xref: /spdk/lib/nvmf/nvmf.c (revision db35950a13859f8680bb97c961337d3317a974b2)
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/endian.h"
41 #include "spdk/string.h"
42 #include "spdk/log.h"
43 #include "spdk_internal/usdt.h"
44 
45 #include "nvmf_internal.h"
46 #include "transport.h"
47 
48 SPDK_LOG_REGISTER_COMPONENT(nvmf)
49 
50 #define SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS 1024
51 #define SPDK_NVMF_DEFAULT_ACCEPT_POLL_RATE_US 10000
52 
53 static TAILQ_HEAD(, spdk_nvmf_tgt) g_nvmf_tgts = TAILQ_HEAD_INITIALIZER(g_nvmf_tgts);
54 
55 typedef void (*nvmf_qpair_disconnect_cpl)(void *ctx, int status);
56 static void nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf);
57 
58 /* supplied to a single call to nvmf_qpair_disconnect */
59 struct nvmf_qpair_disconnect_ctx {
60 	struct spdk_nvmf_qpair *qpair;
61 	struct spdk_nvmf_ctrlr *ctrlr;
62 	nvmf_qpair_disconnect_cb cb_fn;
63 	struct spdk_thread *thread;
64 	void *ctx;
65 	uint16_t qid;
66 };
67 
68 /*
69  * There are several times when we need to iterate through the list of all qpairs and selectively delete them.
70  * In order to do this sequentially without overlap, we must provide a context to recover the next qpair from
71  * to enable calling nvmf_qpair_disconnect on the next desired qpair.
72  */
73 struct nvmf_qpair_disconnect_many_ctx {
74 	struct spdk_nvmf_subsystem *subsystem;
75 	struct spdk_nvmf_poll_group *group;
76 	spdk_nvmf_poll_group_mod_done cpl_fn;
77 	void *cpl_ctx;
78 	uint32_t count;
79 };
80 
81 static void
82 nvmf_qpair_set_state(struct spdk_nvmf_qpair *qpair,
83 		     enum spdk_nvmf_qpair_state state)
84 {
85 	assert(qpair != NULL);
86 	assert(qpair->group->thread == spdk_get_thread());
87 
88 	qpair->state = state;
89 }
90 
91 static int
92 nvmf_poll_group_poll(void *ctx)
93 {
94 	struct spdk_nvmf_poll_group *group = ctx;
95 	int rc;
96 	int count = 0;
97 	struct spdk_nvmf_transport_poll_group *tgroup;
98 
99 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
100 		rc = nvmf_transport_poll_group_poll(tgroup);
101 		if (rc < 0) {
102 			return SPDK_POLLER_BUSY;
103 		}
104 		count += rc;
105 	}
106 
107 	return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
108 }
109 
110 static int
111 nvmf_tgt_create_poll_group(void *io_device, void *ctx_buf)
112 {
113 	struct spdk_nvmf_tgt *tgt = io_device;
114 	struct spdk_nvmf_poll_group *group = ctx_buf;
115 	struct spdk_nvmf_transport *transport;
116 	struct spdk_thread *thread = spdk_get_thread();
117 	uint32_t sid;
118 
119 	SPDK_DTRACE_PROBE1(nvmf_create_poll_group, spdk_thread_get_id(thread));
120 
121 	TAILQ_INIT(&group->tgroups);
122 	TAILQ_INIT(&group->qpairs);
123 
124 	TAILQ_FOREACH(transport, &tgt->transports, link) {
125 		nvmf_poll_group_add_transport(group, transport);
126 	}
127 
128 	group->num_sgroups = tgt->max_subsystems;
129 	group->sgroups = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem_poll_group));
130 	if (!group->sgroups) {
131 		return -ENOMEM;
132 	}
133 
134 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
135 		struct spdk_nvmf_subsystem *subsystem;
136 
137 		subsystem = tgt->subsystems[sid];
138 		if (!subsystem) {
139 			continue;
140 		}
141 
142 		if (nvmf_poll_group_add_subsystem(group, subsystem, NULL, NULL) != 0) {
143 			nvmf_tgt_destroy_poll_group(io_device, ctx_buf);
144 			return -1;
145 		}
146 	}
147 
148 	pthread_mutex_lock(&tgt->mutex);
149 	TAILQ_INSERT_TAIL(&tgt->poll_groups, group, link);
150 	pthread_mutex_unlock(&tgt->mutex);
151 
152 	group->poller = SPDK_POLLER_REGISTER(nvmf_poll_group_poll, group, 0);
153 	group->thread = thread;
154 
155 	return 0;
156 }
157 
158 static void
159 nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf)
160 {
161 	struct spdk_nvmf_tgt *tgt = io_device;
162 	struct spdk_nvmf_poll_group *group = ctx_buf;
163 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
164 	struct spdk_nvmf_subsystem_poll_group *sgroup;
165 	uint32_t sid, nsid;
166 
167 	SPDK_DTRACE_PROBE1(nvmf_destroy_poll_group, spdk_thread_get_id(group->thread));
168 
169 	pthread_mutex_lock(&tgt->mutex);
170 	TAILQ_REMOVE(&tgt->poll_groups, group, link);
171 	pthread_mutex_unlock(&tgt->mutex);
172 
173 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
174 		TAILQ_REMOVE(&group->tgroups, tgroup, link);
175 		nvmf_transport_poll_group_destroy(tgroup);
176 	}
177 
178 	for (sid = 0; sid < group->num_sgroups; sid++) {
179 		sgroup = &group->sgroups[sid];
180 
181 		for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
182 			if (sgroup->ns_info[nsid].channel) {
183 				spdk_put_io_channel(sgroup->ns_info[nsid].channel);
184 				sgroup->ns_info[nsid].channel = NULL;
185 			}
186 		}
187 
188 		free(sgroup->ns_info);
189 	}
190 
191 	free(group->sgroups);
192 
193 	spdk_poller_unregister(&group->poller);
194 
195 	if (group->destroy_cb_fn) {
196 		group->destroy_cb_fn(group->destroy_cb_arg, 0);
197 	}
198 }
199 
200 static void
201 _nvmf_tgt_disconnect_next_qpair(void *ctx)
202 {
203 	struct spdk_nvmf_qpair *qpair;
204 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
205 	struct spdk_nvmf_poll_group *group = qpair_ctx->group;
206 	struct spdk_io_channel *ch;
207 	int rc = 0;
208 
209 	qpair = TAILQ_FIRST(&group->qpairs);
210 
211 	if (qpair) {
212 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_tgt_disconnect_next_qpair, ctx);
213 	}
214 
215 	if (!qpair || rc != 0) {
216 		/* When the refcount from the channels reaches 0, nvmf_tgt_destroy_poll_group will be called. */
217 		ch = spdk_io_channel_from_ctx(group);
218 		spdk_put_io_channel(ch);
219 		free(qpair_ctx);
220 	}
221 }
222 
223 static void
224 nvmf_tgt_destroy_poll_group_qpairs(struct spdk_nvmf_poll_group *group)
225 {
226 	struct nvmf_qpair_disconnect_many_ctx *ctx;
227 
228 	SPDK_DTRACE_PROBE1(nvmf_destroy_poll_group_qpairs, spdk_thread_get_id(group->thread));
229 
230 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
231 	if (!ctx) {
232 		SPDK_ERRLOG("Failed to allocate memory for destroy poll group ctx\n");
233 		return;
234 	}
235 
236 	ctx->group = group;
237 	_nvmf_tgt_disconnect_next_qpair(ctx);
238 }
239 
240 static int
241 nvmf_tgt_accept(void *ctx)
242 {
243 	struct spdk_nvmf_tgt *tgt = ctx;
244 	struct spdk_nvmf_transport *transport, *tmp;
245 	int count = 0;
246 
247 	TAILQ_FOREACH_SAFE(transport, &tgt->transports, link, tmp) {
248 		count += nvmf_transport_accept(transport);
249 	}
250 
251 	return count;
252 }
253 
254 struct spdk_nvmf_tgt *
255 spdk_nvmf_tgt_create(struct spdk_nvmf_target_opts *opts)
256 {
257 	struct spdk_nvmf_tgt *tgt, *tmp_tgt;
258 	uint32_t acceptor_poll_rate;
259 
260 	if (strnlen(opts->name, NVMF_TGT_NAME_MAX_LENGTH) == NVMF_TGT_NAME_MAX_LENGTH) {
261 		SPDK_ERRLOG("Provided target name exceeds the max length of %u.\n", NVMF_TGT_NAME_MAX_LENGTH);
262 		return NULL;
263 	}
264 
265 	TAILQ_FOREACH(tmp_tgt, &g_nvmf_tgts, link) {
266 		if (!strncmp(opts->name, tmp_tgt->name, NVMF_TGT_NAME_MAX_LENGTH)) {
267 			SPDK_ERRLOG("Provided target name must be unique.\n");
268 			return NULL;
269 		}
270 	}
271 
272 	tgt = calloc(1, sizeof(*tgt));
273 	if (!tgt) {
274 		return NULL;
275 	}
276 
277 	snprintf(tgt->name, NVMF_TGT_NAME_MAX_LENGTH, "%s", opts->name);
278 
279 	if (!opts || !opts->max_subsystems) {
280 		tgt->max_subsystems = SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS;
281 	} else {
282 		tgt->max_subsystems = opts->max_subsystems;
283 	}
284 
285 	if (!opts || !opts->acceptor_poll_rate) {
286 		acceptor_poll_rate = SPDK_NVMF_DEFAULT_ACCEPT_POLL_RATE_US;
287 	} else {
288 		acceptor_poll_rate = opts->acceptor_poll_rate;
289 	}
290 
291 	if (!opts) {
292 		tgt->crdt[0] = 0;
293 		tgt->crdt[1] = 0;
294 		tgt->crdt[2] = 0;
295 	} else {
296 		tgt->crdt[0] = opts->crdt[0];
297 		tgt->crdt[1] = opts->crdt[1];
298 		tgt->crdt[2] = opts->crdt[2];
299 	}
300 
301 	tgt->discovery_genctr = 0;
302 	TAILQ_INIT(&tgt->transports);
303 	TAILQ_INIT(&tgt->poll_groups);
304 
305 	tgt->subsystems = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem *));
306 	if (!tgt->subsystems) {
307 		free(tgt);
308 		return NULL;
309 	}
310 
311 	pthread_mutex_init(&tgt->mutex, NULL);
312 
313 	tgt->accept_poller = SPDK_POLLER_REGISTER(nvmf_tgt_accept, tgt, acceptor_poll_rate);
314 	if (!tgt->accept_poller) {
315 		pthread_mutex_destroy(&tgt->mutex);
316 		free(tgt->subsystems);
317 		free(tgt);
318 		return NULL;
319 	}
320 
321 	spdk_io_device_register(tgt,
322 				nvmf_tgt_create_poll_group,
323 				nvmf_tgt_destroy_poll_group,
324 				sizeof(struct spdk_nvmf_poll_group),
325 				tgt->name);
326 
327 	TAILQ_INSERT_HEAD(&g_nvmf_tgts, tgt, link);
328 
329 	return tgt;
330 }
331 
332 static void
333 _nvmf_tgt_destroy_next_transport(void *ctx)
334 {
335 	struct spdk_nvmf_tgt *tgt = ctx;
336 	struct spdk_nvmf_transport *transport;
337 
338 	if (!TAILQ_EMPTY(&tgt->transports)) {
339 		transport = TAILQ_FIRST(&tgt->transports);
340 		TAILQ_REMOVE(&tgt->transports, transport, link);
341 		spdk_nvmf_transport_destroy(transport, _nvmf_tgt_destroy_next_transport, tgt);
342 	} else {
343 		spdk_nvmf_tgt_destroy_done_fn *destroy_cb_fn = tgt->destroy_cb_fn;
344 		void *destroy_cb_arg = tgt->destroy_cb_arg;
345 
346 		pthread_mutex_destroy(&tgt->mutex);
347 		free(tgt);
348 
349 		if (destroy_cb_fn) {
350 			destroy_cb_fn(destroy_cb_arg, 0);
351 		}
352 	}
353 }
354 
355 static void
356 nvmf_tgt_destroy_cb(void *io_device)
357 {
358 	struct spdk_nvmf_tgt *tgt = io_device;
359 	uint32_t i;
360 
361 	if (tgt->subsystems) {
362 		for (i = 0; i < tgt->max_subsystems; i++) {
363 			if (tgt->subsystems[i]) {
364 				nvmf_subsystem_remove_all_listeners(tgt->subsystems[i], true);
365 				spdk_nvmf_subsystem_destroy(tgt->subsystems[i]);
366 			}
367 		}
368 		free(tgt->subsystems);
369 	}
370 
371 	_nvmf_tgt_destroy_next_transport(tgt);
372 }
373 
374 void
375 spdk_nvmf_tgt_destroy(struct spdk_nvmf_tgt *tgt,
376 		      spdk_nvmf_tgt_destroy_done_fn cb_fn,
377 		      void *cb_arg)
378 {
379 	tgt->destroy_cb_fn = cb_fn;
380 	tgt->destroy_cb_arg = cb_arg;
381 
382 	spdk_poller_unregister(&tgt->accept_poller);
383 
384 	TAILQ_REMOVE(&g_nvmf_tgts, tgt, link);
385 
386 	spdk_io_device_unregister(tgt, nvmf_tgt_destroy_cb);
387 }
388 
389 const char *
390 spdk_nvmf_tgt_get_name(struct spdk_nvmf_tgt *tgt)
391 {
392 	return tgt->name;
393 }
394 
395 struct spdk_nvmf_tgt *
396 spdk_nvmf_get_tgt(const char *name)
397 {
398 	struct spdk_nvmf_tgt *tgt;
399 	uint32_t num_targets = 0;
400 
401 	TAILQ_FOREACH(tgt, &g_nvmf_tgts, link) {
402 		if (name) {
403 			if (!strncmp(tgt->name, name, NVMF_TGT_NAME_MAX_LENGTH)) {
404 				return tgt;
405 			}
406 		}
407 		num_targets++;
408 	}
409 
410 	/*
411 	 * special case. If there is only one target and
412 	 * no name was specified, return the only available
413 	 * target. If there is more than one target, name must
414 	 * be specified.
415 	 */
416 	if (!name && num_targets == 1) {
417 		return TAILQ_FIRST(&g_nvmf_tgts);
418 	}
419 
420 	return NULL;
421 }
422 
423 struct spdk_nvmf_tgt *
424 spdk_nvmf_get_first_tgt(void)
425 {
426 	return TAILQ_FIRST(&g_nvmf_tgts);
427 }
428 
429 struct spdk_nvmf_tgt *
430 spdk_nvmf_get_next_tgt(struct spdk_nvmf_tgt *prev)
431 {
432 	return TAILQ_NEXT(prev, link);
433 }
434 
435 static void
436 nvmf_write_subsystem_config_json(struct spdk_json_write_ctx *w,
437 				 struct spdk_nvmf_subsystem *subsystem)
438 {
439 	struct spdk_nvmf_host *host;
440 	struct spdk_nvmf_subsystem_listener *listener;
441 	const struct spdk_nvme_transport_id *trid;
442 	struct spdk_nvmf_ns *ns;
443 	struct spdk_nvmf_ns_opts ns_opts;
444 	uint32_t max_namespaces;
445 	char uuid_str[SPDK_UUID_STRING_LEN];
446 	const char *adrfam;
447 
448 	if (spdk_nvmf_subsystem_get_type(subsystem) != SPDK_NVMF_SUBTYPE_NVME) {
449 		return;
450 	}
451 
452 	/* { */
453 	spdk_json_write_object_begin(w);
454 	spdk_json_write_named_string(w, "method", "nvmf_create_subsystem");
455 
456 	/*     "params" : { */
457 	spdk_json_write_named_object_begin(w, "params");
458 	spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
459 	spdk_json_write_named_bool(w, "allow_any_host", spdk_nvmf_subsystem_get_allow_any_host(subsystem));
460 	spdk_json_write_named_string(w, "serial_number", spdk_nvmf_subsystem_get_sn(subsystem));
461 	spdk_json_write_named_string(w, "model_number", spdk_nvmf_subsystem_get_mn(subsystem));
462 
463 	max_namespaces = spdk_nvmf_subsystem_get_max_namespaces(subsystem);
464 	if (max_namespaces != 0) {
465 		spdk_json_write_named_uint32(w, "max_namespaces", max_namespaces);
466 	}
467 
468 	spdk_json_write_named_uint32(w, "min_cntlid", spdk_nvmf_subsystem_get_min_cntlid(subsystem));
469 	spdk_json_write_named_uint32(w, "max_cntlid", spdk_nvmf_subsystem_get_max_cntlid(subsystem));
470 
471 	/*     } "params" */
472 	spdk_json_write_object_end(w);
473 
474 	/* } */
475 	spdk_json_write_object_end(w);
476 
477 	for (listener = spdk_nvmf_subsystem_get_first_listener(subsystem); listener != NULL;
478 	     listener = spdk_nvmf_subsystem_get_next_listener(subsystem, listener)) {
479 		trid = spdk_nvmf_subsystem_listener_get_trid(listener);
480 
481 		adrfam = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
482 
483 		spdk_json_write_object_begin(w);
484 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_listener");
485 
486 		/*     "params" : { */
487 		spdk_json_write_named_object_begin(w, "params");
488 
489 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
490 
491 		/*     "listen_address" : { */
492 		spdk_json_write_named_object_begin(w, "listen_address");
493 
494 		spdk_json_write_named_string(w, "trtype", trid->trstring);
495 		if (adrfam) {
496 			spdk_json_write_named_string(w, "adrfam", adrfam);
497 		}
498 
499 		spdk_json_write_named_string(w, "traddr", trid->traddr);
500 		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
501 		/*     } "listen_address" */
502 		spdk_json_write_object_end(w);
503 
504 		/*     } "params" */
505 		spdk_json_write_object_end(w);
506 
507 		/* } */
508 		spdk_json_write_object_end(w);
509 	}
510 
511 	for (host = spdk_nvmf_subsystem_get_first_host(subsystem); host != NULL;
512 	     host = spdk_nvmf_subsystem_get_next_host(subsystem, host)) {
513 
514 		spdk_json_write_object_begin(w);
515 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_host");
516 
517 		/*     "params" : { */
518 		spdk_json_write_named_object_begin(w, "params");
519 
520 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
521 		spdk_json_write_named_string(w, "host", spdk_nvmf_host_get_nqn(host));
522 
523 		/*     } "params" */
524 		spdk_json_write_object_end(w);
525 
526 		/* } */
527 		spdk_json_write_object_end(w);
528 	}
529 
530 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
531 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
532 		spdk_nvmf_ns_get_opts(ns, &ns_opts, sizeof(ns_opts));
533 
534 		spdk_json_write_object_begin(w);
535 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_ns");
536 
537 		/*     "params" : { */
538 		spdk_json_write_named_object_begin(w, "params");
539 
540 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
541 
542 		/*     "namespace" : { */
543 		spdk_json_write_named_object_begin(w, "namespace");
544 
545 		spdk_json_write_named_uint32(w, "nsid", spdk_nvmf_ns_get_id(ns));
546 		spdk_json_write_named_string(w, "bdev_name", spdk_bdev_get_name(spdk_nvmf_ns_get_bdev(ns)));
547 
548 		if (!spdk_mem_all_zero(ns_opts.nguid, sizeof(ns_opts.nguid))) {
549 			SPDK_STATIC_ASSERT(sizeof(ns_opts.nguid) == sizeof(uint64_t) * 2, "size mismatch");
550 			spdk_json_write_named_string_fmt(w, "nguid", "%016"PRIX64"%016"PRIX64, from_be64(&ns_opts.nguid[0]),
551 							 from_be64(&ns_opts.nguid[8]));
552 		}
553 
554 		if (!spdk_mem_all_zero(ns_opts.eui64, sizeof(ns_opts.eui64))) {
555 			SPDK_STATIC_ASSERT(sizeof(ns_opts.eui64) == sizeof(uint64_t), "size mismatch");
556 			spdk_json_write_named_string_fmt(w, "eui64", "%016"PRIX64, from_be64(&ns_opts.eui64));
557 		}
558 
559 		if (!spdk_mem_all_zero(&ns_opts.uuid, sizeof(ns_opts.uuid))) {
560 			spdk_uuid_fmt_lower(uuid_str, sizeof(uuid_str), &ns_opts.uuid);
561 			spdk_json_write_named_string(w, "uuid",  uuid_str);
562 		}
563 
564 		/*     "namespace" */
565 		spdk_json_write_object_end(w);
566 
567 		/*     } "params" */
568 		spdk_json_write_object_end(w);
569 
570 		/* } */
571 		spdk_json_write_object_end(w);
572 	}
573 }
574 
575 void
576 spdk_nvmf_tgt_write_config_json(struct spdk_json_write_ctx *w, struct spdk_nvmf_tgt *tgt)
577 {
578 	struct spdk_nvmf_subsystem *subsystem;
579 	struct spdk_nvmf_transport *transport;
580 
581 	spdk_json_write_object_begin(w);
582 	spdk_json_write_named_string(w, "method", "nvmf_set_max_subsystems");
583 
584 	spdk_json_write_named_object_begin(w, "params");
585 	spdk_json_write_named_uint32(w, "max_subsystems", tgt->max_subsystems);
586 	spdk_json_write_object_end(w);
587 
588 	spdk_json_write_object_end(w);
589 
590 	spdk_json_write_object_begin(w);
591 	spdk_json_write_named_string(w, "method", "nvmf_set_crdt");
592 	spdk_json_write_named_object_begin(w, "params");
593 	spdk_json_write_named_uint32(w, "crdt1", tgt->crdt[0]);
594 	spdk_json_write_named_uint32(w, "crdt2", tgt->crdt[1]);
595 	spdk_json_write_named_uint32(w, "crdt3", tgt->crdt[2]);
596 	spdk_json_write_object_end(w);
597 	spdk_json_write_object_end(w);
598 
599 	/* write transports */
600 	TAILQ_FOREACH(transport, &tgt->transports, link) {
601 		spdk_json_write_object_begin(w);
602 		spdk_json_write_named_string(w, "method", "nvmf_create_transport");
603 
604 		spdk_json_write_named_object_begin(w, "params");
605 		spdk_json_write_named_string(w, "trtype", transport->ops->name);
606 		spdk_json_write_named_uint32(w, "max_queue_depth", transport->opts.max_queue_depth);
607 		spdk_json_write_named_uint32(w, "max_io_qpairs_per_ctrlr",
608 					     transport->opts.max_qpairs_per_ctrlr - 1);
609 		spdk_json_write_named_uint32(w, "in_capsule_data_size", transport->opts.in_capsule_data_size);
610 		spdk_json_write_named_uint32(w, "max_io_size", transport->opts.max_io_size);
611 		spdk_json_write_named_uint32(w, "io_unit_size", transport->opts.io_unit_size);
612 		spdk_json_write_named_uint32(w, "max_aq_depth", transport->opts.max_aq_depth);
613 		spdk_json_write_named_uint32(w, "num_shared_buffers", transport->opts.num_shared_buffers);
614 		spdk_json_write_named_uint32(w, "buf_cache_size", transport->opts.buf_cache_size);
615 		spdk_json_write_named_bool(w, "dif_insert_or_strip", transport->opts.dif_insert_or_strip);
616 		if (transport->ops->dump_opts) {
617 			transport->ops->dump_opts(transport, w);
618 		}
619 		spdk_json_write_named_uint32(w, "abort_timeout_sec", transport->opts.abort_timeout_sec);
620 		spdk_json_write_object_end(w);
621 
622 		spdk_json_write_object_end(w);
623 	}
624 
625 	subsystem = spdk_nvmf_subsystem_get_first(tgt);
626 	while (subsystem) {
627 		nvmf_write_subsystem_config_json(w, subsystem);
628 		subsystem = spdk_nvmf_subsystem_get_next(subsystem);
629 	}
630 }
631 
632 static void
633 nvmf_listen_opts_copy(struct spdk_nvmf_listen_opts *opts,
634 		      const struct spdk_nvmf_listen_opts *opts_src, size_t opts_size)
635 {
636 	assert(opts);
637 	assert(opts_src);
638 
639 	opts->opts_size = opts_size;
640 
641 #define SET_FIELD(field) \
642     if (offsetof(struct spdk_nvmf_listen_opts, field) + sizeof(opts->field) <= opts_size) { \
643                  opts->field = opts_src->field; \
644     } \
645 
646 	SET_FIELD(transport_specific);
647 #undef SET_FIELD
648 
649 	/* Do not remove this statement, you should always update this statement when you adding a new field,
650 	 * and do not forget to add the SET_FIELD statement for your added field. */
651 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listen_opts) == 16, "Incorrect size");
652 }
653 
654 void
655 spdk_nvmf_listen_opts_init(struct spdk_nvmf_listen_opts *opts, size_t opts_size)
656 {
657 	struct spdk_nvmf_listen_opts opts_local = {};
658 
659 	/* local version of opts should have defaults set here */
660 
661 	nvmf_listen_opts_copy(opts, &opts_local, opts_size);
662 }
663 
664 int
665 spdk_nvmf_tgt_listen_ext(struct spdk_nvmf_tgt *tgt, const struct spdk_nvme_transport_id *trid,
666 			 struct spdk_nvmf_listen_opts *opts)
667 {
668 	struct spdk_nvmf_transport *transport;
669 	int rc;
670 	struct spdk_nvmf_listen_opts opts_local = {};
671 
672 	if (!opts) {
673 		SPDK_ERRLOG("opts should not be NULL\n");
674 		return -EINVAL;
675 	}
676 
677 	if (!opts->opts_size) {
678 		SPDK_ERRLOG("The opts_size in opts structure should not be zero\n");
679 		return -EINVAL;
680 	}
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 	nvmf_listen_opts_copy(&opts_local, opts, opts->opts_size);
690 	rc = spdk_nvmf_transport_listen(transport, trid, &opts_local);
691 	if (rc < 0) {
692 		SPDK_ERRLOG("Unable to listen on address '%s'\n", trid->traddr);
693 	}
694 
695 	return rc;
696 }
697 
698 int
699 spdk_nvmf_tgt_stop_listen(struct spdk_nvmf_tgt *tgt,
700 			  struct spdk_nvme_transport_id *trid)
701 {
702 	struct spdk_nvmf_transport *transport;
703 	int rc;
704 
705 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
706 	if (!transport) {
707 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
708 			    trid->trstring);
709 		return -EINVAL;
710 	}
711 
712 	rc = spdk_nvmf_transport_stop_listen(transport, trid);
713 	if (rc < 0) {
714 		SPDK_ERRLOG("Failed to stop listening on address '%s'\n", trid->traddr);
715 		return rc;
716 	}
717 	return 0;
718 }
719 
720 struct spdk_nvmf_tgt_add_transport_ctx {
721 	struct spdk_nvmf_tgt *tgt;
722 	struct spdk_nvmf_transport *transport;
723 	spdk_nvmf_tgt_add_transport_done_fn cb_fn;
724 	void *cb_arg;
725 };
726 
727 static void
728 _nvmf_tgt_add_transport_done(struct spdk_io_channel_iter *i, int status)
729 {
730 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
731 
732 	ctx->cb_fn(ctx->cb_arg, status);
733 
734 	free(ctx);
735 }
736 
737 static void
738 _nvmf_tgt_add_transport(struct spdk_io_channel_iter *i)
739 {
740 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
741 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
742 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
743 	int rc;
744 
745 	rc = nvmf_poll_group_add_transport(group, ctx->transport);
746 	spdk_for_each_channel_continue(i, rc);
747 }
748 
749 void spdk_nvmf_tgt_add_transport(struct spdk_nvmf_tgt *tgt,
750 				 struct spdk_nvmf_transport *transport,
751 				 spdk_nvmf_tgt_add_transport_done_fn cb_fn,
752 				 void *cb_arg)
753 {
754 	struct spdk_nvmf_tgt_add_transport_ctx *ctx;
755 
756 	if (spdk_nvmf_tgt_get_transport(tgt, transport->ops->name)) {
757 		cb_fn(cb_arg, -EEXIST);
758 		return; /* transport already created */
759 	}
760 
761 	transport->tgt = tgt;
762 	TAILQ_INSERT_TAIL(&tgt->transports, transport, link);
763 
764 	ctx = calloc(1, sizeof(*ctx));
765 	if (!ctx) {
766 		cb_fn(cb_arg, -ENOMEM);
767 		return;
768 	}
769 
770 	ctx->tgt = tgt;
771 	ctx->transport = transport;
772 	ctx->cb_fn = cb_fn;
773 	ctx->cb_arg = cb_arg;
774 
775 	spdk_for_each_channel(tgt,
776 			      _nvmf_tgt_add_transport,
777 			      ctx,
778 			      _nvmf_tgt_add_transport_done);
779 }
780 
781 struct spdk_nvmf_subsystem *
782 spdk_nvmf_tgt_find_subsystem(struct spdk_nvmf_tgt *tgt, const char *subnqn)
783 {
784 	struct spdk_nvmf_subsystem	*subsystem;
785 	uint32_t sid;
786 
787 	if (!subnqn) {
788 		return NULL;
789 	}
790 
791 	/* Ensure that subnqn is null terminated */
792 	if (!memchr(subnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
793 		SPDK_ERRLOG("Connect SUBNQN is not null terminated\n");
794 		return NULL;
795 	}
796 
797 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
798 		subsystem = tgt->subsystems[sid];
799 		if (subsystem == NULL) {
800 			continue;
801 		}
802 
803 		if (strcmp(subnqn, subsystem->subnqn) == 0) {
804 			return subsystem;
805 		}
806 	}
807 
808 	return NULL;
809 }
810 
811 struct spdk_nvmf_transport *
812 spdk_nvmf_tgt_get_transport(struct spdk_nvmf_tgt *tgt, const char *transport_name)
813 {
814 	struct spdk_nvmf_transport *transport;
815 
816 	TAILQ_FOREACH(transport, &tgt->transports, link) {
817 		if (!strncasecmp(transport->ops->name, transport_name, SPDK_NVMF_TRSTRING_MAX_LEN)) {
818 			return transport;
819 		}
820 	}
821 	return NULL;
822 }
823 
824 struct nvmf_new_qpair_ctx {
825 	struct spdk_nvmf_qpair *qpair;
826 	struct spdk_nvmf_poll_group *group;
827 };
828 
829 static void
830 _nvmf_poll_group_add(void *_ctx)
831 {
832 	struct nvmf_new_qpair_ctx *ctx = _ctx;
833 	struct spdk_nvmf_qpair *qpair = ctx->qpair;
834 	struct spdk_nvmf_poll_group *group = ctx->group;
835 
836 	free(_ctx);
837 
838 	if (spdk_nvmf_poll_group_add(group, qpair) != 0) {
839 		SPDK_ERRLOG("Unable to add the qpair to a poll group.\n");
840 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
841 	}
842 }
843 
844 void
845 spdk_nvmf_tgt_new_qpair(struct spdk_nvmf_tgt *tgt, struct spdk_nvmf_qpair *qpair)
846 {
847 	struct spdk_nvmf_poll_group *group;
848 	struct nvmf_new_qpair_ctx *ctx;
849 
850 	group = spdk_nvmf_get_optimal_poll_group(qpair);
851 	if (group == NULL) {
852 		if (tgt->next_poll_group == NULL) {
853 			tgt->next_poll_group = TAILQ_FIRST(&tgt->poll_groups);
854 			if (tgt->next_poll_group == NULL) {
855 				SPDK_ERRLOG("No poll groups exist.\n");
856 				spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
857 				return;
858 			}
859 		}
860 		group = tgt->next_poll_group;
861 		tgt->next_poll_group = TAILQ_NEXT(group, link);
862 	}
863 
864 	ctx = calloc(1, sizeof(*ctx));
865 	if (!ctx) {
866 		SPDK_ERRLOG("Unable to send message to poll group.\n");
867 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
868 		return;
869 	}
870 
871 	ctx->qpair = qpair;
872 	ctx->group = group;
873 
874 	spdk_thread_send_msg(group->thread, _nvmf_poll_group_add, ctx);
875 }
876 
877 struct spdk_nvmf_poll_group *
878 spdk_nvmf_poll_group_create(struct spdk_nvmf_tgt *tgt)
879 {
880 	struct spdk_io_channel *ch;
881 
882 	ch = spdk_get_io_channel(tgt);
883 	if (!ch) {
884 		SPDK_ERRLOG("Unable to get I/O channel for target\n");
885 		return NULL;
886 	}
887 
888 	return spdk_io_channel_get_ctx(ch);
889 }
890 
891 void
892 spdk_nvmf_poll_group_destroy(struct spdk_nvmf_poll_group *group,
893 			     spdk_nvmf_poll_group_destroy_done_fn cb_fn,
894 			     void *cb_arg)
895 {
896 	assert(group->destroy_cb_fn == NULL);
897 	group->destroy_cb_fn = cb_fn;
898 	group->destroy_cb_arg = cb_arg;
899 
900 	/* This function will put the io_channel associated with this poll group */
901 	nvmf_tgt_destroy_poll_group_qpairs(group);
902 }
903 
904 int
905 spdk_nvmf_poll_group_add(struct spdk_nvmf_poll_group *group,
906 			 struct spdk_nvmf_qpair *qpair)
907 {
908 	int rc = -1;
909 	struct spdk_nvmf_transport_poll_group *tgroup;
910 
911 	TAILQ_INIT(&qpair->outstanding);
912 	qpair->group = group;
913 	qpair->ctrlr = NULL;
914 	qpair->disconnect_started = false;
915 
916 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
917 		if (tgroup->transport == qpair->transport) {
918 			rc = nvmf_transport_poll_group_add(tgroup, qpair);
919 			break;
920 		}
921 	}
922 
923 	/* We add the qpair to the group only it is succesfully added into the tgroup */
924 	if (rc == 0) {
925 		SPDK_DTRACE_PROBE2(nvmf_poll_group_add_qpair, qpair, spdk_thread_get_id(group->thread));
926 		TAILQ_INSERT_TAIL(&group->qpairs, qpair, link);
927 		nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ACTIVE);
928 	}
929 
930 	return rc;
931 }
932 
933 static void
934 _nvmf_ctrlr_destruct(void *ctx)
935 {
936 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
937 
938 	nvmf_ctrlr_destruct(ctrlr);
939 }
940 
941 static void
942 _nvmf_transport_qpair_fini_complete(void *cb_ctx)
943 {
944 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = cb_ctx;
945 
946 	if (qpair_ctx->cb_fn) {
947 		spdk_thread_send_msg(qpair_ctx->thread, qpair_ctx->cb_fn, qpair_ctx->ctx);
948 	}
949 	free(qpair_ctx);
950 }
951 
952 static void
953 _nvmf_transport_qpair_fini(void *ctx)
954 {
955 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
956 
957 	spdk_nvmf_poll_group_remove(qpair_ctx->qpair);
958 	nvmf_transport_qpair_fini(qpair_ctx->qpair, _nvmf_transport_qpair_fini_complete, qpair_ctx);
959 }
960 
961 static void
962 _nvmf_ctrlr_free_from_qpair(void *ctx)
963 {
964 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
965 	struct spdk_nvmf_ctrlr *ctrlr = qpair_ctx->ctrlr;
966 	uint32_t count;
967 
968 	spdk_bit_array_clear(ctrlr->qpair_mask, qpair_ctx->qid);
969 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
970 	if (count == 0) {
971 		ctrlr->in_destruct = true;
972 		spdk_thread_send_msg(ctrlr->subsys->thread, _nvmf_ctrlr_destruct, ctrlr);
973 	}
974 
975 	spdk_thread_send_msg(qpair_ctx->thread, _nvmf_transport_qpair_fini, qpair_ctx);
976 }
977 
978 void
979 spdk_nvmf_poll_group_remove(struct spdk_nvmf_qpair *qpair)
980 {
981 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
982 	struct spdk_nvmf_transport_poll_group *tgroup;
983 	struct spdk_nvmf_request *req, *tmp;
984 	struct spdk_nvmf_subsystem_poll_group *sgroup;
985 	int rc;
986 
987 	SPDK_DTRACE_PROBE2(nvmf_poll_group_remove_qpair, qpair,
988 			   spdk_thread_get_id(qpair->group->thread));
989 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ERROR);
990 
991 	/* Find the tgroup and remove the qpair from the tgroup */
992 	TAILQ_FOREACH(tgroup, &qpair->group->tgroups, link) {
993 		if (tgroup->transport == qpair->transport) {
994 			rc = nvmf_transport_poll_group_remove(tgroup, qpair);
995 			if (rc && (rc != ENOTSUP)) {
996 				SPDK_ERRLOG("Cannot remove qpair=%p from transport group=%p\n",
997 					    qpair, tgroup);
998 			}
999 			break;
1000 		}
1001 	}
1002 
1003 	if (ctrlr) {
1004 		sgroup = &qpair->group->sgroups[ctrlr->subsys->id];
1005 		TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1006 			if (req->qpair == qpair) {
1007 				TAILQ_REMOVE(&sgroup->queued, req, link);
1008 				if (nvmf_transport_req_free(req)) {
1009 					SPDK_ERRLOG("Transport request free error!\n");
1010 				}
1011 			}
1012 		}
1013 	}
1014 
1015 	TAILQ_REMOVE(&qpair->group->qpairs, qpair, link);
1016 	qpair->group = NULL;
1017 }
1018 
1019 static void
1020 _nvmf_qpair_destroy(void *ctx, int status)
1021 {
1022 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1023 	struct spdk_nvmf_qpair *qpair = qpair_ctx->qpair;
1024 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
1025 
1026 	assert(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
1027 	qpair_ctx->qid = qpair->qid;
1028 
1029 	if (ctrlr) {
1030 		if (0 == qpair->qid) {
1031 			assert(qpair->group->stat.current_admin_qpairs > 0);
1032 			qpair->group->stat.current_admin_qpairs--;
1033 		} else {
1034 			assert(qpair->group->stat.current_io_qpairs > 0);
1035 			qpair->group->stat.current_io_qpairs--;
1036 		}
1037 	}
1038 
1039 	if (!ctrlr || !ctrlr->thread) {
1040 		spdk_nvmf_poll_group_remove(qpair);
1041 		nvmf_transport_qpair_fini(qpair, _nvmf_transport_qpair_fini_complete, qpair_ctx);
1042 		return;
1043 	}
1044 
1045 	qpair_ctx->ctrlr = ctrlr;
1046 	spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_free_from_qpair, qpair_ctx);
1047 }
1048 
1049 static void
1050 _nvmf_qpair_disconnect_msg(void *ctx)
1051 {
1052 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1053 
1054 	spdk_nvmf_qpair_disconnect(qpair_ctx->qpair, qpair_ctx->cb_fn, qpair_ctx->ctx);
1055 	free(ctx);
1056 }
1057 
1058 int
1059 spdk_nvmf_qpair_disconnect(struct spdk_nvmf_qpair *qpair, nvmf_qpair_disconnect_cb cb_fn, void *ctx)
1060 {
1061 	struct spdk_nvmf_poll_group *group = qpair->group;
1062 	struct nvmf_qpair_disconnect_ctx *qpair_ctx;
1063 
1064 	if (__atomic_test_and_set(&qpair->disconnect_started, __ATOMIC_RELAXED)) {
1065 		if (cb_fn) {
1066 			cb_fn(ctx);
1067 		}
1068 		return 0;
1069 	}
1070 
1071 	/* If we get a qpair in the uninitialized state, we can just destroy it immediately */
1072 	if (qpair->state == SPDK_NVMF_QPAIR_UNINITIALIZED) {
1073 		nvmf_transport_qpair_fini(qpair, NULL, NULL);
1074 		if (cb_fn) {
1075 			cb_fn(ctx);
1076 		}
1077 		return 0;
1078 	}
1079 
1080 	assert(group != NULL);
1081 	if (spdk_get_thread() != group->thread) {
1082 		/* clear the atomic so we can set it on the next call on the proper thread. */
1083 		__atomic_clear(&qpair->disconnect_started, __ATOMIC_RELAXED);
1084 		qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1085 		if (!qpair_ctx) {
1086 			SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1087 			return -ENOMEM;
1088 		}
1089 		qpair_ctx->qpair = qpair;
1090 		qpair_ctx->cb_fn = cb_fn;
1091 		qpair_ctx->thread = group->thread;
1092 		qpair_ctx->ctx = ctx;
1093 		spdk_thread_send_msg(group->thread, _nvmf_qpair_disconnect_msg, qpair_ctx);
1094 		return 0;
1095 	}
1096 
1097 	SPDK_DTRACE_PROBE2(nvmf_qpair_disconnect, qpair, spdk_thread_get_id(group->thread));
1098 	assert(qpair->state == SPDK_NVMF_QPAIR_ACTIVE);
1099 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_DEACTIVATING);
1100 
1101 	qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1102 	if (!qpair_ctx) {
1103 		SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1104 		return -ENOMEM;
1105 	}
1106 
1107 	qpair_ctx->qpair = qpair;
1108 	qpair_ctx->cb_fn = cb_fn;
1109 	qpair_ctx->thread = group->thread;
1110 	qpair_ctx->ctx = ctx;
1111 
1112 	/* Check for outstanding I/O */
1113 	if (!TAILQ_EMPTY(&qpair->outstanding)) {
1114 		SPDK_DTRACE_PROBE2(nvmf_poll_group_drain_qpair, qpair, spdk_thread_get_id(group->thread));
1115 		qpair->state_cb = _nvmf_qpair_destroy;
1116 		qpair->state_cb_arg = qpair_ctx;
1117 		nvmf_qpair_free_aer(qpair);
1118 		return 0;
1119 	}
1120 
1121 	_nvmf_qpair_destroy(qpair_ctx, 0);
1122 
1123 	return 0;
1124 }
1125 
1126 int
1127 spdk_nvmf_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
1128 			      struct spdk_nvme_transport_id *trid)
1129 {
1130 	return nvmf_transport_qpair_get_peer_trid(qpair, trid);
1131 }
1132 
1133 int
1134 spdk_nvmf_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
1135 			       struct spdk_nvme_transport_id *trid)
1136 {
1137 	return nvmf_transport_qpair_get_local_trid(qpair, trid);
1138 }
1139 
1140 int
1141 spdk_nvmf_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
1142 				struct spdk_nvme_transport_id *trid)
1143 {
1144 	return nvmf_transport_qpair_get_listen_trid(qpair, trid);
1145 }
1146 
1147 int
1148 nvmf_poll_group_add_transport(struct spdk_nvmf_poll_group *group,
1149 			      struct spdk_nvmf_transport *transport)
1150 {
1151 	struct spdk_nvmf_transport_poll_group *tgroup;
1152 
1153 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1154 		if (tgroup->transport == transport) {
1155 			/* Transport already in the poll group */
1156 			return 0;
1157 		}
1158 	}
1159 
1160 	tgroup = nvmf_transport_poll_group_create(transport);
1161 	if (!tgroup) {
1162 		SPDK_ERRLOG("Unable to create poll group for transport\n");
1163 		return -1;
1164 	}
1165 
1166 	tgroup->group = group;
1167 	TAILQ_INSERT_TAIL(&group->tgroups, tgroup, link);
1168 
1169 	return 0;
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 			if (ctrlr->admin_qpair->group == group) {
1322 				nvmf_ctrlr_async_event_ns_notice(ctrlr);
1323 				nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
1324 			}
1325 		}
1326 	}
1327 
1328 	return 0;
1329 }
1330 
1331 int
1332 nvmf_poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1333 				 struct spdk_nvmf_subsystem *subsystem)
1334 {
1335 	return poll_group_update_subsystem(group, subsystem);
1336 }
1337 
1338 int
1339 nvmf_poll_group_add_subsystem(struct spdk_nvmf_poll_group *group,
1340 			      struct spdk_nvmf_subsystem *subsystem,
1341 			      spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1342 {
1343 	int rc = 0;
1344 	struct spdk_nvmf_subsystem_poll_group *sgroup = &group->sgroups[subsystem->id];
1345 	uint32_t i;
1346 
1347 	TAILQ_INIT(&sgroup->queued);
1348 
1349 	rc = poll_group_update_subsystem(group, subsystem);
1350 	if (rc) {
1351 		nvmf_poll_group_remove_subsystem(group, subsystem, NULL, NULL);
1352 		goto fini;
1353 	}
1354 
1355 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1356 
1357 	for (i = 0; i < sgroup->num_ns; i++) {
1358 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1359 	}
1360 
1361 fini:
1362 	if (cb_fn) {
1363 		cb_fn(cb_arg, rc);
1364 	}
1365 
1366 	return rc;
1367 }
1368 
1369 static void
1370 _nvmf_poll_group_remove_subsystem_cb(void *ctx, int status)
1371 {
1372 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1373 	struct spdk_nvmf_subsystem *subsystem;
1374 	struct spdk_nvmf_poll_group *group;
1375 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1376 	spdk_nvmf_poll_group_mod_done cpl_fn = NULL;
1377 	void *cpl_ctx = NULL;
1378 	uint32_t nsid;
1379 
1380 	group = qpair_ctx->group;
1381 	subsystem = qpair_ctx->subsystem;
1382 	cpl_fn = qpair_ctx->cpl_fn;
1383 	cpl_ctx = qpair_ctx->cpl_ctx;
1384 	sgroup = &group->sgroups[subsystem->id];
1385 
1386 	if (status) {
1387 		goto fini;
1388 	}
1389 
1390 	for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
1391 		if (sgroup->ns_info[nsid].channel) {
1392 			spdk_put_io_channel(sgroup->ns_info[nsid].channel);
1393 			sgroup->ns_info[nsid].channel = NULL;
1394 		}
1395 	}
1396 
1397 	sgroup->num_ns = 0;
1398 	free(sgroup->ns_info);
1399 	sgroup->ns_info = NULL;
1400 fini:
1401 	free(qpair_ctx);
1402 	if (cpl_fn) {
1403 		cpl_fn(cpl_ctx, status);
1404 	}
1405 }
1406 
1407 static void nvmf_poll_group_remove_subsystem_msg(void *ctx);
1408 
1409 static void
1410 remove_subsystem_qpair_cb(void *ctx)
1411 {
1412 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1413 
1414 	assert(qpair_ctx->count > 0);
1415 	qpair_ctx->count--;
1416 	if (qpair_ctx->count == 0) {
1417 		/* All of the asynchronous callbacks for this context have been
1418 		 * completed.  Call nvmf_poll_group_remove_subsystem_msg() again
1419 		 * to check if all associated qpairs for this subsystem have
1420 		 * been removed from the poll group.
1421 		 */
1422 		nvmf_poll_group_remove_subsystem_msg(ctx);
1423 	}
1424 }
1425 
1426 static void
1427 nvmf_poll_group_remove_subsystem_msg(void *ctx)
1428 {
1429 	struct spdk_nvmf_qpair *qpair, *qpair_tmp;
1430 	struct spdk_nvmf_subsystem *subsystem;
1431 	struct spdk_nvmf_poll_group *group;
1432 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1433 	bool qpairs_found = false;
1434 	int rc = 0;
1435 
1436 	group = qpair_ctx->group;
1437 	subsystem = qpair_ctx->subsystem;
1438 
1439 	/* Initialize count to 1.  This acts like a ref count, to ensure that if spdk_nvmf_qpair_disconnect
1440 	 * immediately invokes the callback (i.e. the qpairs is already in process of being disconnected)
1441 	 * that we don't recursively call nvmf_poll_group_remove_subsystem_msg before we've iterated the
1442 	 * full list of qpairs.
1443 	 */
1444 	qpair_ctx->count = 1;
1445 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, qpair_tmp) {
1446 		if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
1447 			qpairs_found = true;
1448 			qpair_ctx->count++;
1449 			rc = spdk_nvmf_qpair_disconnect(qpair, remove_subsystem_qpair_cb, ctx);
1450 			if (rc) {
1451 				break;
1452 			}
1453 		}
1454 	}
1455 	qpair_ctx->count--;
1456 
1457 	if (!qpairs_found) {
1458 		_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
1459 		return;
1460 	}
1461 
1462 	if (qpair_ctx->count == 0 || rc) {
1463 		/* If count == 0, it means there were some qpairs in the poll group but they
1464 		 * were already in process of being disconnected.  So we send a message to this
1465 		 * same thread so that this function executes again later.  We won't actually
1466 		 * invoke the remove_subsystem_cb until all of the qpairs are actually removed
1467 		 * from the poll group.
1468 		 */
1469 		spdk_thread_send_msg(spdk_get_thread(), nvmf_poll_group_remove_subsystem_msg, ctx);
1470 	}
1471 }
1472 
1473 void
1474 nvmf_poll_group_remove_subsystem(struct spdk_nvmf_poll_group *group,
1475 				 struct spdk_nvmf_subsystem *subsystem,
1476 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1477 {
1478 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1479 	struct nvmf_qpair_disconnect_many_ctx *ctx;
1480 	uint32_t i;
1481 
1482 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
1483 	if (!ctx) {
1484 		SPDK_ERRLOG("Unable to allocate memory for context to remove poll subsystem\n");
1485 		if (cb_fn) {
1486 			cb_fn(cb_arg, -1);
1487 		}
1488 		return;
1489 	}
1490 
1491 	ctx->group = group;
1492 	ctx->subsystem = subsystem;
1493 	ctx->cpl_fn = cb_fn;
1494 	ctx->cpl_ctx = cb_arg;
1495 
1496 	sgroup = &group->sgroups[subsystem->id];
1497 	sgroup->state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1498 
1499 	for (i = 0; i < sgroup->num_ns; i++) {
1500 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1501 	}
1502 
1503 	nvmf_poll_group_remove_subsystem_msg(ctx);
1504 }
1505 
1506 void
1507 nvmf_poll_group_pause_subsystem(struct spdk_nvmf_poll_group *group,
1508 				struct spdk_nvmf_subsystem *subsystem,
1509 				uint32_t nsid,
1510 				spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1511 {
1512 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1513 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info = NULL;
1514 	int rc = 0;
1515 
1516 	if (subsystem->id >= group->num_sgroups) {
1517 		rc = -1;
1518 		goto fini;
1519 	}
1520 
1521 	sgroup = &group->sgroups[subsystem->id];
1522 	if (sgroup == NULL) {
1523 		rc = -1;
1524 		goto fini;
1525 	}
1526 
1527 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSED) {
1528 		goto fini;
1529 	}
1530 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1531 
1532 	/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
1533 	if (nsid - 1 < sgroup->num_ns) {
1534 		ns_info  = &sgroup->ns_info[nsid - 1];
1535 		ns_info->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1536 	}
1537 
1538 	if (sgroup->mgmt_io_outstanding > 0) {
1539 		assert(sgroup->cb_fn == NULL);
1540 		sgroup->cb_fn = cb_fn;
1541 		assert(sgroup->cb_arg == NULL);
1542 		sgroup->cb_arg = cb_arg;
1543 		return;
1544 	}
1545 
1546 	if (ns_info != NULL && ns_info->io_outstanding > 0) {
1547 		assert(sgroup->cb_fn == NULL);
1548 		sgroup->cb_fn = cb_fn;
1549 		assert(sgroup->cb_arg == NULL);
1550 		sgroup->cb_arg = cb_arg;
1551 		return;
1552 	}
1553 
1554 	assert(sgroup->mgmt_io_outstanding == 0);
1555 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
1556 fini:
1557 	if (cb_fn) {
1558 		cb_fn(cb_arg, rc);
1559 	}
1560 }
1561 
1562 void
1563 nvmf_poll_group_resume_subsystem(struct spdk_nvmf_poll_group *group,
1564 				 struct spdk_nvmf_subsystem *subsystem,
1565 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1566 {
1567 	struct spdk_nvmf_request *req, *tmp;
1568 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1569 	int rc = 0;
1570 	uint32_t i;
1571 
1572 	if (subsystem->id >= group->num_sgroups) {
1573 		rc = -1;
1574 		goto fini;
1575 	}
1576 
1577 	sgroup = &group->sgroups[subsystem->id];
1578 
1579 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_ACTIVE) {
1580 		goto fini;
1581 	}
1582 
1583 	rc = poll_group_update_subsystem(group, subsystem);
1584 	if (rc) {
1585 		goto fini;
1586 	}
1587 
1588 	for (i = 0; i < sgroup->num_ns; i++) {
1589 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1590 	}
1591 
1592 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1593 
1594 	/* Release all queued requests */
1595 	TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1596 		TAILQ_REMOVE(&sgroup->queued, req, link);
1597 		spdk_nvmf_request_exec(req);
1598 	}
1599 fini:
1600 	if (cb_fn) {
1601 		cb_fn(cb_arg, rc);
1602 	}
1603 }
1604 
1605 
1606 struct spdk_nvmf_poll_group *
1607 spdk_nvmf_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
1608 {
1609 	struct spdk_nvmf_transport_poll_group *tgroup;
1610 
1611 	tgroup = nvmf_transport_get_optimal_poll_group(qpair->transport, qpair);
1612 
1613 	if (tgroup == NULL) {
1614 		return NULL;
1615 	}
1616 
1617 	return tgroup->group;
1618 }
1619 
1620 int
1621 spdk_nvmf_poll_group_get_stat(struct spdk_nvmf_tgt *tgt,
1622 			      struct spdk_nvmf_poll_group_stat *stat)
1623 {
1624 	struct spdk_io_channel *ch;
1625 	struct spdk_nvmf_poll_group *group;
1626 
1627 	SPDK_ERRLOG("spdk_nvmf_poll_group_get_stat is deprecated and will be removed\n");
1628 
1629 	if (tgt == NULL || stat == NULL) {
1630 		return -EINVAL;
1631 	}
1632 
1633 	ch = spdk_get_io_channel(tgt);
1634 	group = spdk_io_channel_get_ctx(ch);
1635 	*stat = group->stat;
1636 	spdk_put_io_channel(ch);
1637 	return 0;
1638 }
1639 
1640 void
1641 spdk_nvmf_poll_group_dump_stat(struct spdk_nvmf_poll_group *group, struct spdk_json_write_ctx *w)
1642 {
1643 	struct spdk_nvmf_transport_poll_group *tgroup;
1644 
1645 	spdk_json_write_object_begin(w);
1646 
1647 	spdk_json_write_named_string(w, "name", spdk_thread_get_name(spdk_get_thread()));
1648 	spdk_json_write_named_uint32(w, "admin_qpairs", group->stat.admin_qpairs);
1649 	spdk_json_write_named_uint32(w, "io_qpairs", group->stat.io_qpairs);
1650 	spdk_json_write_named_uint32(w, "current_admin_qpairs", group->stat.current_admin_qpairs);
1651 	spdk_json_write_named_uint32(w, "current_io_qpairs", group->stat.current_io_qpairs);
1652 	spdk_json_write_named_uint64(w, "pending_bdev_io", group->stat.pending_bdev_io);
1653 
1654 	spdk_json_write_named_array_begin(w, "transports");
1655 
1656 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1657 		spdk_json_write_object_begin(w);
1658 		/*
1659 		 * The trtype field intentionally contains a transport name as this is more informative.
1660 		 * The field has not been renamed for backward compatibility.
1661 		 */
1662 		spdk_json_write_named_string(w, "trtype", spdk_nvmf_get_transport_name(tgroup->transport));
1663 
1664 		if (tgroup->transport->ops->poll_group_dump_stat) {
1665 			tgroup->transport->ops->poll_group_dump_stat(tgroup, w);
1666 		}
1667 
1668 		spdk_json_write_object_end(w);
1669 	}
1670 
1671 	spdk_json_write_array_end(w);
1672 	spdk_json_write_object_end(w);
1673 }
1674