xref: /spdk/lib/nvmf/nvmf.c (revision 32999ab917f67af61872f868585fd3d78ad6fb8a)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2018-2019, 2021 Mellanox Technologies LTD. All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "spdk/bdev.h"
37 #include "spdk/bit_array.h"
38 #include "spdk/thread.h"
39 #include "spdk/nvmf.h"
40 #include "spdk/trace.h"
41 #include "spdk/endian.h"
42 #include "spdk/string.h"
43 
44 #include "spdk/log.h"
45 
46 #include "nvmf_internal.h"
47 #include "transport.h"
48 
49 SPDK_LOG_REGISTER_COMPONENT(nvmf)
50 
51 #define SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS 1024
52 #define SPDK_NVMF_DEFAULT_ACCEPT_POLL_RATE_US 10000
53 
54 static TAILQ_HEAD(, spdk_nvmf_tgt) g_nvmf_tgts = TAILQ_HEAD_INITIALIZER(g_nvmf_tgts);
55 
56 typedef void (*nvmf_qpair_disconnect_cpl)(void *ctx, int status);
57 static void nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf);
58 
59 /* supplied to a single call to nvmf_qpair_disconnect */
60 struct nvmf_qpair_disconnect_ctx {
61 	struct spdk_nvmf_qpair *qpair;
62 	struct spdk_nvmf_ctrlr *ctrlr;
63 	nvmf_qpair_disconnect_cb cb_fn;
64 	struct spdk_thread *thread;
65 	void *ctx;
66 	uint16_t qid;
67 };
68 
69 /*
70  * There are several times when we need to iterate through the list of all qpairs and selectively delete them.
71  * In order to do this sequentially without overlap, we must provide a context to recover the next qpair from
72  * to enable calling nvmf_qpair_disconnect on the next desired qpair.
73  */
74 struct nvmf_qpair_disconnect_many_ctx {
75 	struct spdk_nvmf_subsystem *subsystem;
76 	struct spdk_nvmf_poll_group *group;
77 	spdk_nvmf_poll_group_mod_done cpl_fn;
78 	void *cpl_ctx;
79 	uint32_t count;
80 };
81 
82 static void
83 nvmf_qpair_set_state(struct spdk_nvmf_qpair *qpair,
84 		     enum spdk_nvmf_qpair_state state)
85 {
86 	assert(qpair != NULL);
87 	assert(qpair->group->thread == spdk_get_thread());
88 
89 	qpair->state = state;
90 }
91 
92 static int
93 nvmf_poll_group_poll(void *ctx)
94 {
95 	struct spdk_nvmf_poll_group *group = ctx;
96 	int rc;
97 	int count = 0;
98 	struct spdk_nvmf_transport_poll_group *tgroup;
99 
100 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
101 		rc = nvmf_transport_poll_group_poll(tgroup);
102 		if (rc < 0) {
103 			return SPDK_POLLER_BUSY;
104 		}
105 		count += rc;
106 	}
107 
108 	return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
109 }
110 
111 static int
112 nvmf_tgt_create_poll_group(void *io_device, void *ctx_buf)
113 {
114 	struct spdk_nvmf_tgt *tgt = io_device;
115 	struct spdk_nvmf_poll_group *group = ctx_buf;
116 	struct spdk_nvmf_transport *transport;
117 	uint32_t sid;
118 
119 	TAILQ_INIT(&group->tgroups);
120 	TAILQ_INIT(&group->qpairs);
121 
122 	TAILQ_FOREACH(transport, &tgt->transports, link) {
123 		nvmf_poll_group_add_transport(group, transport);
124 	}
125 
126 	group->num_sgroups = tgt->max_subsystems;
127 	group->sgroups = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem_poll_group));
128 	if (!group->sgroups) {
129 		return -ENOMEM;
130 	}
131 
132 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
133 		struct spdk_nvmf_subsystem *subsystem;
134 
135 		subsystem = tgt->subsystems[sid];
136 		if (!subsystem) {
137 			continue;
138 		}
139 
140 		if (nvmf_poll_group_add_subsystem(group, subsystem, NULL, NULL) != 0) {
141 			nvmf_tgt_destroy_poll_group(io_device, ctx_buf);
142 			return -1;
143 		}
144 	}
145 
146 	pthread_mutex_lock(&tgt->mutex);
147 	TAILQ_INSERT_TAIL(&tgt->poll_groups, group, link);
148 	pthread_mutex_unlock(&tgt->mutex);
149 
150 	group->poller = SPDK_POLLER_REGISTER(nvmf_poll_group_poll, group, 0);
151 	group->thread = spdk_get_thread();
152 
153 	return 0;
154 }
155 
156 static void
157 nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf)
158 {
159 	struct spdk_nvmf_tgt *tgt = io_device;
160 	struct spdk_nvmf_poll_group *group = ctx_buf;
161 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
162 	struct spdk_nvmf_subsystem_poll_group *sgroup;
163 	uint32_t sid, nsid;
164 
165 	pthread_mutex_lock(&tgt->mutex);
166 	TAILQ_REMOVE(&tgt->poll_groups, group, link);
167 	pthread_mutex_unlock(&tgt->mutex);
168 
169 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
170 		TAILQ_REMOVE(&group->tgroups, tgroup, link);
171 		nvmf_transport_poll_group_destroy(tgroup);
172 	}
173 
174 	for (sid = 0; sid < group->num_sgroups; sid++) {
175 		sgroup = &group->sgroups[sid];
176 
177 		for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
178 			if (sgroup->ns_info[nsid].channel) {
179 				spdk_put_io_channel(sgroup->ns_info[nsid].channel);
180 				sgroup->ns_info[nsid].channel = NULL;
181 			}
182 		}
183 
184 		free(sgroup->ns_info);
185 	}
186 
187 	free(group->sgroups);
188 
189 	spdk_poller_unregister(&group->poller);
190 
191 	if (group->destroy_cb_fn) {
192 		group->destroy_cb_fn(group->destroy_cb_arg, 0);
193 	}
194 }
195 
196 static void
197 _nvmf_tgt_disconnect_next_qpair(void *ctx)
198 {
199 	struct spdk_nvmf_qpair *qpair;
200 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
201 	struct spdk_nvmf_poll_group *group = qpair_ctx->group;
202 	struct spdk_io_channel *ch;
203 	int rc = 0;
204 
205 	qpair = TAILQ_FIRST(&group->qpairs);
206 
207 	if (qpair) {
208 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_tgt_disconnect_next_qpair, ctx);
209 	}
210 
211 	if (!qpair || rc != 0) {
212 		/* When the refcount from the channels reaches 0, nvmf_tgt_destroy_poll_group will be called. */
213 		ch = spdk_io_channel_from_ctx(group);
214 		spdk_put_io_channel(ch);
215 		free(qpair_ctx);
216 	}
217 }
218 
219 static void
220 nvmf_tgt_destroy_poll_group_qpairs(struct spdk_nvmf_poll_group *group)
221 {
222 	struct nvmf_qpair_disconnect_many_ctx *ctx;
223 
224 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
225 
226 	if (!ctx) {
227 		SPDK_ERRLOG("Failed to allocate memory for destroy poll group ctx\n");
228 		return;
229 	}
230 
231 	ctx->group = group;
232 	_nvmf_tgt_disconnect_next_qpair(ctx);
233 }
234 
235 static int
236 nvmf_tgt_accept(void *ctx)
237 {
238 	struct spdk_nvmf_tgt *tgt = ctx;
239 	struct spdk_nvmf_transport *transport, *tmp;
240 	int count = 0;
241 
242 	TAILQ_FOREACH_SAFE(transport, &tgt->transports, link, tmp) {
243 		count += nvmf_transport_accept(transport);
244 	}
245 
246 	return count;
247 }
248 
249 struct spdk_nvmf_tgt *
250 spdk_nvmf_tgt_create(struct spdk_nvmf_target_opts *opts)
251 {
252 	struct spdk_nvmf_tgt *tgt, *tmp_tgt;
253 	uint32_t acceptor_poll_rate;
254 
255 	if (strnlen(opts->name, NVMF_TGT_NAME_MAX_LENGTH) == NVMF_TGT_NAME_MAX_LENGTH) {
256 		SPDK_ERRLOG("Provided target name exceeds the max length of %u.\n", NVMF_TGT_NAME_MAX_LENGTH);
257 		return NULL;
258 	}
259 
260 	TAILQ_FOREACH(tmp_tgt, &g_nvmf_tgts, link) {
261 		if (!strncmp(opts->name, tmp_tgt->name, NVMF_TGT_NAME_MAX_LENGTH)) {
262 			SPDK_ERRLOG("Provided target name must be unique.\n");
263 			return NULL;
264 		}
265 	}
266 
267 	tgt = calloc(1, sizeof(*tgt));
268 	if (!tgt) {
269 		return NULL;
270 	}
271 
272 	snprintf(tgt->name, NVMF_TGT_NAME_MAX_LENGTH, "%s", opts->name);
273 
274 	if (!opts || !opts->max_subsystems) {
275 		tgt->max_subsystems = SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS;
276 	} else {
277 		tgt->max_subsystems = opts->max_subsystems;
278 	}
279 
280 	if (!opts || !opts->acceptor_poll_rate) {
281 		acceptor_poll_rate = SPDK_NVMF_DEFAULT_ACCEPT_POLL_RATE_US;
282 	} else {
283 		acceptor_poll_rate = opts->acceptor_poll_rate;
284 	}
285 
286 	tgt->discovery_genctr = 0;
287 	TAILQ_INIT(&tgt->transports);
288 	TAILQ_INIT(&tgt->poll_groups);
289 
290 	tgt->subsystems = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem *));
291 	if (!tgt->subsystems) {
292 		free(tgt);
293 		return NULL;
294 	}
295 
296 	pthread_mutex_init(&tgt->mutex, NULL);
297 
298 	tgt->accept_poller = SPDK_POLLER_REGISTER(nvmf_tgt_accept, tgt, acceptor_poll_rate);
299 	if (!tgt->accept_poller) {
300 		pthread_mutex_destroy(&tgt->mutex);
301 		free(tgt->subsystems);
302 		free(tgt);
303 		return NULL;
304 	}
305 
306 	spdk_io_device_register(tgt,
307 				nvmf_tgt_create_poll_group,
308 				nvmf_tgt_destroy_poll_group,
309 				sizeof(struct spdk_nvmf_poll_group),
310 				tgt->name);
311 
312 	TAILQ_INSERT_HEAD(&g_nvmf_tgts, tgt, link);
313 
314 	return tgt;
315 }
316 
317 static void
318 _nvmf_tgt_destroy_next_transport(void *ctx)
319 {
320 	struct spdk_nvmf_tgt *tgt = ctx;
321 	struct spdk_nvmf_transport *transport;
322 
323 	if (!TAILQ_EMPTY(&tgt->transports)) {
324 		transport = TAILQ_FIRST(&tgt->transports);
325 		TAILQ_REMOVE(&tgt->transports, transport, link);
326 		spdk_nvmf_transport_destroy(transport, _nvmf_tgt_destroy_next_transport, tgt);
327 	} else {
328 		spdk_nvmf_tgt_destroy_done_fn *destroy_cb_fn = tgt->destroy_cb_fn;
329 		void *destroy_cb_arg = tgt->destroy_cb_arg;
330 
331 		pthread_mutex_destroy(&tgt->mutex);
332 		free(tgt);
333 
334 		if (destroy_cb_fn) {
335 			destroy_cb_fn(destroy_cb_arg, 0);
336 		}
337 	}
338 }
339 
340 static void
341 nvmf_tgt_destroy_cb(void *io_device)
342 {
343 	struct spdk_nvmf_tgt *tgt = io_device;
344 	uint32_t i;
345 
346 	if (tgt->subsystems) {
347 		for (i = 0; i < tgt->max_subsystems; i++) {
348 			if (tgt->subsystems[i]) {
349 				nvmf_subsystem_remove_all_listeners(tgt->subsystems[i], true);
350 				spdk_nvmf_subsystem_destroy(tgt->subsystems[i]);
351 			}
352 		}
353 		free(tgt->subsystems);
354 	}
355 
356 	_nvmf_tgt_destroy_next_transport(tgt);
357 }
358 
359 void
360 spdk_nvmf_tgt_destroy(struct spdk_nvmf_tgt *tgt,
361 		      spdk_nvmf_tgt_destroy_done_fn cb_fn,
362 		      void *cb_arg)
363 {
364 	tgt->destroy_cb_fn = cb_fn;
365 	tgt->destroy_cb_arg = cb_arg;
366 
367 	spdk_poller_unregister(&tgt->accept_poller);
368 
369 	TAILQ_REMOVE(&g_nvmf_tgts, tgt, link);
370 
371 	spdk_io_device_unregister(tgt, nvmf_tgt_destroy_cb);
372 }
373 
374 const char *
375 spdk_nvmf_tgt_get_name(struct spdk_nvmf_tgt *tgt)
376 {
377 	return tgt->name;
378 }
379 
380 struct spdk_nvmf_tgt *
381 spdk_nvmf_get_tgt(const char *name)
382 {
383 	struct spdk_nvmf_tgt *tgt;
384 	uint32_t num_targets = 0;
385 
386 	TAILQ_FOREACH(tgt, &g_nvmf_tgts, link) {
387 		if (name) {
388 			if (!strncmp(tgt->name, name, NVMF_TGT_NAME_MAX_LENGTH)) {
389 				return tgt;
390 			}
391 		}
392 		num_targets++;
393 	}
394 
395 	/*
396 	 * special case. If there is only one target and
397 	 * no name was specified, return the only available
398 	 * target. If there is more than one target, name must
399 	 * be specified.
400 	 */
401 	if (!name && num_targets == 1) {
402 		return TAILQ_FIRST(&g_nvmf_tgts);
403 	}
404 
405 	return NULL;
406 }
407 
408 struct spdk_nvmf_tgt *
409 spdk_nvmf_get_first_tgt(void)
410 {
411 	return TAILQ_FIRST(&g_nvmf_tgts);
412 }
413 
414 struct spdk_nvmf_tgt *
415 spdk_nvmf_get_next_tgt(struct spdk_nvmf_tgt *prev)
416 {
417 	return TAILQ_NEXT(prev, link);
418 }
419 
420 static void
421 nvmf_write_subsystem_config_json(struct spdk_json_write_ctx *w,
422 				 struct spdk_nvmf_subsystem *subsystem)
423 {
424 	struct spdk_nvmf_host *host;
425 	struct spdk_nvmf_subsystem_listener *listener;
426 	const struct spdk_nvme_transport_id *trid;
427 	struct spdk_nvmf_ns *ns;
428 	struct spdk_nvmf_ns_opts ns_opts;
429 	uint32_t max_namespaces;
430 	char uuid_str[SPDK_UUID_STRING_LEN];
431 	const char *adrfam;
432 
433 	if (spdk_nvmf_subsystem_get_type(subsystem) != SPDK_NVMF_SUBTYPE_NVME) {
434 		return;
435 	}
436 
437 	/* { */
438 	spdk_json_write_object_begin(w);
439 	spdk_json_write_named_string(w, "method", "nvmf_create_subsystem");
440 
441 	/*     "params" : { */
442 	spdk_json_write_named_object_begin(w, "params");
443 	spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
444 	spdk_json_write_named_bool(w, "allow_any_host", spdk_nvmf_subsystem_get_allow_any_host(subsystem));
445 	spdk_json_write_named_string(w, "serial_number", spdk_nvmf_subsystem_get_sn(subsystem));
446 	spdk_json_write_named_string(w, "model_number", spdk_nvmf_subsystem_get_mn(subsystem));
447 
448 	max_namespaces = spdk_nvmf_subsystem_get_max_namespaces(subsystem);
449 	if (max_namespaces != 0) {
450 		spdk_json_write_named_uint32(w, "max_namespaces", max_namespaces);
451 	}
452 
453 	/*     } "params" */
454 	spdk_json_write_object_end(w);
455 
456 	/* } */
457 	spdk_json_write_object_end(w);
458 
459 	for (listener = spdk_nvmf_subsystem_get_first_listener(subsystem); listener != NULL;
460 	     listener = spdk_nvmf_subsystem_get_next_listener(subsystem, listener)) {
461 		trid = spdk_nvmf_subsystem_listener_get_trid(listener);
462 
463 		adrfam = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
464 
465 		spdk_json_write_object_begin(w);
466 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_listener");
467 
468 		/*     "params" : { */
469 		spdk_json_write_named_object_begin(w, "params");
470 
471 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
472 
473 		/*     "listen_address" : { */
474 		spdk_json_write_named_object_begin(w, "listen_address");
475 
476 		spdk_json_write_named_string(w, "trtype", trid->trstring);
477 		if (adrfam) {
478 			spdk_json_write_named_string(w, "adrfam", adrfam);
479 		}
480 
481 		spdk_json_write_named_string(w, "traddr", trid->traddr);
482 		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
483 		/*     } "listen_address" */
484 		spdk_json_write_object_end(w);
485 
486 		/*     } "params" */
487 		spdk_json_write_object_end(w);
488 
489 		/* } */
490 		spdk_json_write_object_end(w);
491 	}
492 
493 	for (host = spdk_nvmf_subsystem_get_first_host(subsystem); host != NULL;
494 	     host = spdk_nvmf_subsystem_get_next_host(subsystem, host)) {
495 
496 		spdk_json_write_object_begin(w);
497 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_host");
498 
499 		/*     "params" : { */
500 		spdk_json_write_named_object_begin(w, "params");
501 
502 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
503 		spdk_json_write_named_string(w, "host", spdk_nvmf_host_get_nqn(host));
504 
505 		/*     } "params" */
506 		spdk_json_write_object_end(w);
507 
508 		/* } */
509 		spdk_json_write_object_end(w);
510 	}
511 
512 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
513 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
514 		spdk_nvmf_ns_get_opts(ns, &ns_opts, sizeof(ns_opts));
515 
516 		spdk_json_write_object_begin(w);
517 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_ns");
518 
519 		/*     "params" : { */
520 		spdk_json_write_named_object_begin(w, "params");
521 
522 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
523 
524 		/*     "namespace" : { */
525 		spdk_json_write_named_object_begin(w, "namespace");
526 
527 		spdk_json_write_named_uint32(w, "nsid", spdk_nvmf_ns_get_id(ns));
528 		spdk_json_write_named_string(w, "bdev_name", spdk_bdev_get_name(spdk_nvmf_ns_get_bdev(ns)));
529 
530 		if (!spdk_mem_all_zero(ns_opts.nguid, sizeof(ns_opts.nguid))) {
531 			SPDK_STATIC_ASSERT(sizeof(ns_opts.nguid) == sizeof(uint64_t) * 2, "size mismatch");
532 			spdk_json_write_named_string_fmt(w, "nguid", "%016"PRIX64"%016"PRIX64, from_be64(&ns_opts.nguid[0]),
533 							 from_be64(&ns_opts.nguid[8]));
534 		}
535 
536 		if (!spdk_mem_all_zero(ns_opts.eui64, sizeof(ns_opts.eui64))) {
537 			SPDK_STATIC_ASSERT(sizeof(ns_opts.eui64) == sizeof(uint64_t), "size mismatch");
538 			spdk_json_write_named_string_fmt(w, "eui64", "%016"PRIX64, from_be64(&ns_opts.eui64));
539 		}
540 
541 		if (!spdk_mem_all_zero(&ns_opts.uuid, sizeof(ns_opts.uuid))) {
542 			spdk_uuid_fmt_lower(uuid_str, sizeof(uuid_str), &ns_opts.uuid);
543 			spdk_json_write_named_string(w, "uuid",  uuid_str);
544 		}
545 
546 		/*     "namespace" */
547 		spdk_json_write_object_end(w);
548 
549 		/*     } "params" */
550 		spdk_json_write_object_end(w);
551 
552 		/* } */
553 		spdk_json_write_object_end(w);
554 	}
555 }
556 
557 void
558 spdk_nvmf_tgt_write_config_json(struct spdk_json_write_ctx *w, struct spdk_nvmf_tgt *tgt)
559 {
560 	struct spdk_nvmf_subsystem *subsystem;
561 	struct spdk_nvmf_transport *transport;
562 
563 	spdk_json_write_object_begin(w);
564 	spdk_json_write_named_string(w, "method", "nvmf_set_max_subsystems");
565 
566 	spdk_json_write_named_object_begin(w, "params");
567 	spdk_json_write_named_uint32(w, "max_subsystems", tgt->max_subsystems);
568 	spdk_json_write_object_end(w);
569 
570 	spdk_json_write_object_end(w);
571 
572 	/* write transports */
573 	TAILQ_FOREACH(transport, &tgt->transports, link) {
574 		spdk_json_write_object_begin(w);
575 		spdk_json_write_named_string(w, "method", "nvmf_create_transport");
576 
577 		spdk_json_write_named_object_begin(w, "params");
578 		spdk_json_write_named_string(w, "trtype", transport->ops->name);
579 		spdk_json_write_named_uint32(w, "max_queue_depth", transport->opts.max_queue_depth);
580 		spdk_json_write_named_uint32(w, "max_io_qpairs_per_ctrlr",
581 					     transport->opts.max_qpairs_per_ctrlr - 1);
582 		spdk_json_write_named_uint32(w, "in_capsule_data_size", transport->opts.in_capsule_data_size);
583 		spdk_json_write_named_uint32(w, "max_io_size", transport->opts.max_io_size);
584 		spdk_json_write_named_uint32(w, "io_unit_size", transport->opts.io_unit_size);
585 		spdk_json_write_named_uint32(w, "max_aq_depth", transport->opts.max_aq_depth);
586 		spdk_json_write_named_uint32(w, "num_shared_buffers", transport->opts.num_shared_buffers);
587 		spdk_json_write_named_uint32(w, "buf_cache_size", transport->opts.buf_cache_size);
588 		spdk_json_write_named_bool(w, "dif_insert_or_strip", transport->opts.dif_insert_or_strip);
589 		if (transport->ops->dump_opts) {
590 			transport->ops->dump_opts(transport, w);
591 		}
592 		spdk_json_write_named_uint32(w, "abort_timeout_sec", transport->opts.abort_timeout_sec);
593 		spdk_json_write_object_end(w);
594 
595 		spdk_json_write_object_end(w);
596 	}
597 
598 	subsystem = spdk_nvmf_subsystem_get_first(tgt);
599 	while (subsystem) {
600 		nvmf_write_subsystem_config_json(w, subsystem);
601 		subsystem = spdk_nvmf_subsystem_get_next(subsystem);
602 	}
603 }
604 
605 static void
606 nvmf_listen_opts_copy(struct spdk_nvmf_listen_opts *opts,
607 		      const struct spdk_nvmf_listen_opts *opts_src, size_t opts_size)
608 {
609 	assert(opts);
610 	assert(opts_src);
611 
612 	opts->opts_size = opts_size;
613 
614 #define SET_FIELD(field) \
615     if (offsetof(struct spdk_nvmf_listen_opts, field) + sizeof(opts->field) <= opts_size) { \
616                  opts->field = opts_src->field; \
617     } \
618 
619 	SET_FIELD(transport_specific);
620 #undef SET_FIELD
621 
622 	/* Do not remove this statement, you should always update this statement when you adding a new field,
623 	 * and do not forget to add the SET_FIELD statement for your added field. */
624 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listen_opts) == 16, "Incorrect size");
625 }
626 
627 void
628 spdk_nvmf_listen_opts_init(struct spdk_nvmf_listen_opts *opts, size_t opts_size)
629 {
630 	struct spdk_nvmf_listen_opts opts_local = {};
631 
632 	/* local version of opts should have defaults set here */
633 
634 	nvmf_listen_opts_copy(opts, &opts_local, opts_size);
635 }
636 
637 int
638 spdk_nvmf_tgt_listen_ext(struct spdk_nvmf_tgt *tgt, const struct spdk_nvme_transport_id *trid,
639 			 struct spdk_nvmf_listen_opts *opts)
640 {
641 	struct spdk_nvmf_transport *transport;
642 	int rc;
643 	struct spdk_nvmf_listen_opts opts_local = {};
644 
645 	if (!opts) {
646 		SPDK_ERRLOG("opts should not be NULL\n");
647 		return -EINVAL;
648 	}
649 
650 	if (!opts->opts_size) {
651 		SPDK_ERRLOG("The opts_size in opts structure should not be zero\n");
652 		return -EINVAL;
653 	}
654 
655 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
656 	if (!transport) {
657 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
658 			    trid->trstring);
659 		return -EINVAL;
660 	}
661 
662 	nvmf_listen_opts_copy(&opts_local, opts, opts->opts_size);
663 	rc = spdk_nvmf_transport_listen(transport, trid, &opts_local);
664 	if (rc < 0) {
665 		SPDK_ERRLOG("Unable to listen on address '%s'\n", trid->traddr);
666 	}
667 
668 	return rc;
669 }
670 
671 int
672 spdk_nvmf_tgt_stop_listen(struct spdk_nvmf_tgt *tgt,
673 			  struct spdk_nvme_transport_id *trid)
674 {
675 	struct spdk_nvmf_transport *transport;
676 	int rc;
677 
678 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
679 	if (!transport) {
680 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
681 			    trid->trstring);
682 		return -EINVAL;
683 	}
684 
685 	rc = spdk_nvmf_transport_stop_listen(transport, trid);
686 	if (rc < 0) {
687 		SPDK_ERRLOG("Failed to stop listening on address '%s'\n", trid->traddr);
688 		return rc;
689 	}
690 	return 0;
691 }
692 
693 struct spdk_nvmf_tgt_add_transport_ctx {
694 	struct spdk_nvmf_tgt *tgt;
695 	struct spdk_nvmf_transport *transport;
696 	spdk_nvmf_tgt_add_transport_done_fn cb_fn;
697 	void *cb_arg;
698 };
699 
700 static void
701 _nvmf_tgt_add_transport_done(struct spdk_io_channel_iter *i, int status)
702 {
703 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
704 
705 	ctx->cb_fn(ctx->cb_arg, status);
706 
707 	free(ctx);
708 }
709 
710 static void
711 _nvmf_tgt_add_transport(struct spdk_io_channel_iter *i)
712 {
713 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
714 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
715 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
716 	int rc;
717 
718 	rc = nvmf_poll_group_add_transport(group, ctx->transport);
719 	spdk_for_each_channel_continue(i, rc);
720 }
721 
722 void spdk_nvmf_tgt_add_transport(struct spdk_nvmf_tgt *tgt,
723 				 struct spdk_nvmf_transport *transport,
724 				 spdk_nvmf_tgt_add_transport_done_fn cb_fn,
725 				 void *cb_arg)
726 {
727 	struct spdk_nvmf_tgt_add_transport_ctx *ctx;
728 
729 	if (spdk_nvmf_tgt_get_transport(tgt, transport->ops->name)) {
730 		cb_fn(cb_arg, -EEXIST);
731 		return; /* transport already created */
732 	}
733 
734 	transport->tgt = tgt;
735 	TAILQ_INSERT_TAIL(&tgt->transports, transport, link);
736 
737 	ctx = calloc(1, sizeof(*ctx));
738 	if (!ctx) {
739 		cb_fn(cb_arg, -ENOMEM);
740 		return;
741 	}
742 
743 	ctx->tgt = tgt;
744 	ctx->transport = transport;
745 	ctx->cb_fn = cb_fn;
746 	ctx->cb_arg = cb_arg;
747 
748 	spdk_for_each_channel(tgt,
749 			      _nvmf_tgt_add_transport,
750 			      ctx,
751 			      _nvmf_tgt_add_transport_done);
752 }
753 
754 struct spdk_nvmf_subsystem *
755 spdk_nvmf_tgt_find_subsystem(struct spdk_nvmf_tgt *tgt, const char *subnqn)
756 {
757 	struct spdk_nvmf_subsystem	*subsystem;
758 	uint32_t sid;
759 
760 	if (!subnqn) {
761 		return NULL;
762 	}
763 
764 	/* Ensure that subnqn is null terminated */
765 	if (!memchr(subnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
766 		SPDK_ERRLOG("Connect SUBNQN is not null terminated\n");
767 		return NULL;
768 	}
769 
770 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
771 		subsystem = tgt->subsystems[sid];
772 		if (subsystem == NULL) {
773 			continue;
774 		}
775 
776 		if (strcmp(subnqn, subsystem->subnqn) == 0) {
777 			return subsystem;
778 		}
779 	}
780 
781 	return NULL;
782 }
783 
784 struct spdk_nvmf_transport *
785 spdk_nvmf_tgt_get_transport(struct spdk_nvmf_tgt *tgt, const char *transport_name)
786 {
787 	struct spdk_nvmf_transport *transport;
788 
789 	TAILQ_FOREACH(transport, &tgt->transports, link) {
790 		if (!strncasecmp(transport->ops->name, transport_name, SPDK_NVMF_TRSTRING_MAX_LEN)) {
791 			return transport;
792 		}
793 	}
794 	return NULL;
795 }
796 
797 struct nvmf_new_qpair_ctx {
798 	struct spdk_nvmf_qpair *qpair;
799 	struct spdk_nvmf_poll_group *group;
800 };
801 
802 static void
803 _nvmf_poll_group_add(void *_ctx)
804 {
805 	struct nvmf_new_qpair_ctx *ctx = _ctx;
806 	struct spdk_nvmf_qpair *qpair = ctx->qpair;
807 	struct spdk_nvmf_poll_group *group = ctx->group;
808 
809 	free(_ctx);
810 
811 	if (spdk_nvmf_poll_group_add(group, qpair) != 0) {
812 		SPDK_ERRLOG("Unable to add the qpair to a poll group.\n");
813 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
814 	}
815 }
816 
817 void
818 spdk_nvmf_tgt_new_qpair(struct spdk_nvmf_tgt *tgt, struct spdk_nvmf_qpair *qpair)
819 {
820 	struct spdk_nvmf_poll_group *group;
821 	struct nvmf_new_qpair_ctx *ctx;
822 
823 	group = spdk_nvmf_get_optimal_poll_group(qpair);
824 	if (group == NULL) {
825 		if (tgt->next_poll_group == NULL) {
826 			tgt->next_poll_group = TAILQ_FIRST(&tgt->poll_groups);
827 			if (tgt->next_poll_group == NULL) {
828 				SPDK_ERRLOG("No poll groups exist.\n");
829 				spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
830 				return;
831 			}
832 		}
833 		group = tgt->next_poll_group;
834 		tgt->next_poll_group = TAILQ_NEXT(group, link);
835 	}
836 
837 	ctx = calloc(1, sizeof(*ctx));
838 	if (!ctx) {
839 		SPDK_ERRLOG("Unable to send message to poll group.\n");
840 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
841 		return;
842 	}
843 
844 	ctx->qpair = qpair;
845 	ctx->group = group;
846 
847 	spdk_thread_send_msg(group->thread, _nvmf_poll_group_add, ctx);
848 }
849 
850 struct spdk_nvmf_poll_group *
851 spdk_nvmf_poll_group_create(struct spdk_nvmf_tgt *tgt)
852 {
853 	struct spdk_io_channel *ch;
854 
855 	ch = spdk_get_io_channel(tgt);
856 	if (!ch) {
857 		SPDK_ERRLOG("Unable to get I/O channel for target\n");
858 		return NULL;
859 	}
860 
861 	return spdk_io_channel_get_ctx(ch);
862 }
863 
864 void
865 spdk_nvmf_poll_group_destroy(struct spdk_nvmf_poll_group *group,
866 			     spdk_nvmf_poll_group_destroy_done_fn cb_fn,
867 			     void *cb_arg)
868 {
869 	assert(group->destroy_cb_fn == NULL);
870 	group->destroy_cb_fn = cb_fn;
871 	group->destroy_cb_arg = cb_arg;
872 
873 	/* This function will put the io_channel associated with this poll group */
874 	nvmf_tgt_destroy_poll_group_qpairs(group);
875 }
876 
877 int
878 spdk_nvmf_poll_group_add(struct spdk_nvmf_poll_group *group,
879 			 struct spdk_nvmf_qpair *qpair)
880 {
881 	int rc = -1;
882 	struct spdk_nvmf_transport_poll_group *tgroup;
883 
884 	TAILQ_INIT(&qpair->outstanding);
885 	qpair->group = group;
886 	qpair->ctrlr = NULL;
887 	qpair->disconnect_started = false;
888 
889 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
890 		if (tgroup->transport == qpair->transport) {
891 			rc = nvmf_transport_poll_group_add(tgroup, qpair);
892 			break;
893 		}
894 	}
895 
896 	/* We add the qpair to the group only it is succesfully added into the tgroup */
897 	if (rc == 0) {
898 		TAILQ_INSERT_TAIL(&group->qpairs, qpair, link);
899 		nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ACTIVE);
900 	}
901 
902 	return rc;
903 }
904 
905 static void
906 _nvmf_ctrlr_destruct(void *ctx)
907 {
908 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
909 
910 	nvmf_ctrlr_destruct(ctrlr);
911 }
912 
913 static void
914 _nvmf_transport_qpair_fini_complete(void *cb_ctx)
915 {
916 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = cb_ctx;
917 
918 	if (qpair_ctx->cb_fn) {
919 		spdk_thread_send_msg(qpair_ctx->thread, qpair_ctx->cb_fn, qpair_ctx->ctx);
920 	}
921 	free(qpair_ctx);
922 }
923 
924 static void
925 _nvmf_transport_qpair_fini(void *ctx)
926 {
927 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
928 
929 	nvmf_transport_qpair_fini(qpair_ctx->qpair, _nvmf_transport_qpair_fini_complete, qpair_ctx);
930 }
931 
932 static void
933 _nvmf_ctrlr_free_from_qpair(void *ctx)
934 {
935 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
936 	struct spdk_nvmf_ctrlr *ctrlr = qpair_ctx->ctrlr;
937 	uint32_t count;
938 
939 	spdk_bit_array_clear(ctrlr->qpair_mask, qpair_ctx->qid);
940 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
941 	if (count == 0) {
942 		ctrlr->in_destruct = true;
943 		spdk_thread_send_msg(ctrlr->subsys->thread, _nvmf_ctrlr_destruct, ctrlr);
944 	}
945 
946 	spdk_thread_send_msg(qpair_ctx->thread, _nvmf_transport_qpair_fini, qpair_ctx);
947 }
948 
949 void
950 spdk_nvmf_poll_group_remove(struct spdk_nvmf_qpair *qpair)
951 {
952 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
953 	struct spdk_nvmf_transport_poll_group *tgroup;
954 	struct spdk_nvmf_request *req, *tmp;
955 	struct spdk_nvmf_subsystem_poll_group *sgroup;
956 	int rc;
957 
958 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ERROR);
959 
960 	/* Find the tgroup and remove the qpair from the tgroup */
961 	TAILQ_FOREACH(tgroup, &qpair->group->tgroups, link) {
962 		if (tgroup->transport == qpair->transport) {
963 			rc = nvmf_transport_poll_group_remove(tgroup, qpair);
964 			if (rc && (rc != ENOTSUP)) {
965 				SPDK_ERRLOG("Cannot remove qpair=%p from transport group=%p\n",
966 					    qpair, tgroup);
967 			}
968 			break;
969 		}
970 	}
971 
972 	if (ctrlr) {
973 		sgroup = &qpair->group->sgroups[ctrlr->subsys->id];
974 		TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
975 			if (req->qpair == qpair) {
976 				TAILQ_REMOVE(&sgroup->queued, req, link);
977 				if (nvmf_transport_req_free(req)) {
978 					SPDK_ERRLOG("Transport request free error!\n");
979 				}
980 			}
981 		}
982 	}
983 
984 	TAILQ_REMOVE(&qpair->group->qpairs, qpair, link);
985 	qpair->group = NULL;
986 }
987 
988 static void
989 _nvmf_qpair_destroy(void *ctx, int status)
990 {
991 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
992 	struct spdk_nvmf_qpair *qpair = qpair_ctx->qpair;
993 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
994 
995 	assert(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
996 	qpair_ctx->qid = qpair->qid;
997 
998 	spdk_nvmf_poll_group_remove(qpair);
999 
1000 	if (!ctrlr || !ctrlr->thread) {
1001 		nvmf_transport_qpair_fini(qpair, _nvmf_transport_qpair_fini_complete, qpair_ctx);
1002 		return;
1003 	}
1004 
1005 	qpair_ctx->ctrlr = ctrlr;
1006 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_free_from_qpair, qpair_ctx);
1007 }
1008 
1009 static void
1010 _nvmf_qpair_disconnect_msg(void *ctx)
1011 {
1012 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1013 
1014 	spdk_nvmf_qpair_disconnect(qpair_ctx->qpair, qpair_ctx->cb_fn, qpair_ctx->ctx);
1015 	free(ctx);
1016 }
1017 
1018 int
1019 spdk_nvmf_qpair_disconnect(struct spdk_nvmf_qpair *qpair, nvmf_qpair_disconnect_cb cb_fn, void *ctx)
1020 {
1021 	struct nvmf_qpair_disconnect_ctx *qpair_ctx;
1022 
1023 	if (__atomic_test_and_set(&qpair->disconnect_started, __ATOMIC_RELAXED)) {
1024 		if (cb_fn) {
1025 			cb_fn(ctx);
1026 		}
1027 		return 0;
1028 	}
1029 
1030 	/* If we get a qpair in the uninitialized state, we can just destroy it immediately */
1031 	if (qpair->state == SPDK_NVMF_QPAIR_UNINITIALIZED) {
1032 		nvmf_transport_qpair_fini(qpair, NULL, NULL);
1033 		if (cb_fn) {
1034 			cb_fn(ctx);
1035 		}
1036 		return 0;
1037 	}
1038 
1039 	assert(qpair->group != NULL);
1040 	if (spdk_get_thread() != qpair->group->thread) {
1041 		/* clear the atomic so we can set it on the next call on the proper thread. */
1042 		__atomic_clear(&qpair->disconnect_started, __ATOMIC_RELAXED);
1043 		qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1044 		if (!qpair_ctx) {
1045 			SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1046 			return -ENOMEM;
1047 		}
1048 		qpair_ctx->qpair = qpair;
1049 		qpair_ctx->cb_fn = cb_fn;
1050 		qpair_ctx->thread = qpair->group->thread;
1051 		qpair_ctx->ctx = ctx;
1052 		spdk_thread_send_msg(qpair->group->thread, _nvmf_qpair_disconnect_msg, qpair_ctx);
1053 		return 0;
1054 	}
1055 
1056 	assert(qpair->state == SPDK_NVMF_QPAIR_ACTIVE);
1057 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_DEACTIVATING);
1058 
1059 	qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1060 	if (!qpair_ctx) {
1061 		SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1062 		return -ENOMEM;
1063 	}
1064 
1065 	qpair_ctx->qpair = qpair;
1066 	qpair_ctx->cb_fn = cb_fn;
1067 	qpair_ctx->thread = qpair->group->thread;
1068 	qpair_ctx->ctx = ctx;
1069 
1070 	/* Check for outstanding I/O */
1071 	if (!TAILQ_EMPTY(&qpair->outstanding)) {
1072 		qpair->state_cb = _nvmf_qpair_destroy;
1073 		qpair->state_cb_arg = qpair_ctx;
1074 		nvmf_qpair_free_aer(qpair);
1075 		return 0;
1076 	}
1077 
1078 	_nvmf_qpair_destroy(qpair_ctx, 0);
1079 
1080 	return 0;
1081 }
1082 
1083 int
1084 spdk_nvmf_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
1085 			      struct spdk_nvme_transport_id *trid)
1086 {
1087 	return nvmf_transport_qpair_get_peer_trid(qpair, trid);
1088 }
1089 
1090 int
1091 spdk_nvmf_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
1092 			       struct spdk_nvme_transport_id *trid)
1093 {
1094 	return nvmf_transport_qpair_get_local_trid(qpair, trid);
1095 }
1096 
1097 int
1098 spdk_nvmf_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
1099 				struct spdk_nvme_transport_id *trid)
1100 {
1101 	return nvmf_transport_qpair_get_listen_trid(qpair, trid);
1102 }
1103 
1104 int
1105 nvmf_poll_group_add_transport(struct spdk_nvmf_poll_group *group,
1106 			      struct spdk_nvmf_transport *transport)
1107 {
1108 	struct spdk_nvmf_transport_poll_group *tgroup;
1109 
1110 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1111 		if (tgroup->transport == transport) {
1112 			/* Transport already in the poll group */
1113 			return 0;
1114 		}
1115 	}
1116 
1117 	tgroup = nvmf_transport_poll_group_create(transport);
1118 	if (!tgroup) {
1119 		SPDK_ERRLOG("Unable to create poll group for transport\n");
1120 		return -1;
1121 	}
1122 
1123 	tgroup->group = group;
1124 	TAILQ_INSERT_TAIL(&group->tgroups, tgroup, link);
1125 
1126 	return 0;
1127 }
1128 
1129 static int
1130 poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1131 			    struct spdk_nvmf_subsystem *subsystem)
1132 {
1133 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1134 	uint32_t new_num_ns, old_num_ns;
1135 	uint32_t i, j;
1136 	struct spdk_nvmf_ns *ns;
1137 	struct spdk_nvmf_registrant *reg, *tmp;
1138 	struct spdk_io_channel *ch;
1139 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1140 	struct spdk_nvmf_ctrlr *ctrlr;
1141 	bool ns_changed;
1142 
1143 	/* Make sure our poll group has memory for this subsystem allocated */
1144 	if (subsystem->id >= group->num_sgroups) {
1145 		return -ENOMEM;
1146 	}
1147 
1148 	sgroup = &group->sgroups[subsystem->id];
1149 
1150 	/* Make sure the array of namespace information is the correct size */
1151 	new_num_ns = subsystem->max_nsid;
1152 	old_num_ns = sgroup->num_ns;
1153 
1154 	ns_changed = false;
1155 
1156 	if (old_num_ns == 0) {
1157 		if (new_num_ns > 0) {
1158 			/* First allocation */
1159 			sgroup->ns_info = calloc(new_num_ns, sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1160 			if (!sgroup->ns_info) {
1161 				return -ENOMEM;
1162 			}
1163 		}
1164 	} else if (new_num_ns > old_num_ns) {
1165 		void *buf;
1166 
1167 		/* Make the array larger */
1168 		buf = realloc(sgroup->ns_info, new_num_ns * sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1169 		if (!buf) {
1170 			return -ENOMEM;
1171 		}
1172 
1173 		sgroup->ns_info = buf;
1174 
1175 		/* Null out the new namespace information slots */
1176 		for (i = old_num_ns; i < new_num_ns; i++) {
1177 			memset(&sgroup->ns_info[i], 0, sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1178 		}
1179 	} else if (new_num_ns < old_num_ns) {
1180 		void *buf;
1181 
1182 		/* Free the extra I/O channels */
1183 		for (i = new_num_ns; i < old_num_ns; i++) {
1184 			ns_info = &sgroup->ns_info[i];
1185 
1186 			if (ns_info->channel) {
1187 				spdk_put_io_channel(ns_info->channel);
1188 				ns_info->channel = NULL;
1189 			}
1190 		}
1191 
1192 		/* Make the array smaller */
1193 		if (new_num_ns > 0) {
1194 			buf = realloc(sgroup->ns_info, new_num_ns * sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1195 			if (!buf) {
1196 				return -ENOMEM;
1197 			}
1198 			sgroup->ns_info = buf;
1199 		} else {
1200 			free(sgroup->ns_info);
1201 			sgroup->ns_info = NULL;
1202 		}
1203 	}
1204 
1205 	sgroup->num_ns = new_num_ns;
1206 
1207 	/* Detect bdevs that were added or removed */
1208 	for (i = 0; i < sgroup->num_ns; i++) {
1209 		ns = subsystem->ns[i];
1210 		ns_info = &sgroup->ns_info[i];
1211 		ch = ns_info->channel;
1212 
1213 		if (ns == NULL && ch == NULL) {
1214 			/* Both NULL. Leave empty */
1215 		} else if (ns == NULL && ch != NULL) {
1216 			/* There was a channel here, but the namespace is gone. */
1217 			ns_changed = true;
1218 			spdk_put_io_channel(ch);
1219 			ns_info->channel = NULL;
1220 		} else if (ns != NULL && ch == NULL) {
1221 			/* A namespace appeared but there is no channel yet */
1222 			ns_changed = true;
1223 			ch = spdk_bdev_get_io_channel(ns->desc);
1224 			if (ch == NULL) {
1225 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
1226 				return -ENOMEM;
1227 			}
1228 			ns_info->channel = ch;
1229 		} else if (spdk_uuid_compare(&ns_info->uuid, spdk_bdev_get_uuid(ns->bdev)) != 0) {
1230 			/* A namespace was here before, but was replaced by a new one. */
1231 			ns_changed = true;
1232 			spdk_put_io_channel(ns_info->channel);
1233 			memset(ns_info, 0, sizeof(*ns_info));
1234 
1235 			ch = spdk_bdev_get_io_channel(ns->desc);
1236 			if (ch == NULL) {
1237 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
1238 				return -ENOMEM;
1239 			}
1240 			ns_info->channel = ch;
1241 		} else if (ns_info->num_blocks != spdk_bdev_get_num_blocks(ns->bdev)) {
1242 			/* Namespace is still there but size has changed */
1243 			SPDK_DEBUGLOG(nvmf, "Namespace resized: subsystem_id %u,"
1244 				      " nsid %u, pg %p, old %" PRIu64 ", new %" PRIu64 "\n",
1245 				      subsystem->id,
1246 				      ns->nsid,
1247 				      group,
1248 				      ns_info->num_blocks,
1249 				      spdk_bdev_get_num_blocks(ns->bdev));
1250 			ns_changed = true;
1251 		}
1252 
1253 		if (ns == NULL) {
1254 			memset(ns_info, 0, sizeof(*ns_info));
1255 		} else {
1256 			ns_info->uuid = *spdk_bdev_get_uuid(ns->bdev);
1257 			ns_info->num_blocks = spdk_bdev_get_num_blocks(ns->bdev);
1258 			ns_info->crkey = ns->crkey;
1259 			ns_info->rtype = ns->rtype;
1260 			if (ns->holder) {
1261 				ns_info->holder_id = ns->holder->hostid;
1262 			}
1263 
1264 			memset(&ns_info->reg_hostid, 0, SPDK_NVMF_MAX_NUM_REGISTRANTS * sizeof(struct spdk_uuid));
1265 			j = 0;
1266 			TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
1267 				if (j >= SPDK_NVMF_MAX_NUM_REGISTRANTS) {
1268 					SPDK_ERRLOG("Maximum %u registrants can support.\n", SPDK_NVMF_MAX_NUM_REGISTRANTS);
1269 					return -EINVAL;
1270 				}
1271 				ns_info->reg_hostid[j++] = reg->hostid;
1272 			}
1273 		}
1274 	}
1275 
1276 	if (ns_changed) {
1277 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1278 			if (ctrlr->admin_qpair->group == group) {
1279 				nvmf_ctrlr_async_event_ns_notice(ctrlr);
1280 				nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
1281 			}
1282 		}
1283 	}
1284 
1285 	return 0;
1286 }
1287 
1288 int
1289 nvmf_poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1290 				 struct spdk_nvmf_subsystem *subsystem)
1291 {
1292 	return poll_group_update_subsystem(group, subsystem);
1293 }
1294 
1295 int
1296 nvmf_poll_group_add_subsystem(struct spdk_nvmf_poll_group *group,
1297 			      struct spdk_nvmf_subsystem *subsystem,
1298 			      spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1299 {
1300 	int rc = 0;
1301 	struct spdk_nvmf_subsystem_poll_group *sgroup = &group->sgroups[subsystem->id];
1302 	uint32_t i;
1303 
1304 	TAILQ_INIT(&sgroup->queued);
1305 
1306 	rc = poll_group_update_subsystem(group, subsystem);
1307 	if (rc) {
1308 		nvmf_poll_group_remove_subsystem(group, subsystem, NULL, NULL);
1309 		goto fini;
1310 	}
1311 
1312 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1313 
1314 	for (i = 0; i < sgroup->num_ns; i++) {
1315 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1316 	}
1317 
1318 fini:
1319 	if (cb_fn) {
1320 		cb_fn(cb_arg, rc);
1321 	}
1322 
1323 	return rc;
1324 }
1325 
1326 static void
1327 _nvmf_poll_group_remove_subsystem_cb(void *ctx, int status)
1328 {
1329 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1330 	struct spdk_nvmf_subsystem *subsystem;
1331 	struct spdk_nvmf_poll_group *group;
1332 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1333 	spdk_nvmf_poll_group_mod_done cpl_fn = NULL;
1334 	void *cpl_ctx = NULL;
1335 	uint32_t nsid;
1336 
1337 	group = qpair_ctx->group;
1338 	subsystem = qpair_ctx->subsystem;
1339 	cpl_fn = qpair_ctx->cpl_fn;
1340 	cpl_ctx = qpair_ctx->cpl_ctx;
1341 	sgroup = &group->sgroups[subsystem->id];
1342 
1343 	if (status) {
1344 		goto fini;
1345 	}
1346 
1347 	for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
1348 		if (sgroup->ns_info[nsid].channel) {
1349 			spdk_put_io_channel(sgroup->ns_info[nsid].channel);
1350 			sgroup->ns_info[nsid].channel = NULL;
1351 		}
1352 	}
1353 
1354 	sgroup->num_ns = 0;
1355 	free(sgroup->ns_info);
1356 	sgroup->ns_info = NULL;
1357 fini:
1358 	free(qpair_ctx);
1359 	if (cpl_fn) {
1360 		cpl_fn(cpl_ctx, status);
1361 	}
1362 }
1363 
1364 static void
1365 remove_subsystem_qpair_cb(void *ctx)
1366 {
1367 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1368 
1369 	assert(qpair_ctx->count > 0);
1370 	qpair_ctx->count--;
1371 	if (qpair_ctx->count == 0) {
1372 		_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
1373 	}
1374 }
1375 
1376 static void
1377 nvmf_poll_group_remove_subsystem_msg(void *ctx)
1378 {
1379 	struct spdk_nvmf_qpair *qpair, *qpair_tmp;
1380 	struct spdk_nvmf_subsystem *subsystem;
1381 	struct spdk_nvmf_poll_group *group;
1382 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1383 	int rc = 0;
1384 
1385 	group = qpair_ctx->group;
1386 	subsystem = qpair_ctx->subsystem;
1387 
1388 	/* Initialize count to 1.  This acts like a ref count, to ensure that if spdk_nvmf_qpair_disconnect
1389 	 * immediately invokes the callback (i.e. the qpairs is already in process of being disconnected)
1390 	 * that we don't prematurely call _nvmf_poll_group_remove_subsystem_cb() before we've
1391 	 * iterated the full list of qpairs.
1392 	 */
1393 	qpair_ctx->count = 1;
1394 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, qpair_tmp) {
1395 		if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
1396 			qpair_ctx->count++;
1397 			rc = spdk_nvmf_qpair_disconnect(qpair, remove_subsystem_qpair_cb, ctx);
1398 			if (rc) {
1399 				break;
1400 			}
1401 		}
1402 	}
1403 	qpair_ctx->count--;
1404 
1405 	if (qpair_ctx->count == 0 || rc) {
1406 		_nvmf_poll_group_remove_subsystem_cb(ctx, rc);
1407 	}
1408 }
1409 
1410 void
1411 nvmf_poll_group_remove_subsystem(struct spdk_nvmf_poll_group *group,
1412 				 struct spdk_nvmf_subsystem *subsystem,
1413 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1414 {
1415 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1416 	struct nvmf_qpair_disconnect_many_ctx *ctx;
1417 	uint32_t i;
1418 
1419 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
1420 	if (!ctx) {
1421 		SPDK_ERRLOG("Unable to allocate memory for context to remove poll subsystem\n");
1422 		if (cb_fn) {
1423 			cb_fn(cb_arg, -1);
1424 		}
1425 		return;
1426 	}
1427 
1428 	ctx->group = group;
1429 	ctx->subsystem = subsystem;
1430 	ctx->cpl_fn = cb_fn;
1431 	ctx->cpl_ctx = cb_arg;
1432 
1433 	sgroup = &group->sgroups[subsystem->id];
1434 	sgroup->state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1435 
1436 	for (i = 0; i < sgroup->num_ns; i++) {
1437 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1438 	}
1439 
1440 	nvmf_poll_group_remove_subsystem_msg(ctx);
1441 }
1442 
1443 void
1444 nvmf_poll_group_pause_subsystem(struct spdk_nvmf_poll_group *group,
1445 				struct spdk_nvmf_subsystem *subsystem,
1446 				uint32_t nsid,
1447 				spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1448 {
1449 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1450 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info = NULL;
1451 	int rc = 0;
1452 
1453 	if (subsystem->id >= group->num_sgroups) {
1454 		rc = -1;
1455 		goto fini;
1456 	}
1457 
1458 	sgroup = &group->sgroups[subsystem->id];
1459 	if (sgroup == NULL) {
1460 		rc = -1;
1461 		goto fini;
1462 	}
1463 
1464 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSED) {
1465 		goto fini;
1466 	}
1467 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1468 
1469 	/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
1470 	if (nsid - 1 < sgroup->num_ns) {
1471 		ns_info  = &sgroup->ns_info[nsid - 1];
1472 		ns_info->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1473 	}
1474 
1475 	if (sgroup->mgmt_io_outstanding > 0) {
1476 		assert(sgroup->cb_fn == NULL);
1477 		sgroup->cb_fn = cb_fn;
1478 		assert(sgroup->cb_arg == NULL);
1479 		sgroup->cb_arg = cb_arg;
1480 		return;
1481 	}
1482 
1483 	if (ns_info != NULL && ns_info->io_outstanding > 0) {
1484 		assert(sgroup->cb_fn == NULL);
1485 		sgroup->cb_fn = cb_fn;
1486 		assert(sgroup->cb_arg == NULL);
1487 		sgroup->cb_arg = cb_arg;
1488 		return;
1489 	}
1490 
1491 	assert(sgroup->mgmt_io_outstanding == 0);
1492 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
1493 fini:
1494 	if (cb_fn) {
1495 		cb_fn(cb_arg, rc);
1496 	}
1497 }
1498 
1499 void
1500 nvmf_poll_group_resume_subsystem(struct spdk_nvmf_poll_group *group,
1501 				 struct spdk_nvmf_subsystem *subsystem,
1502 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1503 {
1504 	struct spdk_nvmf_request *req, *tmp;
1505 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1506 	int rc = 0;
1507 	uint32_t i;
1508 
1509 	if (subsystem->id >= group->num_sgroups) {
1510 		rc = -1;
1511 		goto fini;
1512 	}
1513 
1514 	sgroup = &group->sgroups[subsystem->id];
1515 
1516 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_ACTIVE) {
1517 		goto fini;
1518 	}
1519 
1520 	rc = poll_group_update_subsystem(group, subsystem);
1521 	if (rc) {
1522 		goto fini;
1523 	}
1524 
1525 	for (i = 0; i < sgroup->num_ns; i++) {
1526 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1527 	}
1528 
1529 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1530 
1531 	/* Release all queued requests */
1532 	TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1533 		TAILQ_REMOVE(&sgroup->queued, req, link);
1534 		spdk_nvmf_request_exec(req);
1535 	}
1536 fini:
1537 	if (cb_fn) {
1538 		cb_fn(cb_arg, rc);
1539 	}
1540 }
1541 
1542 
1543 struct spdk_nvmf_poll_group *
1544 spdk_nvmf_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
1545 {
1546 	struct spdk_nvmf_transport_poll_group *tgroup;
1547 
1548 	tgroup = nvmf_transport_get_optimal_poll_group(qpair->transport, qpair);
1549 
1550 	if (tgroup == NULL) {
1551 		return NULL;
1552 	}
1553 
1554 	return tgroup->group;
1555 }
1556 
1557 int
1558 spdk_nvmf_poll_group_get_stat(struct spdk_nvmf_tgt *tgt,
1559 			      struct spdk_nvmf_poll_group_stat *stat)
1560 {
1561 	struct spdk_io_channel *ch;
1562 	struct spdk_nvmf_poll_group *group;
1563 
1564 	SPDK_ERRLOG("spdk_nvmf_poll_group_get_stat is deprecated and will be removed\n");
1565 
1566 	if (tgt == NULL || stat == NULL) {
1567 		return -EINVAL;
1568 	}
1569 
1570 	ch = spdk_get_io_channel(tgt);
1571 	group = spdk_io_channel_get_ctx(ch);
1572 	*stat = group->stat;
1573 	spdk_put_io_channel(ch);
1574 	return 0;
1575 }
1576 
1577 void
1578 spdk_nvmf_poll_group_dump_stat(struct spdk_nvmf_poll_group *group, struct spdk_json_write_ctx *w)
1579 {
1580 	struct spdk_nvmf_transport_poll_group *tgroup;
1581 
1582 	spdk_json_write_object_begin(w);
1583 
1584 	spdk_json_write_named_string(w, "name", spdk_thread_get_name(spdk_get_thread()));
1585 	spdk_json_write_named_uint32(w, "admin_qpairs", group->stat.admin_qpairs);
1586 	spdk_json_write_named_uint32(w, "io_qpairs", group->stat.io_qpairs);
1587 	spdk_json_write_named_uint64(w, "pending_bdev_io", group->stat.pending_bdev_io);
1588 
1589 	spdk_json_write_named_array_begin(w, "transports");
1590 
1591 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1592 		spdk_json_write_object_begin(w);
1593 		/*
1594 		 * The trtype field intentionally contains a transport name as this is more informative.
1595 		 * The field has not been renamed for backward compatibility.
1596 		 */
1597 		spdk_json_write_named_string(w, "trtype", spdk_nvmf_get_transport_name(tgroup->transport));
1598 
1599 		if (tgroup->transport->ops->poll_group_dump_stat) {
1600 			tgroup->transport->ops->poll_group_dump_stat(tgroup, w);
1601 		}
1602 
1603 		spdk_json_write_object_end(w);
1604 	}
1605 
1606 	spdk_json_write_array_end(w);
1607 	spdk_json_write_object_end(w);
1608 }
1609