xref: /spdk/lib/nvmf/nvmf.c (revision 927f1fd57bd004df581518466ec4c1b8083e5d23)
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  *   Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "spdk/stdinc.h"
36 
37 #include "spdk/bdev.h"
38 #include "spdk/bit_array.h"
39 #include "spdk/thread.h"
40 #include "spdk/nvmf.h"
41 #include "spdk/endian.h"
42 #include "spdk/string.h"
43 #include "spdk/log.h"
44 #include "spdk_internal/usdt.h"
45 
46 #include "nvmf_internal.h"
47 #include "transport.h"
48 
49 SPDK_LOG_REGISTER_COMPONENT(nvmf)
50 
51 #define SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS 1024
52 
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 
57 /* supplied to a single call to nvmf_qpair_disconnect */
58 struct nvmf_qpair_disconnect_ctx {
59 	struct spdk_nvmf_qpair *qpair;
60 	struct spdk_nvmf_ctrlr *ctrlr;
61 	nvmf_qpair_disconnect_cb cb_fn;
62 	struct spdk_thread *thread;
63 	void *ctx;
64 	uint16_t qid;
65 };
66 
67 /*
68  * There are several times when we need to iterate through the list of all qpairs and selectively delete them.
69  * In order to do this sequentially without overlap, we must provide a context to recover the next qpair from
70  * to enable calling nvmf_qpair_disconnect on the next desired qpair.
71  */
72 struct nvmf_qpair_disconnect_many_ctx {
73 	struct spdk_nvmf_subsystem *subsystem;
74 	struct spdk_nvmf_poll_group *group;
75 	spdk_nvmf_poll_group_mod_done cpl_fn;
76 	void *cpl_ctx;
77 	uint32_t count;
78 };
79 
80 static void
81 nvmf_qpair_set_state(struct spdk_nvmf_qpair *qpair,
82 		     enum spdk_nvmf_qpair_state state)
83 {
84 	assert(qpair != NULL);
85 	assert(qpair->group->thread == spdk_get_thread());
86 
87 	qpair->state = state;
88 }
89 
90 static int
91 nvmf_poll_group_poll(void *ctx)
92 {
93 	struct spdk_nvmf_poll_group *group = ctx;
94 	int rc;
95 	int count = 0;
96 	struct spdk_nvmf_transport_poll_group *tgroup;
97 
98 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
99 		rc = nvmf_transport_poll_group_poll(tgroup);
100 		if (rc < 0) {
101 			return SPDK_POLLER_BUSY;
102 		}
103 		count += rc;
104 	}
105 
106 	return count > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
107 }
108 
109 /*
110  * Reset and clean up the poll group (I/O channel code will actually free the
111  * group).
112  */
113 static void
114 nvmf_tgt_cleanup_poll_group(struct spdk_nvmf_poll_group *group)
115 {
116 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
117 	struct spdk_nvmf_subsystem_poll_group *sgroup;
118 	uint32_t sid, nsid;
119 
120 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
121 		TAILQ_REMOVE(&group->tgroups, tgroup, link);
122 		nvmf_transport_poll_group_destroy(tgroup);
123 	}
124 
125 	for (sid = 0; sid < group->num_sgroups; sid++) {
126 		sgroup = &group->sgroups[sid];
127 
128 		assert(sgroup != NULL);
129 
130 		for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
131 			if (sgroup->ns_info[nsid].channel) {
132 				spdk_put_io_channel(sgroup->ns_info[nsid].channel);
133 				sgroup->ns_info[nsid].channel = NULL;
134 			}
135 		}
136 
137 		free(sgroup->ns_info);
138 	}
139 
140 	free(group->sgroups);
141 
142 	spdk_poller_unregister(&group->poller);
143 
144 	if (group->destroy_cb_fn) {
145 		group->destroy_cb_fn(group->destroy_cb_arg, 0);
146 	}
147 }
148 
149 /*
150  * Callback to unregister a poll group from the target, and clean up its state.
151  */
152 static void
153 nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf)
154 {
155 	struct spdk_nvmf_tgt *tgt = io_device;
156 	struct spdk_nvmf_poll_group *group = ctx_buf;
157 
158 	SPDK_DTRACE_PROBE1(nvmf_destroy_poll_group, spdk_thread_get_id(group->thread));
159 
160 	pthread_mutex_lock(&tgt->mutex);
161 	TAILQ_REMOVE(&tgt->poll_groups, group, link);
162 	pthread_mutex_unlock(&tgt->mutex);
163 
164 	nvmf_tgt_cleanup_poll_group(group);
165 }
166 
167 static int
168 nvmf_tgt_create_poll_group(void *io_device, void *ctx_buf)
169 {
170 	struct spdk_nvmf_tgt *tgt = io_device;
171 	struct spdk_nvmf_poll_group *group = ctx_buf;
172 	struct spdk_nvmf_transport *transport;
173 	struct spdk_thread *thread = spdk_get_thread();
174 	uint32_t sid;
175 	int rc;
176 
177 	TAILQ_INIT(&group->tgroups);
178 	TAILQ_INIT(&group->qpairs);
179 	group->thread = thread;
180 
181 	group->poller = SPDK_POLLER_REGISTER(nvmf_poll_group_poll, group, 0);
182 
183 	SPDK_DTRACE_PROBE1(nvmf_create_poll_group, spdk_thread_get_id(thread));
184 
185 	TAILQ_FOREACH(transport, &tgt->transports, link) {
186 		rc = nvmf_poll_group_add_transport(group, transport);
187 		if (rc != 0) {
188 			nvmf_tgt_cleanup_poll_group(group);
189 			return rc;
190 		}
191 	}
192 
193 	group->num_sgroups = tgt->max_subsystems;
194 	group->sgroups = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem_poll_group));
195 	if (!group->sgroups) {
196 		nvmf_tgt_cleanup_poll_group(group);
197 		return -ENOMEM;
198 	}
199 
200 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
201 		struct spdk_nvmf_subsystem *subsystem;
202 
203 		subsystem = tgt->subsystems[sid];
204 		if (!subsystem) {
205 			continue;
206 		}
207 
208 		if (nvmf_poll_group_add_subsystem(group, subsystem, NULL, NULL) != 0) {
209 			nvmf_tgt_cleanup_poll_group(group);
210 			return -1;
211 		}
212 	}
213 
214 	pthread_mutex_lock(&tgt->mutex);
215 	TAILQ_INSERT_TAIL(&tgt->poll_groups, group, link);
216 	pthread_mutex_unlock(&tgt->mutex);
217 
218 	return 0;
219 }
220 
221 static void
222 _nvmf_tgt_disconnect_next_qpair(void *ctx)
223 {
224 	struct spdk_nvmf_qpair *qpair;
225 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
226 	struct spdk_nvmf_poll_group *group = qpair_ctx->group;
227 	struct spdk_io_channel *ch;
228 	int rc = 0;
229 
230 	qpair = TAILQ_FIRST(&group->qpairs);
231 
232 	if (qpair) {
233 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_tgt_disconnect_next_qpair, ctx);
234 	}
235 
236 	if (!qpair || rc != 0) {
237 		/* When the refcount from the channels reaches 0, nvmf_tgt_destroy_poll_group will be called. */
238 		ch = spdk_io_channel_from_ctx(group);
239 		spdk_put_io_channel(ch);
240 		free(qpair_ctx);
241 	}
242 }
243 
244 static void
245 nvmf_tgt_destroy_poll_group_qpairs(struct spdk_nvmf_poll_group *group)
246 {
247 	struct nvmf_qpair_disconnect_many_ctx *ctx;
248 
249 	SPDK_DTRACE_PROBE1(nvmf_destroy_poll_group_qpairs, spdk_thread_get_id(group->thread));
250 
251 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
252 	if (!ctx) {
253 		SPDK_ERRLOG("Failed to allocate memory for destroy poll group ctx\n");
254 		return;
255 	}
256 
257 	ctx->group = group;
258 	_nvmf_tgt_disconnect_next_qpair(ctx);
259 }
260 
261 struct spdk_nvmf_tgt *
262 spdk_nvmf_tgt_create(struct spdk_nvmf_target_opts *opts)
263 {
264 	struct spdk_nvmf_tgt *tgt, *tmp_tgt;
265 
266 	if (strnlen(opts->name, NVMF_TGT_NAME_MAX_LENGTH) == NVMF_TGT_NAME_MAX_LENGTH) {
267 		SPDK_ERRLOG("Provided target name exceeds the max length of %u.\n", NVMF_TGT_NAME_MAX_LENGTH);
268 		return NULL;
269 	}
270 
271 	TAILQ_FOREACH(tmp_tgt, &g_nvmf_tgts, link) {
272 		if (!strncmp(opts->name, tmp_tgt->name, NVMF_TGT_NAME_MAX_LENGTH)) {
273 			SPDK_ERRLOG("Provided target name must be unique.\n");
274 			return NULL;
275 		}
276 	}
277 
278 	tgt = calloc(1, sizeof(*tgt));
279 	if (!tgt) {
280 		return NULL;
281 	}
282 
283 	snprintf(tgt->name, NVMF_TGT_NAME_MAX_LENGTH, "%s", opts->name);
284 
285 	if (!opts || !opts->max_subsystems) {
286 		tgt->max_subsystems = SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS;
287 	} else {
288 		tgt->max_subsystems = opts->max_subsystems;
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 	if (!opts) {
302 		tgt->discovery_filter = SPDK_NVMF_TGT_DISCOVERY_MATCH_ANY;
303 	} else {
304 		tgt->discovery_filter = opts->discovery_filter;
305 	}
306 
307 	tgt->discovery_genctr = 0;
308 	TAILQ_INIT(&tgt->transports);
309 	TAILQ_INIT(&tgt->poll_groups);
310 
311 	tgt->subsystems = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem *));
312 	if (!tgt->subsystems) {
313 		free(tgt);
314 		return NULL;
315 	}
316 
317 	pthread_mutex_init(&tgt->mutex, NULL);
318 
319 	spdk_io_device_register(tgt,
320 				nvmf_tgt_create_poll_group,
321 				nvmf_tgt_destroy_poll_group,
322 				sizeof(struct spdk_nvmf_poll_group),
323 				tgt->name);
324 
325 	TAILQ_INSERT_HEAD(&g_nvmf_tgts, tgt, link);
326 
327 	return tgt;
328 }
329 
330 static void
331 _nvmf_tgt_destroy_next_transport(void *ctx)
332 {
333 	struct spdk_nvmf_tgt *tgt = ctx;
334 	struct spdk_nvmf_transport *transport;
335 
336 	if (!TAILQ_EMPTY(&tgt->transports)) {
337 		transport = TAILQ_FIRST(&tgt->transports);
338 		TAILQ_REMOVE(&tgt->transports, transport, link);
339 		spdk_nvmf_transport_destroy(transport, _nvmf_tgt_destroy_next_transport, tgt);
340 	} else {
341 		spdk_nvmf_tgt_destroy_done_fn *destroy_cb_fn = tgt->destroy_cb_fn;
342 		void *destroy_cb_arg = tgt->destroy_cb_arg;
343 
344 		pthread_mutex_destroy(&tgt->mutex);
345 		free(tgt);
346 
347 		if (destroy_cb_fn) {
348 			destroy_cb_fn(destroy_cb_arg, 0);
349 		}
350 	}
351 }
352 
353 static void
354 nvmf_tgt_destroy_cb(void *io_device)
355 {
356 	struct spdk_nvmf_tgt *tgt = io_device;
357 	uint32_t i;
358 	int rc;
359 
360 	if (tgt->subsystems) {
361 		for (i = 0; i < tgt->max_subsystems; i++) {
362 			if (tgt->subsystems[i]) {
363 				nvmf_subsystem_remove_all_listeners(tgt->subsystems[i], true);
364 
365 				rc = spdk_nvmf_subsystem_destroy(tgt->subsystems[i], nvmf_tgt_destroy_cb, tgt);
366 				if (rc) {
367 					if (rc == -EINPROGRESS) {
368 						/* If rc is -EINPROGRESS, nvmf_tgt_destroy_cb will be called again when subsystem #i
369 						 * is destroyed, nvmf_tgt_destroy_cb will continue to destroy other subsystems if any */
370 						return;
371 					} else {
372 						SPDK_ERRLOG("Failed to destroy subsystem, id %u, rc %d\n", tgt->subsystems[i]->id, rc);
373 						assert(0);
374 					}
375 				}
376 			}
377 		}
378 		free(tgt->subsystems);
379 	}
380 
381 	_nvmf_tgt_destroy_next_transport(tgt);
382 }
383 
384 void
385 spdk_nvmf_tgt_destroy(struct spdk_nvmf_tgt *tgt,
386 		      spdk_nvmf_tgt_destroy_done_fn cb_fn,
387 		      void *cb_arg)
388 {
389 	tgt->destroy_cb_fn = cb_fn;
390 	tgt->destroy_cb_arg = cb_arg;
391 
392 	TAILQ_REMOVE(&g_nvmf_tgts, tgt, link);
393 
394 	spdk_io_device_unregister(tgt, nvmf_tgt_destroy_cb);
395 }
396 
397 const char *
398 spdk_nvmf_tgt_get_name(struct spdk_nvmf_tgt *tgt)
399 {
400 	return tgt->name;
401 }
402 
403 struct spdk_nvmf_tgt *
404 spdk_nvmf_get_tgt(const char *name)
405 {
406 	struct spdk_nvmf_tgt *tgt;
407 	uint32_t num_targets = 0;
408 
409 	TAILQ_FOREACH(tgt, &g_nvmf_tgts, link) {
410 		if (name) {
411 			if (!strncmp(tgt->name, name, NVMF_TGT_NAME_MAX_LENGTH)) {
412 				return tgt;
413 			}
414 		}
415 		num_targets++;
416 	}
417 
418 	/*
419 	 * special case. If there is only one target and
420 	 * no name was specified, return the only available
421 	 * target. If there is more than one target, name must
422 	 * be specified.
423 	 */
424 	if (!name && num_targets == 1) {
425 		return TAILQ_FIRST(&g_nvmf_tgts);
426 	}
427 
428 	return NULL;
429 }
430 
431 struct spdk_nvmf_tgt *
432 spdk_nvmf_get_first_tgt(void)
433 {
434 	return TAILQ_FIRST(&g_nvmf_tgts);
435 }
436 
437 struct spdk_nvmf_tgt *
438 spdk_nvmf_get_next_tgt(struct spdk_nvmf_tgt *prev)
439 {
440 	return TAILQ_NEXT(prev, link);
441 }
442 
443 static void
444 nvmf_write_subsystem_config_json(struct spdk_json_write_ctx *w,
445 				 struct spdk_nvmf_subsystem *subsystem)
446 {
447 	struct spdk_nvmf_host *host;
448 	struct spdk_nvmf_subsystem_listener *listener;
449 	const struct spdk_nvme_transport_id *trid;
450 	struct spdk_nvmf_ns *ns;
451 	struct spdk_nvmf_ns_opts ns_opts;
452 	uint32_t max_namespaces;
453 	char uuid_str[SPDK_UUID_STRING_LEN];
454 
455 	if (spdk_nvmf_subsystem_get_type(subsystem) != SPDK_NVMF_SUBTYPE_NVME) {
456 		return;
457 	}
458 
459 	/* { */
460 	spdk_json_write_object_begin(w);
461 	spdk_json_write_named_string(w, "method", "nvmf_create_subsystem");
462 
463 	/*     "params" : { */
464 	spdk_json_write_named_object_begin(w, "params");
465 	spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
466 	spdk_json_write_named_bool(w, "allow_any_host", spdk_nvmf_subsystem_get_allow_any_host(subsystem));
467 	spdk_json_write_named_string(w, "serial_number", spdk_nvmf_subsystem_get_sn(subsystem));
468 	spdk_json_write_named_string(w, "model_number", spdk_nvmf_subsystem_get_mn(subsystem));
469 
470 	max_namespaces = spdk_nvmf_subsystem_get_max_namespaces(subsystem);
471 	if (max_namespaces != 0) {
472 		spdk_json_write_named_uint32(w, "max_namespaces", max_namespaces);
473 	}
474 
475 	spdk_json_write_named_uint32(w, "min_cntlid", spdk_nvmf_subsystem_get_min_cntlid(subsystem));
476 	spdk_json_write_named_uint32(w, "max_cntlid", spdk_nvmf_subsystem_get_max_cntlid(subsystem));
477 	spdk_json_write_named_bool(w, "ana_reporting", nvmf_subsystem_get_ana_reporting(subsystem));
478 
479 	/*     } "params" */
480 	spdk_json_write_object_end(w);
481 
482 	/* } */
483 	spdk_json_write_object_end(w);
484 
485 	for (listener = spdk_nvmf_subsystem_get_first_listener(subsystem); listener != NULL;
486 	     listener = spdk_nvmf_subsystem_get_next_listener(subsystem, listener)) {
487 		trid = spdk_nvmf_subsystem_listener_get_trid(listener);
488 
489 		spdk_json_write_object_begin(w);
490 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_listener");
491 
492 		/*     "params" : { */
493 		spdk_json_write_named_object_begin(w, "params");
494 
495 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
496 		nvmf_transport_listen_dump_opts(listener->transport, trid, w);
497 
498 		/*     } "params" */
499 		spdk_json_write_object_end(w);
500 
501 		/* } */
502 		spdk_json_write_object_end(w);
503 	}
504 
505 	for (host = spdk_nvmf_subsystem_get_first_host(subsystem); host != NULL;
506 	     host = spdk_nvmf_subsystem_get_next_host(subsystem, host)) {
507 
508 		spdk_json_write_object_begin(w);
509 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_host");
510 
511 		/*     "params" : { */
512 		spdk_json_write_named_object_begin(w, "params");
513 
514 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
515 		spdk_json_write_named_string(w, "host", spdk_nvmf_host_get_nqn(host));
516 
517 		/*     } "params" */
518 		spdk_json_write_object_end(w);
519 
520 		/* } */
521 		spdk_json_write_object_end(w);
522 	}
523 
524 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
525 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
526 		spdk_nvmf_ns_get_opts(ns, &ns_opts, sizeof(ns_opts));
527 
528 		spdk_json_write_object_begin(w);
529 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_ns");
530 
531 		/*     "params" : { */
532 		spdk_json_write_named_object_begin(w, "params");
533 
534 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
535 
536 		/*     "namespace" : { */
537 		spdk_json_write_named_object_begin(w, "namespace");
538 
539 		spdk_json_write_named_uint32(w, "nsid", spdk_nvmf_ns_get_id(ns));
540 		spdk_json_write_named_string(w, "bdev_name", spdk_bdev_get_name(spdk_nvmf_ns_get_bdev(ns)));
541 
542 		if (!spdk_mem_all_zero(ns_opts.nguid, sizeof(ns_opts.nguid))) {
543 			SPDK_STATIC_ASSERT(sizeof(ns_opts.nguid) == sizeof(uint64_t) * 2, "size mismatch");
544 			spdk_json_write_named_string_fmt(w, "nguid", "%016"PRIX64"%016"PRIX64, from_be64(&ns_opts.nguid[0]),
545 							 from_be64(&ns_opts.nguid[8]));
546 		}
547 
548 		if (!spdk_mem_all_zero(ns_opts.eui64, sizeof(ns_opts.eui64))) {
549 			SPDK_STATIC_ASSERT(sizeof(ns_opts.eui64) == sizeof(uint64_t), "size mismatch");
550 			spdk_json_write_named_string_fmt(w, "eui64", "%016"PRIX64, from_be64(&ns_opts.eui64));
551 		}
552 
553 		if (!spdk_mem_all_zero(&ns_opts.uuid, sizeof(ns_opts.uuid))) {
554 			spdk_uuid_fmt_lower(uuid_str, sizeof(uuid_str), &ns_opts.uuid);
555 			spdk_json_write_named_string(w, "uuid",  uuid_str);
556 		}
557 
558 		if (nvmf_subsystem_get_ana_reporting(subsystem)) {
559 			spdk_json_write_named_uint32(w, "anagrpid", ns_opts.anagrpid);
560 		}
561 
562 		/*     "namespace" */
563 		spdk_json_write_object_end(w);
564 
565 		/*     } "params" */
566 		spdk_json_write_object_end(w);
567 
568 		/* } */
569 		spdk_json_write_object_end(w);
570 	}
571 }
572 
573 void
574 spdk_nvmf_tgt_write_config_json(struct spdk_json_write_ctx *w, struct spdk_nvmf_tgt *tgt)
575 {
576 	struct spdk_nvmf_subsystem *subsystem;
577 	struct spdk_nvmf_transport *transport;
578 
579 	spdk_json_write_object_begin(w);
580 	spdk_json_write_named_string(w, "method", "nvmf_set_max_subsystems");
581 
582 	spdk_json_write_named_object_begin(w, "params");
583 	spdk_json_write_named_uint32(w, "max_subsystems", tgt->max_subsystems);
584 	spdk_json_write_object_end(w);
585 
586 	spdk_json_write_object_end(w);
587 
588 	spdk_json_write_object_begin(w);
589 	spdk_json_write_named_string(w, "method", "nvmf_set_crdt");
590 	spdk_json_write_named_object_begin(w, "params");
591 	spdk_json_write_named_uint32(w, "crdt1", tgt->crdt[0]);
592 	spdk_json_write_named_uint32(w, "crdt2", tgt->crdt[1]);
593 	spdk_json_write_named_uint32(w, "crdt3", tgt->crdt[2]);
594 	spdk_json_write_object_end(w);
595 	spdk_json_write_object_end(w);
596 
597 	/* write transports */
598 	TAILQ_FOREACH(transport, &tgt->transports, link) {
599 		spdk_json_write_object_begin(w);
600 		spdk_json_write_named_string(w, "method", "nvmf_create_transport");
601 		nvmf_transport_dump_opts(transport, w, true);
602 		spdk_json_write_object_end(w);
603 	}
604 
605 	subsystem = spdk_nvmf_subsystem_get_first(tgt);
606 	while (subsystem) {
607 		nvmf_write_subsystem_config_json(w, subsystem);
608 		subsystem = spdk_nvmf_subsystem_get_next(subsystem);
609 	}
610 }
611 
612 static void
613 nvmf_listen_opts_copy(struct spdk_nvmf_listen_opts *opts,
614 		      const struct spdk_nvmf_listen_opts *opts_src, size_t opts_size)
615 {
616 	assert(opts);
617 	assert(opts_src);
618 
619 	opts->opts_size = opts_size;
620 
621 #define SET_FIELD(field) \
622     if (offsetof(struct spdk_nvmf_listen_opts, field) + sizeof(opts->field) <= opts_size) { \
623                  opts->field = opts_src->field; \
624     } \
625 
626 	SET_FIELD(transport_specific);
627 #undef SET_FIELD
628 
629 	/* Do not remove this statement, you should always update this statement when you adding a new field,
630 	 * and do not forget to add the SET_FIELD statement for your added field. */
631 	SPDK_STATIC_ASSERT(sizeof(struct spdk_nvmf_listen_opts) == 16, "Incorrect size");
632 }
633 
634 void
635 spdk_nvmf_listen_opts_init(struct spdk_nvmf_listen_opts *opts, size_t opts_size)
636 {
637 	struct spdk_nvmf_listen_opts opts_local = {};
638 
639 	/* local version of opts should have defaults set here */
640 
641 	nvmf_listen_opts_copy(opts, &opts_local, opts_size);
642 }
643 
644 int
645 spdk_nvmf_tgt_listen_ext(struct spdk_nvmf_tgt *tgt, const struct spdk_nvme_transport_id *trid,
646 			 struct spdk_nvmf_listen_opts *opts)
647 {
648 	struct spdk_nvmf_transport *transport;
649 	int rc;
650 	struct spdk_nvmf_listen_opts opts_local = {};
651 
652 	if (!opts) {
653 		SPDK_ERRLOG("opts should not be NULL\n");
654 		return -EINVAL;
655 	}
656 
657 	if (!opts->opts_size) {
658 		SPDK_ERRLOG("The opts_size in opts structure should not be zero\n");
659 		return -EINVAL;
660 	}
661 
662 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
663 	if (!transport) {
664 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
665 			    trid->trstring);
666 		return -EINVAL;
667 	}
668 
669 	nvmf_listen_opts_copy(&opts_local, opts, opts->opts_size);
670 	rc = spdk_nvmf_transport_listen(transport, trid, &opts_local);
671 	if (rc < 0) {
672 		SPDK_ERRLOG("Unable to listen on address '%s'\n", trid->traddr);
673 	}
674 
675 	return rc;
676 }
677 
678 int
679 spdk_nvmf_tgt_stop_listen(struct spdk_nvmf_tgt *tgt,
680 			  struct spdk_nvme_transport_id *trid)
681 {
682 	struct spdk_nvmf_transport *transport;
683 	int rc;
684 
685 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trstring);
686 	if (!transport) {
687 		SPDK_ERRLOG("Unable to find %s transport. The transport must be created first also make sure it is properly registered.\n",
688 			    trid->trstring);
689 		return -EINVAL;
690 	}
691 
692 	rc = spdk_nvmf_transport_stop_listen(transport, trid);
693 	if (rc < 0) {
694 		SPDK_ERRLOG("Failed to stop listening on address '%s'\n", trid->traddr);
695 		return rc;
696 	}
697 	return 0;
698 }
699 
700 struct spdk_nvmf_tgt_add_transport_ctx {
701 	struct spdk_nvmf_tgt *tgt;
702 	struct spdk_nvmf_transport *transport;
703 	spdk_nvmf_tgt_add_transport_done_fn cb_fn;
704 	void *cb_arg;
705 	int status;
706 };
707 
708 static void
709 _nvmf_tgt_remove_transport_done(struct spdk_io_channel_iter *i, int status)
710 {
711 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
712 
713 	ctx->cb_fn(ctx->cb_arg, ctx->status);
714 	free(ctx);
715 }
716 
717 static void
718 _nvmf_tgt_remove_transport(struct spdk_io_channel_iter *i)
719 {
720 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
721 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
722 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
723 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
724 
725 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
726 		if (tgroup->transport == ctx->transport) {
727 			TAILQ_REMOVE(&group->tgroups, tgroup, link);
728 			nvmf_transport_poll_group_destroy(tgroup);
729 		}
730 	}
731 
732 	spdk_for_each_channel_continue(i, 0);
733 }
734 
735 static void
736 _nvmf_tgt_add_transport_done(struct spdk_io_channel_iter *i, int status)
737 {
738 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
739 
740 	if (status) {
741 		ctx->status = status;
742 		spdk_for_each_channel(ctx->tgt,
743 				      _nvmf_tgt_remove_transport,
744 				      ctx,
745 				      _nvmf_tgt_remove_transport_done);
746 		return;
747 	}
748 
749 	ctx->transport->tgt = ctx->tgt;
750 	TAILQ_INSERT_TAIL(&ctx->tgt->transports, ctx->transport, link);
751 	ctx->cb_fn(ctx->cb_arg, status);
752 	free(ctx);
753 }
754 
755 static void
756 _nvmf_tgt_add_transport(struct spdk_io_channel_iter *i)
757 {
758 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
759 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
760 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
761 	int rc;
762 
763 	rc = nvmf_poll_group_add_transport(group, ctx->transport);
764 	spdk_for_each_channel_continue(i, rc);
765 }
766 
767 void spdk_nvmf_tgt_add_transport(struct spdk_nvmf_tgt *tgt,
768 				 struct spdk_nvmf_transport *transport,
769 				 spdk_nvmf_tgt_add_transport_done_fn cb_fn,
770 				 void *cb_arg)
771 {
772 	struct spdk_nvmf_tgt_add_transport_ctx *ctx;
773 
774 	SPDK_DTRACE_PROBE2(nvmf_tgt_add_transport, transport, tgt->name);
775 
776 	if (spdk_nvmf_tgt_get_transport(tgt, transport->ops->name)) {
777 		cb_fn(cb_arg, -EEXIST);
778 		return; /* transport already created */
779 	}
780 
781 	ctx = calloc(1, sizeof(*ctx));
782 	if (!ctx) {
783 		cb_fn(cb_arg, -ENOMEM);
784 		return;
785 	}
786 
787 	ctx->tgt = tgt;
788 	ctx->transport = transport;
789 	ctx->cb_fn = cb_fn;
790 	ctx->cb_arg = cb_arg;
791 
792 	spdk_for_each_channel(tgt,
793 			      _nvmf_tgt_add_transport,
794 			      ctx,
795 			      _nvmf_tgt_add_transport_done);
796 }
797 
798 struct spdk_nvmf_subsystem *
799 spdk_nvmf_tgt_find_subsystem(struct spdk_nvmf_tgt *tgt, const char *subnqn)
800 {
801 	struct spdk_nvmf_subsystem	*subsystem;
802 	uint32_t sid;
803 
804 	if (!subnqn) {
805 		return NULL;
806 	}
807 
808 	/* Ensure that subnqn is null terminated */
809 	if (!memchr(subnqn, '\0', SPDK_NVMF_NQN_MAX_LEN + 1)) {
810 		SPDK_ERRLOG("Connect SUBNQN is not null terminated\n");
811 		return NULL;
812 	}
813 
814 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
815 		subsystem = tgt->subsystems[sid];
816 		if (subsystem == NULL) {
817 			continue;
818 		}
819 
820 		if (strcmp(subnqn, subsystem->subnqn) == 0) {
821 			return subsystem;
822 		}
823 	}
824 
825 	return NULL;
826 }
827 
828 struct spdk_nvmf_transport *
829 spdk_nvmf_tgt_get_transport(struct spdk_nvmf_tgt *tgt, const char *transport_name)
830 {
831 	struct spdk_nvmf_transport *transport;
832 
833 	TAILQ_FOREACH(transport, &tgt->transports, link) {
834 		if (!strncasecmp(transport->ops->name, transport_name, SPDK_NVMF_TRSTRING_MAX_LEN)) {
835 			return transport;
836 		}
837 	}
838 	return NULL;
839 }
840 
841 struct nvmf_new_qpair_ctx {
842 	struct spdk_nvmf_qpair *qpair;
843 	struct spdk_nvmf_poll_group *group;
844 };
845 
846 static void
847 _nvmf_poll_group_add(void *_ctx)
848 {
849 	struct nvmf_new_qpair_ctx *ctx = _ctx;
850 	struct spdk_nvmf_qpair *qpair = ctx->qpair;
851 	struct spdk_nvmf_poll_group *group = ctx->group;
852 
853 	free(_ctx);
854 
855 	if (spdk_nvmf_poll_group_add(group, qpair) != 0) {
856 		SPDK_ERRLOG("Unable to add the qpair to a poll group.\n");
857 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
858 	}
859 }
860 
861 void
862 spdk_nvmf_tgt_new_qpair(struct spdk_nvmf_tgt *tgt, struct spdk_nvmf_qpair *qpair)
863 {
864 	struct spdk_nvmf_poll_group *group;
865 	struct nvmf_new_qpair_ctx *ctx;
866 
867 	group = spdk_nvmf_get_optimal_poll_group(qpair);
868 	if (group == NULL) {
869 		if (tgt->next_poll_group == NULL) {
870 			tgt->next_poll_group = TAILQ_FIRST(&tgt->poll_groups);
871 			if (tgt->next_poll_group == NULL) {
872 				SPDK_ERRLOG("No poll groups exist.\n");
873 				spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
874 				return;
875 			}
876 		}
877 		group = tgt->next_poll_group;
878 		tgt->next_poll_group = TAILQ_NEXT(group, link);
879 	}
880 
881 	ctx = calloc(1, sizeof(*ctx));
882 	if (!ctx) {
883 		SPDK_ERRLOG("Unable to send message to poll group.\n");
884 		spdk_nvmf_qpair_disconnect(qpair, NULL, NULL);
885 		return;
886 	}
887 
888 	ctx->qpair = qpair;
889 	ctx->group = group;
890 
891 	spdk_thread_send_msg(group->thread, _nvmf_poll_group_add, ctx);
892 }
893 
894 struct spdk_nvmf_poll_group *
895 spdk_nvmf_poll_group_create(struct spdk_nvmf_tgt *tgt)
896 {
897 	struct spdk_io_channel *ch;
898 
899 	ch = spdk_get_io_channel(tgt);
900 	if (!ch) {
901 		SPDK_ERRLOG("Unable to get I/O channel for target\n");
902 		return NULL;
903 	}
904 
905 	return spdk_io_channel_get_ctx(ch);
906 }
907 
908 void
909 spdk_nvmf_poll_group_destroy(struct spdk_nvmf_poll_group *group,
910 			     spdk_nvmf_poll_group_destroy_done_fn cb_fn,
911 			     void *cb_arg)
912 {
913 	assert(group->destroy_cb_fn == NULL);
914 	group->destroy_cb_fn = cb_fn;
915 	group->destroy_cb_arg = cb_arg;
916 
917 	/* This function will put the io_channel associated with this poll group */
918 	nvmf_tgt_destroy_poll_group_qpairs(group);
919 }
920 
921 int
922 spdk_nvmf_poll_group_add(struct spdk_nvmf_poll_group *group,
923 			 struct spdk_nvmf_qpair *qpair)
924 {
925 	int rc = -1;
926 	struct spdk_nvmf_transport_poll_group *tgroup;
927 
928 	TAILQ_INIT(&qpair->outstanding);
929 	qpair->group = group;
930 	qpair->ctrlr = NULL;
931 	qpair->disconnect_started = false;
932 
933 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
934 		if (tgroup->transport == qpair->transport) {
935 			rc = nvmf_transport_poll_group_add(tgroup, qpair);
936 			break;
937 		}
938 	}
939 
940 	/* We add the qpair to the group only it is successfully added into the tgroup */
941 	if (rc == 0) {
942 		SPDK_DTRACE_PROBE2(nvmf_poll_group_add_qpair, qpair, spdk_thread_get_id(group->thread));
943 		TAILQ_INSERT_TAIL(&group->qpairs, qpair, link);
944 		nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ACTIVE);
945 	}
946 
947 	return rc;
948 }
949 
950 static void
951 _nvmf_ctrlr_destruct(void *ctx)
952 {
953 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
954 
955 	nvmf_ctrlr_destruct(ctrlr);
956 }
957 
958 static void
959 _nvmf_ctrlr_free_from_qpair(void *ctx)
960 {
961 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
962 	struct spdk_nvmf_ctrlr *ctrlr = qpair_ctx->ctrlr;
963 	uint32_t count;
964 
965 	spdk_bit_array_clear(ctrlr->qpair_mask, qpair_ctx->qid);
966 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
967 	if (count == 0) {
968 		assert(!ctrlr->in_destruct);
969 		SPDK_DEBUGLOG(nvmf, "Last qpair %u, destroy ctrlr 0x%hx\n", qpair_ctx->qid, ctrlr->cntlid);
970 		ctrlr->in_destruct = true;
971 		spdk_thread_send_msg(ctrlr->subsys->thread, _nvmf_ctrlr_destruct, ctrlr);
972 	}
973 	free(qpair_ctx);
974 }
975 
976 static void
977 _nvmf_transport_qpair_fini_complete(void *cb_ctx)
978 {
979 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = cb_ctx;
980 	struct spdk_nvmf_ctrlr *ctrlr;
981 	/* Store cb args since cb_ctx can be freed in _nvmf_ctrlr_free_from_qpair */
982 	nvmf_qpair_disconnect_cb cb_fn = qpair_ctx->cb_fn;
983 	void *cb_arg = qpair_ctx->ctx;
984 	struct spdk_thread *cb_thread = qpair_ctx->thread;
985 
986 	ctrlr = qpair_ctx->ctrlr;
987 	SPDK_DEBUGLOG(nvmf, "Finish destroying qid %u\n", qpair_ctx->qid);
988 
989 	if (ctrlr) {
990 		if (qpair_ctx->qid == 0) {
991 			/* Admin qpair is removed, so set the pointer to NULL.
992 			 * This operation is safe since we are on ctrlr thread now, admin qpair's thread is the same
993 			 * as controller's thread */
994 			assert(ctrlr->thread == spdk_get_thread());
995 			ctrlr->admin_qpair = NULL;
996 		}
997 		/* Free qpair id from controller's bit mask and destroy the controller if it is the last qpair */
998 		if (ctrlr->thread) {
999 			spdk_thread_send_msg(ctrlr->thread, _nvmf_ctrlr_free_from_qpair, qpair_ctx);
1000 		} else {
1001 			_nvmf_ctrlr_free_from_qpair(qpair_ctx);
1002 		}
1003 	} else {
1004 		free(qpair_ctx);
1005 	}
1006 
1007 	if (cb_fn) {
1008 		spdk_thread_send_msg(cb_thread, cb_fn, cb_arg);
1009 	}
1010 }
1011 
1012 void
1013 spdk_nvmf_poll_group_remove(struct spdk_nvmf_qpair *qpair)
1014 {
1015 	struct spdk_nvmf_transport_poll_group *tgroup;
1016 	int rc;
1017 
1018 	SPDK_DTRACE_PROBE2(nvmf_poll_group_remove_qpair, qpair,
1019 			   spdk_thread_get_id(qpair->group->thread));
1020 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ERROR);
1021 
1022 	/* Find the tgroup and remove the qpair from the tgroup */
1023 	TAILQ_FOREACH(tgroup, &qpair->group->tgroups, link) {
1024 		if (tgroup->transport == qpair->transport) {
1025 			rc = nvmf_transport_poll_group_remove(tgroup, qpair);
1026 			if (rc && (rc != ENOTSUP)) {
1027 				SPDK_ERRLOG("Cannot remove qpair=%p from transport group=%p\n",
1028 					    qpair, tgroup);
1029 			}
1030 			break;
1031 		}
1032 	}
1033 
1034 	TAILQ_REMOVE(&qpair->group->qpairs, qpair, link);
1035 	qpair->group = NULL;
1036 }
1037 
1038 static void
1039 _nvmf_qpair_destroy(void *ctx, int status)
1040 {
1041 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1042 	struct spdk_nvmf_qpair *qpair = qpair_ctx->qpair;
1043 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
1044 	struct spdk_nvmf_request *req, *tmp;
1045 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1046 
1047 	assert(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
1048 	qpair_ctx->qid = qpair->qid;
1049 
1050 	if (ctrlr) {
1051 		if (0 == qpair->qid) {
1052 			assert(qpair->group->stat.current_admin_qpairs > 0);
1053 			qpair->group->stat.current_admin_qpairs--;
1054 		} else {
1055 			assert(qpair->group->stat.current_io_qpairs > 0);
1056 			qpair->group->stat.current_io_qpairs--;
1057 		}
1058 
1059 		sgroup = &qpair->group->sgroups[ctrlr->subsys->id];
1060 		TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1061 			if (req->qpair == qpair) {
1062 				TAILQ_REMOVE(&sgroup->queued, req, link);
1063 				if (nvmf_transport_req_free(req)) {
1064 					SPDK_ERRLOG("Transport request free error!/n");
1065 				}
1066 			}
1067 		}
1068 	}
1069 
1070 	qpair_ctx->ctrlr = ctrlr;
1071 	spdk_nvmf_poll_group_remove(qpair);
1072 	nvmf_transport_qpair_fini(qpair, _nvmf_transport_qpair_fini_complete, qpair_ctx);
1073 }
1074 
1075 static void
1076 _nvmf_qpair_disconnect_msg(void *ctx)
1077 {
1078 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
1079 
1080 	spdk_nvmf_qpair_disconnect(qpair_ctx->qpair, qpair_ctx->cb_fn, qpair_ctx->ctx);
1081 	free(ctx);
1082 }
1083 
1084 int
1085 spdk_nvmf_qpair_disconnect(struct spdk_nvmf_qpair *qpair, nvmf_qpair_disconnect_cb cb_fn, void *ctx)
1086 {
1087 	struct spdk_nvmf_poll_group *group = qpair->group;
1088 	struct nvmf_qpair_disconnect_ctx *qpair_ctx;
1089 
1090 	if (__atomic_test_and_set(&qpair->disconnect_started, __ATOMIC_RELAXED)) {
1091 		if (cb_fn) {
1092 			cb_fn(ctx);
1093 		}
1094 		return 0;
1095 	}
1096 
1097 	/* If we get a qpair in the uninitialized state, we can just destroy it immediately */
1098 	if (qpair->state == SPDK_NVMF_QPAIR_UNINITIALIZED) {
1099 		nvmf_transport_qpair_fini(qpair, NULL, NULL);
1100 		if (cb_fn) {
1101 			cb_fn(ctx);
1102 		}
1103 		return 0;
1104 	}
1105 
1106 	assert(group != NULL);
1107 	if (spdk_get_thread() != group->thread) {
1108 		/* clear the atomic so we can set it on the next call on the proper thread. */
1109 		__atomic_clear(&qpair->disconnect_started, __ATOMIC_RELAXED);
1110 		qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1111 		if (!qpair_ctx) {
1112 			SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1113 			return -ENOMEM;
1114 		}
1115 		qpair_ctx->qpair = qpair;
1116 		qpair_ctx->cb_fn = cb_fn;
1117 		qpair_ctx->thread = group->thread;
1118 		qpair_ctx->ctx = ctx;
1119 		spdk_thread_send_msg(group->thread, _nvmf_qpair_disconnect_msg, qpair_ctx);
1120 		return 0;
1121 	}
1122 
1123 	SPDK_DTRACE_PROBE2(nvmf_qpair_disconnect, qpair, spdk_thread_get_id(group->thread));
1124 	assert(qpair->state == SPDK_NVMF_QPAIR_ACTIVE);
1125 	nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_DEACTIVATING);
1126 
1127 	qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
1128 	if (!qpair_ctx) {
1129 		SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
1130 		return -ENOMEM;
1131 	}
1132 
1133 	qpair_ctx->qpair = qpair;
1134 	qpair_ctx->cb_fn = cb_fn;
1135 	qpair_ctx->thread = group->thread;
1136 	qpair_ctx->ctx = ctx;
1137 
1138 	/* Check for outstanding I/O */
1139 	if (!TAILQ_EMPTY(&qpair->outstanding)) {
1140 		SPDK_DTRACE_PROBE2(nvmf_poll_group_drain_qpair, qpair, spdk_thread_get_id(group->thread));
1141 		qpair->state_cb = _nvmf_qpair_destroy;
1142 		qpair->state_cb_arg = qpair_ctx;
1143 		nvmf_qpair_abort_pending_zcopy_reqs(qpair);
1144 		nvmf_qpair_free_aer(qpair);
1145 		return 0;
1146 	}
1147 
1148 	_nvmf_qpair_destroy(qpair_ctx, 0);
1149 
1150 	return 0;
1151 }
1152 
1153 int
1154 spdk_nvmf_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
1155 			      struct spdk_nvme_transport_id *trid)
1156 {
1157 	return nvmf_transport_qpair_get_peer_trid(qpair, trid);
1158 }
1159 
1160 int
1161 spdk_nvmf_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
1162 			       struct spdk_nvme_transport_id *trid)
1163 {
1164 	return nvmf_transport_qpair_get_local_trid(qpair, trid);
1165 }
1166 
1167 int
1168 spdk_nvmf_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
1169 				struct spdk_nvme_transport_id *trid)
1170 {
1171 	return nvmf_transport_qpair_get_listen_trid(qpair, trid);
1172 }
1173 
1174 int
1175 nvmf_poll_group_add_transport(struct spdk_nvmf_poll_group *group,
1176 			      struct spdk_nvmf_transport *transport)
1177 {
1178 	struct spdk_nvmf_transport_poll_group *tgroup;
1179 
1180 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1181 		if (tgroup->transport == transport) {
1182 			/* Transport already in the poll group */
1183 			return 0;
1184 		}
1185 	}
1186 
1187 	tgroup = nvmf_transport_poll_group_create(transport, group);
1188 	if (!tgroup) {
1189 		SPDK_ERRLOG("Unable to create poll group for transport\n");
1190 		return -1;
1191 	}
1192 	SPDK_DTRACE_PROBE2(nvmf_transport_poll_group_create, transport, spdk_thread_get_id(group->thread));
1193 
1194 	tgroup->group = group;
1195 	TAILQ_INSERT_TAIL(&group->tgroups, tgroup, link);
1196 
1197 	return 0;
1198 }
1199 
1200 static int
1201 poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1202 			    struct spdk_nvmf_subsystem *subsystem)
1203 {
1204 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1205 	uint32_t new_num_ns, old_num_ns;
1206 	uint32_t i, j;
1207 	struct spdk_nvmf_ns *ns;
1208 	struct spdk_nvmf_registrant *reg, *tmp;
1209 	struct spdk_io_channel *ch;
1210 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info;
1211 	struct spdk_nvmf_ctrlr *ctrlr;
1212 	bool ns_changed;
1213 
1214 	/* Make sure our poll group has memory for this subsystem allocated */
1215 	if (subsystem->id >= group->num_sgroups) {
1216 		return -ENOMEM;
1217 	}
1218 
1219 	sgroup = &group->sgroups[subsystem->id];
1220 
1221 	/* Make sure the array of namespace information is the correct size */
1222 	new_num_ns = subsystem->max_nsid;
1223 	old_num_ns = sgroup->num_ns;
1224 
1225 	ns_changed = false;
1226 
1227 	if (old_num_ns == 0) {
1228 		if (new_num_ns > 0) {
1229 			/* First allocation */
1230 			sgroup->ns_info = calloc(new_num_ns, sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1231 			if (!sgroup->ns_info) {
1232 				return -ENOMEM;
1233 			}
1234 		}
1235 	} else if (new_num_ns > old_num_ns) {
1236 		void *buf;
1237 
1238 		/* Make the array larger */
1239 		buf = realloc(sgroup->ns_info, new_num_ns * sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1240 		if (!buf) {
1241 			return -ENOMEM;
1242 		}
1243 
1244 		sgroup->ns_info = buf;
1245 
1246 		/* Null out the new namespace information slots */
1247 		for (i = old_num_ns; i < new_num_ns; i++) {
1248 			memset(&sgroup->ns_info[i], 0, sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1249 		}
1250 	} else if (new_num_ns < old_num_ns) {
1251 		void *buf;
1252 
1253 		/* Free the extra I/O channels */
1254 		for (i = new_num_ns; i < old_num_ns; i++) {
1255 			ns_info = &sgroup->ns_info[i];
1256 
1257 			if (ns_info->channel) {
1258 				spdk_put_io_channel(ns_info->channel);
1259 				ns_info->channel = NULL;
1260 			}
1261 		}
1262 
1263 		/* Make the array smaller */
1264 		if (new_num_ns > 0) {
1265 			buf = realloc(sgroup->ns_info, new_num_ns * sizeof(struct spdk_nvmf_subsystem_pg_ns_info));
1266 			if (!buf) {
1267 				return -ENOMEM;
1268 			}
1269 			sgroup->ns_info = buf;
1270 		} else {
1271 			free(sgroup->ns_info);
1272 			sgroup->ns_info = NULL;
1273 		}
1274 	}
1275 
1276 	sgroup->num_ns = new_num_ns;
1277 
1278 	/* Detect bdevs that were added or removed */
1279 	for (i = 0; i < sgroup->num_ns; i++) {
1280 		ns = subsystem->ns[i];
1281 		ns_info = &sgroup->ns_info[i];
1282 		ch = ns_info->channel;
1283 
1284 		if (ns == NULL && ch == NULL) {
1285 			/* Both NULL. Leave empty */
1286 		} else if (ns == NULL && ch != NULL) {
1287 			/* There was a channel here, but the namespace is gone. */
1288 			ns_changed = true;
1289 			spdk_put_io_channel(ch);
1290 			ns_info->channel = NULL;
1291 		} else if (ns != NULL && ch == NULL) {
1292 			/* A namespace appeared but there is no channel yet */
1293 			ns_changed = true;
1294 			ch = spdk_bdev_get_io_channel(ns->desc);
1295 			if (ch == NULL) {
1296 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
1297 				return -ENOMEM;
1298 			}
1299 			ns_info->channel = ch;
1300 		} else if (spdk_uuid_compare(&ns_info->uuid, spdk_bdev_get_uuid(ns->bdev)) != 0) {
1301 			/* A namespace was here before, but was replaced by a new one. */
1302 			ns_changed = true;
1303 			spdk_put_io_channel(ns_info->channel);
1304 			memset(ns_info, 0, sizeof(*ns_info));
1305 
1306 			ch = spdk_bdev_get_io_channel(ns->desc);
1307 			if (ch == NULL) {
1308 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
1309 				return -ENOMEM;
1310 			}
1311 			ns_info->channel = ch;
1312 		} else if (ns_info->num_blocks != spdk_bdev_get_num_blocks(ns->bdev)) {
1313 			/* Namespace is still there but size has changed */
1314 			SPDK_DEBUGLOG(nvmf, "Namespace resized: subsystem_id %u,"
1315 				      " nsid %u, pg %p, old %" PRIu64 ", new %" PRIu64 "\n",
1316 				      subsystem->id,
1317 				      ns->nsid,
1318 				      group,
1319 				      ns_info->num_blocks,
1320 				      spdk_bdev_get_num_blocks(ns->bdev));
1321 			ns_changed = true;
1322 		}
1323 
1324 		if (ns == NULL) {
1325 			memset(ns_info, 0, sizeof(*ns_info));
1326 		} else {
1327 			ns_info->uuid = *spdk_bdev_get_uuid(ns->bdev);
1328 			ns_info->num_blocks = spdk_bdev_get_num_blocks(ns->bdev);
1329 			ns_info->crkey = ns->crkey;
1330 			ns_info->rtype = ns->rtype;
1331 			if (ns->holder) {
1332 				ns_info->holder_id = ns->holder->hostid;
1333 			}
1334 
1335 			memset(&ns_info->reg_hostid, 0, SPDK_NVMF_MAX_NUM_REGISTRANTS * sizeof(struct spdk_uuid));
1336 			j = 0;
1337 			TAILQ_FOREACH_SAFE(reg, &ns->registrants, link, tmp) {
1338 				if (j >= SPDK_NVMF_MAX_NUM_REGISTRANTS) {
1339 					SPDK_ERRLOG("Maximum %u registrants can support.\n", SPDK_NVMF_MAX_NUM_REGISTRANTS);
1340 					return -EINVAL;
1341 				}
1342 				ns_info->reg_hostid[j++] = reg->hostid;
1343 			}
1344 		}
1345 	}
1346 
1347 	if (ns_changed) {
1348 		TAILQ_FOREACH(ctrlr, &subsystem->ctrlrs, link) {
1349 			/* It is possible that a ctrlr was added but the admin_qpair hasn't been
1350 			 * assigned yet.
1351 			 */
1352 			if (!ctrlr->admin_qpair) {
1353 				continue;
1354 			}
1355 			if (ctrlr->admin_qpair->group == group) {
1356 				nvmf_ctrlr_async_event_ns_notice(ctrlr);
1357 				nvmf_ctrlr_async_event_ana_change_notice(ctrlr);
1358 			}
1359 		}
1360 	}
1361 
1362 	return 0;
1363 }
1364 
1365 int
1366 nvmf_poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
1367 				 struct spdk_nvmf_subsystem *subsystem)
1368 {
1369 	return poll_group_update_subsystem(group, subsystem);
1370 }
1371 
1372 int
1373 nvmf_poll_group_add_subsystem(struct spdk_nvmf_poll_group *group,
1374 			      struct spdk_nvmf_subsystem *subsystem,
1375 			      spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1376 {
1377 	int rc = 0;
1378 	struct spdk_nvmf_subsystem_poll_group *sgroup = &group->sgroups[subsystem->id];
1379 	uint32_t i;
1380 
1381 	TAILQ_INIT(&sgroup->queued);
1382 
1383 	rc = poll_group_update_subsystem(group, subsystem);
1384 	if (rc) {
1385 		nvmf_poll_group_remove_subsystem(group, subsystem, NULL, NULL);
1386 		goto fini;
1387 	}
1388 
1389 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1390 
1391 	for (i = 0; i < sgroup->num_ns; i++) {
1392 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1393 	}
1394 
1395 fini:
1396 	if (cb_fn) {
1397 		cb_fn(cb_arg, rc);
1398 	}
1399 
1400 	SPDK_DTRACE_PROBE2(nvmf_poll_group_add_subsystem, spdk_thread_get_id(group->thread),
1401 			   subsystem->subnqn);
1402 
1403 	return rc;
1404 }
1405 
1406 static void
1407 _nvmf_poll_group_remove_subsystem_cb(void *ctx, int status)
1408 {
1409 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1410 	struct spdk_nvmf_subsystem *subsystem;
1411 	struct spdk_nvmf_poll_group *group;
1412 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1413 	spdk_nvmf_poll_group_mod_done cpl_fn = NULL;
1414 	void *cpl_ctx = NULL;
1415 	uint32_t nsid;
1416 
1417 	group = qpair_ctx->group;
1418 	subsystem = qpair_ctx->subsystem;
1419 	cpl_fn = qpair_ctx->cpl_fn;
1420 	cpl_ctx = qpair_ctx->cpl_ctx;
1421 	sgroup = &group->sgroups[subsystem->id];
1422 
1423 	if (status) {
1424 		goto fini;
1425 	}
1426 
1427 	for (nsid = 0; nsid < sgroup->num_ns; nsid++) {
1428 		if (sgroup->ns_info[nsid].channel) {
1429 			spdk_put_io_channel(sgroup->ns_info[nsid].channel);
1430 			sgroup->ns_info[nsid].channel = NULL;
1431 		}
1432 	}
1433 
1434 	sgroup->num_ns = 0;
1435 	free(sgroup->ns_info);
1436 	sgroup->ns_info = NULL;
1437 fini:
1438 	free(qpair_ctx);
1439 	if (cpl_fn) {
1440 		cpl_fn(cpl_ctx, status);
1441 	}
1442 }
1443 
1444 static void nvmf_poll_group_remove_subsystem_msg(void *ctx);
1445 
1446 static void
1447 remove_subsystem_qpair_cb(void *ctx)
1448 {
1449 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1450 
1451 	assert(qpair_ctx->count > 0);
1452 	qpair_ctx->count--;
1453 	if (qpair_ctx->count == 0) {
1454 		/* All of the asynchronous callbacks for this context have been
1455 		 * completed.  Call nvmf_poll_group_remove_subsystem_msg() again
1456 		 * to check if all associated qpairs for this subsystem have
1457 		 * been removed from the poll group.
1458 		 */
1459 		nvmf_poll_group_remove_subsystem_msg(ctx);
1460 	}
1461 }
1462 
1463 static void
1464 nvmf_poll_group_remove_subsystem_msg(void *ctx)
1465 {
1466 	struct spdk_nvmf_qpair *qpair, *qpair_tmp;
1467 	struct spdk_nvmf_subsystem *subsystem;
1468 	struct spdk_nvmf_poll_group *group;
1469 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1470 	bool qpairs_found = false;
1471 	int rc = 0;
1472 
1473 	group = qpair_ctx->group;
1474 	subsystem = qpair_ctx->subsystem;
1475 
1476 	/* Initialize count to 1.  This acts like a ref count, to ensure that if spdk_nvmf_qpair_disconnect
1477 	 * immediately invokes the callback (i.e. the qpairs is already in process of being disconnected)
1478 	 * that we don't recursively call nvmf_poll_group_remove_subsystem_msg before we've iterated the
1479 	 * full list of qpairs.
1480 	 */
1481 	qpair_ctx->count = 1;
1482 	TAILQ_FOREACH_SAFE(qpair, &group->qpairs, link, qpair_tmp) {
1483 		if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
1484 			qpairs_found = true;
1485 			qpair_ctx->count++;
1486 			rc = spdk_nvmf_qpair_disconnect(qpair, remove_subsystem_qpair_cb, ctx);
1487 			if (rc) {
1488 				break;
1489 			}
1490 		}
1491 	}
1492 	qpair_ctx->count--;
1493 
1494 	if (!qpairs_found) {
1495 		_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
1496 		return;
1497 	}
1498 
1499 	if (qpair_ctx->count == 0 || rc) {
1500 		/* If count == 0, it means there were some qpairs in the poll group but they
1501 		 * were already in process of being disconnected.  So we send a message to this
1502 		 * same thread so that this function executes again later.  We won't actually
1503 		 * invoke the remove_subsystem_cb until all of the qpairs are actually removed
1504 		 * from the poll group.
1505 		 */
1506 		spdk_thread_send_msg(spdk_get_thread(), nvmf_poll_group_remove_subsystem_msg, ctx);
1507 	}
1508 }
1509 
1510 void
1511 nvmf_poll_group_remove_subsystem(struct spdk_nvmf_poll_group *group,
1512 				 struct spdk_nvmf_subsystem *subsystem,
1513 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1514 {
1515 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1516 	struct nvmf_qpair_disconnect_many_ctx *ctx;
1517 	uint32_t i;
1518 
1519 	SPDK_DTRACE_PROBE3(nvmf_poll_group_remove_subsystem, group, spdk_thread_get_id(group->thread),
1520 			   subsystem->subnqn);
1521 
1522 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
1523 	if (!ctx) {
1524 		SPDK_ERRLOG("Unable to allocate memory for context to remove poll subsystem\n");
1525 		if (cb_fn) {
1526 			cb_fn(cb_arg, -1);
1527 		}
1528 		return;
1529 	}
1530 
1531 	ctx->group = group;
1532 	ctx->subsystem = subsystem;
1533 	ctx->cpl_fn = cb_fn;
1534 	ctx->cpl_ctx = cb_arg;
1535 
1536 	sgroup = &group->sgroups[subsystem->id];
1537 	sgroup->state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1538 
1539 	for (i = 0; i < sgroup->num_ns; i++) {
1540 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1541 	}
1542 
1543 	nvmf_poll_group_remove_subsystem_msg(ctx);
1544 }
1545 
1546 void
1547 nvmf_poll_group_pause_subsystem(struct spdk_nvmf_poll_group *group,
1548 				struct spdk_nvmf_subsystem *subsystem,
1549 				uint32_t nsid,
1550 				spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1551 {
1552 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1553 	struct spdk_nvmf_subsystem_pg_ns_info *ns_info = NULL;
1554 	int rc = 0;
1555 
1556 	if (subsystem->id >= group->num_sgroups) {
1557 		rc = -1;
1558 		goto fini;
1559 	}
1560 
1561 	sgroup = &group->sgroups[subsystem->id];
1562 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSED) {
1563 		goto fini;
1564 	}
1565 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1566 
1567 	/* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */
1568 	if (nsid - 1 < sgroup->num_ns) {
1569 		ns_info  = &sgroup->ns_info[nsid - 1];
1570 		ns_info->state = SPDK_NVMF_SUBSYSTEM_PAUSING;
1571 	}
1572 
1573 	if (sgroup->mgmt_io_outstanding > 0) {
1574 		assert(sgroup->cb_fn == NULL);
1575 		sgroup->cb_fn = cb_fn;
1576 		assert(sgroup->cb_arg == NULL);
1577 		sgroup->cb_arg = cb_arg;
1578 		return;
1579 	}
1580 
1581 	if (ns_info != NULL && ns_info->io_outstanding > 0) {
1582 		assert(sgroup->cb_fn == NULL);
1583 		sgroup->cb_fn = cb_fn;
1584 		assert(sgroup->cb_arg == NULL);
1585 		sgroup->cb_arg = cb_arg;
1586 		return;
1587 	}
1588 
1589 	assert(sgroup->mgmt_io_outstanding == 0);
1590 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
1591 fini:
1592 	if (cb_fn) {
1593 		cb_fn(cb_arg, rc);
1594 	}
1595 }
1596 
1597 void
1598 nvmf_poll_group_resume_subsystem(struct spdk_nvmf_poll_group *group,
1599 				 struct spdk_nvmf_subsystem *subsystem,
1600 				 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1601 {
1602 	struct spdk_nvmf_request *req, *tmp;
1603 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1604 	int rc = 0;
1605 	uint32_t i;
1606 
1607 	if (subsystem->id >= group->num_sgroups) {
1608 		rc = -1;
1609 		goto fini;
1610 	}
1611 
1612 	sgroup = &group->sgroups[subsystem->id];
1613 
1614 	if (sgroup->state == SPDK_NVMF_SUBSYSTEM_ACTIVE) {
1615 		goto fini;
1616 	}
1617 
1618 	rc = poll_group_update_subsystem(group, subsystem);
1619 	if (rc) {
1620 		goto fini;
1621 	}
1622 
1623 	for (i = 0; i < sgroup->num_ns; i++) {
1624 		sgroup->ns_info[i].state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1625 	}
1626 
1627 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1628 
1629 	/* Release all queued requests */
1630 	TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1631 		TAILQ_REMOVE(&sgroup->queued, req, link);
1632 		if (spdk_nvmf_request_using_zcopy(req)) {
1633 			spdk_nvmf_request_zcopy_start(req);
1634 		} else {
1635 			spdk_nvmf_request_exec(req);
1636 		}
1637 
1638 	}
1639 fini:
1640 	if (cb_fn) {
1641 		cb_fn(cb_arg, rc);
1642 	}
1643 }
1644 
1645 
1646 struct spdk_nvmf_poll_group *
1647 spdk_nvmf_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair)
1648 {
1649 	struct spdk_nvmf_transport_poll_group *tgroup;
1650 
1651 	tgroup = nvmf_transport_get_optimal_poll_group(qpair->transport, qpair);
1652 
1653 	if (tgroup == NULL) {
1654 		return NULL;
1655 	}
1656 
1657 	return tgroup->group;
1658 }
1659 
1660 void
1661 spdk_nvmf_poll_group_dump_stat(struct spdk_nvmf_poll_group *group, struct spdk_json_write_ctx *w)
1662 {
1663 	struct spdk_nvmf_transport_poll_group *tgroup;
1664 
1665 	spdk_json_write_object_begin(w);
1666 
1667 	spdk_json_write_named_string(w, "name", spdk_thread_get_name(spdk_get_thread()));
1668 	spdk_json_write_named_uint32(w, "admin_qpairs", group->stat.admin_qpairs);
1669 	spdk_json_write_named_uint32(w, "io_qpairs", group->stat.io_qpairs);
1670 	spdk_json_write_named_uint32(w, "current_admin_qpairs", group->stat.current_admin_qpairs);
1671 	spdk_json_write_named_uint32(w, "current_io_qpairs", group->stat.current_io_qpairs);
1672 	spdk_json_write_named_uint64(w, "pending_bdev_io", group->stat.pending_bdev_io);
1673 
1674 	spdk_json_write_named_array_begin(w, "transports");
1675 
1676 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
1677 		spdk_json_write_object_begin(w);
1678 		/*
1679 		 * The trtype field intentionally contains a transport name as this is more informative.
1680 		 * The field has not been renamed for backward compatibility.
1681 		 */
1682 		spdk_json_write_named_string(w, "trtype", spdk_nvmf_get_transport_name(tgroup->transport));
1683 
1684 		if (tgroup->transport->ops->poll_group_dump_stat) {
1685 			tgroup->transport->ops->poll_group_dump_stat(tgroup, w);
1686 		}
1687 
1688 		spdk_json_write_object_end(w);
1689 	}
1690 
1691 	spdk_json_write_array_end(w);
1692 	spdk_json_write_object_end(w);
1693 }
1694