xref: /spdk/lib/nvmf/nvmf.c (revision b78e763c1af2ace4c19d2932065a43357e3f5d3e)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   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/conf.h"
39 #include "spdk/thread.h"
40 #include "spdk/nvmf.h"
41 #include "spdk/trace.h"
42 #include "spdk/endian.h"
43 #include "spdk/string.h"
44 
45 #include "spdk_internal/log.h"
46 
47 #include "nvmf_internal.h"
48 #include "transport.h"
49 
50 SPDK_LOG_REGISTER_COMPONENT("nvmf", SPDK_LOG_NVMF)
51 
52 #define SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS 1024
53 
54 typedef void (*nvmf_qpair_disconnect_cpl)(void *ctx, int status);
55 static void spdk_nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf);
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 };
78 
79 static void
80 spdk_nvmf_qpair_set_state(struct spdk_nvmf_qpair *qpair,
81 			  enum spdk_nvmf_qpair_state state)
82 {
83 	assert(qpair != NULL);
84 	assert(qpair->group->thread == spdk_get_thread());
85 
86 	qpair->state = state;
87 }
88 
89 static int
90 spdk_nvmf_poll_group_poll(void *ctx)
91 {
92 	struct spdk_nvmf_poll_group *group = ctx;
93 	int rc;
94 	int count = 0;
95 	struct spdk_nvmf_transport_poll_group *tgroup;
96 
97 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
98 		rc = spdk_nvmf_transport_poll_group_poll(tgroup);
99 		if (rc < 0) {
100 			return -1;
101 		}
102 		count += rc;
103 	}
104 
105 	return count;
106 }
107 
108 static int
109 spdk_nvmf_tgt_create_poll_group(void *io_device, void *ctx_buf)
110 {
111 	struct spdk_nvmf_tgt *tgt = io_device;
112 	struct spdk_nvmf_poll_group *group = ctx_buf;
113 	struct spdk_nvmf_transport *transport;
114 	uint32_t sid;
115 
116 	TAILQ_INIT(&group->tgroups);
117 	TAILQ_INIT(&group->qpairs);
118 
119 	TAILQ_FOREACH(transport, &tgt->transports, link) {
120 		spdk_nvmf_poll_group_add_transport(group, transport);
121 	}
122 
123 	group->num_sgroups = tgt->max_subsystems;
124 	group->sgroups = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem_poll_group));
125 	if (!group->sgroups) {
126 		return -ENOMEM;
127 	}
128 
129 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
130 		struct spdk_nvmf_subsystem *subsystem;
131 
132 		subsystem = tgt->subsystems[sid];
133 		if (!subsystem) {
134 			continue;
135 		}
136 
137 		if (spdk_nvmf_poll_group_add_subsystem(group, subsystem, NULL, NULL) != 0) {
138 			spdk_nvmf_tgt_destroy_poll_group(io_device, ctx_buf);
139 			return -1;
140 		}
141 	}
142 
143 	group->poller = spdk_poller_register(spdk_nvmf_poll_group_poll, group, 0);
144 	group->thread = spdk_get_thread();
145 
146 	return 0;
147 }
148 
149 static void
150 spdk_nvmf_tgt_destroy_poll_group(void *io_device, void *ctx_buf)
151 {
152 	struct spdk_nvmf_poll_group *group = ctx_buf;
153 	struct spdk_nvmf_transport_poll_group *tgroup, *tmp;
154 	struct spdk_nvmf_subsystem_poll_group *sgroup;
155 	uint32_t sid, nsid;
156 
157 	TAILQ_FOREACH_SAFE(tgroup, &group->tgroups, link, tmp) {
158 		TAILQ_REMOVE(&group->tgroups, tgroup, link);
159 		spdk_nvmf_transport_poll_group_destroy(tgroup);
160 	}
161 
162 	for (sid = 0; sid < group->num_sgroups; sid++) {
163 		sgroup = &group->sgroups[sid];
164 
165 		for (nsid = 0; nsid < sgroup->num_channels; nsid++) {
166 			if (sgroup->channels[nsid]) {
167 				spdk_put_io_channel(sgroup->channels[nsid]);
168 				sgroup->channels[nsid] = NULL;
169 			}
170 		}
171 
172 		free(sgroup->channels);
173 	}
174 
175 	free(group->sgroups);
176 }
177 
178 static void
179 _nvmf_tgt_disconnect_next_qpair(void *ctx)
180 {
181 	struct spdk_nvmf_qpair *qpair;
182 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
183 	struct spdk_nvmf_poll_group *group = qpair_ctx->group;
184 	struct spdk_io_channel *ch;
185 	int rc = 0;
186 
187 	qpair = TAILQ_FIRST(&group->qpairs);
188 
189 	if (qpair) {
190 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_tgt_disconnect_next_qpair, ctx);
191 	}
192 
193 	if (!qpair || rc != 0) {
194 		/* When the refcount from the channels reaches 0, spdk_nvmf_tgt_destroy_poll_group will be called. */
195 		ch = spdk_io_channel_from_ctx(group);
196 		spdk_put_io_channel(ch);
197 		free(qpair_ctx);
198 	}
199 }
200 
201 static void
202 spdk_nvmf_tgt_destroy_poll_group_qpairs(struct spdk_nvmf_poll_group *group)
203 {
204 	struct nvmf_qpair_disconnect_many_ctx *ctx;
205 
206 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
207 
208 	if (!ctx) {
209 		SPDK_ERRLOG("Failed to allocate memory for destroy poll group ctx\n");
210 		return;
211 	}
212 
213 	spdk_poller_unregister(&group->poller);
214 
215 	ctx->group = group;
216 	_nvmf_tgt_disconnect_next_qpair(ctx);
217 }
218 
219 struct spdk_nvmf_tgt *
220 spdk_nvmf_tgt_create(uint32_t max_subsystems)
221 {
222 	struct spdk_nvmf_tgt *tgt;
223 
224 	tgt = calloc(1, sizeof(*tgt));
225 	if (!tgt) {
226 		return NULL;
227 	}
228 
229 	if (!max_subsystems) {
230 		tgt->max_subsystems = SPDK_NVMF_DEFAULT_MAX_SUBSYSTEMS;
231 	} else {
232 		tgt->max_subsystems = max_subsystems;
233 	}
234 
235 	tgt->discovery_genctr = 0;
236 	tgt->discovery_log_page = NULL;
237 	tgt->discovery_log_page_size = 0;
238 	TAILQ_INIT(&tgt->transports);
239 
240 	tgt->subsystems = calloc(tgt->max_subsystems, sizeof(struct spdk_nvmf_subsystem *));
241 	if (!tgt->subsystems) {
242 		free(tgt);
243 		return NULL;
244 	}
245 
246 	spdk_io_device_register(tgt,
247 				spdk_nvmf_tgt_create_poll_group,
248 				spdk_nvmf_tgt_destroy_poll_group,
249 				sizeof(struct spdk_nvmf_poll_group),
250 				"nvmf_tgt");
251 
252 	return tgt;
253 }
254 
255 static void
256 spdk_nvmf_tgt_destroy_cb(void *io_device)
257 {
258 	struct spdk_nvmf_tgt *tgt = io_device;
259 	struct spdk_nvmf_transport *transport, *transport_tmp;
260 	spdk_nvmf_tgt_destroy_done_fn		*destroy_cb_fn;
261 	void					*destroy_cb_arg;
262 	uint32_t i;
263 
264 	if (tgt->discovery_log_page) {
265 		free(tgt->discovery_log_page);
266 	}
267 
268 	if (tgt->subsystems) {
269 		for (i = 0; i < tgt->max_subsystems; i++) {
270 			if (tgt->subsystems[i]) {
271 				spdk_nvmf_subsystem_destroy(tgt->subsystems[i]);
272 			}
273 		}
274 		free(tgt->subsystems);
275 	}
276 
277 	TAILQ_FOREACH_SAFE(transport, &tgt->transports, link, transport_tmp) {
278 		TAILQ_REMOVE(&tgt->transports, transport, link);
279 		spdk_nvmf_transport_destroy(transport);
280 	}
281 
282 	destroy_cb_fn = tgt->destroy_cb_fn;
283 	destroy_cb_arg = tgt->destroy_cb_arg;
284 
285 	free(tgt);
286 
287 	if (destroy_cb_fn) {
288 		destroy_cb_fn(destroy_cb_arg, 0);
289 	}
290 }
291 
292 void
293 spdk_nvmf_tgt_destroy(struct spdk_nvmf_tgt *tgt,
294 		      spdk_nvmf_tgt_destroy_done_fn cb_fn,
295 		      void *cb_arg)
296 {
297 	tgt->destroy_cb_fn = cb_fn;
298 	tgt->destroy_cb_arg = cb_arg;
299 
300 	spdk_io_device_unregister(tgt, spdk_nvmf_tgt_destroy_cb);
301 }
302 
303 static void
304 spdk_nvmf_write_subsystem_config_json(struct spdk_json_write_ctx *w,
305 				      struct spdk_nvmf_subsystem *subsystem)
306 {
307 	struct spdk_nvmf_host *host;
308 	struct spdk_nvmf_listener *listener;
309 	const struct spdk_nvme_transport_id *trid;
310 	struct spdk_nvmf_ns *ns;
311 	struct spdk_nvmf_ns_opts ns_opts;
312 	uint32_t max_namespaces;
313 	char uuid_str[SPDK_UUID_STRING_LEN];
314 	const char *trtype;
315 	const char *adrfam;
316 
317 	if (spdk_nvmf_subsystem_get_type(subsystem) != SPDK_NVMF_SUBTYPE_NVME) {
318 		return;
319 	}
320 
321 	/* { */
322 	spdk_json_write_object_begin(w);
323 	spdk_json_write_named_string(w, "method", "nvmf_subsystem_create");
324 
325 	/*     "params" : { */
326 	spdk_json_write_named_object_begin(w, "params");
327 	spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
328 	spdk_json_write_named_bool(w, "allow_any_host", spdk_nvmf_subsystem_get_allow_any_host(subsystem));
329 	spdk_json_write_named_string(w, "serial_number", spdk_nvmf_subsystem_get_sn(subsystem));
330 
331 	max_namespaces = spdk_nvmf_subsystem_get_max_namespaces(subsystem);
332 	if (max_namespaces != 0) {
333 		spdk_json_write_named_uint32(w, "max_namespaces", max_namespaces);
334 	}
335 
336 	/*     } "params" */
337 	spdk_json_write_object_end(w);
338 
339 	/* } */
340 	spdk_json_write_object_end(w);
341 
342 	for (listener = spdk_nvmf_subsystem_get_first_listener(subsystem); listener != NULL;
343 	     listener = spdk_nvmf_subsystem_get_next_listener(subsystem, listener)) {
344 		trid = spdk_nvmf_listener_get_trid(listener);
345 
346 		trtype = spdk_nvme_transport_id_trtype_str(trid->trtype);
347 		adrfam = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
348 
349 		spdk_json_write_object_begin(w);
350 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_listener");
351 
352 		/*     "params" : { */
353 		spdk_json_write_named_object_begin(w, "params");
354 
355 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
356 
357 		/*     "listen_address" : { */
358 		spdk_json_write_named_object_begin(w, "listen_address");
359 
360 		spdk_json_write_named_string(w, "trtype", trtype);
361 		if (adrfam) {
362 			spdk_json_write_named_string(w, "adrfam", adrfam);
363 		}
364 
365 		spdk_json_write_named_string(w, "traddr", trid->traddr);
366 		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
367 		/*     } "listen_address" */
368 		spdk_json_write_object_end(w);
369 
370 		/*     } "params" */
371 		spdk_json_write_object_end(w);
372 
373 		/* } */
374 		spdk_json_write_object_end(w);
375 	}
376 
377 	for (host = spdk_nvmf_subsystem_get_first_host(subsystem); host != NULL;
378 	     host = spdk_nvmf_subsystem_get_next_host(subsystem, host)) {
379 
380 		spdk_json_write_object_begin(w);
381 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_host");
382 
383 		/*     "params" : { */
384 		spdk_json_write_named_object_begin(w, "params");
385 
386 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
387 		spdk_json_write_named_string(w, "host", spdk_nvmf_host_get_nqn(host));
388 
389 		/*     } "params" */
390 		spdk_json_write_object_end(w);
391 
392 		/* } */
393 		spdk_json_write_object_end(w);
394 	}
395 
396 	for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL;
397 	     ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) {
398 		spdk_nvmf_ns_get_opts(ns, &ns_opts, sizeof(ns_opts));
399 
400 		spdk_json_write_object_begin(w);
401 		spdk_json_write_named_string(w, "method", "nvmf_subsystem_add_ns");
402 
403 		/*     "params" : { */
404 		spdk_json_write_named_object_begin(w, "params");
405 
406 		spdk_json_write_named_string(w, "nqn", spdk_nvmf_subsystem_get_nqn(subsystem));
407 
408 		/*     "namespace" : { */
409 		spdk_json_write_named_object_begin(w, "namespace");
410 
411 		spdk_json_write_named_uint32(w, "nsid", spdk_nvmf_ns_get_id(ns));
412 		spdk_json_write_named_string(w, "bdev_name", spdk_bdev_get_name(spdk_nvmf_ns_get_bdev(ns)));
413 
414 		if (!spdk_mem_all_zero(ns_opts.nguid, sizeof(ns_opts.nguid))) {
415 			SPDK_STATIC_ASSERT(sizeof(ns_opts.nguid) == sizeof(uint64_t) * 2, "size mismatch");
416 			spdk_json_write_named_string_fmt(w, "nguid", "%016"PRIX64"%016"PRIX64, from_be64(&ns_opts.nguid[0]),
417 							 from_be64(&ns_opts.nguid[8]));
418 		}
419 
420 		if (!spdk_mem_all_zero(ns_opts.eui64, sizeof(ns_opts.eui64))) {
421 			SPDK_STATIC_ASSERT(sizeof(ns_opts.eui64) == sizeof(uint64_t), "size mismatch");
422 			spdk_json_write_named_string_fmt(w, "eui64", "%016"PRIX64, from_be64(&ns_opts.eui64));
423 		}
424 
425 		if (!spdk_mem_all_zero(&ns_opts.uuid, sizeof(ns_opts.uuid))) {
426 			spdk_uuid_fmt_lower(uuid_str, sizeof(uuid_str), &ns_opts.uuid);
427 			spdk_json_write_named_string(w, "uuid",  uuid_str);
428 		}
429 
430 		/*     "namespace" */
431 		spdk_json_write_object_end(w);
432 
433 		/*     } "params" */
434 		spdk_json_write_object_end(w);
435 
436 		/* } */
437 		spdk_json_write_object_end(w);
438 	}
439 }
440 
441 void
442 spdk_nvmf_tgt_write_config_json(struct spdk_json_write_ctx *w, struct spdk_nvmf_tgt *tgt)
443 {
444 	struct spdk_nvmf_subsystem *subsystem;
445 	struct spdk_nvmf_transport *transport;
446 
447 	spdk_json_write_object_begin(w);
448 	spdk_json_write_named_string(w, "method", "set_nvmf_target_max_subsystems");
449 
450 	spdk_json_write_named_object_begin(w, "params");
451 	spdk_json_write_named_uint32(w, "max_subsystems", tgt->max_subsystems);
452 	spdk_json_write_object_end(w);
453 
454 	spdk_json_write_object_end(w);
455 
456 	/* write transports */
457 	TAILQ_FOREACH(transport, &tgt->transports, link) {
458 		spdk_json_write_object_begin(w);
459 		spdk_json_write_named_string(w, "method", "nvmf_create_transport");
460 
461 		spdk_json_write_named_object_begin(w, "params");
462 		spdk_json_write_named_string(w, "trtype", spdk_nvme_transport_id_trtype_str(transport->ops->type));
463 		spdk_json_write_named_uint32(w, "max_queue_depth", transport->opts.max_queue_depth);
464 		spdk_json_write_named_uint32(w, "max_qpairs_per_ctrlr", transport->opts.max_qpairs_per_ctrlr);
465 		spdk_json_write_named_uint32(w, "in_capsule_data_size", transport->opts.in_capsule_data_size);
466 		spdk_json_write_named_uint32(w, "max_io_size", transport->opts.max_io_size);
467 		spdk_json_write_named_uint32(w, "io_unit_size", transport->opts.io_unit_size);
468 		spdk_json_write_named_uint32(w, "max_aq_depth", transport->opts.max_aq_depth);
469 		spdk_json_write_object_end(w);
470 
471 		spdk_json_write_object_end(w);
472 	}
473 
474 	subsystem = spdk_nvmf_subsystem_get_first(tgt);
475 	while (subsystem) {
476 		spdk_nvmf_write_subsystem_config_json(w, subsystem);
477 		subsystem = spdk_nvmf_subsystem_get_next(subsystem);
478 	}
479 }
480 
481 void
482 spdk_nvmf_tgt_listen(struct spdk_nvmf_tgt *tgt,
483 		     struct spdk_nvme_transport_id *trid,
484 		     spdk_nvmf_tgt_listen_done_fn cb_fn,
485 		     void *cb_arg)
486 {
487 	struct spdk_nvmf_transport *transport;
488 	const char *trtype;
489 	int rc;
490 
491 	transport = spdk_nvmf_tgt_get_transport(tgt, trid->trtype);
492 	if (!transport) {
493 		trtype = spdk_nvme_transport_id_trtype_str(trid->trtype);
494 		if (trtype != NULL) {
495 			SPDK_ERRLOG("Unable to listen on transport %s. The transport must be created first.\n", trtype);
496 		} else {
497 			SPDK_ERRLOG("The specified trtype %d is unknown. Please make sure that it is properly registered.\n",
498 				    trid->trtype);
499 		}
500 		cb_fn(cb_arg, -EINVAL);
501 		return;
502 	}
503 
504 	rc = spdk_nvmf_transport_listen(transport, trid);
505 	if (rc < 0) {
506 		SPDK_ERRLOG("Unable to listen on address '%s'\n", trid->traddr);
507 		cb_fn(cb_arg, rc);
508 		return;
509 	}
510 
511 	tgt->discovery_genctr++;
512 
513 	cb_fn(cb_arg, 0);
514 }
515 
516 struct spdk_nvmf_tgt_add_transport_ctx {
517 	struct spdk_nvmf_tgt *tgt;
518 	struct spdk_nvmf_transport *transport;
519 	spdk_nvmf_tgt_add_transport_done_fn cb_fn;
520 	void *cb_arg;
521 };
522 
523 static void
524 _spdk_nvmf_tgt_add_transport_done(struct spdk_io_channel_iter *i, int status)
525 {
526 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
527 
528 	ctx->cb_fn(ctx->cb_arg, status);
529 
530 	free(ctx);
531 }
532 
533 static void
534 _spdk_nvmf_tgt_add_transport(struct spdk_io_channel_iter *i)
535 {
536 	struct spdk_nvmf_tgt_add_transport_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
537 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
538 	struct spdk_nvmf_poll_group *group = spdk_io_channel_get_ctx(ch);
539 	int rc;
540 
541 	rc = spdk_nvmf_poll_group_add_transport(group, ctx->transport);
542 	spdk_for_each_channel_continue(i, rc);
543 }
544 
545 void spdk_nvmf_tgt_add_transport(struct spdk_nvmf_tgt *tgt,
546 				 struct spdk_nvmf_transport *transport,
547 				 spdk_nvmf_tgt_add_transport_done_fn cb_fn,
548 				 void *cb_arg)
549 {
550 	struct spdk_nvmf_tgt_add_transport_ctx *ctx;
551 
552 	if (spdk_nvmf_tgt_get_transport(tgt, transport->ops->type)) {
553 		cb_fn(cb_arg, -EEXIST);
554 		return; /* transport already created */
555 	}
556 
557 	transport->tgt = tgt;
558 	TAILQ_INSERT_TAIL(&tgt->transports, transport, link);
559 
560 	ctx = calloc(1, sizeof(*ctx));
561 	if (!ctx) {
562 		cb_fn(cb_arg, -ENOMEM);
563 		return;
564 	}
565 
566 	ctx->tgt = tgt;
567 	ctx->transport = transport;
568 	ctx->cb_fn = cb_fn;
569 	ctx->cb_arg = cb_arg;
570 
571 	spdk_for_each_channel(tgt,
572 			      _spdk_nvmf_tgt_add_transport,
573 			      ctx,
574 			      _spdk_nvmf_tgt_add_transport_done);
575 }
576 
577 struct spdk_nvmf_subsystem *
578 spdk_nvmf_tgt_find_subsystem(struct spdk_nvmf_tgt *tgt, const char *subnqn)
579 {
580 	struct spdk_nvmf_subsystem	*subsystem;
581 	uint32_t sid;
582 
583 	if (!subnqn) {
584 		return NULL;
585 	}
586 
587 	for (sid = 0; sid < tgt->max_subsystems; sid++) {
588 		subsystem = tgt->subsystems[sid];
589 		if (subsystem == NULL) {
590 			continue;
591 		}
592 
593 		if (strcmp(subnqn, subsystem->subnqn) == 0) {
594 			return subsystem;
595 		}
596 	}
597 
598 	return NULL;
599 }
600 
601 struct spdk_nvmf_transport *
602 spdk_nvmf_tgt_get_transport(struct spdk_nvmf_tgt *tgt, enum spdk_nvme_transport_type type)
603 {
604 	struct spdk_nvmf_transport *transport;
605 
606 	TAILQ_FOREACH(transport, &tgt->transports, link) {
607 		if (transport->ops->type == type) {
608 			return transport;
609 		}
610 	}
611 
612 	return NULL;
613 }
614 
615 void
616 spdk_nvmf_tgt_accept(struct spdk_nvmf_tgt *tgt, new_qpair_fn cb_fn)
617 {
618 	struct spdk_nvmf_transport *transport, *tmp;
619 
620 	TAILQ_FOREACH_SAFE(transport, &tgt->transports, link, tmp) {
621 		spdk_nvmf_transport_accept(transport, cb_fn);
622 	}
623 }
624 
625 struct spdk_nvmf_poll_group *
626 spdk_nvmf_poll_group_create(struct spdk_nvmf_tgt *tgt)
627 {
628 	struct spdk_io_channel *ch;
629 
630 	ch = spdk_get_io_channel(tgt);
631 	if (!ch) {
632 		SPDK_ERRLOG("Unable to get I/O channel for target\n");
633 		return NULL;
634 	}
635 
636 	return spdk_io_channel_get_ctx(ch);
637 }
638 
639 void
640 spdk_nvmf_poll_group_destroy(struct spdk_nvmf_poll_group *group)
641 {
642 	/* This function will put the io_channel associated with this poll group */
643 	spdk_nvmf_tgt_destroy_poll_group_qpairs(group);
644 }
645 
646 int
647 spdk_nvmf_poll_group_add(struct spdk_nvmf_poll_group *group,
648 			 struct spdk_nvmf_qpair *qpair)
649 {
650 	int rc = -1;
651 	struct spdk_nvmf_transport_poll_group *tgroup;
652 
653 	TAILQ_INIT(&qpair->outstanding);
654 	qpair->group = group;
655 	spdk_nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ACTIVATING);
656 
657 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
658 		if (tgroup->transport == qpair->transport) {
659 			rc = spdk_nvmf_transport_poll_group_add(tgroup, qpair);
660 			break;
661 		}
662 	}
663 
664 	/* We add the qpair to the group only it is succesfully added into the tgroup */
665 	if (rc == 0) {
666 		TAILQ_INSERT_TAIL(&group->qpairs, qpair, link);
667 		spdk_nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ACTIVE);
668 	}
669 
670 	return rc;
671 }
672 
673 static
674 void _nvmf_ctrlr_destruct(void *ctx)
675 {
676 	struct spdk_nvmf_ctrlr *ctrlr = ctx;
677 
678 	spdk_nvmf_ctrlr_destruct(ctrlr);
679 }
680 
681 static void
682 _spdk_nvmf_ctrlr_free_from_qpair(void *ctx)
683 {
684 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
685 	struct spdk_nvmf_ctrlr *ctrlr = qpair_ctx->ctrlr;
686 	uint32_t count;
687 
688 	spdk_bit_array_clear(ctrlr->qpair_mask, qpair_ctx->qid);
689 	count = spdk_bit_array_count_set(ctrlr->qpair_mask);
690 	if (count == 0) {
691 		spdk_bit_array_free(&ctrlr->qpair_mask);
692 
693 		spdk_thread_send_msg(ctrlr->subsys->thread, _nvmf_ctrlr_destruct, ctrlr);
694 	}
695 
696 	if (qpair_ctx->cb_fn) {
697 		spdk_thread_send_msg(qpair_ctx->thread, qpair_ctx->cb_fn, qpair_ctx->ctx);
698 	}
699 	free(qpair_ctx);
700 }
701 
702 static void
703 _spdk_nvmf_qpair_destroy(void *ctx, int status)
704 {
705 	struct nvmf_qpair_disconnect_ctx *qpair_ctx = ctx;
706 	struct spdk_nvmf_qpair *qpair = qpair_ctx->qpair;
707 	struct spdk_nvmf_ctrlr *ctrlr = qpair->ctrlr;
708 	struct spdk_nvmf_transport_poll_group *tgroup;
709 	struct spdk_nvmf_request *req, *tmp;
710 	struct spdk_nvmf_subsystem_poll_group *sgroup;
711 	int rc;
712 
713 	assert(qpair->state == SPDK_NVMF_QPAIR_DEACTIVATING);
714 	spdk_nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_ERROR);
715 	qpair_ctx->qid = qpair->qid;
716 
717 	/* Find the tgroup and remove the qpair from the tgroup */
718 	TAILQ_FOREACH(tgroup, &qpair->group->tgroups, link) {
719 		if (tgroup->transport == qpair->transport) {
720 			rc = spdk_nvmf_transport_poll_group_remove(tgroup, qpair);
721 			if (rc && (rc != ENOTSUP)) {
722 				SPDK_ERRLOG("Cannot remove qpair=%p from transport group=%p\n",
723 					    qpair, tgroup);
724 			}
725 			break;
726 		}
727 	}
728 
729 	if (ctrlr) {
730 		sgroup = &qpair->group->sgroups[ctrlr->subsys->id];
731 		TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
732 			if (req->qpair == qpair) {
733 				TAILQ_REMOVE(&sgroup->queued, req, link);
734 				if (spdk_nvmf_transport_req_free(req)) {
735 					SPDK_ERRLOG("Transport request free error!\n");
736 				}
737 			}
738 		}
739 	}
740 
741 	TAILQ_REMOVE(&qpair->group->qpairs, qpair, link);
742 	qpair->group = NULL;
743 
744 	spdk_nvmf_transport_qpair_fini(qpair);
745 
746 	if (!ctrlr || !ctrlr->thread) {
747 		if (qpair_ctx->cb_fn) {
748 			spdk_thread_send_msg(qpair_ctx->thread, qpair_ctx->cb_fn, qpair_ctx->ctx);
749 		}
750 		free(qpair_ctx);
751 		return;
752 	}
753 
754 	qpair_ctx->ctrlr = ctrlr;
755 	spdk_thread_send_msg(ctrlr->thread, _spdk_nvmf_ctrlr_free_from_qpair, qpair_ctx);
756 
757 }
758 
759 int
760 spdk_nvmf_qpair_disconnect(struct spdk_nvmf_qpair *qpair, nvmf_qpair_disconnect_cb cb_fn, void *ctx)
761 {
762 	struct nvmf_qpair_disconnect_ctx *qpair_ctx;
763 
764 	/* If we get a qpair in the uninitialized state, we can just destroy it immediately */
765 	if (qpair->state == SPDK_NVMF_QPAIR_UNINITIALIZED) {
766 		spdk_nvmf_transport_qpair_fini(qpair);
767 		if (cb_fn) {
768 			cb_fn(ctx);
769 		}
770 		return 0;
771 	}
772 
773 	/* The queue pair must be disconnected from the thread that owns it */
774 	assert(qpair->group->thread == spdk_get_thread());
775 
776 	if (qpair->state != SPDK_NVMF_QPAIR_ACTIVE) {
777 		/* This can occur if the connection is killed by the target,
778 		 * which results in a notification that the connection
779 		 * died. Send a message to defer the processing of this
780 		 * callback. This allows the stack to unwind in the case
781 		 * where a bunch of connections are disconnected in
782 		 * a loop. */
783 		if (cb_fn) {
784 			spdk_thread_send_msg(qpair->group->thread, cb_fn, ctx);
785 		}
786 		return 0;
787 	}
788 
789 	assert(qpair->state == SPDK_NVMF_QPAIR_ACTIVE);
790 	spdk_nvmf_qpair_set_state(qpair, SPDK_NVMF_QPAIR_DEACTIVATING);
791 
792 	qpair_ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_ctx));
793 	if (!qpair_ctx) {
794 		SPDK_ERRLOG("Unable to allocate context for nvmf_qpair_disconnect\n");
795 		return -ENOMEM;
796 	}
797 
798 	qpair_ctx->qpair = qpair;
799 	qpair_ctx->cb_fn = cb_fn;
800 	qpair_ctx->thread = qpair->group->thread;
801 	qpair_ctx->ctx = ctx;
802 
803 	/* Check for outstanding I/O */
804 	if (!TAILQ_EMPTY(&qpair->outstanding)) {
805 		qpair->state_cb = _spdk_nvmf_qpair_destroy;
806 		qpair->state_cb_arg = qpair_ctx;
807 		spdk_nvmf_qpair_free_aer(qpair);
808 		return 0;
809 	}
810 
811 	_spdk_nvmf_qpair_destroy(qpair_ctx, 0);
812 
813 	return 0;
814 }
815 
816 int
817 spdk_nvmf_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair,
818 			      struct spdk_nvme_transport_id *trid)
819 {
820 	return spdk_nvmf_transport_qpair_get_peer_trid(qpair, trid);
821 }
822 
823 int
824 spdk_nvmf_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair,
825 			       struct spdk_nvme_transport_id *trid)
826 {
827 	return spdk_nvmf_transport_qpair_get_local_trid(qpair, trid);
828 }
829 
830 int
831 spdk_nvmf_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair,
832 				struct spdk_nvme_transport_id *trid)
833 {
834 	return spdk_nvmf_transport_qpair_get_listen_trid(qpair, trid);
835 }
836 
837 int
838 spdk_nvmf_poll_group_add_transport(struct spdk_nvmf_poll_group *group,
839 				   struct spdk_nvmf_transport *transport)
840 {
841 	struct spdk_nvmf_transport_poll_group *tgroup;
842 
843 	TAILQ_FOREACH(tgroup, &group->tgroups, link) {
844 		if (tgroup->transport == transport) {
845 			/* Transport already in the poll group */
846 			return 0;
847 		}
848 	}
849 
850 	tgroup = spdk_nvmf_transport_poll_group_create(transport);
851 	if (!tgroup) {
852 		SPDK_ERRLOG("Unable to create poll group for transport\n");
853 		return -1;
854 	}
855 
856 	TAILQ_INSERT_TAIL(&group->tgroups, tgroup, link);
857 
858 	return 0;
859 }
860 
861 static int
862 poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
863 			    struct spdk_nvmf_subsystem *subsystem)
864 {
865 	struct spdk_nvmf_subsystem_poll_group *sgroup;
866 	uint32_t new_num_channels, old_num_channels;
867 	uint32_t i;
868 	struct spdk_nvmf_ns *ns;
869 
870 	/* Make sure our poll group has memory for this subsystem allocated */
871 	if (subsystem->id >= group->num_sgroups) {
872 		return -ENOMEM;
873 	}
874 
875 	sgroup = &group->sgroups[subsystem->id];
876 
877 	/* Make sure the array of channels is the correct size */
878 	new_num_channels = subsystem->max_nsid;
879 	old_num_channels = sgroup->num_channels;
880 
881 	if (old_num_channels == 0) {
882 		if (new_num_channels > 0) {
883 			/* First allocation */
884 			sgroup->channels = calloc(new_num_channels, sizeof(sgroup->channels[0]));
885 			if (!sgroup->channels) {
886 				return -ENOMEM;
887 			}
888 		}
889 	} else if (new_num_channels > old_num_channels) {
890 		void *buf;
891 
892 		/* Make the array larger */
893 		buf = realloc(sgroup->channels, new_num_channels * sizeof(sgroup->channels[0]));
894 		if (!buf) {
895 			return -ENOMEM;
896 		}
897 
898 		sgroup->channels = buf;
899 
900 		/* Null out the new channels slots */
901 		for (i = old_num_channels; i < new_num_channels; i++) {
902 			sgroup->channels[i] = NULL;
903 		}
904 	} else if (new_num_channels < old_num_channels) {
905 		void *buf;
906 
907 		/* Free the extra I/O channels */
908 		for (i = new_num_channels; i < old_num_channels; i++) {
909 			if (sgroup->channels[i]) {
910 				spdk_put_io_channel(sgroup->channels[i]);
911 				sgroup->channels[i] = NULL;
912 			}
913 		}
914 
915 		/* Make the array smaller */
916 		if (new_num_channels > 0) {
917 			buf = realloc(sgroup->channels, new_num_channels * sizeof(sgroup->channels[0]));
918 			if (!buf) {
919 				return -ENOMEM;
920 			}
921 			sgroup->channels = buf;
922 		} else {
923 			free(sgroup->channels);
924 			sgroup->channels = NULL;
925 		}
926 	}
927 
928 	sgroup->num_channels = new_num_channels;
929 
930 	/* Detect bdevs that were added or removed */
931 	for (i = 0; i < sgroup->num_channels; i++) {
932 		ns = subsystem->ns[i];
933 		if (ns == NULL && sgroup->channels[i] == NULL) {
934 			/* Both NULL. Leave empty */
935 		} else if (ns == NULL && sgroup->channels[i] != NULL) {
936 			/* There was a channel here, but the namespace is gone. */
937 			spdk_put_io_channel(sgroup->channels[i]);
938 			sgroup->channels[i] = NULL;
939 		} else if (ns != NULL && sgroup->channels[i] == NULL) {
940 			/* A namespace appeared but there is no channel yet */
941 			sgroup->channels[i] = spdk_bdev_get_io_channel(ns->desc);
942 			if (sgroup->channels[i] == NULL) {
943 				SPDK_ERRLOG("Could not allocate I/O channel.\n");
944 				return -ENOMEM;
945 			}
946 		} else {
947 			/* A namespace was present before and didn't change. */
948 		}
949 	}
950 
951 	return 0;
952 }
953 
954 int
955 spdk_nvmf_poll_group_update_subsystem(struct spdk_nvmf_poll_group *group,
956 				      struct spdk_nvmf_subsystem *subsystem)
957 {
958 	return poll_group_update_subsystem(group, subsystem);
959 }
960 
961 int
962 spdk_nvmf_poll_group_add_subsystem(struct spdk_nvmf_poll_group *group,
963 				   struct spdk_nvmf_subsystem *subsystem,
964 				   spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
965 {
966 	int rc = 0;
967 	struct spdk_nvmf_subsystem_poll_group *sgroup = &group->sgroups[subsystem->id];
968 
969 	TAILQ_INIT(&sgroup->queued);
970 
971 	rc = poll_group_update_subsystem(group, subsystem);
972 	if (rc) {
973 		spdk_nvmf_poll_group_remove_subsystem(group, subsystem, NULL, NULL);
974 		goto fini;
975 	}
976 
977 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
978 fini:
979 	if (cb_fn) {
980 		cb_fn(cb_arg, rc);
981 	}
982 
983 	return rc;
984 }
985 
986 static void
987 _nvmf_poll_group_remove_subsystem_cb(void *ctx, int status)
988 {
989 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
990 	struct spdk_nvmf_subsystem *subsystem;
991 	struct spdk_nvmf_poll_group *group;
992 	struct spdk_nvmf_subsystem_poll_group *sgroup;
993 	spdk_nvmf_poll_group_mod_done cpl_fn = NULL;
994 	void *cpl_ctx = NULL;
995 	uint32_t nsid;
996 
997 	group = qpair_ctx->group;
998 	subsystem = qpair_ctx->subsystem;
999 	cpl_fn = qpair_ctx->cpl_fn;
1000 	cpl_ctx = qpair_ctx->cpl_ctx;
1001 	sgroup = &group->sgroups[subsystem->id];
1002 
1003 	if (status) {
1004 		goto fini;
1005 	}
1006 
1007 	for (nsid = 0; nsid < sgroup->num_channels; nsid++) {
1008 		if (sgroup->channels[nsid]) {
1009 			spdk_put_io_channel(sgroup->channels[nsid]);
1010 			sgroup->channels[nsid] = NULL;
1011 		}
1012 	}
1013 
1014 	sgroup->num_channels = 0;
1015 	free(sgroup->channels);
1016 	sgroup->channels = NULL;
1017 fini:
1018 	free(qpair_ctx);
1019 	if (cpl_fn) {
1020 		cpl_fn(cpl_ctx, status);
1021 	}
1022 }
1023 
1024 static void
1025 _nvmf_subsystem_disconnect_next_qpair(void *ctx)
1026 {
1027 	struct spdk_nvmf_qpair *qpair;
1028 	struct nvmf_qpair_disconnect_many_ctx *qpair_ctx = ctx;
1029 	struct spdk_nvmf_subsystem *subsystem;
1030 	struct spdk_nvmf_poll_group *group;
1031 	int rc = 0;
1032 
1033 	group = qpair_ctx->group;
1034 	subsystem = qpair_ctx->subsystem;
1035 
1036 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
1037 		if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
1038 			break;
1039 		}
1040 	}
1041 
1042 	if (qpair) {
1043 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_subsystem_disconnect_next_qpair, qpair_ctx);
1044 	}
1045 
1046 	if (!qpair || rc != 0) {
1047 		_nvmf_poll_group_remove_subsystem_cb(ctx, rc);
1048 	}
1049 	return;
1050 }
1051 
1052 void
1053 spdk_nvmf_poll_group_remove_subsystem(struct spdk_nvmf_poll_group *group,
1054 				      struct spdk_nvmf_subsystem *subsystem,
1055 				      spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1056 {
1057 	struct spdk_nvmf_qpair *qpair;
1058 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1059 	struct nvmf_qpair_disconnect_many_ctx *ctx;
1060 	int rc = 0;
1061 
1062 	ctx = calloc(1, sizeof(struct nvmf_qpair_disconnect_many_ctx));
1063 
1064 	if (!ctx) {
1065 		SPDK_ERRLOG("Unable to allocate memory for context to remove poll subsystem\n");
1066 		goto fini;
1067 	}
1068 
1069 	ctx->group = group;
1070 	ctx->subsystem = subsystem;
1071 	ctx->cpl_fn = cb_fn;
1072 	ctx->cpl_ctx = cb_arg;
1073 
1074 	sgroup = &group->sgroups[subsystem->id];
1075 	sgroup->state = SPDK_NVMF_SUBSYSTEM_INACTIVE;
1076 
1077 	TAILQ_FOREACH(qpair, &group->qpairs, link) {
1078 		if ((qpair->ctrlr != NULL) && (qpair->ctrlr->subsys == subsystem)) {
1079 			break;
1080 		}
1081 	}
1082 
1083 	if (qpair) {
1084 		rc = spdk_nvmf_qpair_disconnect(qpair, _nvmf_subsystem_disconnect_next_qpair, ctx);
1085 	} else {
1086 		/* call the callback immediately. It will handle any channel iteration */
1087 		_nvmf_poll_group_remove_subsystem_cb(ctx, 0);
1088 	}
1089 
1090 	if (rc != 0) {
1091 		free(ctx);
1092 		goto fini;
1093 	}
1094 
1095 	return;
1096 fini:
1097 	if (cb_fn) {
1098 		cb_fn(cb_arg, rc);
1099 	}
1100 }
1101 
1102 void
1103 spdk_nvmf_poll_group_pause_subsystem(struct spdk_nvmf_poll_group *group,
1104 				     struct spdk_nvmf_subsystem *subsystem,
1105 				     spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1106 {
1107 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1108 	int rc = 0;
1109 
1110 	if (subsystem->id >= group->num_sgroups) {
1111 		rc = -1;
1112 		goto fini;
1113 	}
1114 
1115 	sgroup = &group->sgroups[subsystem->id];
1116 	if (sgroup == NULL) {
1117 		rc = -1;
1118 		goto fini;
1119 	}
1120 
1121 	assert(sgroup->state == SPDK_NVMF_SUBSYSTEM_ACTIVE);
1122 	/* TODO: This currently does not quiesce I/O */
1123 	sgroup->state = SPDK_NVMF_SUBSYSTEM_PAUSED;
1124 fini:
1125 	if (cb_fn) {
1126 		cb_fn(cb_arg, rc);
1127 	}
1128 }
1129 
1130 void
1131 spdk_nvmf_poll_group_resume_subsystem(struct spdk_nvmf_poll_group *group,
1132 				      struct spdk_nvmf_subsystem *subsystem,
1133 				      spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg)
1134 {
1135 	struct spdk_nvmf_request *req, *tmp;
1136 	struct spdk_nvmf_subsystem_poll_group *sgroup;
1137 	int rc = 0;
1138 
1139 	if (subsystem->id >= group->num_sgroups) {
1140 		rc = -1;
1141 		goto fini;
1142 	}
1143 
1144 	sgroup = &group->sgroups[subsystem->id];
1145 
1146 	assert(sgroup->state == SPDK_NVMF_SUBSYSTEM_PAUSED);
1147 
1148 	rc = poll_group_update_subsystem(group, subsystem);
1149 	if (rc) {
1150 		goto fini;
1151 	}
1152 
1153 	sgroup->state = SPDK_NVMF_SUBSYSTEM_ACTIVE;
1154 
1155 	/* Release all queued requests */
1156 	TAILQ_FOREACH_SAFE(req, &sgroup->queued, link, tmp) {
1157 		TAILQ_REMOVE(&sgroup->queued, req, link);
1158 		spdk_nvmf_request_exec(req);
1159 	}
1160 fini:
1161 	if (cb_fn) {
1162 		cb_fn(cb_arg, rc);
1163 	}
1164 }
1165