xref: /spdk/lib/nvmf/nvmf.c (revision 510f4c134a21b45ff3a5add9ebc6c6cf7e49aeab)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (c) Intel Corporation. All rights reserved.
3  *   Copyright (c) 2018-2019, 2021 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 "spdk/bdev.h"
10 #include "spdk/bit_array.h"
11 #include "spdk/thread.h"
12 #include "spdk/nvmf.h"
13 #include "spdk/endian.h"
14 #include "spdk/string.h"
15 #include "spdk/log.h"
16 #include "spdk_internal/usdt.h"
17 
18 #include "nvmf_internal.h"
19 #include "transport.h"
20 
21 SPDK_LOG_REGISTER_COMPONENT(nvmf)
22 
23 #define SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS 1024
24 
25 static TAILQ_HEAD(, spdk_nvmf_tgt) g_nvmf_tgts = TAILQ_HEAD_INITIALIZER(g_nvmf_tgts);
26 
27 typedef void (*nvmf_qpair_disconnect_cpl)(void *ctx, int status);
28 
29 /* supplied to a single call to nvmf_qpair_disconnect */
30 struct nvmf_qpair_disconnect_ctx {
31 	struct spdk_nvmf_qpair *qpair;
32 	struct spdk_nvmf_ctrlr *ctrlr;
33 	nvmf_qpair_disconnect_cb cb_fn;
34 	struct spdk_thread *thread;
35 	void *ctx;
36 	uint16_t qid;
37 };
38 
39 /*
40  * There are several times when we need to iterate through the list of all qpairs and selectively delete them.
41  * In order to do this sequentially without overlap, we must provide a context to recover the next qpair from
42  * to enable calling nvmf_qpair_disconnect on the next desired qpair.
43  */
44 struct nvmf_qpair_disconnect_many_ctx {
45 	struct spdk_nvmf_subsystem *subsystem;
46 	struct spdk_nvmf_poll_group *group;
47 	spdk_nvmf_poll_group_mod_done cpl_fn;
48 	void *cpl_ctx;
49 	uint32_t count;
50 };
51 
52 static void
53 nvmf_qpair_set_state(struct spdk_nvmf_qpair *qpair,
54 		     enum spdk_nvmf_qpair_state state)
55 {
56 	assert(qpair != NULL);
57 	assert(qpair->group->thread == spdk_get_thread());
58 
59 	qpair->state = state;
60 }
61 
62 static int
63 nvmf_poll_group_poll(void *ctx)
64 {
65 	struct spdk_nvmf_poll_group *group = ctx;
66 	int rc;
67 	int count = 0;
68 	struct spdk_nvmf_transport_poll_group *tgroup;
69 
70 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
71 		rc = nvmf_transport_poll_group_poll(tgroup);
72 		if (rc < 0) {
73 			return SPDK_POLLER_BUSY;
74 		}
75 		count += rc;
76 	}
77 
78 	return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
79 }
80 
81 /*
82  * Reset and clean up the poll group (I/O channel code will actually free the
83  * group).
84  */
85 static void
86 nvmf_tgt_cleanup_poll_group(struct spdk_nvmf_poll_group *group)
87 {
88 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
89 	struct spdk_nvmf_subsystem_poll_group *sgroup;
90 	uint32_t sid, nsid;
91 
92 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
93 		TAILQ_REMOVE(&group->tgroups, tgroup, link);
94 		nvmf_transport_poll_group_destroy(tgroup);
95 	}
96 
97 	for (sid = 0; sid < group->num_sgroups; sid++) {
98 		sgroup = &group->sgroups[sid];
99 
100 		assert(sgroup != NULL);
101 
102 		for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
103 			if (sgroup->ns_info[nsid].channel) {
104 				spdk_put_io_channel(sgroup->ns_info[nsid].channel);
105 				sgroup->ns_info[nsid].channel = NULL;
106 			}
107 		}
108 
109 		free(sgroup->ns_info);
110 	}
111 
112 	free(group->sgroups);
113 
114 	spdk_poller_unregister(&group->poller);
115 
116 	if (group->destroy_cb_fn) {
117 		group->destroy_cb_fn(group->destroy_cb_arg, 0);
118 	}
119 }
120 
121 /*
122  * Callback to unregister a poll group from the target, and clean up its state.
123  */
124 static void
125 nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf)
126 {
127 	struct spdk_nvmf_tgt *tgt = io_device;
128 	struct spdk_nvmf_poll_group *group = ctx_buf;
129 
130 	SPDK_DTRACE_PROBE1(nvmf_destroy_poll_group, spdk_thread_get_id(group->thread));
131 
132 	pthread_mutex_lock(&tgt->mutex);
133 	TAILQ_REMOVE(&tgt->poll_groups, group, link);
134 	pthread_mutex_unlock(&tgt->mutex);
135 
136 	nvmf_tgt_cleanup_poll_group(group);
137 }
138 
139 static int
140 nvmf_poll_group_add_transport(struct spdk_nvmf_poll_group *group,
141 			      struct spdk_nvmf_transport *transport)
142 {
143 	struct spdk_nvmf_transport_poll_group *tgroup;
144 
145 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
146 		if (tgroup->transport == transport) {
147 			/* Transport already in the poll group */
148 			return 0;
149 		}
150 	}
151 
152 	tgroup = nvmf_transport_poll_group_create(transport, group);
153 	if (!tgroup) {
154 		SPDK_ERRLOG("Unable to create poll group for transport\n");
155 		return -1;
156 	}
157 	SPDK_DTRACE_PROBE2(nvmf_transport_poll_group_create, transport, spdk_thread_get_id(group->thread));
158 
159 	tgroup->group = group;
160 	TAILQ_INSERT_TAIL(&group->tgroups, tgroup, link);
161 
162 	return 0;
163 }
164 
165 static int
166 nvmf_tgt_create_poll_group(void *io_device, void *ctx_buf)
167 {
168 	struct spdk_nvmf_tgt *tgt = io_device;
169 	struct spdk_nvmf_poll_group *group = ctx_buf;
170 	struct spdk_nvmf_transport *transport;
171 	struct spdk_thread *thread = spdk_get_thread();
172 	uint32_t sid;
173 	int rc;
174 
175 	TAILQ_INIT(&group->tgroups);
176 	TAILQ_INIT(&group->qpairs);
177 	group->thread = thread;
178 
179 	group->poller = SPDK_POLLER_REGISTER(nvmf_poll_group_poll, group, 0);
180 
181 	SPDK_DTRACE_PROBE1(nvmf_create_poll_group, spdk_thread_get_id(thread));
182 
183 	TAILQ_FOREACH(transport, &tgt->transports, link) {
184 		rc = nvmf_poll_group_add_transport(group, transport);
185 		if (rc != 0) {
186 			nvmf_tgt_cleanup_poll_group(group);
187 			return rc;
188 		}
189 	}
190 
191 	group->num_sgroups = tgt->max_subsystems;
192 	group->sgroups = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem_poll_group));
193 	if (!group->sgroups) {
194 		nvmf_tgt_cleanup_poll_group(group);
195 		return -ENOMEM;
196 	}
197 
198 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
199 		struct spdk_nvmf_subsystem *subsystem;
200 
201 		subsystem = tgt->subsystems[sid];
202 		if (!subsystem) {
203 			continue;
204 		}
205 
206 		if (nvmf_poll_group_add_subsystem(group, subsystem, NULL, NULL) != 0) {
207 			nvmf_tgt_cleanup_poll_group(group);
208 			return -1;
209 		}
210 	}
211 
212 	pthread_mutex_lock(&tgt->mutex);
213 	TAILQ_INSERT_TAIL(&tgt->poll_groups, group, link);
214 	pthread_mutex_unlock(&tgt->mutex);
215 
216 	return 0;
217 }
218 
219 static void
220 _nvmf_tgt_disconnect_next_qpair(void *ctx)
221 {
222 	struct spdk_nvmf_qpair *qpair;
223 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
224 	struct spdk_nvmf_poll_group *group = qpair_ctx->group;
225 	struct spdk_io_channel *ch;
226 	int rc = 0;
227 
228 	qpair = TAILQ_FIRST(&group->qpairs);
229 
230 	if (qpair) {
231 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_tgt_disconnect_next_qpair, ctx);
232 	}
233 
234 	if (!qpair || rc != 0) {
235 		/* When the refcount from the channels reaches 0, nvmf_tgt_destroy_poll_group will be called. */
236 		ch = spdk_io_channel_from_ctx(group);
237 		spdk_put_io_channel(ch);
238 		free(qpair_ctx);
239 	}
240 }
241 
242 static void
243 nvmf_tgt_destroy_poll_group_qpairs(struct spdk_nvmf_poll_group *group)
244 {
245 	struct nvmf_qpair_disconnect_many_ctx *ctx;
246 
247 	SPDK_DTRACE_PROBE1(nvmf_destroy_poll_group_qpairs, spdk_thread_get_id(group->thread));
248 
249 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
250 	if (!ctx) {
251 		SPDK_ERRLOG("Failed to allocate memory for destroy poll group ctx\n");
252 		return;
253 	}
254 
255 	ctx->group = group;
256 	_nvmf_tgt_disconnect_next_qpair(ctx);
257 }
258 
259 struct spdk_nvmf_tgt *
260 spdk_nvmf_tgt_create(struct spdk_nvmf_target_opts *opts)
261 {
262 	struct spdk_nvmf_tgt *tgt, *tmp_tgt;
263 
264 	if (strnlen(opts->name, NVMF_TGT_NAME_MAX_LENGTH) == NVMF_TGT_NAME_MAX_LENGTH) {
265 		SPDK_ERRLOG("Provided target name exceeds the max length of %u.\n", NVMF_TGT_NAME_MAX_LENGTH);
266 		return NULL;
267 	}
268 
269 	TAILQ_FOREACH(tmp_tgt, &g_nvmf_tgts, link) {
270 		if (!strncmp(opts->name, tmp_tgt->name, NVMF_TGT_NAME_MAX_LENGTH)) {
271 			SPDK_ERRLOG("Provided target name must be unique.\n");
272 			return NULL;
273 		}
274 	}
275 
276 	tgt = calloc(1, sizeof(*tgt));
277 	if (!tgt) {
278 		return NULL;
279 	}
280 
281 	snprintf(tgt->name, NVMF_TGT_NAME_MAX_LENGTH, "%s", opts->name);
282 
283 	if (!opts || !opts->max_subsystems) {
284 		tgt->max_subsystems = SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS;
285 	} else {
286 		tgt->max_subsystems = opts->max_subsystems;
287 	}
288 
289 	if (!opts) {
290 		tgt->crdt[0] = 0;
291 		tgt->crdt[1] = 0;
292 		tgt->crdt[2] = 0;
293 	} else {
294 		tgt->crdt[0] = opts->crdt[0];
295 		tgt->crdt[1] = opts->crdt[1];
296 		tgt->crdt[2] = opts->crdt[2];
297 	}
298 
299 	if (!opts) {
300 		tgt->discovery_filter = SPDK_NVMF_TGT_DISCOVERY_MATCH_ANY;
301 	} else {
302 		tgt->discovery_filter = opts->discovery_filter;
303 	}
304 
305 	tgt->discovery_genctr = 0;
306 	TAILQ_INIT(&tgt->transports);
307 	TAILQ_INIT(&tgt->poll_groups);
308 
309 	tgt->subsystems = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem *));
310 	if (!tgt->subsystems) {
311 		free(tgt);
312 		return NULL;
313 	}
314 
315 	pthread_mutex_init(&tgt->mutex, NULL);
316 
317 	spdk_io_device_register(tgt,
318 				nvmf_tgt_create_poll_group,
319 				nvmf_tgt_destroy_poll_group,
320 				sizeof(struct spdk_nvmf_poll_group),
321 				tgt->name);
322 
323 	TAILQ_INSERT_HEAD(&g_nvmf_tgts, tgt, link);
324 
325 	return tgt;
326 }
327 
328 static void
329 _nvmf_tgt_destroy_next_transport(void *ctx)
330 {
331 	struct spdk_nvmf_tgt *tgt = ctx;
332 	struct spdk_nvmf_transport *transport;
333 
334 	if (!TAILQ_EMPTY(&tgt->transports)) {
335 		transport = TAILQ_FIRST(&tgt->transports);
336 		TAILQ_REMOVE(&tgt->transports, transport, link);
337 		spdk_nvmf_transport_destroy(transport, _nvmf_tgt_destroy_next_transport, tgt);
338 	} else {
339 		spdk_nvmf_tgt_destroy_done_fn *destroy_cb_fn = tgt->destroy_cb_fn;
340 		void *destroy_cb_arg = tgt->destroy_cb_arg;
341 
342 		pthread_mutex_destroy(&tgt->mutex);
343 		free(tgt);
344 
345 		if (destroy_cb_fn) {
346 			destroy_cb_fn(destroy_cb_arg, 0);
347 		}
348 	}
349 }
350 
351 static void
352 nvmf_tgt_destroy_cb(void *io_device)
353 {
354 	struct spdk_nvmf_tgt *tgt = io_device;
355 	uint32_t i;
356 	int rc;
357 
358 	if (tgt->subsystems) {
359 		for (i = 0; i < tgt->max_subsystems; i++) {
360 			if (tgt->subsystems[i]) {
361 				nvmf_subsystem_remove_all_listeners(tgt->subsystems[i], true);
362 
363 				rc = spdk_nvmf_subsystem_destroy(tgt->subsystems[i], nvmf_tgt_destroy_cb, tgt);
364 				if (rc) {
365 					if (rc == -EINPROGRESS) {
366 						/* If rc is -EINPROGRESS, nvmf_tgt_destroy_cb will be called again when subsystem #i
367 						 * is destroyed, nvmf_tgt_destroy_cb will continue to destroy other subsystems if any */
368 						return;
369 					} else {
370 						SPDK_ERRLOG("Failed to destroy subsystem %s, rc %d\n", tgt->subsystems[i]->subnqn, rc);
371 					}
372 				}
373 			}
374 		}
375 		free(tgt->subsystems);
376 	}
377 
378 	_nvmf_tgt_destroy_next_transport(tgt);
379 }
380 
381 void
382 spdk_nvmf_tgt_destroy(struct spdk_nvmf_tgt *tgt,
383 		      spdk_nvmf_tgt_destroy_done_fn cb_fn,
384 		      void *cb_arg)
385 {
386 	tgt->destroy_cb_fn = cb_fn;
387 	tgt->destroy_cb_arg = cb_arg;
388 
389 	TAILQ_REMOVE(&g_nvmf_tgts, tgt, link);
390 
391 	spdk_io_device_unregister(tgt, nvmf_tgt_destroy_cb);
392 }
393 
394 const char *
395 spdk_nvmf_tgt_get_name(struct spdk_nvmf_tgt *tgt)
396 {
397 	return tgt->name;
398 }
399 
400 struct spdk_nvmf_tgt *
401 spdk_nvmf_get_tgt(const char *name)
402 {
403 	struct spdk_nvmf_tgt *tgt;
404 	uint32_t num_targets = 0;
405 
406 	TAILQ_FOREACH(tgt, &g_nvmf_tgts, link) {
407 		if (name) {
408 			if (!strncmp(tgt->name, name, NVMF_TGT_NAME_MAX_LENGTH)) {
409 				return tgt;
410 			}
411 		}
412 		num_targets++;
413 	}
414 
415 	/*
416 	 * special case. If there is only one target and
417 	 * no name was specified, return the only available
418 	 * target. If there is more than one target, name must
419 	 * be specified.
420 	 */
421 	if (!name && num_targets == 1) {
422 		return TAILQ_FIRST(&g_nvmf_tgts);
423 	}
424 
425 	return NULL;
426 }
427 
428 struct spdk_nvmf_tgt *
429 spdk_nvmf_get_first_tgt(void)
430 {
431 	return TAILQ_FIRST(&g_nvmf_tgts);
432 }
433 
434 struct spdk_nvmf_tgt *
435 spdk_nvmf_get_next_tgt(struct spdk_nvmf_tgt *prev)
436 {
437 	return TAILQ_NEXT(prev, link);
438 }
439 
440 static void
441 nvmf_write_subsystem_config_json(struct spdk_json_write_ctx *w,
442 				 struct spdk_nvmf_subsystem *subsystem)
443 {
444 	struct spdk_nvmf_host *host;
445 	struct spdk_nvmf_subsystem_listener *listener;
446 	const struct spdk_nvme_transport_id *trid;
447 	struct spdk_nvmf_ns *ns;
448 	struct spdk_nvmf_ns_opts ns_opts;
449 	uint32_t max_namespaces;
450 	char uuid_str[SPDK_UUID_STRING_LEN];
451 
452 	if (spdk_nvmf_subsystem_get_type(subsystem) != SPDK_NVMF_SUBTYPE_NVME) {
453 		return;
454 	}
455 
456 	/* { */
457 	spdk_json_write_object_begin(w);
458 	spdk_json_write_named_string(w, "method", "nvmf_create_subsystem");
459 
460 	/*     "params" : { */
461 	spdk_json_write_named_object_begin(w, "params");
462 	spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
463 	spdk_json_write_named_bool(w, "allow_any_host", spdk_nvmf_subsystem_get_allow_any_host(subsystem));
464 	spdk_json_write_named_string(w, "serial_number", spdk_nvmf_subsystem_get_sn(subsystem));
465 	spdk_json_write_named_string(w, "model_number", spdk_nvmf_subsystem_get_mn(subsystem));
466 
467 	max_namespaces = spdk_nvmf_subsystem_get_max_namespaces(subsystem);
468 	if (max_namespaces != 0) {
469 		spdk_json_write_named_uint32(w, "max_namespaces", max_namespaces);
470 	}
471 
472 	spdk_json_write_named_uint32(w, "min_cntlid", spdk_nvmf_subsystem_get_min_cntlid(subsystem));
473 	spdk_json_write_named_uint32(w, "max_cntlid", spdk_nvmf_subsystem_get_max_cntlid(subsystem));
474 	spdk_json_write_named_bool(w, "ana_reporting", nvmf_subsystem_get_ana_reporting(subsystem));
475 
476 	/*     } "params" */
477 	spdk_json_write_object_end(w);
478 
479 	/* } */
480 	spdk_json_write_object_end(w);
481 
482 	for (listener = spdk_nvmf_subsystem_get_first_listener(subsystem); listener != NULL;
483 	     listener = spdk_nvmf_subsystem_get_next_listener(subsystem, listener)) {
484 		trid = spdk_nvmf_subsystem_listener_get_trid(listener);
485 
486 		spdk_json_write_object_begin(w);
487 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_listener");
488 
489 		/*     "params" : { */
490 		spdk_json_write_named_object_begin(w, "params");
491 
492 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
493 		nvmf_transport_listen_dump_opts(listener->transport, trid, w);
494 
495 		/*     } "params" */
496 		spdk_json_write_object_end(w);
497 
498 		/* } */
499 		spdk_json_write_object_end(w);
500 	}
501 
502 	for (host = spdk_nvmf_subsystem_get_first_host(subsystem); host != NULL;
503 	     host = spdk_nvmf_subsystem_get_next_host(subsystem, host)) {
504 
505 		spdk_json_write_object_begin(w);
506 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_host");
507 
508 		/*     "params" : { */
509 		spdk_json_write_named_object_begin(w, "params");
510 
511 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
512 		spdk_json_write_named_string(w, "host", spdk_nvmf_host_get_nqn(host));
513 
514 		/*     } "params" */
515 		spdk_json_write_object_end(w);
516 
517 		/* } */
518 		spdk_json_write_object_end(w);
519 	}
520 
521 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
522 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
523 		spdk_nvmf_ns_get_opts(ns, &ns_opts, sizeof(ns_opts));
524 
525 		spdk_json_write_object_begin(w);
526 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_ns");
527 
528 		/*     "params" : { */
529 		spdk_json_write_named_object_begin(w, "params");
530 
531 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
532 
533 		/*     "namespace" : { */
534 		spdk_json_write_named_object_begin(w, "namespace");
535 
536 		spdk_json_write_named_uint32(w, "nsid", spdk_nvmf_ns_get_id(ns));
537 		spdk_json_write_named_string(w, "bdev_name", spdk_bdev_get_name(spdk_nvmf_ns_get_bdev(ns)));
538 
539 		if (!spdk_mem_all_zero(ns_opts.nguid, sizeof(ns_opts.nguid))) {
540 			SPDK_STATIC_ASSERT(sizeof(ns_opts.nguid) == sizeof(uint64_t) * 2, "size mismatch");
541 			spdk_json_write_named_string_fmt(w, "nguid", "%016"PRIX64"%016"PRIX64, from_be64(&ns_opts.nguid[0]),
542 							 from_be64(&ns_opts.nguid[8]));
543 		}
544 
545 		if (!spdk_mem_all_zero(ns_opts.eui64, sizeof(ns_opts.eui64))) {
546 			SPDK_STATIC_ASSERT(sizeof(ns_opts.eui64) == sizeof(uint64_t), "size mismatch");
547 			spdk_json_write_named_string_fmt(w, "eui64", "%016"PRIX64, from_be64(&ns_opts.eui64));
548 		}
549 
550 		if (!spdk_mem_all_zero(&ns_opts.uuid, sizeof(ns_opts.uuid))) {
551 			spdk_uuid_fmt_lower(uuid_str, sizeof(uuid_str), &ns_opts.uuid);
552 			spdk_json_write_named_string(w, "uuid",  uuid_str);
553 		}
554 
555 		if (nvmf_subsystem_get_ana_reporting(subsystem)) {
556 			spdk_json_write_named_uint32(w, "anagrpid", ns_opts.anagrpid);
557 		}
558 
559 		/*     "namespace" */
560 		spdk_json_write_object_end(w);
561 
562 		/*     } "params" */
563 		spdk_json_write_object_end(w);
564 
565 		/* } */
566 		spdk_json_write_object_end(w);
567 	}
568 }
569 
570 void
571 spdk_nvmf_tgt_write_config_json(struct spdk_json_write_ctx *w, struct spdk_nvmf_tgt *tgt)
572 {
573 	struct spdk_nvmf_subsystem *subsystem;
574 	struct spdk_nvmf_transport *transport;
575 
576 	spdk_json_write_object_begin(w);
577 	spdk_json_write_named_string(w, "method", "nvmf_set_max_subsystems");
578 
579 	spdk_json_write_named_object_begin(w, "params");
580 	spdk_json_write_named_uint32(w, "max_subsystems", tgt->max_subsystems);
581 	spdk_json_write_object_end(w);
582 
583 	spdk_json_write_object_end(w);
584 
585 	spdk_json_write_object_begin(w);
586 	spdk_json_write_named_string(w, "method", "nvmf_set_crdt");
587 	spdk_json_write_named_object_begin(w, "params");
588 	spdk_json_write_named_uint32(w, "crdt1", tgt->crdt[0]);
589 	spdk_json_write_named_uint32(w, "crdt2", tgt->crdt[1]);
590 	spdk_json_write_named_uint32(w, "crdt3", tgt->crdt[2]);
591 	spdk_json_write_object_end(w);
592 	spdk_json_write_object_end(w);
593 
594 	/* write transports */
595 	TAILQ_FOREACH(transport, &tgt->transports, link) {
596 		spdk_json_write_object_begin(w);
597 		spdk_json_write_named_string(w, "method", "nvmf_create_transport");
598 		nvmf_transport_dump_opts(transport, w, true);
599 		spdk_json_write_object_end(w);
600 	}
601 
602 	subsystem = spdk_nvmf_subsystem_get_first(tgt);
603 	while (subsystem) {
604 		nvmf_write_subsystem_config_json(w, subsystem);
605 		subsystem = spdk_nvmf_subsystem_get_next(subsystem);
606 	}
607 }
608 
609 static void
610 nvmf_listen_opts_copy(struct spdk_nvmf_listen_opts *opts,
611 		      const struct spdk_nvmf_listen_opts *opts_src, size_t opts_size)
612 {
613 	assert(opts);
614 	assert(opts_src);
615 
616 	opts->opts_size = opts_size;
617 
618 #define SET_FIELD(field) \
619     if (offsetof(struct spdk_nvmf_listen_opts, field) + sizeof(opts->field) <= opts_size) { \
620                  opts->field = opts_src->field; \
621     } \
622 
623 	SET_FIELD(transport_specific);
624 #undef SET_FIELD
625 
626 	/* Do not remove this statement, you should always update this statement when you adding a new field,
627 	 * and do not forget to add the SET_FIELD statement for your added field. */
628 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listen_opts) == 16, "Incorrect size");
629 }
630 
631 void
632 spdk_nvmf_listen_opts_init(struct spdk_nvmf_listen_opts *opts, size_t opts_size)
633 {
634 	struct spdk_nvmf_listen_opts opts_local = {};
635 
636 	/* local version of opts should have defaults set here */
637 
638 	nvmf_listen_opts_copy(opts, &opts_local, opts_size);
639 }
640 
641 int
642 spdk_nvmf_tgt_listen_ext(struct spdk_nvmf_tgt *tgt, const struct spdk_nvme_transport_id *trid,
643 			 struct spdk_nvmf_listen_opts *opts)
644 {
645 	struct spdk_nvmf_transport *transport;
646 	int rc;
647 	struct spdk_nvmf_listen_opts opts_local = {};
648 
649 	if (!opts) {
650 		SPDK_ERRLOG("opts should not be NULL\n");
651 		return -EINVAL;
652 	}
653 
654 	if (!opts->opts_size) {
655 		SPDK_ERRLOG("The opts_size in opts structure should not be zero\n");
656 		return -EINVAL;
657 	}
658 
659 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
660 	if (!transport) {
661 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
662 			    trid->trstring);
663 		return -EINVAL;
664 	}
665 
666 	nvmf_listen_opts_copy(&opts_local, opts, opts->opts_size);
667 	rc = spdk_nvmf_transport_listen(transport, trid, &opts_local);
668 	if (rc < 0) {
669 		SPDK_ERRLOG("Unable to listen on address '%s'\n", trid->traddr);
670 	}
671 
672 	return rc;
673 }
674 
675 int
676 spdk_nvmf_tgt_stop_listen(struct spdk_nvmf_tgt *tgt,
677 			  struct spdk_nvme_transport_id *trid)
678 {
679 	struct spdk_nvmf_transport *transport;
680 	int rc;
681 
682 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
683 	if (!transport) {
684 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
685 			    trid->trstring);
686 		return -EINVAL;
687 	}
688 
689 	rc = spdk_nvmf_transport_stop_listen(transport, trid);
690 	if (rc < 0) {
691 		SPDK_ERRLOG("Failed to stop listening on address '%s'\n", trid->traddr);
692 		return rc;
693 	}
694 	return 0;
695 }
696 
697 struct spdk_nvmf_tgt_add_transport_ctx {
698 	struct spdk_nvmf_tgt *tgt;
699 	struct spdk_nvmf_transport *transport;
700 	spdk_nvmf_tgt_add_transport_done_fn cb_fn;
701 	void *cb_arg;
702 	int status;
703 };
704 
705 static void
706 _nvmf_tgt_remove_transport_done(struct spdk_io_channel_iter *i, int status)
707 {
708 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
709 
710 	ctx->cb_fn(ctx->cb_arg, ctx->status);
711 	free(ctx);
712 }
713 
714 static void
715 _nvmf_tgt_remove_transport(struct spdk_io_channel_iter *i)
716 {
717 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
718 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
719 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
720 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
721 
722 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
723 		if (tgroup->transport == ctx->transport) {
724 			TAILQ_REMOVE(&group->tgroups, tgroup, link);
725 			nvmf_transport_poll_group_destroy(tgroup);
726 		}
727 	}
728 
729 	spdk_for_each_channel_continue(i, 0);
730 }
731 
732 static void
733 _nvmf_tgt_add_transport_done(struct spdk_io_channel_iter *i, int status)
734 {
735 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
736 
737 	if (status) {
738 		ctx->status = status;
739 		spdk_for_each_channel(ctx->tgt,
740 				      _nvmf_tgt_remove_transport,
741 				      ctx,
742 				      _nvmf_tgt_remove_transport_done);
743 		return;
744 	}
745 
746 	ctx->transport->tgt = ctx->tgt;
747 	TAILQ_INSERT_TAIL(&ctx->tgt->transports, ctx->transport, link);
748 	ctx->cb_fn(ctx->cb_arg, status);
749 	free(ctx);
750 }
751 
752 static void
753 _nvmf_tgt_add_transport(struct spdk_io_channel_iter *i)
754 {
755 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
756 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
757 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
758 	int rc;
759 
760 	rc = nvmf_poll_group_add_transport(group, ctx->transport);
761 	spdk_for_each_channel_continue(i, rc);
762 }
763 
764 void
765 spdk_nvmf_tgt_add_transport(struct spdk_nvmf_tgt *tgt,
766 			    struct spdk_nvmf_transport *transport,
767 			    spdk_nvmf_tgt_add_transport_done_fn cb_fn,
768 			    void *cb_arg)
769 {
770 	struct spdk_nvmf_tgt_add_transport_ctx *ctx;
771 
772 	SPDK_DTRACE_PROBE2(nvmf_tgt_add_transport, transport, tgt->name);
773 
774 	if (spdk_nvmf_tgt_get_transport(tgt, transport->ops->name)) {
775 		cb_fn(cb_arg, -EEXIST);
776 		return; /* transport already created */
777 	}
778 
779 	ctx = calloc(1, sizeof(*ctx));
780 	if (!ctx) {
781 		cb_fn(cb_arg, -ENOMEM);
782 		return;
783 	}
784 
785 	ctx->tgt = tgt;
786 	ctx->transport = transport;
787 	ctx->cb_fn = cb_fn;
788 	ctx->cb_arg = cb_arg;
789 
790 	spdk_for_each_channel(tgt,
791 			      _nvmf_tgt_add_transport,
792 			      ctx,
793 			      _nvmf_tgt_add_transport_done);
794 }
795 
796 struct spdk_nvmf_subsystem *
797 spdk_nvmf_tgt_find_subsystem(struct spdk_nvmf_tgt *tgt, const char *subnqn)
798 {
799 	struct spdk_nvmf_subsystem	*subsystem;
800 	uint32_t sid;
801 
802 	if (!subnqn) {
803 		return NULL;
804 	}
805 
806 	/* Ensure that subnqn is null terminated */
807 	if (!memchr(subnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
808 		SPDK_ERRLOG("Connect SUBNQN is not null terminated\n");
809 		return NULL;
810 	}
811 
812 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
813 		subsystem = tgt->subsystems[sid];
814 		if (subsystem == NULL) {
815 			continue;
816 		}
817 
818 		if (strcmp(subnqn, subsystem->subnqn) == 0) {
819 			return subsystem;
820 		}
821 	}
822 
823 	return NULL;
824 }
825 
826 struct spdk_nvmf_transport *
827 spdk_nvmf_tgt_get_transport(struct spdk_nvmf_tgt *tgt, const char *transport_name)
828 {
829 	struct spdk_nvmf_transport *transport;
830 
831 	TAILQ_FOREACH(transport, &tgt->transports, link) {
832 		if (!strncasecmp(transport->ops->name, transport_name, SPDK_NVMF_TRSTRING_MAX_LEN)) {
833 			return transport;
834 		}
835 	}
836 	return NULL;
837 }
838 
839 struct nvmf_new_qpair_ctx {
840 	struct spdk_nvmf_qpair *qpair;
841 	struct spdk_nvmf_poll_group *group;
842 };
843 
844 static void
845 _nvmf_poll_group_add(void *_ctx)
846 {
847 	struct nvmf_new_qpair_ctx *ctx = _ctx;
848 	struct spdk_nvmf_qpair *qpair = ctx->qpair;
849 	struct spdk_nvmf_poll_group *group = ctx->group;
850 
851 	free(_ctx);
852 
853 	if (spdk_nvmf_poll_group_add(group, qpair) != 0) {
854 		SPDK_ERRLOG("Unable to add the qpair to a poll group.\n");
855 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
856 	}
857 }
858 
859 void
860 spdk_nvmf_tgt_new_qpair(struct spdk_nvmf_tgt *tgt, struct spdk_nvmf_qpair *qpair)
861 {
862 	struct spdk_nvmf_poll_group *group;
863 	struct nvmf_new_qpair_ctx *ctx;
864 
865 	group = spdk_nvmf_get_optimal_poll_group(qpair);
866 	if (group == NULL) {
867 		if (tgt->next_poll_group == NULL) {
868 			tgt->next_poll_group = TAILQ_FIRST(&tgt->poll_groups);
869 			if (tgt->next_poll_group == NULL) {
870 				SPDK_ERRLOG("No poll groups exist.\n");
871 				spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
872 				return;
873 			}
874 		}
875 		group = tgt->next_poll_group;
876 		tgt->next_poll_group = TAILQ_NEXT(group, link);
877 	}
878 
879 	ctx = calloc(1, sizeof(*ctx));
880 	if (!ctx) {
881 		SPDK_ERRLOG("Unable to send message to poll group.\n");
882 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
883 		return;
884 	}
885 
886 	ctx->qpair = qpair;
887 	ctx->group = group;
888 
889 	spdk_thread_send_msg(group->thread, _nvmf_poll_group_add, ctx);
890 }
891 
892 struct spdk_nvmf_poll_group *
893 spdk_nvmf_poll_group_create(struct spdk_nvmf_tgt *tgt)
894 {
895 	struct spdk_io_channel *ch;
896 
897 	ch = spdk_get_io_channel(tgt);
898 	if (!ch) {
899 		SPDK_ERRLOG("Unable to get I/O channel for target\n");
900 		return NULL;
901 	}
902 
903 	return spdk_io_channel_get_ctx(ch);
904 }
905 
906 void
907 spdk_nvmf_poll_group_destroy(struct spdk_nvmf_poll_group *group,
908 			     spdk_nvmf_poll_group_destroy_done_fn cb_fn,
909 			     void *cb_arg)
910 {
911 	assert(group->destroy_cb_fn == NULL);
912 	group->destroy_cb_fn = cb_fn;
913 	group->destroy_cb_arg = cb_arg;
914 
915 	/* This function will put the io_channel associated with this poll group */
916 	nvmf_tgt_destroy_poll_group_qpairs(group);
917 }
918 
919 int
920 spdk_nvmf_poll_group_add(struct spdk_nvmf_poll_group *group,
921 			 struct spdk_nvmf_qpair *qpair)
922 {
923 	int rc = -1;
924 	struct spdk_nvmf_transport_poll_group *tgroup;
925 
926 	TAILQ_INIT(&qpair->outstanding);
927 	qpair->group = group;
928 	qpair->ctrlr = NULL;
929 	qpair->disconnect_started = false;
930 
931 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
932 		if (tgroup->transport == qpair->transport) {
933 			rc = nvmf_transport_poll_group_add(tgroup, qpair);
934 			break;
935 		}
936 	}
937 
938 	/* We add the qpair to the group only it is successfully added into the tgroup */
939 	if (rc == 0) {
940 		SPDK_DTRACE_PROBE2(nvmf_poll_group_add_qpair, qpair, spdk_thread_get_id(group->thread));
941 		TAILQ_INSERT_TAIL(&group->qpairs, qpair, link);
942 		nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ACTIVE);
943 	}
944 
945 	return rc;
946 }
947 
948 static void
949 _nvmf_ctrlr_destruct(void *ctx)
950 {
951 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
952 
953 	nvmf_ctrlr_destruct(ctrlr);
954 }
955 
956 static void
957 _nvmf_ctrlr_free_from_qpair(void *ctx)
958 {
959 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
960 	struct spdk_nvmf_ctrlr *ctrlr = qpair_ctx->ctrlr;
961 	uint32_t count;
962 
963 	spdk_bit_array_clear(ctrlr->qpair_mask, qpair_ctx->qid);
964 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
965 	if (count == 0) {
966 		assert(!ctrlr->in_destruct);
967 		SPDK_DEBUGLOG(nvmf, "Last qpair %u, destroy ctrlr 0x%hx\n", qpair_ctx->qid, ctrlr->cntlid);
968 		ctrlr->in_destruct = true;
969 		spdk_thread_send_msg(ctrlr->subsys->thread, _nvmf_ctrlr_destruct, ctrlr);
970 	}
971 	free(qpair_ctx);
972 }
973 
974 static void
975 _nvmf_transport_qpair_fini_complete(void *cb_ctx)
976 {
977 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = cb_ctx;
978 	struct spdk_nvmf_ctrlr *ctrlr;
979 	/* Store cb args since cb_ctx can be freed in _nvmf_ctrlr_free_from_qpair */
980 	nvmf_qpair_disconnect_cb cb_fn = qpair_ctx->cb_fn;
981 	void *cb_arg = qpair_ctx->ctx;
982 	struct spdk_thread *cb_thread = qpair_ctx->thread;
983 
984 	ctrlr = qpair_ctx->ctrlr;
985 	SPDK_DEBUGLOG(nvmf, "Finish destroying qid %u\n", qpair_ctx->qid);
986 
987 	if (ctrlr) {
988 		if (qpair_ctx->qid == 0) {
989 			/* Admin qpair is removed, so set the pointer to NULL.
990 			 * This operation is safe since we are on ctrlr thread now, admin qpair's thread is the same
991 			 * as controller's thread */
992 			assert(ctrlr->thread == spdk_get_thread());
993 			ctrlr->admin_qpair = NULL;
994 		}
995 		/* Free qpair id from controller's bit mask and destroy the controller if it is the last qpair */
996 		if (ctrlr->thread) {
997 			spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_free_from_qpair, qpair_ctx);
998 		} else {
999 			_nvmf_ctrlr_free_from_qpair(qpair_ctx);
1000 		}
1001 	} else {
1002 		free(qpair_ctx);
1003 	}
1004 
1005 	if (cb_fn) {
1006 		spdk_thread_send_msg(cb_thread, cb_fn, cb_arg);
1007 	}
1008 }
1009 
1010 void
1011 spdk_nvmf_poll_group_remove(struct spdk_nvmf_qpair *qpair)
1012 {
1013 	struct spdk_nvmf_transport_poll_group *tgroup;
1014 	int rc;
1015 
1016 	SPDK_DTRACE_PROBE2(nvmf_poll_group_remove_qpair, qpair,
1017 			   spdk_thread_get_id(qpair->group->thread));
1018 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ERROR);
1019 
1020 	/* Find the tgroup and remove the qpair from the tgroup */
1021 	TAILQ_FOREACH(tgroup, &qpair->group->tgroups, link) {
1022 		if (tgroup->transport == qpair->transport) {
1023 			rc = nvmf_transport_poll_group_remove(tgroup, qpair);
1024 			if (rc && (rc != ENOTSUP)) {
1025 				SPDK_ERRLOG("Cannot remove qpair=%p from transport group=%p\n",
1026 					    qpair, tgroup);
1027 			}
1028 			break;
1029 		}
1030 	}
1031 
1032 	TAILQ_REMOVE(&qpair->group->qpairs, qpair, link);
1033 	qpair->group = NULL;
1034 }
1035 
1036 static void
1037 _nvmf_qpair_destroy(void *ctx, int status)
1038 {
1039 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1040 	struct spdk_nvmf_qpair *qpair = qpair_ctx->qpair;
1041 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
1042 	struct spdk_nvmf_request *req, *tmp;
1043 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1044 
1045 	assert(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
1046 	qpair_ctx->qid = qpair->qid;
1047 
1048 	if (ctrlr) {
1049 		if (0 == qpair->qid) {
1050 			assert(qpair->group->stat.current_admin_qpairs > 0);
1051 			qpair->group->stat.current_admin_qpairs--;
1052 		} else {
1053 			assert(qpair->group->stat.current_io_qpairs > 0);
1054 			qpair->group->stat.current_io_qpairs--;
1055 		}
1056 
1057 		sgroup = &qpair->group->sgroups[ctrlr->subsys->id];
1058 		TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1059 			if (req->qpair == qpair) {
1060 				TAILQ_REMOVE(&sgroup->queued, req, link);
1061 				if (nvmf_transport_req_free(req)) {
1062 					SPDK_ERRLOG("Transport request free error!/n");
1063 				}
1064 			}
1065 		}
1066 	}
1067 
1068 	qpair_ctx->ctrlr = ctrlr;
1069 	spdk_nvmf_poll_group_remove(qpair);
1070 	nvmf_transport_qpair_fini(qpair, _nvmf_transport_qpair_fini_complete, qpair_ctx);
1071 }
1072 
1073 static void
1074 _nvmf_qpair_disconnect_msg(void *ctx)
1075 {
1076 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1077 
1078 	spdk_nvmf_qpair_disconnect(qpair_ctx->qpair, qpair_ctx->cb_fn, qpair_ctx->ctx);
1079 	free(ctx);
1080 }
1081 
1082 int
1083 spdk_nvmf_qpair_disconnect(struct spdk_nvmf_qpair *qpair, nvmf_qpair_disconnect_cb cb_fn, void *ctx)
1084 {
1085 	struct spdk_nvmf_poll_group *group = qpair->group;
1086 	struct nvmf_qpair_disconnect_ctx *qpair_ctx;
1087 
1088 	if (__atomic_test_and_set(&qpair->disconnect_started, __ATOMIC_RELAXED)) {
1089 		if (cb_fn) {
1090 			cb_fn(ctx);
1091 		}
1092 		return 0;
1093 	}
1094 
1095 	/* If we get a qpair in the uninitialized state, we can just destroy it immediately */
1096 	if (qpair->state == SPDK_NVMF_QPAIR_UNINITIALIZED) {
1097 		nvmf_transport_qpair_fini(qpair, NULL, NULL);
1098 		if (cb_fn) {
1099 			cb_fn(ctx);
1100 		}
1101 		return 0;
1102 	}
1103 
1104 	assert(group != NULL);
1105 	if (spdk_get_thread() != group->thread) {
1106 		/* clear the atomic so we can set it on the next call on the proper thread. */
1107 		__atomic_clear(&qpair->disconnect_started, __ATOMIC_RELAXED);
1108 		qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1109 		if (!qpair_ctx) {
1110 			SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1111 			return -ENOMEM;
1112 		}
1113 		qpair_ctx->qpair = qpair;
1114 		qpair_ctx->cb_fn = cb_fn;
1115 		qpair_ctx->thread = group->thread;
1116 		qpair_ctx->ctx = ctx;
1117 		spdk_thread_send_msg(group->thread, _nvmf_qpair_disconnect_msg, qpair_ctx);
1118 		return 0;
1119 	}
1120 
1121 	SPDK_DTRACE_PROBE2(nvmf_qpair_disconnect, qpair, spdk_thread_get_id(group->thread));
1122 	assert(qpair->state == SPDK_NVMF_QPAIR_ACTIVE);
1123 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_DEACTIVATING);
1124 
1125 	qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1126 	if (!qpair_ctx) {
1127 		SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1128 		return -ENOMEM;
1129 	}
1130 
1131 	qpair_ctx->qpair = qpair;
1132 	qpair_ctx->cb_fn = cb_fn;
1133 	qpair_ctx->thread = group->thread;
1134 	qpair_ctx->ctx = ctx;
1135 
1136 	/* Check for outstanding I/O */
1137 	if (!TAILQ_EMPTY(&qpair->outstanding)) {
1138 		SPDK_DTRACE_PROBE2(nvmf_poll_group_drain_qpair, qpair, spdk_thread_get_id(group->thread));
1139 		qpair->state_cb = _nvmf_qpair_destroy;
1140 		qpair->state_cb_arg = qpair_ctx;
1141 		nvmf_qpair_abort_pending_zcopy_reqs(qpair);
1142 		nvmf_qpair_free_aer(qpair);
1143 		return 0;
1144 	}
1145 
1146 	_nvmf_qpair_destroy(qpair_ctx, 0);
1147 
1148 	return 0;
1149 }
1150 
1151 int
1152 spdk_nvmf_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
1153 			      struct spdk_nvme_transport_id *trid)
1154 {
1155 	return nvmf_transport_qpair_get_peer_trid(qpair, trid);
1156 }
1157 
1158 int
1159 spdk_nvmf_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
1160 			       struct spdk_nvme_transport_id *trid)
1161 {
1162 	return nvmf_transport_qpair_get_local_trid(qpair, trid);
1163 }
1164 
1165 int
1166 spdk_nvmf_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
1167 				struct spdk_nvme_transport_id *trid)
1168 {
1169 	return nvmf_transport_qpair_get_listen_trid(qpair, trid);
1170 }
1171 
1172 static int
1173 poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1174 			    struct spdk_nvmf_subsystem *subsystem)
1175 {
1176 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1177 	uint32_t new_num_ns, old_num_ns;
1178 	uint32_t i, j;
1179 	struct spdk_nvmf_ns *ns;
1180 	struct spdk_nvmf_registrant *reg, *tmp;
1181 	struct spdk_io_channel *ch;
1182 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1183 	struct spdk_nvmf_ctrlr *ctrlr;
1184 	bool ns_changed;
1185 
1186 	/* Make sure our poll group has memory for this subsystem allocated */
1187 	if (subsystem->id >= group->num_sgroups) {
1188 		return -ENOMEM;
1189 	}
1190 
1191 	sgroup = &group->sgroups[subsystem->id];
1192 
1193 	/* Make sure the array of namespace information is the correct size */
1194 	new_num_ns = subsystem->max_nsid;
1195 	old_num_ns = sgroup->num_ns;
1196 
1197 	ns_changed = false;
1198 
1199 	if (old_num_ns == 0) {
1200 		if (new_num_ns > 0) {
1201 			/* First allocation */
1202 			sgroup->ns_info = calloc(new_num_ns, sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1203 			if (!sgroup->ns_info) {
1204 				return -ENOMEM;
1205 			}
1206 		}
1207 	} else if (new_num_ns > old_num_ns) {
1208 		void *buf;
1209 
1210 		/* Make the array larger */
1211 		buf = realloc(sgroup->ns_info, new_num_ns * sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1212 		if (!buf) {
1213 			return -ENOMEM;
1214 		}
1215 
1216 		sgroup->ns_info = buf;
1217 
1218 		/* Null out the new namespace information slots */
1219 		for (i = old_num_ns; i < new_num_ns; i++) {
1220 			memset(&sgroup->ns_info[i], 0, sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1221 		}
1222 	} else if (new_num_ns < old_num_ns) {
1223 		void *buf;
1224 
1225 		/* Free the extra I/O channels */
1226 		for (i = new_num_ns; i < old_num_ns; i++) {
1227 			ns_info = &sgroup->ns_info[i];
1228 
1229 			if (ns_info->channel) {
1230 				spdk_put_io_channel(ns_info->channel);
1231 				ns_info->channel = NULL;
1232 			}
1233 		}
1234 
1235 		/* Make the array smaller */
1236 		if (new_num_ns > 0) {
1237 			buf = realloc(sgroup->ns_info, new_num_ns * sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1238 			if (!buf) {
1239 				return -ENOMEM;
1240 			}
1241 			sgroup->ns_info = buf;
1242 		} else {
1243 			free(sgroup->ns_info);
1244 			sgroup->ns_info = NULL;
1245 		}
1246 	}
1247 
1248 	sgroup->num_ns = new_num_ns;
1249 
1250 	/* Detect bdevs that were added or removed */
1251 	for (i = 0; i < sgroup->num_ns; i++) {
1252 		ns = subsystem->ns[i];
1253 		ns_info = &sgroup->ns_info[i];
1254 		ch = ns_info->channel;
1255 
1256 		if (ns == NULL && ch == NULL) {
1257 			/* Both NULL. Leave empty */
1258 		} else if (ns == NULL && ch != NULL) {
1259 			/* There was a channel here, but the namespace is gone. */
1260 			ns_changed = true;
1261 			spdk_put_io_channel(ch);
1262 			ns_info->channel = NULL;
1263 		} else if (ns != NULL && ch == NULL) {
1264 			/* A namespace appeared but there is no channel yet */
1265 			ns_changed = true;
1266 			ch = spdk_bdev_get_io_channel(ns->desc);
1267 			if (ch == NULL) {
1268 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
1269 				return -ENOMEM;
1270 			}
1271 			ns_info->channel = ch;
1272 		} else if (spdk_uuid_compare(&ns_info->uuid, spdk_bdev_get_uuid(ns->bdev)) != 0) {
1273 			/* A namespace was here before, but was replaced by a new one. */
1274 			ns_changed = true;
1275 			spdk_put_io_channel(ns_info->channel);
1276 			memset(ns_info, 0, sizeof(*ns_info));
1277 
1278 			ch = spdk_bdev_get_io_channel(ns->desc);
1279 			if (ch == NULL) {
1280 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
1281 				return -ENOMEM;
1282 			}
1283 			ns_info->channel = ch;
1284 		} else if (ns_info->num_blocks != spdk_bdev_get_num_blocks(ns->bdev)) {
1285 			/* Namespace is still there but size has changed */
1286 			SPDK_DEBUGLOG(nvmf, "Namespace resized: subsystem_id %u,"
1287 				      " nsid %u, pg %p, old %" PRIu64 ", new %" PRIu64 "\n",
1288 				      subsystem->id,
1289 				      ns->nsid,
1290 				      group,
1291 				      ns_info->num_blocks,
1292 				      spdk_bdev_get_num_blocks(ns->bdev));
1293 			ns_changed = true;
1294 		}
1295 
1296 		if (ns == NULL) {
1297 			memset(ns_info, 0, sizeof(*ns_info));
1298 		} else {
1299 			ns_info->uuid = *spdk_bdev_get_uuid(ns->bdev);
1300 			ns_info->num_blocks = spdk_bdev_get_num_blocks(ns->bdev);
1301 			ns_info->crkey = ns->crkey;
1302 			ns_info->rtype = ns->rtype;
1303 			if (ns->holder) {
1304 				ns_info->holder_id = ns->holder->hostid;
1305 			}
1306 
1307 			memset(&ns_info->reg_hostid, 0, SPDK_NVMF_MAX_NUM_REGISTRANTS * sizeof(struct spdk_uuid));
1308 			j = 0;
1309 			TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
1310 				if (j >= SPDK_NVMF_MAX_NUM_REGISTRANTS) {
1311 					SPDK_ERRLOG("Maximum %u registrants can support.\n", SPDK_NVMF_MAX_NUM_REGISTRANTS);
1312 					return -EINVAL;
1313 				}
1314 				ns_info->reg_hostid[j++] = reg->hostid;
1315 			}
1316 		}
1317 	}
1318 
1319 	if (ns_changed) {
1320 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1321 			/* It is possible that a ctrlr was added but the admin_qpair hasn't been
1322 			 * assigned yet.
1323 			 */
1324 			if (!ctrlr->admin_qpair) {
1325 				continue;
1326 			}
1327 			if (ctrlr->admin_qpair->group == group) {
1328 				nvmf_ctrlr_async_event_ns_notice(ctrlr);
1329 				nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
1330 			}
1331 		}
1332 	}
1333 
1334 	return 0;
1335 }
1336 
1337 int
1338 nvmf_poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1339 				 struct spdk_nvmf_subsystem *subsystem)
1340 {
1341 	return poll_group_update_subsystem(group, subsystem);
1342 }
1343 
1344 int
1345 nvmf_poll_group_add_subsystem(struct spdk_nvmf_poll_group *group,
1346 			      struct spdk_nvmf_subsystem *subsystem,
1347 			      spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1348 {
1349 	int rc = 0;
1350 	struct spdk_nvmf_subsystem_poll_group *sgroup = &group->sgroups[subsystem->id];
1351 	uint32_t i;
1352 
1353 	TAILQ_INIT(&sgroup->queued);
1354 
1355 	rc = poll_group_update_subsystem(group, subsystem);
1356 	if (rc) {
1357 		nvmf_poll_group_remove_subsystem(group, subsystem, NULL, NULL);
1358 		goto fini;
1359 	}
1360 
1361 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1362 
1363 	for (i = 0; i < sgroup->num_ns; i++) {
1364 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1365 	}
1366 
1367 fini:
1368 	if (cb_fn) {
1369 		cb_fn(cb_arg, rc);
1370 	}
1371 
1372 	SPDK_DTRACE_PROBE2(nvmf_poll_group_add_subsystem, spdk_thread_get_id(group->thread),
1373 			   subsystem->subnqn);
1374 
1375 	return rc;
1376 }
1377 
1378 static void
1379 _nvmf_poll_group_remove_subsystem_cb(void *ctx, int status)
1380 {
1381 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1382 	struct spdk_nvmf_subsystem *subsystem;
1383 	struct spdk_nvmf_poll_group *group;
1384 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1385 	spdk_nvmf_poll_group_mod_done cpl_fn = NULL;
1386 	void *cpl_ctx = NULL;
1387 	uint32_t nsid;
1388 
1389 	group = qpair_ctx->group;
1390 	subsystem = qpair_ctx->subsystem;
1391 	cpl_fn = qpair_ctx->cpl_fn;
1392 	cpl_ctx = qpair_ctx->cpl_ctx;
1393 	sgroup = &group->sgroups[subsystem->id];
1394 
1395 	if (status) {
1396 		goto fini;
1397 	}
1398 
1399 	for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
1400 		if (sgroup->ns_info[nsid].channel) {
1401 			spdk_put_io_channel(sgroup->ns_info[nsid].channel);
1402 			sgroup->ns_info[nsid].channel = NULL;
1403 		}
1404 	}
1405 
1406 	sgroup->num_ns = 0;
1407 	free(sgroup->ns_info);
1408 	sgroup->ns_info = NULL;
1409 fini:
1410 	free(qpair_ctx);
1411 	if (cpl_fn) {
1412 		cpl_fn(cpl_ctx, status);
1413 	}
1414 }
1415 
1416 static void nvmf_poll_group_remove_subsystem_msg(void *ctx);
1417 
1418 static void
1419 remove_subsystem_qpair_cb(void *ctx)
1420 {
1421 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1422 
1423 	assert(qpair_ctx->count > 0);
1424 	qpair_ctx->count--;
1425 	if (qpair_ctx->count == 0) {
1426 		/* All of the asynchronous callbacks for this context have been
1427 		 * completed.  Call nvmf_poll_group_remove_subsystem_msg() again
1428 		 * to check if all associated qpairs for this subsystem have
1429 		 * been removed from the poll group.
1430 		 */
1431 		nvmf_poll_group_remove_subsystem_msg(ctx);
1432 	}
1433 }
1434 
1435 static void
1436 nvmf_poll_group_remove_subsystem_msg(void *ctx)
1437 {
1438 	struct spdk_nvmf_qpair *qpair, *qpair_tmp;
1439 	struct spdk_nvmf_subsystem *subsystem;
1440 	struct spdk_nvmf_poll_group *group;
1441 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1442 	bool qpairs_found = false;
1443 	int rc = 0;
1444 
1445 	group = qpair_ctx->group;
1446 	subsystem = qpair_ctx->subsystem;
1447 
1448 	/* Initialize count to 1.  This acts like a ref count, to ensure that if spdk_nvmf_qpair_disconnect
1449 	 * immediately invokes the callback (i.e. the qpairs is already in process of being disconnected)
1450 	 * that we don't recursively call nvmf_poll_group_remove_subsystem_msg before we've iterated the
1451 	 * full list of qpairs.
1452 	 */
1453 	qpair_ctx->count = 1;
1454 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, qpair_tmp) {
1455 		if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
1456 			qpairs_found = true;
1457 			qpair_ctx->count++;
1458 			rc = spdk_nvmf_qpair_disconnect(qpair, remove_subsystem_qpair_cb, ctx);
1459 			if (rc) {
1460 				break;
1461 			}
1462 		}
1463 	}
1464 	qpair_ctx->count--;
1465 
1466 	if (!qpairs_found) {
1467 		_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
1468 		return;
1469 	}
1470 
1471 	if (qpair_ctx->count == 0 || rc) {
1472 		/* If count == 0, it means there were some qpairs in the poll group but they
1473 		 * were already in process of being disconnected.  So we send a message to this
1474 		 * same thread so that this function executes again later.  We won't actually
1475 		 * invoke the remove_subsystem_cb until all of the qpairs are actually removed
1476 		 * from the poll group.
1477 		 */
1478 		spdk_thread_send_msg(spdk_get_thread(), nvmf_poll_group_remove_subsystem_msg, ctx);
1479 	}
1480 }
1481 
1482 void
1483 nvmf_poll_group_remove_subsystem(struct spdk_nvmf_poll_group *group,
1484 				 struct spdk_nvmf_subsystem *subsystem,
1485 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1486 {
1487 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1488 	struct nvmf_qpair_disconnect_many_ctx *ctx;
1489 	uint32_t i;
1490 
1491 	SPDK_DTRACE_PROBE3(nvmf_poll_group_remove_subsystem, group, spdk_thread_get_id(group->thread),
1492 			   subsystem->subnqn);
1493 
1494 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
1495 	if (!ctx) {
1496 		SPDK_ERRLOG("Unable to allocate memory for context to remove poll subsystem\n");
1497 		if (cb_fn) {
1498 			cb_fn(cb_arg, -1);
1499 		}
1500 		return;
1501 	}
1502 
1503 	ctx->group = group;
1504 	ctx->subsystem = subsystem;
1505 	ctx->cpl_fn = cb_fn;
1506 	ctx->cpl_ctx = cb_arg;
1507 
1508 	sgroup = &group->sgroups[subsystem->id];
1509 	sgroup->state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1510 
1511 	for (i = 0; i < sgroup->num_ns; i++) {
1512 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1513 	}
1514 
1515 	nvmf_poll_group_remove_subsystem_msg(ctx);
1516 }
1517 
1518 void
1519 nvmf_poll_group_pause_subsystem(struct spdk_nvmf_poll_group *group,
1520 				struct spdk_nvmf_subsystem *subsystem,
1521 				uint32_t nsid,
1522 				spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1523 {
1524 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1525 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info = NULL;
1526 	int rc = 0;
1527 	uint32_t i;
1528 
1529 	if (subsystem->id >= group->num_sgroups) {
1530 		rc = -1;
1531 		goto fini;
1532 	}
1533 
1534 	sgroup = &group->sgroups[subsystem->id];
1535 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSED) {
1536 		goto fini;
1537 	}
1538 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1539 
1540 	if (nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1541 		for (i = 0; i < sgroup->num_ns; i++) {
1542 			ns_info = &sgroup->ns_info[i];
1543 			ns_info->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1544 		}
1545 	} else {
1546 		/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
1547 		if (nsid - 1 < sgroup->num_ns) {
1548 			ns_info  = &sgroup->ns_info[nsid - 1];
1549 			ns_info->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1550 		}
1551 	}
1552 
1553 	if (sgroup->mgmt_io_outstanding > 0) {
1554 		assert(sgroup->cb_fn == NULL);
1555 		sgroup->cb_fn = cb_fn;
1556 		assert(sgroup->cb_arg == NULL);
1557 		sgroup->cb_arg = cb_arg;
1558 		return;
1559 	}
1560 
1561 	if (nsid == SPDK_NVME_GLOBAL_NS_TAG) {
1562 		for (i = 0; i < sgroup->num_ns; i++) {
1563 			ns_info = &sgroup->ns_info[i];
1564 
1565 			if (ns_info->io_outstanding > 0) {
1566 				assert(sgroup->cb_fn == NULL);
1567 				sgroup->cb_fn = cb_fn;
1568 				assert(sgroup->cb_arg == NULL);
1569 				sgroup->cb_arg = cb_arg;
1570 				return;
1571 			}
1572 		}
1573 	} else {
1574 		if (ns_info != NULL && ns_info->io_outstanding > 0) {
1575 			assert(sgroup->cb_fn == NULL);
1576 			sgroup->cb_fn = cb_fn;
1577 			assert(sgroup->cb_arg == NULL);
1578 			sgroup->cb_arg = cb_arg;
1579 			return;
1580 		}
1581 	}
1582 
1583 	assert(sgroup->mgmt_io_outstanding == 0);
1584 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
1585 fini:
1586 	if (cb_fn) {
1587 		cb_fn(cb_arg, rc);
1588 	}
1589 }
1590 
1591 void
1592 nvmf_poll_group_resume_subsystem(struct spdk_nvmf_poll_group *group,
1593 				 struct spdk_nvmf_subsystem *subsystem,
1594 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1595 {
1596 	struct spdk_nvmf_request *req, *tmp;
1597 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1598 	int rc = 0;
1599 	uint32_t i;
1600 
1601 	if (subsystem->id >= group->num_sgroups) {
1602 		rc = -1;
1603 		goto fini;
1604 	}
1605 
1606 	sgroup = &group->sgroups[subsystem->id];
1607 
1608 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_ACTIVE) {
1609 		goto fini;
1610 	}
1611 
1612 	rc = poll_group_update_subsystem(group, subsystem);
1613 	if (rc) {
1614 		goto fini;
1615 	}
1616 
1617 	for (i = 0; i < sgroup->num_ns; i++) {
1618 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1619 	}
1620 
1621 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1622 
1623 	/* Release all queued requests */
1624 	TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1625 		TAILQ_REMOVE(&sgroup->queued, req, link);
1626 		if (spdk_nvmf_request_using_zcopy(req)) {
1627 			spdk_nvmf_request_zcopy_start(req);
1628 		} else {
1629 			spdk_nvmf_request_exec(req);
1630 		}
1631 
1632 	}
1633 fini:
1634 	if (cb_fn) {
1635 		cb_fn(cb_arg, rc);
1636 	}
1637 }
1638 
1639 
1640 struct spdk_nvmf_poll_group *
1641 spdk_nvmf_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
1642 {
1643 	struct spdk_nvmf_transport_poll_group *tgroup;
1644 
1645 	tgroup = nvmf_transport_get_optimal_poll_group(qpair->transport, qpair);
1646 
1647 	if (tgroup == NULL) {
1648 		return NULL;
1649 	}
1650 
1651 	return tgroup->group;
1652 }
1653 
1654 void
1655 spdk_nvmf_poll_group_dump_stat(struct spdk_nvmf_poll_group *group, struct spdk_json_write_ctx *w)
1656 {
1657 	struct spdk_nvmf_transport_poll_group *tgroup;
1658 
1659 	spdk_json_write_object_begin(w);
1660 
1661 	spdk_json_write_named_string(w, "name", spdk_thread_get_name(spdk_get_thread()));
1662 	spdk_json_write_named_uint32(w, "admin_qpairs", group->stat.admin_qpairs);
1663 	spdk_json_write_named_uint32(w, "io_qpairs", group->stat.io_qpairs);
1664 	spdk_json_write_named_uint32(w, "current_admin_qpairs", group->stat.current_admin_qpairs);
1665 	spdk_json_write_named_uint32(w, "current_io_qpairs", group->stat.current_io_qpairs);
1666 	spdk_json_write_named_uint64(w, "pending_bdev_io", group->stat.pending_bdev_io);
1667 
1668 	spdk_json_write_named_array_begin(w, "transports");
1669 
1670 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1671 		spdk_json_write_object_begin(w);
1672 		/*
1673 		 * The trtype field intentionally contains a transport name as this is more informative.
1674 		 * The field has not been renamed for backward compatibility.
1675 		 */
1676 		spdk_json_write_named_string(w, "trtype", spdk_nvmf_get_transport_name(tgroup->transport));
1677 
1678 		if (tgroup->transport->ops->poll_group_dump_stat) {
1679 			tgroup->transport->ops->poll_group_dump_stat(tgroup, w);
1680 		}
1681 
1682 		spdk_json_write_object_end(w);
1683 	}
1684 
1685 	spdk_json_write_array_end(w);
1686 	spdk_json_write_object_end(w);
1687 }
1688