
2357 	return ns->opts.nsid;
2358 }
2359 
2360 struct spdk_bdev *
2361 spdk_nvmf_ns_get_bdev(struct spdk_nvmf_ns *ns)
2362 {
2363 	return ns->bdev;
2364 }
2365 
2366 void
2367 spdk_nvmf_ns_get_opts(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_ns_opts *opts,
2368 		      size_t opts_size)
2369 {
2370 	memset(opts, 0, opts_size);
2371 	memcpy(opts, &ns->opts, spdk_min(sizeof(ns->opts), opts_size));
2372 }
2373 
2374 const char *
2375 spdk_nvmf_subsystem_get_sn(const struct spdk_nvmf_subsystem *subsystem)
2376 {
2377 	return subsystem->sn;
2378 }
2379 
2380 int
2381 spdk_nvmf_subsystem_set_sn(struct spdk_nvmf_subsystem *subsystem, const char *sn)
2382 {
2383 	size_t len, max_len;
2384 
2385 	max_len = sizeof(subsystem->sn) - 1;
2386 	len = strlen(sn);
2387 	if (len > max_len) {
2388 		SPDK_DEBUGLOG(nvmf, "Invalid sn \"%s\": length %zu > max %zu\n",
2389 			      sn, len, max_len);
2390 		return -1;
2391 	}
2392 
2393 	if (!nvmf_valid_ascii_string(sn, len)) {
2394 		SPDK_DEBUGLOG(nvmf, "Non-ASCII sn\n");
2395 		SPDK_LOGDUMP(nvmf, "sn", sn, len);
2396 		return -1;
2397 	}
2398 
2399 	snprintf(subsystem->sn, sizeof(subsystem->sn), "%s", sn);
2400 
2401 	return 0;
2402 }
2403 
2404 const char *
2405 spdk_nvmf_subsystem_get_mn(const struct spdk_nvmf_subsystem *subsystem)
2406 {
2407 	return subsystem->mn;
2408 }
2409 
2410 int
2411 spdk_nvmf_subsystem_set_mn(struct spdk_nvmf_subsystem *subsystem, const char *mn)
2412 {
2413 	size_t len, max_len;
2414 
2415 	if (mn == NULL) {
2416 		mn = MODEL_NUMBER_DEFAULT;
2417 	}
2418 	max_len = sizeof(subsystem->mn) - 1;
2419 	len = strlen(mn);
2420 	if (len > max_len) {
2421 		SPDK_DEBUGLOG(nvmf, "Invalid mn \"%s\": length %zu > max %zu\n",
2422 			      mn, len, max_len);
2423 		return -1;
2424 	}
2425 
2426 	if (!nvmf_valid_ascii_string(mn, len)) {
2427 		SPDK_DEBUGLOG(nvmf, "Non-ASCII mn\n");
2428 		SPDK_LOGDUMP(nvmf, "mn", mn, len);
2429 		return -1;
2430 	}
2431 
2432 	snprintf(subsystem->mn, sizeof(subsystem->mn), "%s", mn);
2433 
2434 	return 0;
2435 }
2436 
2437 const char *
2438 spdk_nvmf_subsystem_get_nqn(const struct spdk_nvmf_subsystem *subsystem)
2439 {
2440 	return subsystem->subnqn;
2441 }
2442 
2443 /* We have to use the typedef in the function declaration to appease astyle. */
2444 typedef enum spdk_nvmf_subtype spdk_nvmf_subtype_t;
2445 
2446 spdk_nvmf_subtype_t
2447 spdk_nvmf_subsystem_get_type(struct spdk_nvmf_subsystem *subsystem)
2448 {
2449 	return subsystem->subtype;
2450 }
2451 
2452 uint32_t
2453 spdk_nvmf_subsystem_get_max_nsid(struct spdk_nvmf_subsystem *subsystem)
2454 {
2455 	return subsystem->max_nsid;
2456 }
2457 
2458 int
2459 spdk_nvmf_subsystem_set_cntlid_range(struct spdk_nvmf_subsystem *subsystem,
2460 				     uint16_t min_cntlid, uint16_t max_cntlid)
2461 {
2462 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
2463 		return -EAGAIN;
2464 	}
2465 
2466 	if (min_cntlid > max_cntlid) {
2467 		return -EINVAL;
2468 	}
2469 	/* The spec reserves cntlid values in the range FFF0h to FFFFh. */
2470 	if (min_cntlid < NVMF_MIN_CNTLID || min_cntlid > NVMF_MAX_CNTLID ||
2471 	    max_cntlid < NVMF_MIN_CNTLID || max_cntlid > NVMF_MAX_CNTLID) {
2472 		return -EINVAL;
2473 	}
2474 	subsystem->min_cntlid = min_cntlid;
2475 	subsystem->max_cntlid = max_cntlid;
2476 	if (subsystem->next_cntlid < min_cntlid || subsystem->next_cntlid > max_cntlid - 1) {
2477 		subsystem->next_cntlid = min_cntlid - 1;
2478 	}
2479 
2480 	return 0;
2481 }
2482 
2483 uint16_t
2484 nvmf_subsystem_gen_cntlid(struct spdk_nvmf_subsystem *subsystem)
2485 {
2486 	int count;
2487 
2488 	/*
2489 	 * In the worst case, we might have to try all CNTLID values between min_cntlid and max_cntlid
2490 	 * before we find one that is unused (or find that all values are in use).
2491 	 */
2492 	for (count = 0; count < subsystem->max_cntlid - subsystem->min_cntlid + 1; count++) {
2493 		subsystem->next_cntlid++;
2494 		if (subsystem->next_cntlid > subsystem->max_cntlid) {
2495 			subsystem->next_cntlid = subsystem->min_cntlid;
2496 		}
2497 
2498 		/* Check if a controller with this cntlid currently exists. */
2499 		if (nvmf_subsystem_get_ctrlr(subsystem, subsystem->next_cntlid) == NULL) {
2500 			/* Found unused cntlid */
2501 			return subsystem->next_cntlid;
2502 		}
2503 	}
2504 
2505 	/* All valid cntlid values are in use. */
2506 	return 0xFFFF;
2507 }
2508 
2509 int
2510 nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, struct spdk_nvmf_ctrlr *ctrlr)
2511 {
2512 
2513 	if (ctrlr->dynamic_ctrlr) {
2514 		ctrlr->cntlid = nvmf_subsystem_gen_cntlid(subsystem);
2515 		if (ctrlr->cntlid == 0xFFFF) {
2516 			/* Unable to get a cntlid */
2517 			SPDK_ERRLOG("Reached max simultaneous ctrlrs\n");
2518 			return -EBUSY;
2519 		}
2520 	} else if (nvmf_subsystem_get_ctrlr(subsystem, ctrlr->cntlid) != NULL) {
2521 		SPDK_ERRLOG("Ctrlr with cntlid %u already exist\n", ctrlr->cntlid);
2522 		return -EEXIST;
2523 	}
2524 
2525 	TAILQ_INSERT_TAIL(&subsystem->ctrlrs, ctrlr, link);
2526 
2527 	SPDK_DTRACE_PROBE3(nvmf_subsystem_add_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn);
2528 
2529 	return 0;
2530 }
2531 
2532 void
2533 nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem,
2534 			    struct spdk_nvmf_ctrlr *ctrlr)
2535 {
2536 	SPDK_DTRACE_PROBE3(nvmf_subsystem_remove_ctrlr, subsystem->subnqn, ctrlr, ctrlr->hostnqn);
2537 
2538 	assert(spdk_get_thread() == subsystem->thread);
2539 	assert(subsystem == ctrlr->subsys);
2540 	SPDK_DEBUGLOG(nvmf, "remove ctrlr %p id 0x%x from subsys %p %s\n", ctrlr, ctrlr->cntlid, subsystem,
2541 		      subsystem->subnqn);
2542 	TAILQ_REMOVE(&subsystem->ctrlrs, ctrlr, link);
2543 }
2544 
2545 struct spdk_nvmf_ctrlr *
2546 nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, uint16_t cntlid)
2547 {
2548 	struct spdk_nvmf_ctrlr *ctrlr;
2549 
2550 	TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2551 		if (ctrlr->cntlid == cntlid) {
2552 			return ctrlr;
2553 		}
2554 	}
2555 
2556 	return NULL;
2557 }
2558 
2559 uint32_t
2560 spdk_nvmf_subsystem_get_max_namespaces(const struct spdk_nvmf_subsystem *subsystem)
2561 {
2562 	return subsystem->max_nsid;
2563 }
2564 
2565 uint16_t
2566 spdk_nvmf_subsystem_get_min_cntlid(const struct spdk_nvmf_subsystem *subsystem)
2567 {
2568 	return subsystem->min_cntlid;
2569 }
2570 
2571 uint16_t
2572 spdk_nvmf_subsystem_get_max_cntlid(const struct spdk_nvmf_subsystem *subsystem)
2573 {
2574 	return subsystem->max_cntlid;
2575 }
2576 
2577 struct _nvmf_ns_registrant {
2578 	uint64_t		rkey;
2579 	char			*host_uuid;
2580 };
2581 
2582 struct _nvmf_ns_registrants {
2583 	size_t				num_regs;
2584 	struct _nvmf_ns_registrant	reg[SPDK_NVMF_MAX_NUM_REGISTRANTS];
2585 };
2586 
2587 struct _nvmf_ns_reservation {
2588 	bool					ptpl_activated;
2589 	enum spdk_nvme_reservation_type		rtype;
2590 	uint64_t				crkey;
2591 	char					*bdev_uuid;
2592 	char					*holder_uuid;
2593 	struct _nvmf_ns_registrants		regs;
2594 };
2595 
2596 static const struct spdk_json_object_decoder nvmf_ns_pr_reg_decoders[] = {
2597 	{"rkey", offsetof(struct _nvmf_ns_registrant, rkey), spdk_json_decode_uint64},
2598 	{"host_uuid", offsetof(struct _nvmf_ns_registrant, host_uuid), spdk_json_decode_string},
2599 };
2600 
2601 static int
2602 nvmf_decode_ns_pr_reg(const struct spdk_json_val *val, void *out)
2603 {
2604 	struct _nvmf_ns_registrant *reg = out;
2605 
2606 	return spdk_json_decode_object(val, nvmf_ns_pr_reg_decoders,
2607 				       SPDK_COUNTOF(nvmf_ns_pr_reg_decoders), reg);
2608 }
2609 
2610 static int
2611 nvmf_decode_ns_pr_regs(const struct spdk_json_val *val, void *out)
2612 {
2613 	struct _nvmf_ns_registrants *regs = out;
2614 
2615 	return spdk_json_decode_array(val, nvmf_decode_ns_pr_reg, regs->reg,
2616 				      SPDK_NVMF_MAX_NUM_REGISTRANTS, &regs->num_regs,
2617 				      sizeof(struct _nvmf_ns_registrant));
2618 }
2619 
2620 static const struct spdk_json_object_decoder nvmf_ns_pr_decoders[] = {
2621 	{"ptpl", offsetof(struct _nvmf_ns_reservation, ptpl_activated), spdk_json_decode_bool, true},
2622 	{"rtype", offsetof(struct _nvmf_ns_reservation, rtype), spdk_json_decode_uint32, true},
2623 	{"crkey", offsetof(struct _nvmf_ns_reservation, crkey), spdk_json_decode_uint64, true},
2624 	{"bdev_uuid", offsetof(struct _nvmf_ns_reservation, bdev_uuid), spdk_json_decode_string},
2625 	{"holder_uuid", offsetof(struct _nvmf_ns_reservation, holder_uuid), spdk_json_decode_string, true},
2626 	{"registrants", offsetof(struct _nvmf_ns_reservation, regs), nvmf_decode_ns_pr_regs},
2627 };
2628 
2629 static int
2630 nvmf_ns_reservation_load_json(const struct spdk_nvmf_ns *ns,
2631 			      struct spdk_nvmf_reservation_info *info)
2632 {
2633 	size_t json_size;
2634 	ssize_t values_cnt, rc;
2635 	void *json = NULL, *end;
2636 	struct spdk_json_val *values = NULL;
2637 	struct _nvmf_ns_reservation res = {};
2638 	const char *file = ns->ptpl_file;
2639 	uint32_t i;
2640 
2641 	/* It's not an error if the file does not exist */
2642 	if (access(file, F_OK) != 0) {
2643 		SPDK_DEBUGLOG(nvmf, "File %s does not exist\n", file);
2644 		return 0;
2645 	}
2646 
2647 	/* Load all persist file contents into a local buffer */
2648 	json = spdk_posix_file_load_from_name(file, &json_size);
2649 	if (!json) {
2650 		SPDK_ERRLOG("Load persist file %s failed\n", file);
2651 		return -ENOMEM;
2652 	}
2653 
2654 	rc = spdk_json_parse(json, json_size, NULL, 0, &end, 0);
2655 	if (rc < 0) {
2656 		SPDK_NOTICELOG("Parsing JSON configuration failed (%zd)\n", rc);
2657 		goto exit;
2658 	}
2659 
2660 	values_cnt = rc;
2661 	values = calloc(values_cnt, sizeof(struct spdk_json_val));
2662 	if (values == NULL) {
2663 		goto exit;
2664 	}
2665 
2666 	rc = spdk_json_parse(json, json_size, values, values_cnt, &end, 0);
2667 	if (rc != values_cnt) {
2668 		SPDK_ERRLOG("Parsing JSON configuration failed (%zd)\n", rc);
2669 		goto exit;
2670 	}
2671 
2672 	/* Decode json */
2673 	if (spdk_json_decode_object(values, nvmf_ns_pr_decoders,
2674 				    SPDK_COUNTOF(nvmf_ns_pr_decoders),
2675 				    &res)) {
2676 		SPDK_ERRLOG("Invalid objects in the persist file %s\n", file);
2677 		rc = -EINVAL;
2678 		goto exit;
2679 	}
2680 
2681 	if (res.regs.num_regs > SPDK_NVMF_MAX_NUM_REGISTRANTS) {
2682 		SPDK_ERRLOG("Can only support up to %u registrants\n", SPDK_NVMF_MAX_NUM_REGISTRANTS);
2683 		rc = -ERANGE;
2684 		goto exit;
2685 	}
2686 
2687 	rc = 0;
2688 	info->ptpl_activated = res.ptpl_activated;
2689 	info->rtype = res.rtype;
2690 	info->crkey = res.crkey;
2691 	snprintf(info->bdev_uuid, sizeof(info->bdev_uuid), "%s", res.bdev_uuid);
2692 	snprintf(info->holder_uuid, sizeof(info->holder_uuid), "%s", res.holder_uuid);
2693 	info->num_regs = res.regs.num_regs;
2694 	for (i = 0; i < res.regs.num_regs; i++) {
2695 		info->registrants[i].rkey = res.regs.reg[i].rkey;
2696 		snprintf(info->registrants[i].host_uuid, sizeof(info->registrants[i].host_uuid), "%s",
2697 			 res.regs.reg[i].host_uuid);
2698 	}
2699 
2700 exit:
2701 	free(json);
2702 	free(values);
2703 	free(res.bdev_uuid);
2704 	free(res.holder_uuid);
2705 	for (i = 0; i < res.regs.num_regs; i++) {
2706 		free(res.regs.reg[i].host_uuid);
2707 	}
2708 
2709 	return rc;
2710 }
2711 
2712 static bool nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns);
2713 
2714 static int
2715 nvmf_ns_reservation_restore(struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info)
2716 {
2717 	uint32_t i;
2718 	struct spdk_nvmf_registrant *reg, *holder = NULL;
2719 	struct spdk_uuid bdev_uuid, holder_uuid;
2720 	bool rkey_flag = false;
2721 
2722 	SPDK_DEBUGLOG(nvmf, "NSID %u, PTPL %u, Number of registrants %u\n",
2723 		      ns->nsid, info->ptpl_activated, info->num_regs);
2724 
2725 	/* it's not an error */
2726 	if (!info->ptpl_activated || !info->num_regs) {
2727 		return 0;
2728 	}
2729 
2730 	/* Check info->crkey exist or not in info->registrants[i].rkey */
2731 	for (i = 0; i < info->num_regs; i++) {
2732 		if (info->crkey == info->registrants[i].rkey) {
2733 			rkey_flag = true;
2734 		}
2735 	}
2736 	if (!rkey_flag && info->crkey != 0) {
2737 		return -EINVAL;
2738 	}
2739 
2740 	spdk_uuid_parse(&bdev_uuid, info->bdev_uuid);
2741 	if (spdk_uuid_compare(&bdev_uuid, spdk_bdev_get_uuid(ns->bdev))) {
2742 		SPDK_ERRLOG("Existing bdev UUID is not same with configuration file\n");
2743 		return -EINVAL;
2744 	}
2745 
2746 	ns->crkey = info->crkey;
2747 	ns->rtype = info->rtype;
2748 	ns->ptpl_activated = info->ptpl_activated;
2749 	spdk_uuid_parse(&holder_uuid, info->holder_uuid);
2750 
2751 	SPDK_DEBUGLOG(nvmf, "Bdev UUID %s\n", info->bdev_uuid);
2752 	if (info->rtype) {
2753 		SPDK_DEBUGLOG(nvmf, "Holder UUID %s, RTYPE %u, RKEY 0x%"PRIx64"\n",
2754 			      info->holder_uuid, info->rtype, info->crkey);
2755 	}
2756 
2757 	for (i = 0; i < info->num_regs; i++) {
2758 		reg = calloc(1, sizeof(*reg));
2759 		if (!reg) {
2760 			return -ENOMEM;
2761 		}
2762 		spdk_uuid_parse(&reg->hostid, info->registrants[i].host_uuid);
2763 		reg->rkey = info->registrants[i].rkey;
2764 		TAILQ_INSERT_TAIL(&ns->registrants, reg, link);
2765 		if (info->crkey != 0 && !spdk_uuid_compare(&holder_uuid, &reg->hostid)) {
2766 			holder = reg;
2767 		}
2768 		SPDK_DEBUGLOG(nvmf, "Registrant RKEY 0x%"PRIx64", Host UUID %s\n",
2769 			      info->registrants[i].rkey, info->registrants[i].host_uuid);
2770 	}
2771 
2772 	if (nvmf_ns_reservation_all_registrants_type(ns)) {
2773 		ns->holder = TAILQ_FIRST(&ns->registrants);
2774 	} else {
2775 		ns->holder = holder;
2776 	}
2777 
2778 	return 0;
2779 }
2780 
2781 static int
2782 nvmf_ns_json_write_cb(void *cb_ctx, const void *data, size_t size)
2783 {
2784 	char *file = cb_ctx;
2785 	size_t rc;
2786 	FILE *fd;
2787 
2788 	fd = fopen(file, "w");
2789 	if (!fd) {
2790 		SPDK_ERRLOG("Can't open file %s for write\n", file);
2791 		return -ENOENT;
2792 	}
2793 	rc = fwrite(data, 1, size, fd);
2794 	fclose(fd);
2795 
2796 	return rc == size ? 0 : -1;
2797 }
2798 
2799 static int
2800 nvmf_ns_reservation_update_json(const struct spdk_nvmf_ns *ns,
2801 				const struct spdk_nvmf_reservation_info *info)
2802 {
2803 	const char *file = ns->ptpl_file;
2804 	struct spdk_json_write_ctx *w;
2805 	uint32_t i;
2806 	int rc = 0;
2807 
2808 	w = spdk_json_write_begin(nvmf_ns_json_write_cb, (void *)file, 0);
2809 	if (w == NULL) {
2810 		return -ENOMEM;
2811 	}
2812 	/* clear the configuration file */
2813 	if (!info->ptpl_activated) {
2814 		goto exit;
2815 	}
2816 
2817 	spdk_json_write_object_begin(w);
2818 	spdk_json_write_named_bool(w, "ptpl", info->ptpl_activated);
2819 	spdk_json_write_named_uint32(w, "rtype", info->rtype);
2820 	spdk_json_write_named_uint64(w, "crkey", info->crkey);
2821 	spdk_json_write_named_string(w, "bdev_uuid", info->bdev_uuid);
2822 	spdk_json_write_named_string(w, "holder_uuid", info->holder_uuid);
2823 
2824 	spdk_json_write_named_array_begin(w, "registrants");
2825 	for (i = 0; i < info->num_regs; i++) {
2826 		spdk_json_write_object_begin(w);
2827 		spdk_json_write_named_uint64(w, "rkey", info->registrants[i].rkey);
2828 		spdk_json_write_named_string(w, "host_uuid", info->registrants[i].host_uuid);
2829 		spdk_json_write_object_end(w);
2830 	}
2831 	spdk_json_write_array_end(w);
2832 	spdk_json_write_object_end(w);
2833 
2834 exit:
2835 	rc = spdk_json_write_end(w);
2836 	return rc;
2837 }
2838 
2839 static int
2840 nvmf_ns_update_reservation_info(struct spdk_nvmf_ns *ns)
2841 {
2842 	struct spdk_nvmf_reservation_info info;
2843 	struct spdk_nvmf_registrant *reg, *tmp;
2844 	uint32_t i = 0;
2845 
2846 	assert(ns != NULL);
2847 
2848 	if (!ns->bdev || !nvmf_ns_is_ptpl_capable(ns)) {
2849 		return 0;
2850 	}
2851 
2852 	memset(&info, 0, sizeof(info));
2853 	spdk_uuid_fmt_lower(info.bdev_uuid, sizeof(info.bdev_uuid), spdk_bdev_get_uuid(ns->bdev));
2854 
2855 	if (ns->rtype) {
2856 		info.rtype = ns->rtype;
2857 		info.crkey = ns->crkey;
2858 		if (!nvmf_ns_reservation_all_registrants_type(ns)) {
2859 			assert(ns->holder != NULL);
2860 			spdk_uuid_fmt_lower(info.holder_uuid, sizeof(info.holder_uuid), &ns->holder->hostid);
2861 		}
2862 	}
2863 
2864 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2865 		spdk_uuid_fmt_lower(info.registrants[i].host_uuid, sizeof(info.registrants[i].host_uuid),
2866 				    &reg->hostid);
2867 		info.registrants[i++].rkey = reg->rkey;
2868 	}
2869 
2870 	info.num_regs = i;
2871 	info.ptpl_activated = ns->ptpl_activated;
2872 
2873 	return nvmf_ns_reservation_update(ns, &info);
2874 }
2875 
2876 static struct spdk_nvmf_registrant *
2877 nvmf_ns_reservation_get_registrant(struct spdk_nvmf_ns *ns,
2878 				   struct spdk_uuid *uuid)
2879 {
2880 	struct spdk_nvmf_registrant *reg, *tmp;
2881 
2882 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2883 		if (!spdk_uuid_compare(&reg->hostid, uuid)) {
2884 			return reg;
2885 		}
2886 	}
2887 
2888 	return NULL;
2889 }
2890 
2891 /* Generate reservation notice log to registered HostID controllers */
2892 static void
2893 nvmf_subsystem_gen_ctrlr_notification(struct spdk_nvmf_subsystem *subsystem,
2894 				      struct spdk_nvmf_ns *ns,
2895 				      struct spdk_uuid *hostid_list,
2896 				      uint32_t num_hostid,
2897 				      enum spdk_nvme_reservation_notification_log_page_type type)
2898 {
2899 	struct spdk_nvmf_ctrlr *ctrlr;
2900 	uint32_t i;
2901 
2902 	for (i = 0; i < num_hostid; i++) {
2903 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
2904 			if (!spdk_uuid_compare(&ctrlr->hostid, &hostid_list[i])) {
2905 				nvmf_ctrlr_reservation_notice_log(ctrlr, ns, type);
2906 			}
2907 		}
2908 	}
2909 }
2910 
2911 /* Get all registrants' hostid other than the controller who issued the command */
2912 static uint32_t
2913 nvmf_ns_reservation_get_all_other_hostid(struct spdk_nvmf_ns *ns,
2914 		struct spdk_uuid *hostid_list,
2915 		uint32_t max_num_hostid,
2916 		struct spdk_uuid *current_hostid)
2917 {
2918 	struct spdk_nvmf_registrant *reg, *tmp;
2919 	uint32_t num_hostid = 0;
2920 
2921 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
2922 		if (spdk_uuid_compare(&reg->hostid, current_hostid)) {
2923 			if (num_hostid == max_num_hostid) {
2924 				assert(false);
2925 				return max_num_hostid;
2926 			}
2927 			hostid_list[num_hostid++] = reg->hostid;
2928 		}
2929 	}
2930 
2931 	return num_hostid;
2932 }
2933 
2934 /* Calculate the unregistered HostID list according to list
2935  * prior to execute preempt command and list after executing
2936  * preempt command.
2937  */
2938 static uint32_t
2939 nvmf_ns_reservation_get_unregistered_hostid(struct spdk_uuid *old_hostid_list,
2940 		uint32_t old_num_hostid,
2941 		struct spdk_uuid *remaining_hostid_list,
2942 		uint32_t remaining_num_hostid)
2943 {
2944 	struct spdk_uuid temp_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
2945 	uint32_t i, j, num_hostid = 0;
2946 	bool found;
2947 
2948 	if (!remaining_num_hostid) {
2949 		return old_num_hostid;
2950 	}
2951 
2952 	for (i = 0; i < old_num_hostid; i++) {
2953 		found = false;
2954 		for (j = 0; j < remaining_num_hostid; j++) {
2955 			if (!spdk_uuid_compare(&old_hostid_list[i], &remaining_hostid_list[j])) {
2956 				found = true;
2957 				break;
2958 			}
2959 		}
2960 		if (!found) {
2961 			spdk_uuid_copy(&temp_hostid_list[num_hostid++], &old_hostid_list[i]);
2962 		}
2963 	}
2964 
2965 	if (num_hostid) {
2966 		memcpy(old_hostid_list, temp_hostid_list, sizeof(struct spdk_uuid) * num_hostid);
2967 	}
2968 
2969 	return num_hostid;
2970 }
2971 
2972 /* current reservation type is all registrants or not */
2973 static bool
2974 nvmf_ns_reservation_all_registrants_type(struct spdk_nvmf_ns *ns)
2975 {
2976 	return (ns->rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_ALL_REGS ||
2977 		ns->rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS);
2978 }
2979 
2980 /* current registrant is reservation holder or not */
2981 static bool
2982 nvmf_ns_reservation_registrant_is_holder(struct spdk_nvmf_ns *ns,
2983 		struct spdk_nvmf_registrant *reg)
2984 {
2985 	if (!reg) {
2986 		return false;
2987 	}
2988 
2989 	if (nvmf_ns_reservation_all_registrants_type(ns)) {
2990 		return true;
2991 	}
2992 
2993 	return (ns->holder == reg);
2994 }
2995 
2996 static int
2997 nvmf_ns_reservation_add_registrant(struct spdk_nvmf_ns *ns,
2998 				   struct spdk_nvmf_ctrlr *ctrlr,
2999 				   uint64_t nrkey)
3000 {
3001 	struct spdk_nvmf_registrant *reg;
3002 
3003 	reg = calloc(1, sizeof(*reg));
3004 	if (!reg) {
3005 		return -ENOMEM;
3006 	}
3007 
3008 	reg->rkey = nrkey;
3009 	/* set hostid for the registrant */
3010 	spdk_uuid_copy(&reg->hostid, &ctrlr->hostid);
3011 	TAILQ_INSERT_TAIL(&ns->registrants, reg, link);
3012 	ns->gen++;
3013 
3014 	return 0;
3015 }
3016 
3017 static void
3018 nvmf_ns_reservation_release_reservation(struct spdk_nvmf_ns *ns)
3019 {
3020 	ns->rtype = 0;
3021 	ns->crkey = 0;
3022 	ns->holder = NULL;
3023 }
3024 
3025 /* release the reservation if the last registrant was removed */
3026 static void
3027 nvmf_ns_reservation_check_release_on_remove_registrant(struct spdk_nvmf_ns *ns,
3028 		struct spdk_nvmf_registrant *reg)
3029 {
3030 	struct spdk_nvmf_registrant *next_reg;
3031 
3032 	/* no reservation holder */
3033 	if (!ns->holder) {
3034 		assert(ns->rtype == 0);
3035 		return;
3036 	}
3037 
3038 	next_reg = TAILQ_FIRST(&ns->registrants);
3039 	if (next_reg && nvmf_ns_reservation_all_registrants_type(ns)) {
3040 		/* the next valid registrant is the new holder now */
3041 		ns->holder = next_reg;
3042 	} else if (nvmf_ns_reservation_registrant_is_holder(ns, reg)) {
3043 		/* release the reservation */
3044 		nvmf_ns_reservation_release_reservation(ns);
3045 	}
3046 }
3047 
3048 static void
3049 nvmf_ns_reservation_remove_registrant(struct spdk_nvmf_ns *ns,
3050 				      struct spdk_nvmf_registrant *reg)
3051 {
3052 	TAILQ_REMOVE(&ns->registrants, reg, link);
3053 	nvmf_ns_reservation_check_release_on_remove_registrant(ns, reg);
3054 	free(reg);
3055 	ns->gen++;
3056 	return;
3057 }
3058 
3059 static uint32_t
3060 nvmf_ns_reservation_remove_registrants_by_key(struct spdk_nvmf_ns *ns,
3061 		uint64_t rkey)
3062 {
3063 	struct spdk_nvmf_registrant *reg, *tmp;
3064 	uint32_t count = 0;
3065 
3066 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3067 		if (reg->rkey == rkey) {
3068 			nvmf_ns_reservation_remove_registrant(ns, reg);
3069 			count++;
3070 		}
3071 	}
3072 	return count;
3073 }
3074 
3075 static uint32_t
3076 nvmf_ns_reservation_remove_all_other_registrants(struct spdk_nvmf_ns *ns,
3077 		struct spdk_nvmf_registrant *reg)
3078 {
3079 	struct spdk_nvmf_registrant *reg_tmp, *reg_tmp2;
3080 	uint32_t count = 0;
3081 
3082 	TAILQ_FOREACH_SAFE(reg_tmp, &ns->registrants, link, reg_tmp2) {
3083 		if (reg_tmp != reg) {
3084 			nvmf_ns_reservation_remove_registrant(ns, reg_tmp);
3085 			count++;
3086 		}
3087 	}
3088 	return count;
3089 }
3090 
3091 static uint32_t
3092 nvmf_ns_reservation_clear_all_registrants(struct spdk_nvmf_ns *ns)
3093 {
3094 	struct spdk_nvmf_registrant *reg, *reg_tmp;
3095 	uint32_t count = 0;
3096 
3097 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, reg_tmp) {
3098 		nvmf_ns_reservation_remove_registrant(ns, reg);
3099 		count++;
3100 	}
3101 	return count;
3102 }
3103 
3104 static void
3105 nvmf_ns_reservation_acquire_reservation(struct spdk_nvmf_ns *ns, uint64_t rkey,
3106 					enum spdk_nvme_reservation_type rtype,
3107 					struct spdk_nvmf_registrant *holder)
3108 {
3109 	ns->rtype = rtype;
3110 	ns->crkey = rkey;
3111 	assert(ns->holder == NULL);
3112 	ns->holder = holder;
3113 }
3114 
3115 static bool
3116 nvmf_ns_reservation_register(struct spdk_nvmf_ns *ns,
3117 			     struct spdk_nvmf_ctrlr *ctrlr,
3118 			     struct spdk_nvmf_request *req)
3119 {
3120 	struct spdk_nvme_reservation_register_data key = { 0 };
3121 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3122 	uint8_t rrega, iekey, cptpl, rtype;
3123 	struct spdk_nvmf_registrant *reg;
3124 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3125 	bool update_sgroup = false;
3126 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3127 	uint32_t num_hostid = 0;
3128 	int rc;
3129 
3130 	rrega = cmd->cdw10_bits.resv_register.rrega;
3131 	iekey = cmd->cdw10_bits.resv_register.iekey;
3132 	cptpl = cmd->cdw10_bits.resv_register.cptpl;
3133 
3134 	if (req->iovcnt > 0 && req->length >= sizeof(key)) {
3135 		struct spdk_iov_xfer ix;
3136 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3137 		spdk_iov_xfer_to_buf(&ix, &key, sizeof(key));
3138 	} else {
3139 		SPDK_ERRLOG("No key provided. Failing request.\n");
3140 		status = SPDK_NVME_SC_INVALID_FIELD;
3141 		goto exit;
3142 	}
3143 
3144 	SPDK_DEBUGLOG(nvmf, "REGISTER: RREGA %u, IEKEY %u, CPTPL %u, "
3145 		      "NRKEY 0x%"PRIx64", NRKEY 0x%"PRIx64"\n",
3146 		      rrega, iekey, cptpl, key.crkey, key.nrkey);
3147 
3148 	if (cptpl == SPDK_NVME_RESERVE_PTPL_CLEAR_POWER_ON) {
3149 		/* True to OFF state, and need to be updated in the configuration file */
3150 		if (ns->ptpl_activated) {
3151 			ns->ptpl_activated = 0;
3152 			update_sgroup = true;
3153 		}
3154 	} else if (cptpl == SPDK_NVME_RESERVE_PTPL_PERSIST_POWER_LOSS) {
3155 		if (!nvmf_ns_is_ptpl_capable(ns)) {
3156 			status = SPDK_NVME_SC_INVALID_FIELD;
3157 			goto exit;
3158 		} else if (ns->ptpl_activated == 0) {
3159 			ns->ptpl_activated = 1;
3160 			update_sgroup = true;
3161 		}
3162 	}
3163 
3164 	/* current Host Identifier has registrant or not */
3165 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3166 
3167 	switch (rrega) {
3168 	case SPDK_NVME_RESERVE_REGISTER_KEY:
3169 		if (!reg) {
3170 			/* register new controller */
3171 			if (key.nrkey == 0) {
3172 				SPDK_ERRLOG("Can't register zeroed new key\n");
3173 				status = SPDK_NVME_SC_INVALID_FIELD;
3174 				goto exit;
3175 			}
3176 			rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey);
3177 			if (rc < 0) {
3178 				status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3179 				goto exit;
3180 			}
3181 			update_sgroup = true;
3182 		} else {
3183 			/* register with same key is not an error */
3184 			if (reg->rkey != key.nrkey) {
3185 				SPDK_ERRLOG("The same host already register a "
3186 					    "key with 0x%"PRIx64"\n",
3187 					    reg->rkey);
3188 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3189 				goto exit;
3190 			}
3191 		}
3192 		break;
3193 	case SPDK_NVME_RESERVE_UNREGISTER_KEY:
3194 		if (!reg || (!iekey && reg->rkey != key.crkey)) {
3195 			SPDK_ERRLOG("No registrant or current key doesn't match "
3196 				    "with existing registrant key\n");
3197 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3198 			goto exit;
3199 		}
3200 
3201 		rtype = ns->rtype;
3202 		num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3203 				SPDK_NVMF_MAX_NUM_REGISTRANTS,
3204 				&ctrlr->hostid);
3205 
3206 		nvmf_ns_reservation_remove_registrant(ns, reg);
3207 
3208 		if (!ns->rtype && num_hostid && (rtype == SPDK_NVME_RESERVE_WRITE_EXCLUSIVE_REG_ONLY ||
3209 						 rtype == SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_REG_ONLY)) {
3210 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3211 							      hostid_list,
3212 							      num_hostid,
3213 							      SPDK_NVME_RESERVATION_RELEASED);
3214 		}
3215 		update_sgroup = true;
3216 		break;
3217 	case SPDK_NVME_RESERVE_REPLACE_KEY:
3218 		if (key.nrkey == 0) {
3219 			SPDK_ERRLOG("Can't register zeroed new key\n");
3220 			status = SPDK_NVME_SC_INVALID_FIELD;
3221 			goto exit;
3222 		}
3223 		/* Registrant exists */
3224 		if (reg) {
3225 			if (!iekey && reg->rkey != key.crkey) {
3226 				SPDK_ERRLOG("Current key doesn't match "
3227 					    "existing registrant key\n");
3228 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3229 				goto exit;
3230 			}
3231 			if (reg->rkey == key.nrkey) {
3232 				goto exit;
3233 			}
3234 			reg->rkey = key.nrkey;
3235 		} else if (iekey) { /* No registrant but IEKEY is set */
3236 			/* new registrant */
3237 			rc = nvmf_ns_reservation_add_registrant(ns, ctrlr, key.nrkey);
3238 			if (rc < 0) {
3239 				status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3240 				goto exit;
3241 			}
3242 		} else { /* No registrant */
3243 			SPDK_ERRLOG("No registrant\n");
3244 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3245 			goto exit;
3246 
3247 		}
3248 		update_sgroup = true;
3249 		break;
3250 	default:
3251 		status = SPDK_NVME_SC_INVALID_FIELD;
3252 		goto exit;
3253 	}
3254 
3255 exit:
3256 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3257 	req->rsp->nvme_cpl.status.sc = status;
3258 	return update_sgroup;
3259 }
3260 
3261 static bool
3262 nvmf_ns_reservation_acquire(struct spdk_nvmf_ns *ns,
3263 			    struct spdk_nvmf_ctrlr *ctrlr,
3264 			    struct spdk_nvmf_request *req)
3265 {
3266 	struct spdk_nvme_reservation_acquire_data key = { 0 };
3267 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3268 	uint8_t racqa, iekey, rtype;
3269 	struct spdk_nvmf_registrant *reg;
3270 	bool all_regs = false;
3271 	uint32_t count = 0;
3272 	bool update_sgroup = true;
3273 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3274 	uint32_t num_hostid = 0;
3275 	struct spdk_uuid new_hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3276 	uint32_t new_num_hostid = 0;
3277 	bool reservation_released = false;
3278 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3279 
3280 	racqa = cmd->cdw10_bits.resv_acquire.racqa;
3281 	iekey = cmd->cdw10_bits.resv_acquire.iekey;
3282 	rtype = cmd->cdw10_bits.resv_acquire.rtype;
3283 
3284 	if (req->iovcnt > 0 && req->length >= sizeof(key)) {
3285 		struct spdk_iov_xfer ix;
3286 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3287 		spdk_iov_xfer_to_buf(&ix, &key, sizeof(key));
3288 	} else {
3289 		SPDK_ERRLOG("No key provided. Failing request.\n");
3290 		status = SPDK_NVME_SC_INVALID_FIELD;
3291 		goto exit;
3292 	}
3293 
3294 	SPDK_DEBUGLOG(nvmf, "ACQUIRE: RACQA %u, IEKEY %u, RTYPE %u, "
3295 		      "NRKEY 0x%"PRIx64", PRKEY 0x%"PRIx64"\n",
3296 		      racqa, iekey, rtype, key.crkey, key.prkey);
3297 
3298 	if (iekey || rtype > SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS_ALL_REGS) {
3299 		SPDK_ERRLOG("Ignore existing key field set to 1\n");
3300 		status = SPDK_NVME_SC_INVALID_FIELD;
3301 		update_sgroup = false;
3302 		goto exit;
3303 	}
3304 
3305 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3306 	/* must be registrant and CRKEY must match */
3307 	if (!reg || reg->rkey != key.crkey) {
3308 		SPDK_ERRLOG("No registrant or current key doesn't match "
3309 			    "with existing registrant key\n");
3310 		status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3311 		update_sgroup = false;
3312 		goto exit;
3313 	}
3314 
3315 	all_regs = nvmf_ns_reservation_all_registrants_type(ns);
3316 
3317 	switch (racqa) {
3318 	case SPDK_NVME_RESERVE_ACQUIRE:
3319 		/* it's not an error for the holder to acquire same reservation type again */
3320 		if (nvmf_ns_reservation_registrant_is_holder(ns, reg) && ns->rtype == rtype) {
3321 			/* do nothing */
3322 			update_sgroup = false;
3323 		} else if (ns->holder == NULL) {
3324 			/* first time to acquire the reservation */
3325 			nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg);
3326 		} else {
3327 			SPDK_ERRLOG("Invalid rtype or current registrant is not holder\n");
3328 			status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3329 			update_sgroup = false;
3330 			goto exit;
3331 		}
3332 		break;
3333 	case SPDK_NVME_RESERVE_PREEMPT:
3334 		/* no reservation holder */
3335 		if (!ns->holder) {
3336 			/* unregister with PRKEY */
3337 			nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3338 			break;
3339 		}
3340 		num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3341 				SPDK_NVMF_MAX_NUM_REGISTRANTS,
3342 				&ctrlr->hostid);
3343 
3344 		/* only 1 reservation holder and reservation key is valid */
3345 		if (!all_regs) {
3346 			/* preempt itself */
3347 			if (nvmf_ns_reservation_registrant_is_holder(ns, reg) &&
3348 			    ns->crkey == key.prkey) {
3349 				ns->rtype = rtype;
3350 				reservation_released = true;
3351 				break;
3352 			}
3353 
3354 			if (ns->crkey == key.prkey) {
3355 				nvmf_ns_reservation_remove_registrant(ns, ns->holder);
3356 				nvmf_ns_reservation_acquire_reservation(ns, key.crkey, rtype, reg);
3357 				reservation_released = true;
3358 			} else if (key.prkey != 0) {
3359 				nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3360 			} else {
3361 				/* PRKEY is zero */
3362 				SPDK_ERRLOG("Current PRKEY is zero\n");
3363 				status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3364 				update_sgroup = false;
3365 				goto exit;
3366 			}
3367 		} else {
3368 			/* release all other registrants except for the current one */
3369 			if (key.prkey == 0) {
3370 				nvmf_ns_reservation_remove_all_other_registrants(ns, reg);
3371 				assert(ns->holder == reg);
3372 			} else {
3373 				count = nvmf_ns_reservation_remove_registrants_by_key(ns, key.prkey);
3374 				if (count == 0) {
3375 					SPDK_ERRLOG("PRKEY doesn't match any registrant\n");
3376 					status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3377 					update_sgroup = false;
3378 					goto exit;
3379 				}
3380 			}
3381 		}
3382 		break;
3383 	default:
3384 		status = SPDK_NVME_SC_INVALID_FIELD;
3385 		update_sgroup = false;
3386 		break;
3387 	}
3388 
3389 exit:
3390 	if (update_sgroup && racqa == SPDK_NVME_RESERVE_PREEMPT) {
3391 		new_num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, new_hostid_list,
3392 				 SPDK_NVMF_MAX_NUM_REGISTRANTS,
3393 				 &ctrlr->hostid);
3394 		/* Preempt notification occurs on the unregistered controllers
3395 		 * other than the controller who issued the command.
3396 		 */
3397 		num_hostid = nvmf_ns_reservation_get_unregistered_hostid(hostid_list,
3398 				num_hostid,
3399 				new_hostid_list,
3400 				new_num_hostid);
3401 		if (num_hostid) {
3402 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3403 							      hostid_list,
3404 							      num_hostid,
3405 							      SPDK_NVME_REGISTRATION_PREEMPTED);
3406 
3407 		}
3408 		/* Reservation released notification occurs on the
3409 		 * controllers which are the remaining registrants other than
3410 		 * the controller who issued the command.
3411 		 */
3412 		if (reservation_released && new_num_hostid) {
3413 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3414 							      new_hostid_list,
3415 							      new_num_hostid,
3416 							      SPDK_NVME_RESERVATION_RELEASED);
3417 
3418 		}
3419 	}
3420 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3421 	req->rsp->nvme_cpl.status.sc = status;
3422 	return update_sgroup;
3423 }
3424 
3425 static bool
3426 nvmf_ns_reservation_release(struct spdk_nvmf_ns *ns,
3427 			    struct spdk_nvmf_ctrlr *ctrlr,
3428 			    struct spdk_nvmf_request *req)
3429 {
3430 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3431 	uint8_t rrela, iekey, rtype;
3432 	struct spdk_nvmf_registrant *reg;
3433 	uint64_t crkey = 0;
3434 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3435 	bool update_sgroup = true;
3436 	struct spdk_uuid hostid_list[SPDK_NVMF_MAX_NUM_REGISTRANTS];
3437 	uint32_t num_hostid = 0;
3438 
3439 	rrela = cmd->cdw10_bits.resv_release.rrela;
3440 	iekey = cmd->cdw10_bits.resv_release.iekey;
3441 	rtype = cmd->cdw10_bits.resv_release.rtype;
3442 
3443 	if (req->iovcnt > 0 && req->length >= sizeof(crkey)) {
3444 		struct spdk_iov_xfer ix;
3445 		spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3446 		spdk_iov_xfer_to_buf(&ix, &crkey, sizeof(crkey));
3447 	} else {
3448 		SPDK_ERRLOG("No key provided. Failing request.\n");
3449 		status = SPDK_NVME_SC_INVALID_FIELD;
3450 		goto exit;
3451 	}
3452 
3453 	SPDK_DEBUGLOG(nvmf, "RELEASE: RRELA %u, IEKEY %u, RTYPE %u, "
3454 		      "CRKEY 0x%"PRIx64"\n",  rrela, iekey, rtype, crkey);
3455 
3456 	if (iekey) {
3457 		SPDK_ERRLOG("Ignore existing key field set to 1\n");
3458 		status = SPDK_NVME_SC_INVALID_FIELD;
3459 		update_sgroup = false;
3460 		goto exit;
3461 	}
3462 
3463 	reg = nvmf_ns_reservation_get_registrant(ns, &ctrlr->hostid);
3464 	if (!reg || reg->rkey != crkey) {
3465 		SPDK_ERRLOG("No registrant or current key doesn't match "
3466 			    "with existing registrant key\n");
3467 		status = SPDK_NVME_SC_RESERVATION_CONFLICT;
3468 		update_sgroup = false;
3469 		goto exit;
3470 	}
3471 
3472 	num_hostid = nvmf_ns_reservation_get_all_other_hostid(ns, hostid_list,
3473 			SPDK_NVMF_MAX_NUM_REGISTRANTS,
3474 			&ctrlr->hostid);
3475 
3476 	switch (rrela) {
3477 	case SPDK_NVME_RESERVE_RELEASE:
3478 		if (!ns->holder) {
3479 			SPDK_DEBUGLOG(nvmf, "RELEASE: no holder\n");
3480 			update_sgroup = false;
3481 			goto exit;
3482 		}
3483 		if (ns->rtype != rtype) {
3484 			SPDK_ERRLOG("Type doesn't match\n");
3485 			status = SPDK_NVME_SC_INVALID_FIELD;
3486 			update_sgroup = false;
3487 			goto exit;
3488 		}
3489 		if (!nvmf_ns_reservation_registrant_is_holder(ns, reg)) {
3490 			/* not the reservation holder, this isn't an error */
3491 			update_sgroup = false;
3492 			goto exit;
3493 		}
3494 
3495 		rtype = ns->rtype;
3496 		nvmf_ns_reservation_release_reservation(ns);
3497 
3498 		if (num_hostid && rtype != SPDK_NVME_RESERVE_WRITE_EXCLUSIVE &&
3499 		    rtype != SPDK_NVME_RESERVE_EXCLUSIVE_ACCESS) {
3500 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3501 							      hostid_list,
3502 							      num_hostid,
3503 							      SPDK_NVME_RESERVATION_RELEASED);
3504 		}
3505 		break;
3506 	case SPDK_NVME_RESERVE_CLEAR:
3507 		nvmf_ns_reservation_clear_all_registrants(ns);
3508 		if (num_hostid) {
3509 			nvmf_subsystem_gen_ctrlr_notification(ns->subsystem, ns,
3510 							      hostid_list,
3511 							      num_hostid,
3512 							      SPDK_NVME_RESERVATION_PREEMPTED);
3513 		}
3514 		break;
3515 	default:
3516 		status = SPDK_NVME_SC_INVALID_FIELD;
3517 		update_sgroup = false;
3518 		goto exit;
3519 	}
3520 
3521 exit:
3522 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3523 	req->rsp->nvme_cpl.status.sc = status;
3524 	return update_sgroup;
3525 }
3526 
3527 static void
3528 nvmf_ns_reservation_report(struct spdk_nvmf_ns *ns,
3529 			   struct spdk_nvmf_ctrlr *ctrlr,
3530 			   struct spdk_nvmf_request *req)
3531 {
3532 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3533 	struct spdk_nvmf_registrant *reg, *tmp;
3534 	struct spdk_nvme_reservation_status_extended_data status_data = { 0 };
3535 	struct spdk_iov_xfer ix;
3536 	uint32_t transfer_len;
3537 	uint32_t regctl = 0;
3538 	uint8_t status = SPDK_NVME_SC_SUCCESS;
3539 
3540 	if (req->iovcnt == 0) {
3541 		SPDK_ERRLOG("No data transfer specified for request. "
3542 			    " Unable to transfer back response.\n");
3543 		status = SPDK_NVME_SC_INVALID_FIELD;
3544 		goto exit;
3545 	}
3546 
3547 	if (!cmd->cdw11_bits.resv_report.eds) {
3548 		SPDK_ERRLOG("NVMeoF uses extended controller data structure, "
3549 			    "please set EDS bit in cdw11 and try again\n");
3550 		status = SPDK_NVME_SC_HOSTID_INCONSISTENT_FORMAT;
3551 		goto exit;
3552 	}
3553 
3554 	/* Number of Dwords of the Reservation Status data structure to transfer */
3555 	transfer_len = (cmd->cdw10 + 1) * sizeof(uint32_t);
3556 
3557 	if (transfer_len < sizeof(struct spdk_nvme_reservation_status_extended_data)) {
3558 		status = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3559 		goto exit;
3560 	}
3561 
3562 	spdk_iov_xfer_init(&ix, req->iov, req->iovcnt);
3563 
3564 	status_data.data.gen = ns->gen;
3565 	status_data.data.rtype = ns->rtype;
3566 	status_data.data.ptpls = ns->ptpl_activated;
3567 
3568 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3569 		regctl++;
3570 	}
3571 
3572 	/*
3573 	 * We report the number of registrants as per the spec here, even if
3574 	 * the iov isn't big enough to contain them all. In that case, the
3575 	 * spdk_iov_xfer_from_buf() won't actually copy any of the remaining
3576 	 * data; as it keeps track of the iov cursor itself, it's simplest to
3577 	 * just walk the entire list anyway.
3578 	 */
3579 	status_data.data.regctl = regctl;
3580 
3581 	spdk_iov_xfer_from_buf(&ix, &status_data, sizeof(status_data));
3582 
3583 	TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
3584 		struct spdk_nvme_registered_ctrlr_extended_data ctrlr_data = { 0 };
3585 
3586 		/* Set to 0xffffh for dynamic controller */
3587 		ctrlr_data.cntlid = 0xffff;
3588 		ctrlr_data.rcsts.status = (ns->holder == reg) ? true : false;
3589 		ctrlr_data.rkey = reg->rkey;
3590 		spdk_uuid_copy((struct spdk_uuid *)ctrlr_data.hostid, &reg->hostid);
3591 
3592 		spdk_iov_xfer_from_buf(&ix, &ctrlr_data, sizeof(ctrlr_data));
3593 	}
3594 
3595 exit:
3596 	req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3597 	req->rsp->nvme_cpl.status.sc = status;
3598 	return;
3599 }
3600 
3601 static void
3602 nvmf_ns_reservation_complete(void *ctx)
3603 {
3604 	struct spdk_nvmf_request *req = ctx;
3605 
3606 	spdk_nvmf_request_complete(req);
3607 }
3608 
3609 static void
3610 _nvmf_ns_reservation_update_done(struct spdk_nvmf_subsystem *subsystem,
3611 				 void *cb_arg, int status)
3612 {
3613 	struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)cb_arg;
3614 	struct spdk_nvmf_poll_group *group = req->qpair->group;
3615 
3616 	spdk_thread_send_msg(group->thread, nvmf_ns_reservation_complete, req);
3617 }
3618 
3619 void
3620 nvmf_ns_reservation_request(void *ctx)
3621 {
3622 	struct spdk_nvmf_request *req = (struct spdk_nvmf_request *)ctx;
3623 	struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd;
3624 	struct spdk_nvmf_ctrlr *ctrlr = req->qpair->ctrlr;
3625 	uint32_t nsid;
3626 	struct spdk_nvmf_ns *ns;
3627 	bool update_sgroup = false;
3628 	int status = 0;
3629 
3630 	nsid = cmd->nsid;
3631 	ns = _nvmf_subsystem_get_ns(ctrlr->subsys, nsid);
3632 	assert(ns != NULL);
3633 
3634 	switch (cmd->opc) {
3635 	case SPDK_NVME_OPC_RESERVATION_REGISTER:
3636 		update_sgroup = nvmf_ns_reservation_register(ns, ctrlr, req);
3637 		break;
3638 	case SPDK_NVME_OPC_RESERVATION_ACQUIRE:
3639 		update_sgroup = nvmf_ns_reservation_acquire(ns, ctrlr, req);
3640 		break;
3641 	case SPDK_NVME_OPC_RESERVATION_RELEASE:
3642 		update_sgroup = nvmf_ns_reservation_release(ns, ctrlr, req);
3643 		break;
3644 	case SPDK_NVME_OPC_RESERVATION_REPORT:
3645 		nvmf_ns_reservation_report(ns, ctrlr, req);
3646 		break;
3647 	default:
3648 		break;
3649 	}
3650 
3651 	/* update reservation information to subsystem's poll group */
3652 	if (update_sgroup) {
3653 		if (ns->ptpl_activated || cmd->opc == SPDK_NVME_OPC_RESERVATION_REGISTER) {
3654 			if (nvmf_ns_update_reservation_info(ns) != 0) {
3655 				req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
3656 			}
3657 		}
3658 		status = nvmf_subsystem_update_ns(ctrlr->subsys, _nvmf_ns_reservation_update_done, req);
3659 		if (status == 0) {
3660 			return;
3661 		}
3662 	}
3663 
3664 	_nvmf_ns_reservation_update_done(ctrlr->subsys, req, status);
3665 }
3666 
3667 static bool
3668 nvmf_ns_is_ptpl_capable_json(const struct spdk_nvmf_ns *ns)
3669 {
3670 	return ns->ptpl_file != NULL;
3671 }
3672 
3673 static struct spdk_nvmf_ns_reservation_ops g_reservation_ops = {
3674 	.is_ptpl_capable = nvmf_ns_is_ptpl_capable_json,
3675 	.update = nvmf_ns_reservation_update_json,
3676 	.load = nvmf_ns_reservation_load_json,
3677 };
3678 
3679 bool
3680 nvmf_ns_is_ptpl_capable(const struct spdk_nvmf_ns *ns)
3681 {
3682 	return g_reservation_ops.is_ptpl_capable(ns);
3683 }
3684 
3685 static int
3686 nvmf_ns_reservation_update(const struct spdk_nvmf_ns *ns,
3687 			   const struct spdk_nvmf_reservation_info *info)
3688 {
3689 	return g_reservation_ops.update(ns, info);
3690 }
3691 
3692 static int
3693 nvmf_ns_reservation_load(const struct spdk_nvmf_ns *ns, struct spdk_nvmf_reservation_info *info)
3694 {
3695 	return g_reservation_ops.load(ns, info);
3696 }
3697 
3698 void
3699 spdk_nvmf_set_custom_ns_reservation_ops(const struct spdk_nvmf_ns_reservation_ops *ops)
3700 {
3701 	g_reservation_ops = *ops;
3702 }
3703 
3704 int
3705 spdk_nvmf_subsystem_set_ana_reporting(struct spdk_nvmf_subsystem *subsystem,
3706 				      bool ana_reporting)
3707 {
3708 	if (subsystem->state != SPDK_NVMF_SUBSYSTEM_INACTIVE) {
3709 		return -EAGAIN;
3710 	}
3711 
3712 	subsystem->flags.ana_reporting = ana_reporting;
3713 
3714 	return 0;
3715 }
3716 
3717 bool
3718 spdk_nvmf_subsystem_get_ana_reporting(struct spdk_nvmf_subsystem *subsystem)
3719 {
3720 	return subsystem->flags.ana_reporting;
3721 }
3722 
3723 struct subsystem_listener_update_ctx {
3724 	struct spdk_nvmf_subsystem_listener *listener;
3725 
3726 	spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn;
3727 	void *cb_arg;
3728 };
3729 
3730 static void
3731 subsystem_listener_update_done(struct spdk_io_channel_iter *i, int status)
3732 {
3733 	struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
3734 
3735 	if (ctx->cb_fn) {
3736 		ctx->cb_fn(ctx->cb_arg, status);
3737 	}
3738 	free(ctx);
3739 }
3740 
3741 static void
3742 subsystem_listener_update_on_pg(struct spdk_io_channel_iter *i)
3743 {
3744 	struct subsystem_listener_update_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
3745 	struct spdk_nvmf_subsystem_listener *listener;
3746 	struct spdk_nvmf_poll_group *group;
3747 	struct spdk_nvmf_ctrlr *ctrlr;
3748 
3749 	listener = ctx->listener;
3750 	group = spdk_io_channel_get_ctx(spdk_io_channel_iter_get_channel(i));
3751 
3752 	TAILQ_FOREACH(ctrlr, &listener->subsystem->ctrlrs, link) {
3753 		if (ctrlr->thread != spdk_get_thread()) {
3754 			continue;
3755 		}
3756 
3757 		if (ctrlr->admin_qpair && ctrlr->admin_qpair->group == group && ctrlr->listener == listener) {
3758 			nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
3759 		}
3760 	}
3761 
3762 	spdk_for_each_channel_continue(i, 0);
3763 }
3764 
3765 void
3766 spdk_nvmf_subsystem_set_ana_state(struct spdk_nvmf_subsystem *subsystem,
3767 				  const struct spdk_nvme_transport_id *trid,
3768 				  enum spdk_nvme_ana_state ana_state, uint32_t anagrpid,
3769 				  spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn, void *cb_arg)
3770 {
3771 	struct spdk_nvmf_subsystem_listener *listener;
3772 	struct subsystem_listener_update_ctx *ctx;
3773 	uint32_t i;
3774 
3775 	assert(cb_fn != NULL);
3776 	assert(subsystem->state == SPDK_NVMF_SUBSYSTEM_INACTIVE ||
3777 	       subsystem->state == SPDK_NVMF_SUBSYSTEM_PAUSED);
3778 
3779 	if (!subsystem->flags.ana_reporting) {
3780 		SPDK_ERRLOG("ANA reporting is disabled\n");
3781 		cb_fn(cb_arg, -EINVAL);
3782 		return;
3783 	}
3784 
3785 	/* ANA Change state is not used, ANA Persistent Loss state
3786 	 * is not supported yet.
3787 	 */
3788 	if (!(ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE ||
3789 	      ana_state == SPDK_NVME_ANA_NON_OPTIMIZED_STATE ||
3790 	      ana_state == SPDK_NVME_ANA_INACCESSIBLE_STATE)) {
3791 		SPDK_ERRLOG("ANA state %d is not supported\n", ana_state);
3792 		cb_fn(cb_arg, -ENOTSUP);
3793 		return;
3794 	}
3795 
3796 	if (anagrpid > subsystem->max_nsid) {
3797 		SPDK_ERRLOG("ANA group ID %" PRIu32 " is more than maximum\n", anagrpid);
3798 		cb_fn(cb_arg, -EINVAL);
3799 		return;
3800 	}
3801 
3802 	listener = nvmf_subsystem_find_listener(subsystem, trid);
3803 	if (!listener) {
3804 		SPDK_ERRLOG("Unable to find listener.\n");
3805 		cb_fn(cb_arg, -EINVAL);
3806 		return;
3807 	}
3808 
3809 	if (anagrpid != 0 && listener->ana_state[anagrpid - 1] == ana_state) {
3810 		cb_fn(cb_arg, 0);
3811 		return;
3812 	}
3813 
3814 	ctx = calloc(1, sizeof(*ctx));
3815 	if (!ctx) {
3816 		SPDK_ERRLOG("Unable to allocate context\n");
3817 		cb_fn(cb_arg, -ENOMEM);
3818 		return;
3819 	}
3820 
3821 	for (i = 1; i <= subsystem->max_nsid; i++) {
3822 		if (anagrpid == 0 || i == anagrpid) {
3823 			listener->ana_state[i - 1] = ana_state;
3824 		}
3825 	}
3826 	listener->ana_state_change_count++;
3827 
3828 	ctx->listener = listener;
3829 	ctx->cb_fn = cb_fn;
3830 	ctx->cb_arg = cb_arg;
3831 
3832 	spdk_for_each_channel(subsystem->tgt,
3833 			      subsystem_listener_update_on_pg,
3834 			      ctx,
3835 			      subsystem_listener_update_done);
3836 }
3837 
3838 int
3839 spdk_nvmf_subsystem_get_ana_state(struct spdk_nvmf_subsystem *subsystem,
3840 				  const struct spdk_nvme_transport_id *trid,
3841 				  uint32_t anagrpid,
3842 				  enum spdk_nvme_ana_state *ana_state)
3843 {
3844 	assert(ana_state != NULL);
3845 
3846 	struct spdk_nvmf_subsystem_listener *listener;
3847 
3848 	if (!subsystem->flags.ana_reporting) {
3849 		SPDK_ERRLOG("ANA reporting is disabled\n");
3850 		return -EINVAL;
3851 	}
3852 
3853 	if (anagrpid <= 0 || anagrpid > subsystem->max_nsid) {
3854 		SPDK_ERRLOG("ANA group ID %" PRIu32 " is invalid\n", anagrpid);
3855 		return -EINVAL;
3856 	}
3857 
3858 	listener = nvmf_subsystem_find_listener(subsystem, trid);
3859 	if (!listener) {
3860 		SPDK_ERRLOG("Unable to find listener.\n");
3861 		return -EINVAL;
3862 	}
3863 
3864 	*ana_state = listener->ana_state[anagrpid - 1];
3865 	return 0;
3866 }
3867 
3868 bool
3869 spdk_nvmf_subsystem_is_discovery(struct spdk_nvmf_subsystem *subsystem)
3870 {
3871 	return subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
3872 	       subsystem->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY;
3873 }
3874 
3875 bool
3876 nvmf_nqn_is_discovery(const char *nqn)
3877 {
3878 	return strcmp(nqn, SPDK_NVMF_DISCOVERY_NQN) == 0;
3879 }
3880