xref: /spdk/module/bdev/nvme/bdev_nvme.c (revision dcdab59d332f49b70bf3ad5ac5fef2b91170df2f)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (C) 2016 Intel Corporation. All rights reserved.
3  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
4  *   Copyright (c) 2021-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5  *   Copyright (c) 2022 Dell Inc, or its subsidiaries. All rights reserved.
6  */
7 
8 #include "spdk/stdinc.h"
9 
10 #include "bdev_nvme.h"
11 
12 #include "spdk/accel.h"
13 #include "spdk/config.h"
14 #include "spdk/endian.h"
15 #include "spdk/bdev.h"
16 #include "spdk/json.h"
17 #include "spdk/keyring.h"
18 #include "spdk/likely.h"
19 #include "spdk/nvme.h"
20 #include "spdk/nvme_ocssd.h"
21 #include "spdk/nvme_zns.h"
22 #include "spdk/opal.h"
23 #include "spdk/thread.h"
24 #include "spdk/trace.h"
25 #include "spdk/string.h"
26 #include "spdk/util.h"
27 #include "spdk/uuid.h"
28 
29 #include "spdk/bdev_module.h"
30 #include "spdk/log.h"
31 
32 #include "spdk_internal/usdt.h"
33 #include "spdk_internal/trace_defs.h"
34 
35 #define SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT true
36 #define SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS	(10000)
37 
38 #define NSID_STR_LEN 10
39 
40 #define SPDK_CONTROLLER_NAME_MAX 512
41 
42 static int bdev_nvme_config_json(struct spdk_json_write_ctx *w);
43 
44 struct nvme_bdev_io {
45 	/** array of iovecs to transfer. */
46 	struct iovec *iovs;
47 
48 	/** Number of iovecs in iovs array. */
49 	int iovcnt;
50 
51 	/** Current iovec position. */
52 	int iovpos;
53 
54 	/** Offset in current iovec. */
55 	uint32_t iov_offset;
56 
57 	/** Offset in current iovec. */
58 	uint32_t fused_iov_offset;
59 
60 	/** array of iovecs to transfer. */
61 	struct iovec *fused_iovs;
62 
63 	/** Number of iovecs in iovs array. */
64 	int fused_iovcnt;
65 
66 	/** Current iovec position. */
67 	int fused_iovpos;
68 
69 	/** I/O path the current I/O or admin passthrough is submitted on, or the I/O path
70 	 *  being reset in a reset I/O.
71 	 */
72 	struct nvme_io_path *io_path;
73 
74 	/** Saved status for admin passthru completion event, PI error verification, or intermediate compare-and-write status */
75 	struct spdk_nvme_cpl cpl;
76 
77 	/** Extended IO opts passed by the user to bdev layer and mapped to NVME format */
78 	struct spdk_nvme_ns_cmd_ext_io_opts ext_opts;
79 
80 	/** Keeps track if first of fused commands was submitted */
81 	bool first_fused_submitted;
82 
83 	/** Keeps track if first of fused commands was completed */
84 	bool first_fused_completed;
85 
86 	/* How many times the current I/O was retried. */
87 	int32_t retry_count;
88 
89 	/** Expiration value in ticks to retry the current I/O. */
90 	uint64_t retry_ticks;
91 
92 	/** Temporary pointer to zone report buffer */
93 	struct spdk_nvme_zns_zone_report *zone_report_buf;
94 
95 	/** Keep track of how many zones that have been copied to the spdk_bdev_zone_info struct */
96 	uint64_t handled_zones;
97 
98 	/* Current tsc at submit time. */
99 	uint64_t submit_tsc;
100 
101 	/* Used to put nvme_bdev_io into the list */
102 	TAILQ_ENTRY(nvme_bdev_io) retry_link;
103 };
104 
105 struct nvme_probe_skip_entry {
106 	struct spdk_nvme_transport_id		trid;
107 	TAILQ_ENTRY(nvme_probe_skip_entry)	tailq;
108 };
109 /* All the controllers deleted by users via RPC are skipped by hotplug monitor */
110 static TAILQ_HEAD(, nvme_probe_skip_entry) g_skipped_nvme_ctrlrs = TAILQ_HEAD_INITIALIZER(
111 			g_skipped_nvme_ctrlrs);
112 
113 #define BDEV_NVME_DEFAULT_DIGESTS (SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA256) | \
114 				   SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA384) | \
115 				   SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA512))
116 
117 #define BDEV_NVME_DEFAULT_DHGROUPS (SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_NULL) | \
118 				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_2048) | \
119 				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_3072) | \
120 				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_4096) | \
121 				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_6144) | \
122 				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_8192))
123 
124 static struct spdk_bdev_nvme_opts g_opts = {
125 	.action_on_timeout = SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE,
126 	.timeout_us = 0,
127 	.timeout_admin_us = 0,
128 	.keep_alive_timeout_ms = SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS,
129 	.transport_retry_count = 4,
130 	.arbitration_burst = 0,
131 	.low_priority_weight = 0,
132 	.medium_priority_weight = 0,
133 	.high_priority_weight = 0,
134 	.nvme_adminq_poll_period_us = 10000ULL,
135 	.nvme_ioq_poll_period_us = 0,
136 	.io_queue_requests = 0,
137 	.delay_cmd_submit = SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT,
138 	.bdev_retry_count = 3,
139 	.transport_ack_timeout = 0,
140 	.ctrlr_loss_timeout_sec = 0,
141 	.reconnect_delay_sec = 0,
142 	.fast_io_fail_timeout_sec = 0,
143 	.disable_auto_failback = false,
144 	.generate_uuids = false,
145 	.transport_tos = 0,
146 	.nvme_error_stat = false,
147 	.io_path_stat = false,
148 	.allow_accel_sequence = false,
149 	.dhchap_digests = BDEV_NVME_DEFAULT_DIGESTS,
150 	.dhchap_dhgroups = BDEV_NVME_DEFAULT_DHGROUPS,
151 };
152 
153 #define NVME_HOTPLUG_POLL_PERIOD_MAX			10000000ULL
154 #define NVME_HOTPLUG_POLL_PERIOD_DEFAULT		100000ULL
155 
156 static int g_hot_insert_nvme_controller_index = 0;
157 static uint64_t g_nvme_hotplug_poll_period_us = NVME_HOTPLUG_POLL_PERIOD_DEFAULT;
158 static bool g_nvme_hotplug_enabled = false;
159 struct spdk_thread *g_bdev_nvme_init_thread;
160 static struct spdk_poller *g_hotplug_poller;
161 static struct spdk_poller *g_hotplug_probe_poller;
162 static struct spdk_nvme_probe_ctx *g_hotplug_probe_ctx;
163 
164 static void nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
165 		struct nvme_async_probe_ctx *ctx);
166 static void nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
167 		struct nvme_async_probe_ctx *ctx);
168 static int bdev_nvme_library_init(void);
169 static void bdev_nvme_library_fini(void);
170 static void _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch,
171 				      struct spdk_bdev_io *bdev_io);
172 static void bdev_nvme_submit_request(struct spdk_io_channel *ch,
173 				     struct spdk_bdev_io *bdev_io);
174 static int bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
175 			   void *md, uint64_t lba_count, uint64_t lba,
176 			   uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
177 			   struct spdk_accel_sequence *seq);
178 static int bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
179 				 void *md, uint64_t lba_count, uint64_t lba);
180 static int bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
181 			    void *md, uint64_t lba_count, uint64_t lba,
182 			    uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
183 			    struct spdk_accel_sequence *seq,
184 			    union spdk_bdev_nvme_cdw12 cdw12, union spdk_bdev_nvme_cdw13 cdw13);
185 static int bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
186 				  void *md, uint64_t lba_count,
187 				  uint64_t zslba, uint32_t flags);
188 static int bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
189 			      void *md, uint64_t lba_count, uint64_t lba,
190 			      uint32_t flags);
191 static int bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio,
192 		struct iovec *cmp_iov, int cmp_iovcnt, struct iovec *write_iov,
193 		int write_iovcnt, void *md, uint64_t lba_count, uint64_t lba,
194 		uint32_t flags);
195 static int bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id,
196 				   uint32_t num_zones, struct spdk_bdev_zone_info *info);
197 static int bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
198 				     enum spdk_bdev_zone_action action);
199 static void bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch,
200 				     struct nvme_bdev_io *bio,
201 				     struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes);
202 static int bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
203 				 void *buf, size_t nbytes);
204 static int bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
205 				    void *buf, size_t nbytes, void *md_buf, size_t md_len);
206 static int bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
207 				     struct iovec *iov, int iovcnt, size_t nbytes,
208 				     void *md_buf, size_t md_len);
209 static void bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch,
210 			    struct nvme_bdev_io *bio, struct nvme_bdev_io *bio_to_abort);
211 static void bdev_nvme_reset_io(struct nvme_bdev *nbdev, struct nvme_bdev_io *bio);
212 static int bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
213 static int bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
214 static void remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr);
215 static int nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr);
216 
217 static struct nvme_ns *nvme_ns_alloc(void);
218 static void nvme_ns_free(struct nvme_ns *ns);
219 
220 static int
221 nvme_ns_cmp(struct nvme_ns *ns1, struct nvme_ns *ns2)
222 {
223 	return ns1->id < ns2->id ? -1 : ns1->id > ns2->id;
224 }
225 
226 RB_GENERATE_STATIC(nvme_ns_tree, nvme_ns, node, nvme_ns_cmp);
227 
228 struct spdk_nvme_qpair *
229 bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch)
230 {
231 	struct nvme_ctrlr_channel *ctrlr_ch;
232 
233 	assert(ctrlr_io_ch != NULL);
234 
235 	ctrlr_ch = spdk_io_channel_get_ctx(ctrlr_io_ch);
236 
237 	return ctrlr_ch->qpair->qpair;
238 }
239 
240 static int
241 bdev_nvme_get_ctx_size(void)
242 {
243 	return sizeof(struct nvme_bdev_io);
244 }
245 
246 static struct spdk_bdev_module nvme_if = {
247 	.name = "nvme",
248 	.async_fini = true,
249 	.module_init = bdev_nvme_library_init,
250 	.module_fini = bdev_nvme_library_fini,
251 	.config_json = bdev_nvme_config_json,
252 	.get_ctx_size = bdev_nvme_get_ctx_size,
253 
254 };
255 SPDK_BDEV_MODULE_REGISTER(nvme, &nvme_if)
256 
257 struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs = TAILQ_HEAD_INITIALIZER(g_nvme_bdev_ctrlrs);
258 pthread_mutex_t g_bdev_nvme_mutex = PTHREAD_MUTEX_INITIALIZER;
259 bool g_bdev_nvme_module_finish;
260 
261 struct nvme_bdev_ctrlr *
262 nvme_bdev_ctrlr_get_by_name(const char *name)
263 {
264 	struct nvme_bdev_ctrlr *nbdev_ctrlr;
265 
266 	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
267 		if (strcmp(name, nbdev_ctrlr->name) == 0) {
268 			break;
269 		}
270 	}
271 
272 	return nbdev_ctrlr;
273 }
274 
275 static struct nvme_ctrlr *
276 nvme_bdev_ctrlr_get_ctrlr(struct nvme_bdev_ctrlr *nbdev_ctrlr,
277 			  const struct spdk_nvme_transport_id *trid, const char *hostnqn)
278 {
279 	const struct spdk_nvme_ctrlr_opts *opts;
280 	struct nvme_ctrlr *nvme_ctrlr;
281 
282 	TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
283 		opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
284 		if (spdk_nvme_transport_id_compare(trid, &nvme_ctrlr->active_path_id->trid) == 0 &&
285 		    strcmp(hostnqn, opts->hostnqn) == 0) {
286 			break;
287 		}
288 	}
289 
290 	return nvme_ctrlr;
291 }
292 
293 struct nvme_ctrlr *
294 nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr,
295 				uint16_t cntlid)
296 {
297 	struct nvme_ctrlr *nvme_ctrlr;
298 	const struct spdk_nvme_ctrlr_data *cdata;
299 
300 	TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
301 		cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
302 		if (cdata->cntlid == cntlid) {
303 			break;
304 		}
305 	}
306 
307 	return nvme_ctrlr;
308 }
309 
310 static struct nvme_bdev *
311 nvme_bdev_ctrlr_get_bdev(struct nvme_bdev_ctrlr *nbdev_ctrlr, uint32_t nsid)
312 {
313 	struct nvme_bdev *bdev;
314 
315 	pthread_mutex_lock(&g_bdev_nvme_mutex);
316 	TAILQ_FOREACH(bdev, &nbdev_ctrlr->bdevs, tailq) {
317 		if (bdev->nsid == nsid) {
318 			break;
319 		}
320 	}
321 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
322 
323 	return bdev;
324 }
325 
326 struct nvme_ns *
327 nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid)
328 {
329 	struct nvme_ns ns;
330 
331 	assert(nsid > 0);
332 
333 	ns.id = nsid;
334 	return RB_FIND(nvme_ns_tree, &nvme_ctrlr->namespaces, &ns);
335 }
336 
337 struct nvme_ns *
338 nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr)
339 {
340 	return RB_MIN(nvme_ns_tree, &nvme_ctrlr->namespaces);
341 }
342 
343 struct nvme_ns *
344 nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns)
345 {
346 	if (ns == NULL) {
347 		return NULL;
348 	}
349 
350 	return RB_NEXT(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
351 }
352 
353 static struct nvme_ctrlr *
354 nvme_ctrlr_get(const struct spdk_nvme_transport_id *trid, const char *hostnqn)
355 {
356 	struct nvme_bdev_ctrlr	*nbdev_ctrlr;
357 	struct nvme_ctrlr	*nvme_ctrlr = NULL;
358 
359 	pthread_mutex_lock(&g_bdev_nvme_mutex);
360 	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
361 		nvme_ctrlr = nvme_bdev_ctrlr_get_ctrlr(nbdev_ctrlr, trid, hostnqn);
362 		if (nvme_ctrlr != NULL) {
363 			break;
364 		}
365 	}
366 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
367 
368 	return nvme_ctrlr;
369 }
370 
371 struct nvme_ctrlr *
372 nvme_ctrlr_get_by_name(const char *name)
373 {
374 	struct nvme_bdev_ctrlr *nbdev_ctrlr;
375 	struct nvme_ctrlr *nvme_ctrlr = NULL;
376 
377 	if (name == NULL) {
378 		return NULL;
379 	}
380 
381 	pthread_mutex_lock(&g_bdev_nvme_mutex);
382 	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
383 	if (nbdev_ctrlr != NULL) {
384 		nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
385 	}
386 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
387 
388 	return nvme_ctrlr;
389 }
390 
391 void
392 nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx)
393 {
394 	struct nvme_bdev_ctrlr *nbdev_ctrlr;
395 
396 	pthread_mutex_lock(&g_bdev_nvme_mutex);
397 	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
398 		fn(nbdev_ctrlr, ctx);
399 	}
400 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
401 }
402 
403 struct nvme_ctrlr_channel_iter {
404 	nvme_ctrlr_for_each_channel_msg fn;
405 	nvme_ctrlr_for_each_channel_done cpl;
406 	struct spdk_io_channel_iter *i;
407 	void *ctx;
408 };
409 
410 void
411 nvme_ctrlr_for_each_channel_continue(struct nvme_ctrlr_channel_iter *iter, int status)
412 {
413 	spdk_for_each_channel_continue(iter->i, status);
414 }
415 
416 static void
417 nvme_ctrlr_each_channel_msg(struct spdk_io_channel_iter *i)
418 {
419 	struct nvme_ctrlr_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
420 	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
421 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
422 	struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(ch);
423 
424 	iter->i = i;
425 	iter->fn(iter, nvme_ctrlr, ctrlr_ch, iter->ctx);
426 }
427 
428 static void
429 nvme_ctrlr_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
430 {
431 	struct nvme_ctrlr_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
432 	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
433 
434 	iter->i = i;
435 	iter->cpl(nvme_ctrlr, iter->ctx, status);
436 
437 	free(iter);
438 }
439 
440 void
441 nvme_ctrlr_for_each_channel(struct nvme_ctrlr *nvme_ctrlr,
442 			    nvme_ctrlr_for_each_channel_msg fn, void *ctx,
443 			    nvme_ctrlr_for_each_channel_done cpl)
444 {
445 	struct nvme_ctrlr_channel_iter *iter;
446 
447 	assert(nvme_ctrlr != NULL && fn != NULL);
448 
449 	iter = calloc(1, sizeof(struct nvme_ctrlr_channel_iter));
450 	if (iter == NULL) {
451 		SPDK_ERRLOG("Unable to allocate iterator\n");
452 		assert(false);
453 		return;
454 	}
455 
456 	iter->fn = fn;
457 	iter->cpl = cpl;
458 	iter->ctx = ctx;
459 
460 	spdk_for_each_channel(nvme_ctrlr, nvme_ctrlr_each_channel_msg,
461 			      iter, nvme_ctrlr_each_channel_cpl);
462 }
463 
464 struct nvme_bdev_channel_iter {
465 	nvme_bdev_for_each_channel_msg fn;
466 	nvme_bdev_for_each_channel_done cpl;
467 	struct spdk_io_channel_iter *i;
468 	void *ctx;
469 };
470 
471 void
472 nvme_bdev_for_each_channel_continue(struct nvme_bdev_channel_iter *iter, int status)
473 {
474 	spdk_for_each_channel_continue(iter->i, status);
475 }
476 
477 static void
478 nvme_bdev_each_channel_msg(struct spdk_io_channel_iter *i)
479 {
480 	struct nvme_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
481 	struct nvme_bdev *nbdev = spdk_io_channel_iter_get_io_device(i);
482 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
483 	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
484 
485 	iter->i = i;
486 	iter->fn(iter, nbdev, nbdev_ch, iter->ctx);
487 }
488 
489 static void
490 nvme_bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
491 {
492 	struct nvme_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
493 	struct nvme_bdev *nbdev = spdk_io_channel_iter_get_io_device(i);
494 
495 	iter->i = i;
496 	iter->cpl(nbdev, iter->ctx, status);
497 
498 	free(iter);
499 }
500 
501 void
502 nvme_bdev_for_each_channel(struct nvme_bdev *nbdev,
503 			   nvme_bdev_for_each_channel_msg fn, void *ctx,
504 			   nvme_bdev_for_each_channel_done cpl)
505 {
506 	struct nvme_bdev_channel_iter *iter;
507 
508 	assert(nbdev != NULL && fn != NULL);
509 
510 	iter = calloc(1, sizeof(struct nvme_bdev_channel_iter));
511 	if (iter == NULL) {
512 		SPDK_ERRLOG("Unable to allocate iterator\n");
513 		assert(false);
514 		return;
515 	}
516 
517 	iter->fn = fn;
518 	iter->cpl = cpl;
519 	iter->ctx = ctx;
520 
521 	spdk_for_each_channel(nbdev, nvme_bdev_each_channel_msg, iter,
522 			      nvme_bdev_each_channel_cpl);
523 }
524 
525 void
526 nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid, struct spdk_json_write_ctx *w)
527 {
528 	const char *trtype_str;
529 	const char *adrfam_str;
530 
531 	trtype_str = spdk_nvme_transport_id_trtype_str(trid->trtype);
532 	if (trtype_str) {
533 		spdk_json_write_named_string(w, "trtype", trtype_str);
534 	}
535 
536 	adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
537 	if (adrfam_str) {
538 		spdk_json_write_named_string(w, "adrfam", adrfam_str);
539 	}
540 
541 	if (trid->traddr[0] != '\0') {
542 		spdk_json_write_named_string(w, "traddr", trid->traddr);
543 	}
544 
545 	if (trid->trsvcid[0] != '\0') {
546 		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
547 	}
548 
549 	if (trid->subnqn[0] != '\0') {
550 		spdk_json_write_named_string(w, "subnqn", trid->subnqn);
551 	}
552 }
553 
554 static void
555 nvme_bdev_ctrlr_delete(struct nvme_bdev_ctrlr *nbdev_ctrlr,
556 		       struct nvme_ctrlr *nvme_ctrlr)
557 {
558 	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_delete, nvme_ctrlr->nbdev_ctrlr->name);
559 	pthread_mutex_lock(&g_bdev_nvme_mutex);
560 
561 	TAILQ_REMOVE(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
562 	if (!TAILQ_EMPTY(&nbdev_ctrlr->ctrlrs)) {
563 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
564 
565 		return;
566 	}
567 	TAILQ_REMOVE(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
568 
569 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
570 
571 	assert(TAILQ_EMPTY(&nbdev_ctrlr->bdevs));
572 
573 	free(nbdev_ctrlr->name);
574 	free(nbdev_ctrlr);
575 }
576 
577 static void
578 _nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
579 {
580 	struct nvme_path_id *path_id, *tmp_path;
581 	struct nvme_ns *ns, *tmp_ns;
582 
583 	free(nvme_ctrlr->copied_ana_desc);
584 	spdk_free(nvme_ctrlr->ana_log_page);
585 
586 	if (nvme_ctrlr->opal_dev) {
587 		spdk_opal_dev_destruct(nvme_ctrlr->opal_dev);
588 		nvme_ctrlr->opal_dev = NULL;
589 	}
590 
591 	if (nvme_ctrlr->nbdev_ctrlr) {
592 		nvme_bdev_ctrlr_delete(nvme_ctrlr->nbdev_ctrlr, nvme_ctrlr);
593 	}
594 
595 	RB_FOREACH_SAFE(ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp_ns) {
596 		RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
597 		nvme_ns_free(ns);
598 	}
599 
600 	TAILQ_FOREACH_SAFE(path_id, &nvme_ctrlr->trids, link, tmp_path) {
601 		TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
602 		free(path_id);
603 	}
604 
605 	pthread_mutex_destroy(&nvme_ctrlr->mutex);
606 	spdk_keyring_put_key(nvme_ctrlr->psk);
607 	spdk_keyring_put_key(nvme_ctrlr->dhchap_key);
608 	spdk_keyring_put_key(nvme_ctrlr->dhchap_ctrlr_key);
609 	free(nvme_ctrlr);
610 
611 	pthread_mutex_lock(&g_bdev_nvme_mutex);
612 	if (g_bdev_nvme_module_finish && TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
613 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
614 		spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
615 		spdk_bdev_module_fini_done();
616 		return;
617 	}
618 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
619 }
620 
621 static int
622 nvme_detach_poller(void *arg)
623 {
624 	struct nvme_ctrlr *nvme_ctrlr = arg;
625 	int rc;
626 
627 	rc = spdk_nvme_detach_poll_async(nvme_ctrlr->detach_ctx);
628 	if (rc != -EAGAIN) {
629 		spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
630 		_nvme_ctrlr_delete(nvme_ctrlr);
631 	}
632 
633 	return SPDK_POLLER_BUSY;
634 }
635 
636 static void
637 nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
638 {
639 	int rc;
640 
641 	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
642 
643 	/* First, unregister the adminq poller, as the driver will poll adminq if necessary */
644 	spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);
645 
646 	/* If we got here, the reset/detach poller cannot be active */
647 	assert(nvme_ctrlr->reset_detach_poller == NULL);
648 	nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(nvme_detach_poller,
649 					  nvme_ctrlr, 1000);
650 	if (nvme_ctrlr->reset_detach_poller == NULL) {
651 		SPDK_ERRLOG("Failed to register detach poller\n");
652 		goto error;
653 	}
654 
655 	rc = spdk_nvme_detach_async(nvme_ctrlr->ctrlr, &nvme_ctrlr->detach_ctx);
656 	if (rc != 0) {
657 		SPDK_ERRLOG("Failed to detach the NVMe controller\n");
658 		goto error;
659 	}
660 
661 	return;
662 error:
663 	/* We don't have a good way to handle errors here, so just do what we can and delete the
664 	 * controller without detaching the underlying NVMe device.
665 	 */
666 	spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
667 	_nvme_ctrlr_delete(nvme_ctrlr);
668 }
669 
670 static void
671 nvme_ctrlr_unregister_cb(void *io_device)
672 {
673 	struct nvme_ctrlr *nvme_ctrlr = io_device;
674 
675 	nvme_ctrlr_delete(nvme_ctrlr);
676 }
677 
678 static void
679 nvme_ctrlr_unregister(void *ctx)
680 {
681 	struct nvme_ctrlr *nvme_ctrlr = ctx;
682 
683 	spdk_io_device_unregister(nvme_ctrlr, nvme_ctrlr_unregister_cb);
684 }
685 
686 static bool
687 nvme_ctrlr_can_be_unregistered(struct nvme_ctrlr *nvme_ctrlr)
688 {
689 	if (!nvme_ctrlr->destruct) {
690 		return false;
691 	}
692 
693 	if (nvme_ctrlr->ref > 0) {
694 		return false;
695 	}
696 
697 	if (nvme_ctrlr->resetting) {
698 		return false;
699 	}
700 
701 	if (nvme_ctrlr->ana_log_page_updating) {
702 		return false;
703 	}
704 
705 	if (nvme_ctrlr->io_path_cache_clearing) {
706 		return false;
707 	}
708 
709 	return true;
710 }
711 
712 static void
713 nvme_ctrlr_release(struct nvme_ctrlr *nvme_ctrlr)
714 {
715 	pthread_mutex_lock(&nvme_ctrlr->mutex);
716 	SPDK_DTRACE_PROBE2(bdev_nvme_ctrlr_release, nvme_ctrlr->nbdev_ctrlr->name, nvme_ctrlr->ref);
717 
718 	assert(nvme_ctrlr->ref > 0);
719 	nvme_ctrlr->ref--;
720 
721 	if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
722 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
723 		return;
724 	}
725 
726 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
727 
728 	spdk_thread_exec_msg(nvme_ctrlr->thread, nvme_ctrlr_unregister, nvme_ctrlr);
729 }
730 
731 static void
732 bdev_nvme_clear_current_io_path(struct nvme_bdev_channel *nbdev_ch)
733 {
734 	nbdev_ch->current_io_path = NULL;
735 	nbdev_ch->rr_counter = 0;
736 }
737 
738 static struct nvme_io_path *
739 _bdev_nvme_get_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
740 {
741 	struct nvme_io_path *io_path;
742 
743 	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
744 		if (io_path->nvme_ns == nvme_ns) {
745 			break;
746 		}
747 	}
748 
749 	return io_path;
750 }
751 
752 static struct nvme_io_path *
753 nvme_io_path_alloc(void)
754 {
755 	struct nvme_io_path *io_path;
756 
757 	io_path = calloc(1, sizeof(*io_path));
758 	if (io_path == NULL) {
759 		SPDK_ERRLOG("Failed to alloc io_path.\n");
760 		return NULL;
761 	}
762 
763 	if (g_opts.io_path_stat) {
764 		io_path->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
765 		if (io_path->stat == NULL) {
766 			free(io_path);
767 			SPDK_ERRLOG("Failed to alloc io_path stat.\n");
768 			return NULL;
769 		}
770 		spdk_bdev_reset_io_stat(io_path->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
771 	}
772 
773 	return io_path;
774 }
775 
776 static void
777 nvme_io_path_free(struct nvme_io_path *io_path)
778 {
779 	free(io_path->stat);
780 	free(io_path);
781 }
782 
783 static int
784 _bdev_nvme_add_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
785 {
786 	struct nvme_io_path *io_path;
787 	struct spdk_io_channel *ch;
788 	struct nvme_ctrlr_channel *ctrlr_ch;
789 	struct nvme_qpair *nvme_qpair;
790 
791 	io_path = nvme_io_path_alloc();
792 	if (io_path == NULL) {
793 		return -ENOMEM;
794 	}
795 
796 	io_path->nvme_ns = nvme_ns;
797 
798 	ch = spdk_get_io_channel(nvme_ns->ctrlr);
799 	if (ch == NULL) {
800 		nvme_io_path_free(io_path);
801 		SPDK_ERRLOG("Failed to alloc io_channel.\n");
802 		return -ENOMEM;
803 	}
804 
805 	ctrlr_ch = spdk_io_channel_get_ctx(ch);
806 
807 	nvme_qpair = ctrlr_ch->qpair;
808 	assert(nvme_qpair != NULL);
809 
810 	io_path->qpair = nvme_qpair;
811 	TAILQ_INSERT_TAIL(&nvme_qpair->io_path_list, io_path, tailq);
812 
813 	io_path->nbdev_ch = nbdev_ch;
814 	STAILQ_INSERT_TAIL(&nbdev_ch->io_path_list, io_path, stailq);
815 
816 	bdev_nvme_clear_current_io_path(nbdev_ch);
817 
818 	return 0;
819 }
820 
821 static void
822 bdev_nvme_clear_retry_io_path(struct nvme_bdev_channel *nbdev_ch,
823 			      struct nvme_io_path *io_path)
824 {
825 	struct nvme_bdev_io *bio;
826 
827 	TAILQ_FOREACH(bio, &nbdev_ch->retry_io_list, retry_link) {
828 		if (bio->io_path == io_path) {
829 			bio->io_path = NULL;
830 		}
831 	}
832 }
833 
834 static void
835 _bdev_nvme_delete_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *io_path)
836 {
837 	struct spdk_io_channel *ch;
838 	struct nvme_qpair *nvme_qpair;
839 	struct nvme_ctrlr_channel *ctrlr_ch;
840 	struct nvme_bdev *nbdev;
841 
842 	nbdev = spdk_io_channel_get_io_device(spdk_io_channel_from_ctx(nbdev_ch));
843 
844 	/* Add the statistics to nvme_ns before this path is destroyed. */
845 	pthread_mutex_lock(&nbdev->mutex);
846 	if (nbdev->ref != 0 && io_path->nvme_ns->stat != NULL && io_path->stat != NULL) {
847 		spdk_bdev_add_io_stat(io_path->nvme_ns->stat, io_path->stat);
848 	}
849 	pthread_mutex_unlock(&nbdev->mutex);
850 
851 	bdev_nvme_clear_current_io_path(nbdev_ch);
852 	bdev_nvme_clear_retry_io_path(nbdev_ch, io_path);
853 
854 	STAILQ_REMOVE(&nbdev_ch->io_path_list, io_path, nvme_io_path, stailq);
855 	io_path->nbdev_ch = NULL;
856 
857 	nvme_qpair = io_path->qpair;
858 	assert(nvme_qpair != NULL);
859 
860 	ctrlr_ch = nvme_qpair->ctrlr_ch;
861 	assert(ctrlr_ch != NULL);
862 
863 	ch = spdk_io_channel_from_ctx(ctrlr_ch);
864 	spdk_put_io_channel(ch);
865 
866 	/* After an io_path is removed, I/Os submitted to it may complete and update statistics
867 	 * of the io_path. To avoid heap-use-after-free error from this case, do not free the
868 	 * io_path here but free the io_path when the associated qpair is freed. It is ensured
869 	 * that all I/Os submitted to the io_path are completed when the associated qpair is freed.
870 	 */
871 }
872 
873 static void
874 _bdev_nvme_delete_io_paths(struct nvme_bdev_channel *nbdev_ch)
875 {
876 	struct nvme_io_path *io_path, *tmp_io_path;
877 
878 	STAILQ_FOREACH_SAFE(io_path, &nbdev_ch->io_path_list, stailq, tmp_io_path) {
879 		_bdev_nvme_delete_io_path(nbdev_ch, io_path);
880 	}
881 }
882 
883 static int
884 bdev_nvme_create_bdev_channel_cb(void *io_device, void *ctx_buf)
885 {
886 	struct nvme_bdev_channel *nbdev_ch = ctx_buf;
887 	struct nvme_bdev *nbdev = io_device;
888 	struct nvme_ns *nvme_ns;
889 	int rc;
890 
891 	STAILQ_INIT(&nbdev_ch->io_path_list);
892 	TAILQ_INIT(&nbdev_ch->retry_io_list);
893 
894 	pthread_mutex_lock(&nbdev->mutex);
895 
896 	nbdev_ch->mp_policy = nbdev->mp_policy;
897 	nbdev_ch->mp_selector = nbdev->mp_selector;
898 	nbdev_ch->rr_min_io = nbdev->rr_min_io;
899 
900 	TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
901 		rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
902 		if (rc != 0) {
903 			pthread_mutex_unlock(&nbdev->mutex);
904 
905 			_bdev_nvme_delete_io_paths(nbdev_ch);
906 			return rc;
907 		}
908 	}
909 	pthread_mutex_unlock(&nbdev->mutex);
910 
911 	return 0;
912 }
913 
914 /* If cpl != NULL, complete the bdev_io with nvme status based on 'cpl'.
915  * If cpl == NULL, complete the bdev_io with bdev status based on 'status'.
916  */
917 static inline void
918 __bdev_nvme_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status,
919 			const struct spdk_nvme_cpl *cpl)
920 {
921 	spdk_trace_record(TRACE_BDEV_NVME_IO_DONE, 0, 0, (uintptr_t)bdev_io->driver_ctx,
922 			  (uintptr_t)bdev_io);
923 	if (cpl) {
924 		spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc);
925 	} else {
926 		spdk_bdev_io_complete(bdev_io, status);
927 	}
928 }
929 
930 static void bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch);
931 
932 static void
933 bdev_nvme_destroy_bdev_channel_cb(void *io_device, void *ctx_buf)
934 {
935 	struct nvme_bdev_channel *nbdev_ch = ctx_buf;
936 
937 	bdev_nvme_abort_retry_ios(nbdev_ch);
938 	_bdev_nvme_delete_io_paths(nbdev_ch);
939 }
940 
941 static inline bool
942 bdev_nvme_io_type_is_admin(enum spdk_bdev_io_type io_type)
943 {
944 	switch (io_type) {
945 	case SPDK_BDEV_IO_TYPE_RESET:
946 	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
947 	case SPDK_BDEV_IO_TYPE_ABORT:
948 		return true;
949 	default:
950 		break;
951 	}
952 
953 	return false;
954 }
955 
956 static inline bool
957 nvme_ns_is_active(struct nvme_ns *nvme_ns)
958 {
959 	if (spdk_unlikely(nvme_ns->ana_state_updating)) {
960 		return false;
961 	}
962 
963 	if (spdk_unlikely(nvme_ns->ns == NULL)) {
964 		return false;
965 	}
966 
967 	return true;
968 }
969 
970 static inline bool
971 nvme_ns_is_accessible(struct nvme_ns *nvme_ns)
972 {
973 	if (spdk_unlikely(!nvme_ns_is_active(nvme_ns))) {
974 		return false;
975 	}
976 
977 	switch (nvme_ns->ana_state) {
978 	case SPDK_NVME_ANA_OPTIMIZED_STATE:
979 	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
980 		return true;
981 	default:
982 		break;
983 	}
984 
985 	return false;
986 }
987 
988 static inline bool
989 nvme_qpair_is_connected(struct nvme_qpair *nvme_qpair)
990 {
991 	if (spdk_unlikely(nvme_qpair->qpair == NULL)) {
992 		return false;
993 	}
994 
995 	if (spdk_unlikely(spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
996 			  SPDK_NVME_QPAIR_FAILURE_NONE)) {
997 		return false;
998 	}
999 
1000 	if (spdk_unlikely(nvme_qpair->ctrlr_ch->reset_iter != NULL)) {
1001 		return false;
1002 	}
1003 
1004 	return true;
1005 }
1006 
1007 static inline bool
1008 nvme_io_path_is_available(struct nvme_io_path *io_path)
1009 {
1010 	if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
1011 		return false;
1012 	}
1013 
1014 	if (spdk_unlikely(!nvme_ns_is_accessible(io_path->nvme_ns))) {
1015 		return false;
1016 	}
1017 
1018 	return true;
1019 }
1020 
1021 static inline bool
1022 nvme_ctrlr_is_failed(struct nvme_ctrlr *nvme_ctrlr)
1023 {
1024 	if (nvme_ctrlr->destruct) {
1025 		return true;
1026 	}
1027 
1028 	if (nvme_ctrlr->fast_io_fail_timedout) {
1029 		return true;
1030 	}
1031 
1032 	if (nvme_ctrlr->resetting) {
1033 		if (nvme_ctrlr->opts.reconnect_delay_sec != 0) {
1034 			return false;
1035 		} else {
1036 			return true;
1037 		}
1038 	}
1039 
1040 	if (nvme_ctrlr->reconnect_is_delayed) {
1041 		return false;
1042 	}
1043 
1044 	if (nvme_ctrlr->disabled) {
1045 		return true;
1046 	}
1047 
1048 	if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
1049 		return true;
1050 	} else {
1051 		return false;
1052 	}
1053 }
1054 
1055 static bool
1056 nvme_ctrlr_is_available(struct nvme_ctrlr *nvme_ctrlr)
1057 {
1058 	if (nvme_ctrlr->destruct) {
1059 		return false;
1060 	}
1061 
1062 	if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
1063 		return false;
1064 	}
1065 
1066 	if (nvme_ctrlr->resetting || nvme_ctrlr->reconnect_is_delayed) {
1067 		return false;
1068 	}
1069 
1070 	if (nvme_ctrlr->disabled) {
1071 		return false;
1072 	}
1073 
1074 	return true;
1075 }
1076 
1077 /* Simulate circular linked list. */
1078 static inline struct nvme_io_path *
1079 nvme_io_path_get_next(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *prev_path)
1080 {
1081 	struct nvme_io_path *next_path;
1082 
1083 	if (prev_path != NULL) {
1084 		next_path = STAILQ_NEXT(prev_path, stailq);
1085 		if (next_path != NULL) {
1086 			return next_path;
1087 		}
1088 	}
1089 
1090 	return STAILQ_FIRST(&nbdev_ch->io_path_list);
1091 }
1092 
1093 static struct nvme_io_path *
1094 _bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
1095 {
1096 	struct nvme_io_path *io_path, *start, *non_optimized = NULL;
1097 
1098 	start = nvme_io_path_get_next(nbdev_ch, nbdev_ch->current_io_path);
1099 
1100 	io_path = start;
1101 	do {
1102 		if (spdk_likely(nvme_io_path_is_available(io_path))) {
1103 			switch (io_path->nvme_ns->ana_state) {
1104 			case SPDK_NVME_ANA_OPTIMIZED_STATE:
1105 				nbdev_ch->current_io_path = io_path;
1106 				return io_path;
1107 			case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1108 				if (non_optimized == NULL) {
1109 					non_optimized = io_path;
1110 				}
1111 				break;
1112 			default:
1113 				assert(false);
1114 				break;
1115 			}
1116 		}
1117 		io_path = nvme_io_path_get_next(nbdev_ch, io_path);
1118 	} while (io_path != start);
1119 
1120 	if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
1121 		/* We come here only if there is no optimized path. Cache even non_optimized
1122 		 * path for load balance across multiple non_optimized paths.
1123 		 */
1124 		nbdev_ch->current_io_path = non_optimized;
1125 	}
1126 
1127 	return non_optimized;
1128 }
1129 
1130 static struct nvme_io_path *
1131 _bdev_nvme_find_io_path_min_qd(struct nvme_bdev_channel *nbdev_ch)
1132 {
1133 	struct nvme_io_path *io_path;
1134 	struct nvme_io_path *optimized = NULL, *non_optimized = NULL;
1135 	uint32_t opt_min_qd = UINT32_MAX, non_opt_min_qd = UINT32_MAX;
1136 	uint32_t num_outstanding_reqs;
1137 
1138 	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
1139 		if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
1140 			/* The device is currently resetting. */
1141 			continue;
1142 		}
1143 
1144 		if (spdk_unlikely(!nvme_ns_is_active(io_path->nvme_ns))) {
1145 			continue;
1146 		}
1147 
1148 		num_outstanding_reqs = spdk_nvme_qpair_get_num_outstanding_reqs(io_path->qpair->qpair);
1149 		switch (io_path->nvme_ns->ana_state) {
1150 		case SPDK_NVME_ANA_OPTIMIZED_STATE:
1151 			if (num_outstanding_reqs < opt_min_qd) {
1152 				opt_min_qd = num_outstanding_reqs;
1153 				optimized = io_path;
1154 			}
1155 			break;
1156 		case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1157 			if (num_outstanding_reqs < non_opt_min_qd) {
1158 				non_opt_min_qd = num_outstanding_reqs;
1159 				non_optimized = io_path;
1160 			}
1161 			break;
1162 		default:
1163 			break;
1164 		}
1165 	}
1166 
1167 	/* don't cache io path for BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH selector */
1168 	if (optimized != NULL) {
1169 		return optimized;
1170 	}
1171 
1172 	return non_optimized;
1173 }
1174 
1175 static inline struct nvme_io_path *
1176 bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
1177 {
1178 	if (spdk_likely(nbdev_ch->current_io_path != NULL)) {
1179 		if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE) {
1180 			return nbdev_ch->current_io_path;
1181 		} else if (nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
1182 			if (++nbdev_ch->rr_counter < nbdev_ch->rr_min_io) {
1183 				return nbdev_ch->current_io_path;
1184 			}
1185 			nbdev_ch->rr_counter = 0;
1186 		}
1187 	}
1188 
1189 	if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE ||
1190 	    nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
1191 		return _bdev_nvme_find_io_path(nbdev_ch);
1192 	} else {
1193 		return _bdev_nvme_find_io_path_min_qd(nbdev_ch);
1194 	}
1195 }
1196 
1197 /* Return true if there is any io_path whose qpair is active or ctrlr is not failed,
1198  * or false otherwise.
1199  *
1200  * If any io_path has an active qpair but find_io_path() returned NULL, its namespace
1201  * is likely to be non-accessible now but may become accessible.
1202  *
1203  * If any io_path has an unfailed ctrlr but find_io_path() returned NULL, the ctrlr
1204  * is likely to be resetting now but the reset may succeed. A ctrlr is set to unfailed
1205  * when starting to reset it but it is set to failed when the reset failed. Hence, if
1206  * a ctrlr is unfailed, it is likely that it works fine or is resetting.
1207  */
1208 static bool
1209 any_io_path_may_become_available(struct nvme_bdev_channel *nbdev_ch)
1210 {
1211 	struct nvme_io_path *io_path;
1212 
1213 	if (nbdev_ch->resetting) {
1214 		return false;
1215 	}
1216 
1217 	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
1218 		if (io_path->nvme_ns->ana_transition_timedout) {
1219 			continue;
1220 		}
1221 
1222 		if (nvme_qpair_is_connected(io_path->qpair) ||
1223 		    !nvme_ctrlr_is_failed(io_path->qpair->ctrlr)) {
1224 			return true;
1225 		}
1226 	}
1227 
1228 	return false;
1229 }
1230 
1231 static void
1232 bdev_nvme_retry_io(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
1233 {
1234 	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
1235 	struct spdk_io_channel *ch;
1236 
1237 	if (nbdev_io->io_path != NULL && nvme_io_path_is_available(nbdev_io->io_path)) {
1238 		_bdev_nvme_submit_request(nbdev_ch, bdev_io);
1239 	} else {
1240 		ch = spdk_io_channel_from_ctx(nbdev_ch);
1241 		bdev_nvme_submit_request(ch, bdev_io);
1242 	}
1243 }
1244 
1245 static int
1246 bdev_nvme_retry_ios(void *arg)
1247 {
1248 	struct nvme_bdev_channel *nbdev_ch = arg;
1249 	struct nvme_bdev_io *bio, *tmp_bio;
1250 	uint64_t now, delay_us;
1251 
1252 	now = spdk_get_ticks();
1253 
1254 	TAILQ_FOREACH_SAFE(bio, &nbdev_ch->retry_io_list, retry_link, tmp_bio) {
1255 		if (bio->retry_ticks > now) {
1256 			break;
1257 		}
1258 
1259 		TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
1260 
1261 		bdev_nvme_retry_io(nbdev_ch, spdk_bdev_io_from_ctx(bio));
1262 	}
1263 
1264 	spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1265 
1266 	bio = TAILQ_FIRST(&nbdev_ch->retry_io_list);
1267 	if (bio != NULL) {
1268 		delay_us = (bio->retry_ticks - now) * SPDK_SEC_TO_USEC / spdk_get_ticks_hz();
1269 
1270 		nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
1271 					    delay_us);
1272 	}
1273 
1274 	return SPDK_POLLER_BUSY;
1275 }
1276 
1277 static void
1278 bdev_nvme_queue_retry_io(struct nvme_bdev_channel *nbdev_ch,
1279 			 struct nvme_bdev_io *bio, uint64_t delay_ms)
1280 {
1281 	struct nvme_bdev_io *tmp_bio;
1282 
1283 	bio->retry_ticks = spdk_get_ticks() + delay_ms * spdk_get_ticks_hz() / 1000ULL;
1284 
1285 	TAILQ_FOREACH_REVERSE(tmp_bio, &nbdev_ch->retry_io_list, retry_io_head, retry_link) {
1286 		if (tmp_bio->retry_ticks <= bio->retry_ticks) {
1287 			TAILQ_INSERT_AFTER(&nbdev_ch->retry_io_list, tmp_bio, bio,
1288 					   retry_link);
1289 			return;
1290 		}
1291 	}
1292 
1293 	/* No earlier I/Os were found. This I/O must be the new head. */
1294 	TAILQ_INSERT_HEAD(&nbdev_ch->retry_io_list, bio, retry_link);
1295 
1296 	spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1297 
1298 	nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
1299 				    delay_ms * 1000ULL);
1300 }
1301 
1302 static void
1303 bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch)
1304 {
1305 	struct nvme_bdev_io *bio, *tmp_bio;
1306 
1307 	TAILQ_FOREACH_SAFE(bio, &nbdev_ch->retry_io_list, retry_link, tmp_bio) {
1308 		TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
1309 		__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_ABORTED, NULL);
1310 	}
1311 
1312 	spdk_poller_unregister(&nbdev_ch->retry_io_poller);
1313 }
1314 
1315 static int
1316 bdev_nvme_abort_retry_io(struct nvme_bdev_channel *nbdev_ch,
1317 			 struct nvme_bdev_io *bio_to_abort)
1318 {
1319 	struct nvme_bdev_io *bio;
1320 
1321 	TAILQ_FOREACH(bio, &nbdev_ch->retry_io_list, retry_link) {
1322 		if (bio == bio_to_abort) {
1323 			TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
1324 			__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_ABORTED, NULL);
1325 			return 0;
1326 		}
1327 	}
1328 
1329 	return -ENOENT;
1330 }
1331 
1332 static void
1333 bdev_nvme_update_nvme_error_stat(struct spdk_bdev_io *bdev_io, const struct spdk_nvme_cpl *cpl)
1334 {
1335 	struct nvme_bdev *nbdev;
1336 	uint16_t sct, sc;
1337 
1338 	assert(spdk_nvme_cpl_is_error(cpl));
1339 
1340 	nbdev = bdev_io->bdev->ctxt;
1341 
1342 	if (nbdev->err_stat == NULL) {
1343 		return;
1344 	}
1345 
1346 	sct = cpl->status.sct;
1347 	sc = cpl->status.sc;
1348 
1349 	pthread_mutex_lock(&nbdev->mutex);
1350 
1351 	nbdev->err_stat->status_type[sct]++;
1352 	switch (sct) {
1353 	case SPDK_NVME_SCT_GENERIC:
1354 	case SPDK_NVME_SCT_COMMAND_SPECIFIC:
1355 	case SPDK_NVME_SCT_MEDIA_ERROR:
1356 	case SPDK_NVME_SCT_PATH:
1357 		nbdev->err_stat->status[sct][sc]++;
1358 		break;
1359 	default:
1360 		break;
1361 	}
1362 
1363 	pthread_mutex_unlock(&nbdev->mutex);
1364 }
1365 
1366 static inline void
1367 bdev_nvme_update_io_path_stat(struct nvme_bdev_io *bio)
1368 {
1369 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1370 	uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
1371 	uint32_t blocklen = bdev_io->bdev->blocklen;
1372 	struct spdk_bdev_io_stat *stat;
1373 	uint64_t tsc_diff;
1374 
1375 	if (bio->io_path->stat == NULL) {
1376 		return;
1377 	}
1378 
1379 	tsc_diff = spdk_get_ticks() - bio->submit_tsc;
1380 	stat = bio->io_path->stat;
1381 
1382 	switch (bdev_io->type) {
1383 	case SPDK_BDEV_IO_TYPE_READ:
1384 		stat->bytes_read += num_blocks * blocklen;
1385 		stat->num_read_ops++;
1386 		stat->read_latency_ticks += tsc_diff;
1387 		if (stat->max_read_latency_ticks < tsc_diff) {
1388 			stat->max_read_latency_ticks = tsc_diff;
1389 		}
1390 		if (stat->min_read_latency_ticks > tsc_diff) {
1391 			stat->min_read_latency_ticks = tsc_diff;
1392 		}
1393 		break;
1394 	case SPDK_BDEV_IO_TYPE_WRITE:
1395 		stat->bytes_written += num_blocks * blocklen;
1396 		stat->num_write_ops++;
1397 		stat->write_latency_ticks += tsc_diff;
1398 		if (stat->max_write_latency_ticks < tsc_diff) {
1399 			stat->max_write_latency_ticks = tsc_diff;
1400 		}
1401 		if (stat->min_write_latency_ticks > tsc_diff) {
1402 			stat->min_write_latency_ticks = tsc_diff;
1403 		}
1404 		break;
1405 	case SPDK_BDEV_IO_TYPE_UNMAP:
1406 		stat->bytes_unmapped += num_blocks * blocklen;
1407 		stat->num_unmap_ops++;
1408 		stat->unmap_latency_ticks += tsc_diff;
1409 		if (stat->max_unmap_latency_ticks < tsc_diff) {
1410 			stat->max_unmap_latency_ticks = tsc_diff;
1411 		}
1412 		if (stat->min_unmap_latency_ticks > tsc_diff) {
1413 			stat->min_unmap_latency_ticks = tsc_diff;
1414 		}
1415 		break;
1416 	case SPDK_BDEV_IO_TYPE_ZCOPY:
1417 		/* Track the data in the start phase only */
1418 		if (!bdev_io->u.bdev.zcopy.start) {
1419 			break;
1420 		}
1421 		if (bdev_io->u.bdev.zcopy.populate) {
1422 			stat->bytes_read += num_blocks * blocklen;
1423 			stat->num_read_ops++;
1424 			stat->read_latency_ticks += tsc_diff;
1425 			if (stat->max_read_latency_ticks < tsc_diff) {
1426 				stat->max_read_latency_ticks = tsc_diff;
1427 			}
1428 			if (stat->min_read_latency_ticks > tsc_diff) {
1429 				stat->min_read_latency_ticks = tsc_diff;
1430 			}
1431 		} else {
1432 			stat->bytes_written += num_blocks * blocklen;
1433 			stat->num_write_ops++;
1434 			stat->write_latency_ticks += tsc_diff;
1435 			if (stat->max_write_latency_ticks < tsc_diff) {
1436 				stat->max_write_latency_ticks = tsc_diff;
1437 			}
1438 			if (stat->min_write_latency_ticks > tsc_diff) {
1439 				stat->min_write_latency_ticks = tsc_diff;
1440 			}
1441 		}
1442 		break;
1443 	case SPDK_BDEV_IO_TYPE_COPY:
1444 		stat->bytes_copied += num_blocks * blocklen;
1445 		stat->num_copy_ops++;
1446 		stat->copy_latency_ticks += tsc_diff;
1447 		if (stat->max_copy_latency_ticks < tsc_diff) {
1448 			stat->max_copy_latency_ticks = tsc_diff;
1449 		}
1450 		if (stat->min_copy_latency_ticks > tsc_diff) {
1451 			stat->min_copy_latency_ticks = tsc_diff;
1452 		}
1453 		break;
1454 	default:
1455 		break;
1456 	}
1457 }
1458 
1459 static bool
1460 bdev_nvme_check_retry_io(struct nvme_bdev_io *bio,
1461 			 const struct spdk_nvme_cpl *cpl,
1462 			 struct nvme_bdev_channel *nbdev_ch,
1463 			 uint64_t *_delay_ms)
1464 {
1465 	struct nvme_io_path *io_path = bio->io_path;
1466 	struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
1467 	const struct spdk_nvme_ctrlr_data *cdata;
1468 
1469 	if (spdk_nvme_cpl_is_path_error(cpl) ||
1470 	    spdk_nvme_cpl_is_aborted_sq_deletion(cpl) ||
1471 	    !nvme_io_path_is_available(io_path) ||
1472 	    !nvme_ctrlr_is_available(nvme_ctrlr)) {
1473 		bdev_nvme_clear_current_io_path(nbdev_ch);
1474 		bio->io_path = NULL;
1475 		if (spdk_nvme_cpl_is_ana_error(cpl)) {
1476 			if (nvme_ctrlr_read_ana_log_page(nvme_ctrlr) == 0) {
1477 				io_path->nvme_ns->ana_state_updating = true;
1478 			}
1479 		}
1480 		if (!any_io_path_may_become_available(nbdev_ch)) {
1481 			return false;
1482 		}
1483 		*_delay_ms = 0;
1484 	} else {
1485 		bio->retry_count++;
1486 
1487 		cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
1488 
1489 		if (cpl->status.crd != 0) {
1490 			*_delay_ms = cdata->crdt[cpl->status.crd] * 100;
1491 		} else {
1492 			*_delay_ms = 0;
1493 		}
1494 	}
1495 
1496 	return true;
1497 }
1498 
1499 static inline void
1500 bdev_nvme_io_complete_nvme_status(struct nvme_bdev_io *bio,
1501 				  const struct spdk_nvme_cpl *cpl)
1502 {
1503 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1504 	struct nvme_bdev_channel *nbdev_ch;
1505 	uint64_t delay_ms;
1506 
1507 	assert(!bdev_nvme_io_type_is_admin(bdev_io->type));
1508 
1509 	if (spdk_likely(spdk_nvme_cpl_is_success(cpl))) {
1510 		bdev_nvme_update_io_path_stat(bio);
1511 		goto complete;
1512 	}
1513 
1514 	/* Update error counts before deciding if retry is needed.
1515 	 * Hence, error counts may be more than the number of I/O errors.
1516 	 */
1517 	bdev_nvme_update_nvme_error_stat(bdev_io, cpl);
1518 
1519 	if (cpl->status.dnr != 0 || spdk_nvme_cpl_is_aborted_by_request(cpl) ||
1520 	    (g_opts.bdev_retry_count != -1 && bio->retry_count >= g_opts.bdev_retry_count)) {
1521 		goto complete;
1522 	}
1523 
1524 	/* At this point we don't know whether the sequence was successfully executed or not, so we
1525 	 * cannot retry the IO */
1526 	if (bdev_io->u.bdev.accel_sequence != NULL) {
1527 		goto complete;
1528 	}
1529 
1530 	nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
1531 
1532 	if (bdev_nvme_check_retry_io(bio, cpl, nbdev_ch, &delay_ms)) {
1533 		bdev_nvme_queue_retry_io(nbdev_ch, bio, delay_ms);
1534 		return;
1535 	}
1536 
1537 complete:
1538 	bio->retry_count = 0;
1539 	bio->submit_tsc = 0;
1540 	bdev_io->u.bdev.accel_sequence = NULL;
1541 	__bdev_nvme_io_complete(bdev_io, 0, cpl);
1542 }
1543 
1544 static inline void
1545 bdev_nvme_io_complete(struct nvme_bdev_io *bio, int rc)
1546 {
1547 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1548 	struct nvme_bdev_channel *nbdev_ch;
1549 	enum spdk_bdev_io_status io_status;
1550 
1551 	assert(!bdev_nvme_io_type_is_admin(bdev_io->type));
1552 
1553 	switch (rc) {
1554 	case 0:
1555 		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1556 		break;
1557 	case -ENOMEM:
1558 		io_status = SPDK_BDEV_IO_STATUS_NOMEM;
1559 		break;
1560 	case -ENXIO:
1561 		if (g_opts.bdev_retry_count == -1 || bio->retry_count < g_opts.bdev_retry_count) {
1562 			nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
1563 
1564 			bdev_nvme_clear_current_io_path(nbdev_ch);
1565 			bio->io_path = NULL;
1566 
1567 			if (any_io_path_may_become_available(nbdev_ch)) {
1568 				bdev_nvme_queue_retry_io(nbdev_ch, bio, 1000ULL);
1569 				return;
1570 			}
1571 		}
1572 
1573 	/* fallthrough */
1574 	default:
1575 		spdk_accel_sequence_abort(bdev_io->u.bdev.accel_sequence);
1576 		bdev_io->u.bdev.accel_sequence = NULL;
1577 		io_status = SPDK_BDEV_IO_STATUS_FAILED;
1578 		break;
1579 	}
1580 
1581 	bio->retry_count = 0;
1582 	bio->submit_tsc = 0;
1583 	__bdev_nvme_io_complete(bdev_io, io_status, NULL);
1584 }
1585 
1586 static inline void
1587 bdev_nvme_admin_complete(struct nvme_bdev_io *bio, int rc)
1588 {
1589 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
1590 	enum spdk_bdev_io_status io_status;
1591 
1592 	switch (rc) {
1593 	case 0:
1594 		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1595 		break;
1596 	case -ENOMEM:
1597 		io_status = SPDK_BDEV_IO_STATUS_NOMEM;
1598 		break;
1599 	case -ENXIO:
1600 	/* fallthrough */
1601 	default:
1602 		io_status = SPDK_BDEV_IO_STATUS_FAILED;
1603 		break;
1604 	}
1605 
1606 	__bdev_nvme_io_complete(bdev_io, io_status, NULL);
1607 }
1608 
1609 static void
1610 bdev_nvme_clear_io_path_caches_done(struct nvme_ctrlr *nvme_ctrlr,
1611 				    void *ctx, int status)
1612 {
1613 	pthread_mutex_lock(&nvme_ctrlr->mutex);
1614 
1615 	assert(nvme_ctrlr->io_path_cache_clearing == true);
1616 	nvme_ctrlr->io_path_cache_clearing = false;
1617 
1618 	if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
1619 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
1620 		return;
1621 	}
1622 
1623 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
1624 
1625 	nvme_ctrlr_unregister(nvme_ctrlr);
1626 }
1627 
1628 static void
1629 _bdev_nvme_clear_io_path_cache(struct nvme_qpair *nvme_qpair)
1630 {
1631 	struct nvme_io_path *io_path;
1632 
1633 	TAILQ_FOREACH(io_path, &nvme_qpair->io_path_list, tailq) {
1634 		if (io_path->nbdev_ch == NULL) {
1635 			continue;
1636 		}
1637 		bdev_nvme_clear_current_io_path(io_path->nbdev_ch);
1638 	}
1639 }
1640 
1641 static void
1642 bdev_nvme_clear_io_path_cache(struct nvme_ctrlr_channel_iter *i,
1643 			      struct nvme_ctrlr *nvme_ctrlr,
1644 			      struct nvme_ctrlr_channel *ctrlr_ch,
1645 			      void *ctx)
1646 {
1647 	assert(ctrlr_ch->qpair != NULL);
1648 
1649 	_bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);
1650 
1651 	nvme_ctrlr_for_each_channel_continue(i, 0);
1652 }
1653 
1654 static void
1655 bdev_nvme_clear_io_path_caches(struct nvme_ctrlr *nvme_ctrlr)
1656 {
1657 	pthread_mutex_lock(&nvme_ctrlr->mutex);
1658 	if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
1659 	    nvme_ctrlr->io_path_cache_clearing) {
1660 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
1661 		return;
1662 	}
1663 
1664 	nvme_ctrlr->io_path_cache_clearing = true;
1665 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
1666 
1667 	nvme_ctrlr_for_each_channel(nvme_ctrlr,
1668 				    bdev_nvme_clear_io_path_cache,
1669 				    NULL,
1670 				    bdev_nvme_clear_io_path_caches_done);
1671 }
1672 
1673 static struct nvme_qpair *
1674 nvme_poll_group_get_qpair(struct nvme_poll_group *group, struct spdk_nvme_qpair *qpair)
1675 {
1676 	struct nvme_qpair *nvme_qpair;
1677 
1678 	TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
1679 		if (nvme_qpair->qpair == qpair) {
1680 			break;
1681 		}
1682 	}
1683 
1684 	return nvme_qpair;
1685 }
1686 
1687 static void nvme_qpair_delete(struct nvme_qpair *nvme_qpair);
1688 
1689 static void
1690 bdev_nvme_disconnected_qpair_cb(struct spdk_nvme_qpair *qpair, void *poll_group_ctx)
1691 {
1692 	struct nvme_poll_group *group = poll_group_ctx;
1693 	struct nvme_qpair *nvme_qpair;
1694 	struct nvme_ctrlr_channel *ctrlr_ch;
1695 	int status;
1696 
1697 	nvme_qpair = nvme_poll_group_get_qpair(group, qpair);
1698 	if (nvme_qpair == NULL) {
1699 		return;
1700 	}
1701 
1702 	if (nvme_qpair->qpair != NULL) {
1703 		spdk_nvme_ctrlr_free_io_qpair(nvme_qpair->qpair);
1704 		nvme_qpair->qpair = NULL;
1705 	}
1706 
1707 	_bdev_nvme_clear_io_path_cache(nvme_qpair);
1708 
1709 	ctrlr_ch = nvme_qpair->ctrlr_ch;
1710 
1711 	if (ctrlr_ch != NULL) {
1712 		if (ctrlr_ch->reset_iter != NULL) {
1713 			/* We are in a full reset sequence. */
1714 			if (ctrlr_ch->connect_poller != NULL) {
1715 				/* qpair was failed to connect. Abort the reset sequence. */
1716 				SPDK_DEBUGLOG(bdev_nvme, "qpair %p was failed to connect. abort the reset ctrlr sequence.\n",
1717 					      qpair);
1718 				spdk_poller_unregister(&ctrlr_ch->connect_poller);
1719 				status = -1;
1720 			} else {
1721 				/* qpair was completed to disconnect. Just move to the next ctrlr_channel. */
1722 				SPDK_DEBUGLOG(bdev_nvme, "qpair %p was disconnected and freed in a reset ctrlr sequence.\n",
1723 					      qpair);
1724 				status = 0;
1725 			}
1726 			nvme_ctrlr_for_each_channel_continue(ctrlr_ch->reset_iter, status);
1727 			ctrlr_ch->reset_iter = NULL;
1728 		} else {
1729 			/* qpair was disconnected unexpectedly. Reset controller for recovery. */
1730 			SPDK_NOTICELOG("qpair %p was disconnected and freed. reset controller.\n", qpair);
1731 			bdev_nvme_failover_ctrlr(nvme_qpair->ctrlr);
1732 		}
1733 	} else {
1734 		/* In this case, ctrlr_channel is already deleted. */
1735 		SPDK_DEBUGLOG(bdev_nvme, "qpair %p was disconnected and freed. delete nvme_qpair.\n", qpair);
1736 		nvme_qpair_delete(nvme_qpair);
1737 	}
1738 }
1739 
1740 static void
1741 bdev_nvme_check_io_qpairs(struct nvme_poll_group *group)
1742 {
1743 	struct nvme_qpair *nvme_qpair;
1744 
1745 	TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
1746 		if (nvme_qpair->qpair == NULL || nvme_qpair->ctrlr_ch == NULL) {
1747 			continue;
1748 		}
1749 
1750 		if (spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
1751 		    SPDK_NVME_QPAIR_FAILURE_NONE) {
1752 			_bdev_nvme_clear_io_path_cache(nvme_qpair);
1753 		}
1754 	}
1755 }
1756 
1757 static int
1758 bdev_nvme_poll(void *arg)
1759 {
1760 	struct nvme_poll_group *group = arg;
1761 	int64_t num_completions;
1762 
1763 	if (group->collect_spin_stat && group->start_ticks == 0) {
1764 		group->start_ticks = spdk_get_ticks();
1765 	}
1766 
1767 	num_completions = spdk_nvme_poll_group_process_completions(group->group, 0,
1768 			  bdev_nvme_disconnected_qpair_cb);
1769 	if (group->collect_spin_stat) {
1770 		if (num_completions > 0) {
1771 			if (group->end_ticks != 0) {
1772 				group->spin_ticks += (group->end_ticks - group->start_ticks);
1773 				group->end_ticks = 0;
1774 			}
1775 			group->start_ticks = 0;
1776 		} else {
1777 			group->end_ticks = spdk_get_ticks();
1778 		}
1779 	}
1780 
1781 	if (spdk_unlikely(num_completions < 0)) {
1782 		bdev_nvme_check_io_qpairs(group);
1783 	}
1784 
1785 	return num_completions > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
1786 }
1787 
1788 static int bdev_nvme_poll_adminq(void *arg);
1789 
1790 static void
1791 bdev_nvme_change_adminq_poll_period(struct nvme_ctrlr *nvme_ctrlr, uint64_t new_period_us)
1792 {
1793 	spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);
1794 
1795 	nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq,
1796 					  nvme_ctrlr, new_period_us);
1797 }
1798 
1799 static int
1800 bdev_nvme_poll_adminq(void *arg)
1801 {
1802 	int32_t rc;
1803 	struct nvme_ctrlr *nvme_ctrlr = arg;
1804 	nvme_ctrlr_disconnected_cb disconnected_cb;
1805 
1806 	assert(nvme_ctrlr != NULL);
1807 
1808 	rc = spdk_nvme_ctrlr_process_admin_completions(nvme_ctrlr->ctrlr);
1809 	if (rc < 0) {
1810 		disconnected_cb = nvme_ctrlr->disconnected_cb;
1811 		nvme_ctrlr->disconnected_cb = NULL;
1812 
1813 		if (disconnected_cb != NULL) {
1814 			bdev_nvme_change_adminq_poll_period(nvme_ctrlr,
1815 							    g_opts.nvme_adminq_poll_period_us);
1816 			disconnected_cb(nvme_ctrlr);
1817 		} else {
1818 			bdev_nvme_failover_ctrlr(nvme_ctrlr);
1819 		}
1820 	} else if (spdk_nvme_ctrlr_get_admin_qp_failure_reason(nvme_ctrlr->ctrlr) !=
1821 		   SPDK_NVME_QPAIR_FAILURE_NONE) {
1822 		bdev_nvme_clear_io_path_caches(nvme_ctrlr);
1823 	}
1824 
1825 	return rc == 0 ? SPDK_POLLER_IDLE : SPDK_POLLER_BUSY;
1826 }
1827 
1828 static void
1829 nvme_bdev_free(void *io_device)
1830 {
1831 	struct nvme_bdev *nvme_disk = io_device;
1832 
1833 	pthread_mutex_destroy(&nvme_disk->mutex);
1834 	free(nvme_disk->disk.name);
1835 	free(nvme_disk->err_stat);
1836 	free(nvme_disk);
1837 }
1838 
1839 static int
1840 bdev_nvme_destruct(void *ctx)
1841 {
1842 	struct nvme_bdev *nvme_disk = ctx;
1843 	struct nvme_ns *nvme_ns, *tmp_nvme_ns;
1844 
1845 	SPDK_DTRACE_PROBE2(bdev_nvme_destruct, nvme_disk->nbdev_ctrlr->name, nvme_disk->nsid);
1846 
1847 	TAILQ_FOREACH_SAFE(nvme_ns, &nvme_disk->nvme_ns_list, tailq, tmp_nvme_ns) {
1848 		pthread_mutex_lock(&nvme_ns->ctrlr->mutex);
1849 
1850 		nvme_ns->bdev = NULL;
1851 
1852 		assert(nvme_ns->id > 0);
1853 
1854 		if (nvme_ctrlr_get_ns(nvme_ns->ctrlr, nvme_ns->id) == NULL) {
1855 			pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
1856 
1857 			nvme_ctrlr_release(nvme_ns->ctrlr);
1858 			nvme_ns_free(nvme_ns);
1859 		} else {
1860 			pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
1861 		}
1862 	}
1863 
1864 	pthread_mutex_lock(&g_bdev_nvme_mutex);
1865 	TAILQ_REMOVE(&nvme_disk->nbdev_ctrlr->bdevs, nvme_disk, tailq);
1866 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
1867 
1868 	spdk_io_device_unregister(nvme_disk, nvme_bdev_free);
1869 
1870 	return 0;
1871 }
1872 
1873 static int
1874 bdev_nvme_create_qpair(struct nvme_qpair *nvme_qpair)
1875 {
1876 	struct nvme_ctrlr *nvme_ctrlr;
1877 	struct spdk_nvme_io_qpair_opts opts;
1878 	struct spdk_nvme_qpair *qpair;
1879 	int rc;
1880 
1881 	nvme_ctrlr = nvme_qpair->ctrlr;
1882 
1883 	spdk_nvme_ctrlr_get_default_io_qpair_opts(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
1884 	opts.delay_cmd_submit = g_opts.delay_cmd_submit;
1885 	opts.create_only = true;
1886 	opts.async_mode = true;
1887 	opts.io_queue_requests = spdk_max(g_opts.io_queue_requests, opts.io_queue_requests);
1888 	g_opts.io_queue_requests = opts.io_queue_requests;
1889 
1890 	qpair = spdk_nvme_ctrlr_alloc_io_qpair(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
1891 	if (qpair == NULL) {
1892 		return -1;
1893 	}
1894 
1895 	SPDK_DTRACE_PROBE3(bdev_nvme_create_qpair, nvme_ctrlr->nbdev_ctrlr->name,
1896 			   spdk_nvme_qpair_get_id(qpair), spdk_thread_get_id(nvme_ctrlr->thread));
1897 
1898 	assert(nvme_qpair->group != NULL);
1899 
1900 	rc = spdk_nvme_poll_group_add(nvme_qpair->group->group, qpair);
1901 	if (rc != 0) {
1902 		SPDK_ERRLOG("Unable to begin polling on NVMe Channel.\n");
1903 		goto err;
1904 	}
1905 
1906 	rc = spdk_nvme_ctrlr_connect_io_qpair(nvme_ctrlr->ctrlr, qpair);
1907 	if (rc != 0) {
1908 		SPDK_ERRLOG("Unable to connect I/O qpair.\n");
1909 		goto err;
1910 	}
1911 
1912 	nvme_qpair->qpair = qpair;
1913 
1914 	if (!g_opts.disable_auto_failback) {
1915 		_bdev_nvme_clear_io_path_cache(nvme_qpair);
1916 	}
1917 
1918 	return 0;
1919 
1920 err:
1921 	spdk_nvme_ctrlr_free_io_qpair(qpair);
1922 
1923 	return rc;
1924 }
1925 
1926 static void bdev_nvme_reset_io_continue(void *cb_arg, int rc);
1927 
1928 static void
1929 bdev_nvme_complete_pending_resets(struct nvme_ctrlr_channel_iter *i,
1930 				  struct nvme_ctrlr *nvme_ctrlr,
1931 				  struct nvme_ctrlr_channel *ctrlr_ch,
1932 				  void *ctx)
1933 {
1934 	int rc = 0;
1935 	struct nvme_bdev_io *bio;
1936 
1937 	if (ctx != NULL) {
1938 		rc = -1;
1939 	}
1940 
1941 	while (!TAILQ_EMPTY(&ctrlr_ch->pending_resets)) {
1942 		bio = TAILQ_FIRST(&ctrlr_ch->pending_resets);
1943 		TAILQ_REMOVE(&ctrlr_ch->pending_resets, bio, retry_link);
1944 
1945 		bdev_nvme_reset_io_continue(bio, rc);
1946 	}
1947 
1948 	nvme_ctrlr_for_each_channel_continue(i, 0);
1949 }
1950 
1951 /* This function marks the current trid as failed by storing the current ticks
1952  * and then sets the next trid to the active trid within a controller if exists.
1953  *
1954  * The purpose of the boolean return value is to request the caller to disconnect
1955  * the current trid now to try connecting the next trid.
1956  */
1957 static bool
1958 bdev_nvme_failover_trid(struct nvme_ctrlr *nvme_ctrlr, bool remove, bool start)
1959 {
1960 	struct nvme_path_id *path_id, *next_path;
1961 	int rc __attribute__((unused));
1962 
1963 	path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
1964 	assert(path_id);
1965 	assert(path_id == nvme_ctrlr->active_path_id);
1966 	next_path = TAILQ_NEXT(path_id, link);
1967 
1968 	/* Update the last failed time. It means the trid is failed if its last
1969 	 * failed time is non-zero.
1970 	 */
1971 	path_id->last_failed_tsc = spdk_get_ticks();
1972 
1973 	if (next_path == NULL) {
1974 		/* There is no alternate trid within a controller. */
1975 		return false;
1976 	}
1977 
1978 	if (!start && nvme_ctrlr->opts.reconnect_delay_sec == 0) {
1979 		/* Connect is not retried in a controller reset sequence. Connecting
1980 		 * the next trid will be done by the next bdev_nvme_failover_ctrlr() call.
1981 		 */
1982 		return false;
1983 	}
1984 
1985 	assert(path_id->trid.trtype != SPDK_NVME_TRANSPORT_PCIE);
1986 
1987 	SPDK_NOTICELOG("Start failover from %s:%s to %s:%s\n", path_id->trid.traddr,
1988 		       path_id->trid.trsvcid,	next_path->trid.traddr, next_path->trid.trsvcid);
1989 
1990 	spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
1991 	nvme_ctrlr->active_path_id = next_path;
1992 	rc = spdk_nvme_ctrlr_set_trid(nvme_ctrlr->ctrlr, &next_path->trid);
1993 	assert(rc == 0);
1994 	TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
1995 	if (!remove) {
1996 		/** Shuffle the old trid to the end of the list and use the new one.
1997 		 * Allows for round robin through multiple connections.
1998 		 */
1999 		TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, path_id, link);
2000 	} else {
2001 		free(path_id);
2002 	}
2003 
2004 	if (start || next_path->last_failed_tsc == 0) {
2005 		/* bdev_nvme_failover_ctrlr() is just called or the next trid is not failed
2006 		 * or used yet. Try the next trid now.
2007 		 */
2008 		return true;
2009 	}
2010 
2011 	if (spdk_get_ticks() > next_path->last_failed_tsc + spdk_get_ticks_hz() *
2012 	    nvme_ctrlr->opts.reconnect_delay_sec) {
2013 		/* Enough backoff passed since the next trid failed. Try the next trid now. */
2014 		return true;
2015 	}
2016 
2017 	/* The next trid will be tried after reconnect_delay_sec seconds. */
2018 	return false;
2019 }
2020 
2021 static bool
2022 bdev_nvme_check_ctrlr_loss_timeout(struct nvme_ctrlr *nvme_ctrlr)
2023 {
2024 	int32_t elapsed;
2025 
2026 	if (nvme_ctrlr->opts.ctrlr_loss_timeout_sec == 0 ||
2027 	    nvme_ctrlr->opts.ctrlr_loss_timeout_sec == -1) {
2028 		return false;
2029 	}
2030 
2031 	elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
2032 	if (elapsed >= nvme_ctrlr->opts.ctrlr_loss_timeout_sec) {
2033 		return true;
2034 	} else {
2035 		return false;
2036 	}
2037 }
2038 
2039 static bool
2040 bdev_nvme_check_fast_io_fail_timeout(struct nvme_ctrlr *nvme_ctrlr)
2041 {
2042 	uint32_t elapsed;
2043 
2044 	if (nvme_ctrlr->opts.fast_io_fail_timeout_sec == 0) {
2045 		return false;
2046 	}
2047 
2048 	elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
2049 	if (elapsed >= nvme_ctrlr->opts.fast_io_fail_timeout_sec) {
2050 		return true;
2051 	} else {
2052 		return false;
2053 	}
2054 }
2055 
2056 static void bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success);
2057 
2058 static void
2059 nvme_ctrlr_disconnect(struct nvme_ctrlr *nvme_ctrlr, nvme_ctrlr_disconnected_cb cb_fn)
2060 {
2061 	int rc;
2062 
2063 	rc = spdk_nvme_ctrlr_disconnect(nvme_ctrlr->ctrlr);
2064 	if (rc != 0) {
2065 		/* Disconnect fails if ctrlr is already resetting or removed. In this case,
2066 		 * fail the reset sequence immediately.
2067 		 */
2068 		bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2069 		return;
2070 	}
2071 
2072 	/* spdk_nvme_ctrlr_disconnect() may complete asynchronously later by polling adminq.
2073 	 * Set callback here to execute the specified operation after ctrlr is really disconnected.
2074 	 */
2075 	assert(nvme_ctrlr->disconnected_cb == NULL);
2076 	nvme_ctrlr->disconnected_cb = cb_fn;
2077 
2078 	/* During disconnection, reduce the period to poll adminq more often. */
2079 	bdev_nvme_change_adminq_poll_period(nvme_ctrlr, 0);
2080 }
2081 
2082 enum bdev_nvme_op_after_reset {
2083 	OP_NONE,
2084 	OP_COMPLETE_PENDING_DESTRUCT,
2085 	OP_DESTRUCT,
2086 	OP_DELAYED_RECONNECT,
2087 	OP_FAILOVER,
2088 };
2089 
2090 typedef enum bdev_nvme_op_after_reset _bdev_nvme_op_after_reset;
2091 
2092 static _bdev_nvme_op_after_reset
2093 bdev_nvme_check_op_after_reset(struct nvme_ctrlr *nvme_ctrlr, bool success)
2094 {
2095 	if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
2096 		/* Complete pending destruct after reset completes. */
2097 		return OP_COMPLETE_PENDING_DESTRUCT;
2098 	} else if (nvme_ctrlr->pending_failover) {
2099 		nvme_ctrlr->pending_failover = false;
2100 		nvme_ctrlr->reset_start_tsc = 0;
2101 		return OP_FAILOVER;
2102 	} else if (success || nvme_ctrlr->opts.reconnect_delay_sec == 0) {
2103 		nvme_ctrlr->reset_start_tsc = 0;
2104 		return OP_NONE;
2105 	} else if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
2106 		return OP_DESTRUCT;
2107 	} else {
2108 		if (bdev_nvme_check_fast_io_fail_timeout(nvme_ctrlr)) {
2109 			nvme_ctrlr->fast_io_fail_timedout = true;
2110 		}
2111 		return OP_DELAYED_RECONNECT;
2112 	}
2113 }
2114 
2115 static int bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug);
2116 static void bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
2117 
2118 static int
2119 bdev_nvme_reconnect_delay_timer_expired(void *ctx)
2120 {
2121 	struct nvme_ctrlr *nvme_ctrlr = ctx;
2122 
2123 	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect_delay, nvme_ctrlr->nbdev_ctrlr->name);
2124 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2125 
2126 	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2127 
2128 	if (!nvme_ctrlr->reconnect_is_delayed) {
2129 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2130 		return SPDK_POLLER_BUSY;
2131 	}
2132 
2133 	nvme_ctrlr->reconnect_is_delayed = false;
2134 
2135 	if (nvme_ctrlr->destruct) {
2136 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2137 		return SPDK_POLLER_BUSY;
2138 	}
2139 
2140 	assert(nvme_ctrlr->resetting == false);
2141 	nvme_ctrlr->resetting = true;
2142 
2143 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2144 
2145 	spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);
2146 
2147 	bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2148 	return SPDK_POLLER_BUSY;
2149 }
2150 
2151 static void
2152 bdev_nvme_start_reconnect_delay_timer(struct nvme_ctrlr *nvme_ctrlr)
2153 {
2154 	spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);
2155 
2156 	assert(nvme_ctrlr->reconnect_is_delayed == false);
2157 	nvme_ctrlr->reconnect_is_delayed = true;
2158 
2159 	assert(nvme_ctrlr->reconnect_delay_timer == NULL);
2160 	nvme_ctrlr->reconnect_delay_timer = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_delay_timer_expired,
2161 					    nvme_ctrlr,
2162 					    nvme_ctrlr->opts.reconnect_delay_sec * SPDK_SEC_TO_USEC);
2163 }
2164 
2165 static void remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr);
2166 
2167 static void
2168 _bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2169 {
2170 	bool success = (ctx == NULL);
2171 	bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
2172 	void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
2173 	enum bdev_nvme_op_after_reset op_after_reset;
2174 
2175 	assert(nvme_ctrlr->thread == spdk_get_thread());
2176 
2177 	nvme_ctrlr->ctrlr_op_cb_fn = NULL;
2178 	nvme_ctrlr->ctrlr_op_cb_arg = NULL;
2179 
2180 	if (!success) {
2181 		SPDK_ERRLOG("Resetting controller failed.\n");
2182 	} else {
2183 		SPDK_NOTICELOG("Resetting controller successful.\n");
2184 	}
2185 
2186 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2187 	nvme_ctrlr->resetting = false;
2188 	nvme_ctrlr->dont_retry = false;
2189 	nvme_ctrlr->in_failover = false;
2190 
2191 	op_after_reset = bdev_nvme_check_op_after_reset(nvme_ctrlr, success);
2192 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2193 
2194 	/* Delay callbacks when the next operation is a failover. */
2195 	if (ctrlr_op_cb_fn && op_after_reset != OP_FAILOVER) {
2196 		ctrlr_op_cb_fn(ctrlr_op_cb_arg, success ? 0 : -1);
2197 	}
2198 
2199 	switch (op_after_reset) {
2200 	case OP_COMPLETE_PENDING_DESTRUCT:
2201 		nvme_ctrlr_unregister(nvme_ctrlr);
2202 		break;
2203 	case OP_DESTRUCT:
2204 		bdev_nvme_delete_ctrlr(nvme_ctrlr, false);
2205 		remove_discovery_entry(nvme_ctrlr);
2206 		break;
2207 	case OP_DELAYED_RECONNECT:
2208 		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_start_reconnect_delay_timer);
2209 		break;
2210 	case OP_FAILOVER:
2211 		nvme_ctrlr->ctrlr_op_cb_fn = ctrlr_op_cb_fn;
2212 		nvme_ctrlr->ctrlr_op_cb_arg = ctrlr_op_cb_arg;
2213 		bdev_nvme_failover_ctrlr(nvme_ctrlr);
2214 		break;
2215 	default:
2216 		break;
2217 	}
2218 }
2219 
2220 static void
2221 bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success)
2222 {
2223 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2224 	if (!success) {
2225 		/* Connecting the active trid failed. Set the next alternate trid to the
2226 		 * active trid if it exists.
2227 		 */
2228 		if (bdev_nvme_failover_trid(nvme_ctrlr, false, false)) {
2229 			/* The next alternate trid exists and is ready to try. Try it now. */
2230 			pthread_mutex_unlock(&nvme_ctrlr->mutex);
2231 
2232 			nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
2233 			return;
2234 		}
2235 
2236 		/* We came here if there is no alternate trid or if the next trid exists but
2237 		 * is not ready to try. We will try the active trid after reconnect_delay_sec
2238 		 * seconds if it is non-zero or at the next reset call otherwise.
2239 		 */
2240 	} else {
2241 		/* Connecting the active trid succeeded. Clear the last failed time because it
2242 		 * means the trid is failed if its last failed time is non-zero.
2243 		 */
2244 		nvme_ctrlr->active_path_id->last_failed_tsc = 0;
2245 	}
2246 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2247 
2248 	/* Make sure we clear any pending resets before returning. */
2249 	nvme_ctrlr_for_each_channel(nvme_ctrlr,
2250 				    bdev_nvme_complete_pending_resets,
2251 				    success ? NULL : (void *)0x1,
2252 				    _bdev_nvme_reset_ctrlr_complete);
2253 }
2254 
2255 static void
2256 bdev_nvme_reset_create_qpairs_failed(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2257 {
2258 	bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2259 }
2260 
2261 static void
2262 bdev_nvme_reset_destroy_qpair(struct nvme_ctrlr_channel_iter *i,
2263 			      struct nvme_ctrlr *nvme_ctrlr,
2264 			      struct nvme_ctrlr_channel *ctrlr_ch, void *ctx)
2265 {
2266 	struct nvme_qpair *nvme_qpair;
2267 
2268 	nvme_qpair = ctrlr_ch->qpair;
2269 	assert(nvme_qpair != NULL);
2270 
2271 	_bdev_nvme_clear_io_path_cache(nvme_qpair);
2272 
2273 	if (nvme_qpair->qpair != NULL) {
2274 		if (nvme_qpair->ctrlr->dont_retry) {
2275 			spdk_nvme_qpair_set_abort_dnr(nvme_qpair->qpair, true);
2276 		}
2277 		spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);
2278 
2279 		/* The current full reset sequence will move to the next
2280 		 * ctrlr_channel after the qpair is actually disconnected.
2281 		 */
2282 		assert(ctrlr_ch->reset_iter == NULL);
2283 		ctrlr_ch->reset_iter = i;
2284 	} else {
2285 		nvme_ctrlr_for_each_channel_continue(i, 0);
2286 	}
2287 }
2288 
2289 static void
2290 bdev_nvme_reset_create_qpairs_done(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2291 {
2292 	if (status == 0) {
2293 		bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, true);
2294 	} else {
2295 		/* Delete the added qpairs and quiesce ctrlr to make the states clean. */
2296 		nvme_ctrlr_for_each_channel(nvme_ctrlr,
2297 					    bdev_nvme_reset_destroy_qpair,
2298 					    NULL,
2299 					    bdev_nvme_reset_create_qpairs_failed);
2300 	}
2301 }
2302 
2303 static int
2304 bdev_nvme_reset_check_qpair_connected(void *ctx)
2305 {
2306 	struct nvme_ctrlr_channel *ctrlr_ch = ctx;
2307 
2308 	if (ctrlr_ch->reset_iter == NULL) {
2309 		/* qpair was already failed to connect and the reset sequence is being aborted. */
2310 		assert(ctrlr_ch->connect_poller == NULL);
2311 		assert(ctrlr_ch->qpair->qpair == NULL);
2312 		return SPDK_POLLER_BUSY;
2313 	}
2314 
2315 	assert(ctrlr_ch->qpair->qpair != NULL);
2316 
2317 	if (!spdk_nvme_qpair_is_connected(ctrlr_ch->qpair->qpair)) {
2318 		return SPDK_POLLER_BUSY;
2319 	}
2320 
2321 	spdk_poller_unregister(&ctrlr_ch->connect_poller);
2322 
2323 	/* qpair was completed to connect. Move to the next ctrlr_channel */
2324 	nvme_ctrlr_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
2325 	ctrlr_ch->reset_iter = NULL;
2326 
2327 	if (!g_opts.disable_auto_failback) {
2328 		_bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);
2329 	}
2330 
2331 	return SPDK_POLLER_BUSY;
2332 }
2333 
2334 static void
2335 bdev_nvme_reset_create_qpair(struct nvme_ctrlr_channel_iter *i,
2336 			     struct nvme_ctrlr *nvme_ctrlr,
2337 			     struct nvme_ctrlr_channel *ctrlr_ch,
2338 			     void *ctx)
2339 {
2340 	int rc;
2341 
2342 	rc = bdev_nvme_create_qpair(ctrlr_ch->qpair);
2343 	if (rc == 0) {
2344 		ctrlr_ch->connect_poller = SPDK_POLLER_REGISTER(bdev_nvme_reset_check_qpair_connected,
2345 					   ctrlr_ch, 0);
2346 
2347 		/* The current full reset sequence will move to the next
2348 		 * ctrlr_channel after the qpair is actually connected.
2349 		 */
2350 		assert(ctrlr_ch->reset_iter == NULL);
2351 		ctrlr_ch->reset_iter = i;
2352 	} else {
2353 		nvme_ctrlr_for_each_channel_continue(i, rc);
2354 	}
2355 }
2356 
2357 static void
2358 nvme_ctrlr_check_namespaces(struct nvme_ctrlr *nvme_ctrlr)
2359 {
2360 	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
2361 	struct nvme_ns *nvme_ns;
2362 
2363 	for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
2364 	     nvme_ns != NULL;
2365 	     nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
2366 		if (!spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
2367 			SPDK_DEBUGLOG(bdev_nvme, "NSID %u was removed during reset.\n", nvme_ns->id);
2368 			/* NS can be added again. Just nullify nvme_ns->ns. */
2369 			nvme_ns->ns = NULL;
2370 		}
2371 	}
2372 }
2373 
2374 
2375 static int
2376 bdev_nvme_reconnect_ctrlr_poll(void *arg)
2377 {
2378 	struct nvme_ctrlr *nvme_ctrlr = arg;
2379 	int rc = -ETIMEDOUT;
2380 
2381 	if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
2382 		/* Mark the ctrlr as failed. The next call to
2383 		 * spdk_nvme_ctrlr_reconnect_poll_async() will then
2384 		 * do the necessary cleanup and return failure.
2385 		 */
2386 		spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
2387 	}
2388 
2389 	rc = spdk_nvme_ctrlr_reconnect_poll_async(nvme_ctrlr->ctrlr);
2390 	if (rc == -EAGAIN) {
2391 		return SPDK_POLLER_BUSY;
2392 	}
2393 
2394 	spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
2395 	if (rc == 0) {
2396 		nvme_ctrlr_check_namespaces(nvme_ctrlr);
2397 
2398 		/* Recreate all of the I/O queue pairs */
2399 		nvme_ctrlr_for_each_channel(nvme_ctrlr,
2400 					    bdev_nvme_reset_create_qpair,
2401 					    NULL,
2402 					    bdev_nvme_reset_create_qpairs_done);
2403 	} else {
2404 		bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
2405 	}
2406 	return SPDK_POLLER_BUSY;
2407 }
2408 
2409 static void
2410 bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2411 {
2412 	spdk_nvme_ctrlr_reconnect_async(nvme_ctrlr->ctrlr);
2413 
2414 	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect, nvme_ctrlr->nbdev_ctrlr->name);
2415 	assert(nvme_ctrlr->reset_detach_poller == NULL);
2416 	nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_ctrlr_poll,
2417 					  nvme_ctrlr, 0);
2418 }
2419 
2420 static void
2421 bdev_nvme_reset_destroy_qpair_done(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2422 {
2423 	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reset, nvme_ctrlr->nbdev_ctrlr->name);
2424 	assert(status == 0);
2425 
2426 	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2427 		bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2428 	} else {
2429 		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
2430 	}
2431 }
2432 
2433 static void
2434 bdev_nvme_reset_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
2435 {
2436 	nvme_ctrlr_for_each_channel(nvme_ctrlr,
2437 				    bdev_nvme_reset_destroy_qpair,
2438 				    NULL,
2439 				    bdev_nvme_reset_destroy_qpair_done);
2440 }
2441 
2442 static void
2443 bdev_nvme_reconnect_ctrlr_now(void *ctx)
2444 {
2445 	struct nvme_ctrlr *nvme_ctrlr = ctx;
2446 
2447 	assert(nvme_ctrlr->resetting == true);
2448 	assert(nvme_ctrlr->thread == spdk_get_thread());
2449 
2450 	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2451 
2452 	spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);
2453 
2454 	bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
2455 }
2456 
2457 static void
2458 _bdev_nvme_reset_ctrlr(void *ctx)
2459 {
2460 	struct nvme_ctrlr *nvme_ctrlr = ctx;
2461 
2462 	assert(nvme_ctrlr->resetting == true);
2463 	assert(nvme_ctrlr->thread == spdk_get_thread());
2464 
2465 	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2466 		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reset_destroy_qpairs);
2467 	} else {
2468 		bdev_nvme_reset_destroy_qpairs(nvme_ctrlr);
2469 	}
2470 }
2471 
2472 static int
2473 bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2474 {
2475 	spdk_msg_fn msg_fn;
2476 
2477 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2478 	if (nvme_ctrlr->destruct) {
2479 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2480 		return -ENXIO;
2481 	}
2482 
2483 	if (nvme_ctrlr->resetting) {
2484 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2485 		SPDK_NOTICELOG("Unable to perform reset, already in progress.\n");
2486 		return -EBUSY;
2487 	}
2488 
2489 	if (nvme_ctrlr->disabled) {
2490 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2491 		SPDK_NOTICELOG("Unable to perform reset. Controller is disabled.\n");
2492 		return -EALREADY;
2493 	}
2494 
2495 	nvme_ctrlr->resetting = true;
2496 	nvme_ctrlr->dont_retry = true;
2497 
2498 	if (nvme_ctrlr->reconnect_is_delayed) {
2499 		SPDK_DEBUGLOG(bdev_nvme, "Reconnect is already scheduled.\n");
2500 		msg_fn = bdev_nvme_reconnect_ctrlr_now;
2501 		nvme_ctrlr->reconnect_is_delayed = false;
2502 	} else {
2503 		msg_fn = _bdev_nvme_reset_ctrlr;
2504 		assert(nvme_ctrlr->reset_start_tsc == 0);
2505 	}
2506 
2507 	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2508 
2509 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2510 
2511 	spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
2512 	return 0;
2513 }
2514 
2515 static int
2516 bdev_nvme_enable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2517 {
2518 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2519 	if (nvme_ctrlr->destruct) {
2520 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2521 		return -ENXIO;
2522 	}
2523 
2524 	if (nvme_ctrlr->resetting) {
2525 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2526 		return -EBUSY;
2527 	}
2528 
2529 	if (!nvme_ctrlr->disabled) {
2530 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2531 		return -EALREADY;
2532 	}
2533 
2534 	nvme_ctrlr->disabled = false;
2535 	nvme_ctrlr->resetting = true;
2536 
2537 	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2538 
2539 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2540 
2541 	spdk_thread_send_msg(nvme_ctrlr->thread, bdev_nvme_reconnect_ctrlr_now, nvme_ctrlr);
2542 	return 0;
2543 }
2544 
2545 static void
2546 _bdev_nvme_disable_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2547 {
2548 	bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
2549 	void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
2550 	enum bdev_nvme_op_after_reset op_after_disable;
2551 
2552 	assert(nvme_ctrlr->thread == spdk_get_thread());
2553 
2554 	nvme_ctrlr->ctrlr_op_cb_fn = NULL;
2555 	nvme_ctrlr->ctrlr_op_cb_arg = NULL;
2556 
2557 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2558 
2559 	nvme_ctrlr->resetting = false;
2560 	nvme_ctrlr->dont_retry = false;
2561 
2562 	op_after_disable = bdev_nvme_check_op_after_reset(nvme_ctrlr, true);
2563 
2564 	nvme_ctrlr->disabled = true;
2565 	spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);
2566 
2567 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2568 
2569 	if (ctrlr_op_cb_fn) {
2570 		ctrlr_op_cb_fn(ctrlr_op_cb_arg, 0);
2571 	}
2572 
2573 	switch (op_after_disable) {
2574 	case OP_COMPLETE_PENDING_DESTRUCT:
2575 		nvme_ctrlr_unregister(nvme_ctrlr);
2576 		break;
2577 	default:
2578 		break;
2579 	}
2580 
2581 }
2582 
2583 static void
2584 bdev_nvme_disable_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr)
2585 {
2586 	/* Make sure we clear any pending resets before returning. */
2587 	nvme_ctrlr_for_each_channel(nvme_ctrlr,
2588 				    bdev_nvme_complete_pending_resets,
2589 				    NULL,
2590 				    _bdev_nvme_disable_ctrlr_complete);
2591 }
2592 
2593 static void
2594 bdev_nvme_disable_destroy_qpairs_done(struct nvme_ctrlr *nvme_ctrlr, void *ctx, int status)
2595 {
2596 	assert(status == 0);
2597 
2598 	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2599 		bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
2600 	} else {
2601 		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_ctrlr_complete);
2602 	}
2603 }
2604 
2605 static void
2606 bdev_nvme_disable_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
2607 {
2608 	nvme_ctrlr_for_each_channel(nvme_ctrlr,
2609 				    bdev_nvme_reset_destroy_qpair,
2610 				    NULL,
2611 				    bdev_nvme_disable_destroy_qpairs_done);
2612 }
2613 
2614 static void
2615 _bdev_nvme_cancel_reconnect_and_disable_ctrlr(void *ctx)
2616 {
2617 	struct nvme_ctrlr *nvme_ctrlr = ctx;
2618 
2619 	assert(nvme_ctrlr->resetting == true);
2620 	assert(nvme_ctrlr->thread == spdk_get_thread());
2621 
2622 	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);
2623 
2624 	bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
2625 }
2626 
2627 static void
2628 _bdev_nvme_disconnect_and_disable_ctrlr(void *ctx)
2629 {
2630 	struct nvme_ctrlr *nvme_ctrlr = ctx;
2631 
2632 	assert(nvme_ctrlr->resetting == true);
2633 	assert(nvme_ctrlr->thread == spdk_get_thread());
2634 
2635 	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
2636 		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_destroy_qpairs);
2637 	} else {
2638 		bdev_nvme_disable_destroy_qpairs(nvme_ctrlr);
2639 	}
2640 }
2641 
2642 static int
2643 bdev_nvme_disable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
2644 {
2645 	spdk_msg_fn msg_fn;
2646 
2647 	pthread_mutex_lock(&nvme_ctrlr->mutex);
2648 	if (nvme_ctrlr->destruct) {
2649 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2650 		return -ENXIO;
2651 	}
2652 
2653 	if (nvme_ctrlr->resetting) {
2654 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2655 		return -EBUSY;
2656 	}
2657 
2658 	if (nvme_ctrlr->disabled) {
2659 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
2660 		return -EALREADY;
2661 	}
2662 
2663 	nvme_ctrlr->resetting = true;
2664 	nvme_ctrlr->dont_retry = true;
2665 
2666 	if (nvme_ctrlr->reconnect_is_delayed) {
2667 		msg_fn = _bdev_nvme_cancel_reconnect_and_disable_ctrlr;
2668 		nvme_ctrlr->reconnect_is_delayed = false;
2669 	} else {
2670 		msg_fn = _bdev_nvme_disconnect_and_disable_ctrlr;
2671 	}
2672 
2673 	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
2674 
2675 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
2676 
2677 	spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
2678 	return 0;
2679 }
2680 
2681 static int
2682 nvme_ctrlr_op(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
2683 	      bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2684 {
2685 	int rc;
2686 
2687 	switch (op) {
2688 	case NVME_CTRLR_OP_RESET:
2689 		rc = bdev_nvme_reset_ctrlr(nvme_ctrlr);
2690 		break;
2691 	case NVME_CTRLR_OP_ENABLE:
2692 		rc = bdev_nvme_enable_ctrlr(nvme_ctrlr);
2693 		break;
2694 	case NVME_CTRLR_OP_DISABLE:
2695 		rc = bdev_nvme_disable_ctrlr(nvme_ctrlr);
2696 		break;
2697 	default:
2698 		rc = -EINVAL;
2699 		break;
2700 	}
2701 
2702 	if (rc == 0) {
2703 		assert(nvme_ctrlr->ctrlr_op_cb_fn == NULL);
2704 		assert(nvme_ctrlr->ctrlr_op_cb_arg == NULL);
2705 		nvme_ctrlr->ctrlr_op_cb_fn = cb_fn;
2706 		nvme_ctrlr->ctrlr_op_cb_arg = cb_arg;
2707 	}
2708 	return rc;
2709 }
2710 
2711 struct nvme_ctrlr_op_rpc_ctx {
2712 	struct nvme_ctrlr *nvme_ctrlr;
2713 	struct spdk_thread *orig_thread;
2714 	enum nvme_ctrlr_op op;
2715 	int rc;
2716 	bdev_nvme_ctrlr_op_cb cb_fn;
2717 	void *cb_arg;
2718 };
2719 
2720 static void
2721 _nvme_ctrlr_op_rpc_complete(void *_ctx)
2722 {
2723 	struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
2724 
2725 	assert(ctx != NULL);
2726 	assert(ctx->cb_fn != NULL);
2727 
2728 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
2729 
2730 	free(ctx);
2731 }
2732 
2733 static void
2734 nvme_ctrlr_op_rpc_complete(void *cb_arg, int rc)
2735 {
2736 	struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;
2737 
2738 	ctx->rc = rc;
2739 
2740 	spdk_thread_send_msg(ctx->orig_thread, _nvme_ctrlr_op_rpc_complete, ctx);
2741 }
2742 
2743 void
2744 nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
2745 		  bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2746 {
2747 	struct nvme_ctrlr_op_rpc_ctx *ctx;
2748 	int rc;
2749 
2750 	assert(cb_fn != NULL);
2751 
2752 	ctx = calloc(1, sizeof(*ctx));
2753 	if (ctx == NULL) {
2754 		SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
2755 		cb_fn(cb_arg, -ENOMEM);
2756 		return;
2757 	}
2758 
2759 	ctx->orig_thread = spdk_get_thread();
2760 	ctx->cb_fn = cb_fn;
2761 	ctx->cb_arg = cb_arg;
2762 
2763 	rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_ctrlr_op_rpc_complete, ctx);
2764 	if (rc == 0) {
2765 		return;
2766 	} else if (rc == -EALREADY) {
2767 		rc = 0;
2768 	}
2769 
2770 	nvme_ctrlr_op_rpc_complete(ctx, rc);
2771 }
2772 
2773 static void nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc);
2774 
2775 static void
2776 _nvme_bdev_ctrlr_op_rpc_continue(void *_ctx)
2777 {
2778 	struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
2779 	struct nvme_ctrlr *prev_nvme_ctrlr, *next_nvme_ctrlr;
2780 	int rc;
2781 
2782 	prev_nvme_ctrlr = ctx->nvme_ctrlr;
2783 	ctx->nvme_ctrlr = NULL;
2784 
2785 	if (ctx->rc != 0) {
2786 		goto complete;
2787 	}
2788 
2789 	next_nvme_ctrlr = TAILQ_NEXT(prev_nvme_ctrlr, tailq);
2790 	if (next_nvme_ctrlr == NULL) {
2791 		goto complete;
2792 	}
2793 
2794 	rc = nvme_ctrlr_op(next_nvme_ctrlr, ctx->op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
2795 	if (rc == 0) {
2796 		ctx->nvme_ctrlr = next_nvme_ctrlr;
2797 		return;
2798 	} else if (rc == -EALREADY) {
2799 		ctx->nvme_ctrlr = next_nvme_ctrlr;
2800 		rc = 0;
2801 	}
2802 
2803 	ctx->rc = rc;
2804 
2805 complete:
2806 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
2807 	free(ctx);
2808 }
2809 
2810 static void
2811 nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc)
2812 {
2813 	struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;
2814 
2815 	ctx->rc = rc;
2816 
2817 	spdk_thread_send_msg(ctx->orig_thread, _nvme_bdev_ctrlr_op_rpc_continue, ctx);
2818 }
2819 
2820 void
2821 nvme_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op,
2822 		       bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
2823 {
2824 	struct nvme_ctrlr_op_rpc_ctx *ctx;
2825 	struct nvme_ctrlr *nvme_ctrlr;
2826 	int rc;
2827 
2828 	assert(cb_fn != NULL);
2829 
2830 	ctx = calloc(1, sizeof(*ctx));
2831 	if (ctx == NULL) {
2832 		SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
2833 		cb_fn(cb_arg, -ENOMEM);
2834 		return;
2835 	}
2836 
2837 	ctx->orig_thread = spdk_get_thread();
2838 	ctx->op = op;
2839 	ctx->cb_fn = cb_fn;
2840 	ctx->cb_arg = cb_arg;
2841 
2842 	nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
2843 	assert(nvme_ctrlr != NULL);
2844 
2845 	rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
2846 	if (rc == 0) {
2847 		ctx->nvme_ctrlr = nvme_ctrlr;
2848 		return;
2849 	} else if (rc == -EALREADY) {
2850 		ctx->nvme_ctrlr = nvme_ctrlr;
2851 		rc = 0;
2852 	}
2853 
2854 	nvme_bdev_ctrlr_op_rpc_continue(ctx, rc);
2855 }
2856 
2857 static int _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio);
2858 
2859 static void
2860 bdev_nvme_unfreeze_bdev_channel_done(struct nvme_bdev *nbdev, void *ctx, int status)
2861 {
2862 	struct nvme_bdev_io *bio = ctx;
2863 	enum spdk_bdev_io_status io_status;
2864 
2865 	if (bio->cpl.cdw0 == 0) {
2866 		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2867 	} else {
2868 		io_status = SPDK_BDEV_IO_STATUS_FAILED;
2869 	}
2870 
2871 	__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), io_status, NULL);
2872 }
2873 
2874 static void
2875 bdev_nvme_unfreeze_bdev_channel(struct nvme_bdev_channel_iter *i,
2876 				struct nvme_bdev *nbdev,
2877 				struct nvme_bdev_channel *nbdev_ch, void *ctx)
2878 {
2879 	bdev_nvme_abort_retry_ios(nbdev_ch);
2880 	nbdev_ch->resetting = false;
2881 
2882 	nvme_bdev_for_each_channel_continue(i, 0);
2883 }
2884 
2885 static void
2886 bdev_nvme_reset_io_complete(struct nvme_bdev_io *bio)
2887 {
2888 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2889 	struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
2890 
2891 	/* Abort all queued I/Os for retry. */
2892 	nvme_bdev_for_each_channel(nbdev,
2893 				   bdev_nvme_unfreeze_bdev_channel,
2894 				   bio,
2895 				   bdev_nvme_unfreeze_bdev_channel_done);
2896 }
2897 
2898 static void
2899 _bdev_nvme_reset_io_continue(void *ctx)
2900 {
2901 	struct nvme_bdev_io *bio = ctx;
2902 	struct nvme_io_path *prev_io_path, *next_io_path;
2903 	int rc;
2904 
2905 	prev_io_path = bio->io_path;
2906 	bio->io_path = NULL;
2907 
2908 	next_io_path = STAILQ_NEXT(prev_io_path, stailq);
2909 	if (next_io_path == NULL) {
2910 		goto complete;
2911 	}
2912 
2913 	rc = _bdev_nvme_reset_io(next_io_path, bio);
2914 	if (rc == 0) {
2915 		return;
2916 	}
2917 
2918 complete:
2919 	bdev_nvme_reset_io_complete(bio);
2920 }
2921 
2922 static void
2923 bdev_nvme_reset_io_continue(void *cb_arg, int rc)
2924 {
2925 	struct nvme_bdev_io *bio = cb_arg;
2926 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2927 
2928 	/* Reset status is initialized as "failed". Set to "success" once we have at least one
2929 	 * successfully reset nvme_ctrlr.
2930 	 */
2931 	if (rc == 0) {
2932 		bio->cpl.cdw0 = 0;
2933 	}
2934 
2935 	spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), _bdev_nvme_reset_io_continue, bio);
2936 }
2937 
2938 static int
2939 _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio)
2940 {
2941 	struct nvme_ctrlr_channel *ctrlr_ch;
2942 	int rc;
2943 
2944 	rc = nvme_ctrlr_op(io_path->qpair->ctrlr, NVME_CTRLR_OP_RESET,
2945 			   bdev_nvme_reset_io_continue, bio);
2946 	if (rc != 0 && rc != -EBUSY) {
2947 		return rc;
2948 	}
2949 
2950 	assert(bio->io_path == NULL);
2951 	bio->io_path = io_path;
2952 
2953 	if (rc == -EBUSY) {
2954 		ctrlr_ch = io_path->qpair->ctrlr_ch;
2955 		assert(ctrlr_ch != NULL);
2956 		/*
2957 		 * Reset call is queued only if it is from the app framework. This is on purpose so that
2958 		 * we don't interfere with the app framework reset strategy. i.e. we are deferring to the
2959 		 * upper level. If they are in the middle of a reset, we won't try to schedule another one.
2960 		 */
2961 		TAILQ_INSERT_TAIL(&ctrlr_ch->pending_resets, bio, retry_link);
2962 	}
2963 
2964 	return 0;
2965 }
2966 
2967 static void
2968 bdev_nvme_freeze_bdev_channel_done(struct nvme_bdev *nbdev, void *ctx, int status)
2969 {
2970 	struct nvme_bdev_io *bio = ctx;
2971 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2972 	struct nvme_bdev_channel *nbdev_ch;
2973 	struct nvme_io_path *io_path;
2974 	int rc;
2975 
2976 	nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
2977 
2978 	/* Initialize with failed status. With multipath it is enough to have at least one successful
2979 	 * nvme_ctrlr reset. If there is none, reset status will remain failed.
2980 	 */
2981 	bio->cpl.cdw0 = 1;
2982 
2983 	/* Reset all nvme_ctrlrs of a bdev controller sequentially. */
2984 	io_path = STAILQ_FIRST(&nbdev_ch->io_path_list);
2985 	assert(io_path != NULL);
2986 
2987 	rc = _bdev_nvme_reset_io(io_path, bio);
2988 	if (rc != 0) {
2989 		/* If the current nvme_ctrlr is disabled, skip it and move to the next nvme_ctrlr. */
2990 		rc = (rc == -EALREADY) ? 0 : rc;
2991 
2992 		bdev_nvme_reset_io_continue(bio, rc);
2993 	}
2994 }
2995 
2996 static void
2997 bdev_nvme_freeze_bdev_channel(struct nvme_bdev_channel_iter *i,
2998 			      struct nvme_bdev *nbdev,
2999 			      struct nvme_bdev_channel *nbdev_ch, void *ctx)
3000 {
3001 	nbdev_ch->resetting = true;
3002 
3003 	nvme_bdev_for_each_channel_continue(i, 0);
3004 }
3005 
3006 static void
3007 bdev_nvme_reset_io(struct nvme_bdev *nbdev, struct nvme_bdev_io *bio)
3008 {
3009 	nvme_bdev_for_each_channel(nbdev,
3010 				   bdev_nvme_freeze_bdev_channel,
3011 				   bio,
3012 				   bdev_nvme_freeze_bdev_channel_done);
3013 }
3014 
3015 static int
3016 bdev_nvme_failover_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool remove)
3017 {
3018 	if (nvme_ctrlr->destruct) {
3019 		/* Don't bother resetting if the controller is in the process of being destructed. */
3020 		return -ENXIO;
3021 	}
3022 
3023 	if (nvme_ctrlr->resetting) {
3024 		if (!nvme_ctrlr->in_failover) {
3025 			SPDK_NOTICELOG("Reset is already in progress. Defer failover until reset completes.\n");
3026 
3027 			/* Defer failover until reset completes. */
3028 			nvme_ctrlr->pending_failover = true;
3029 			return -EINPROGRESS;
3030 		} else {
3031 			SPDK_NOTICELOG("Unable to perform failover, already in progress.\n");
3032 			return -EBUSY;
3033 		}
3034 	}
3035 
3036 	bdev_nvme_failover_trid(nvme_ctrlr, remove, true);
3037 
3038 	if (nvme_ctrlr->reconnect_is_delayed) {
3039 		SPDK_NOTICELOG("Reconnect is already scheduled.\n");
3040 
3041 		/* We rely on the next reconnect for the failover. */
3042 		return -EALREADY;
3043 	}
3044 
3045 	if (nvme_ctrlr->disabled) {
3046 		SPDK_NOTICELOG("Controller is disabled.\n");
3047 
3048 		/* We rely on the enablement for the failover. */
3049 		return -EALREADY;
3050 	}
3051 
3052 	nvme_ctrlr->resetting = true;
3053 	nvme_ctrlr->in_failover = true;
3054 
3055 	assert(nvme_ctrlr->reset_start_tsc == 0);
3056 	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();
3057 
3058 	return 0;
3059 }
3060 
3061 static int
3062 bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
3063 {
3064 	int rc;
3065 
3066 	pthread_mutex_lock(&nvme_ctrlr->mutex);
3067 	rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, false);
3068 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
3069 
3070 	if (rc == 0) {
3071 		spdk_thread_send_msg(nvme_ctrlr->thread, _bdev_nvme_reset_ctrlr, nvme_ctrlr);
3072 	} else if (rc == -EALREADY) {
3073 		rc = 0;
3074 	}
3075 
3076 	return rc;
3077 }
3078 
3079 static int bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks,
3080 			   uint64_t num_blocks);
3081 
3082 static int bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks,
3083 				  uint64_t num_blocks);
3084 
3085 static int bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks,
3086 			  uint64_t src_offset_blocks,
3087 			  uint64_t num_blocks);
3088 
3089 static void
3090 bdev_nvme_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3091 		     bool success)
3092 {
3093 	struct nvme_bdev_io *bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3094 	int ret;
3095 
3096 	if (!success) {
3097 		ret = -EINVAL;
3098 		goto exit;
3099 	}
3100 
3101 	if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
3102 		ret = -ENXIO;
3103 		goto exit;
3104 	}
3105 
3106 	ret = bdev_nvme_readv(bio,
3107 			      bdev_io->u.bdev.iovs,
3108 			      bdev_io->u.bdev.iovcnt,
3109 			      bdev_io->u.bdev.md_buf,
3110 			      bdev_io->u.bdev.num_blocks,
3111 			      bdev_io->u.bdev.offset_blocks,
3112 			      bdev_io->u.bdev.dif_check_flags,
3113 			      bdev_io->u.bdev.memory_domain,
3114 			      bdev_io->u.bdev.memory_domain_ctx,
3115 			      bdev_io->u.bdev.accel_sequence);
3116 
3117 exit:
3118 	if (spdk_unlikely(ret != 0)) {
3119 		bdev_nvme_io_complete(bio, ret);
3120 	}
3121 }
3122 
3123 static inline void
3124 _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
3125 {
3126 	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3127 	struct spdk_bdev *bdev = bdev_io->bdev;
3128 	struct nvme_bdev_io *nbdev_io_to_abort;
3129 	int rc = 0;
3130 
3131 	switch (bdev_io->type) {
3132 	case SPDK_BDEV_IO_TYPE_READ:
3133 		if (bdev_io->u.bdev.iovs && bdev_io->u.bdev.iovs[0].iov_base) {
3134 
3135 			rc = bdev_nvme_readv(nbdev_io,
3136 					     bdev_io->u.bdev.iovs,
3137 					     bdev_io->u.bdev.iovcnt,
3138 					     bdev_io->u.bdev.md_buf,
3139 					     bdev_io->u.bdev.num_blocks,
3140 					     bdev_io->u.bdev.offset_blocks,
3141 					     bdev_io->u.bdev.dif_check_flags,
3142 					     bdev_io->u.bdev.memory_domain,
3143 					     bdev_io->u.bdev.memory_domain_ctx,
3144 					     bdev_io->u.bdev.accel_sequence);
3145 		} else {
3146 			spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb,
3147 					     bdev_io->u.bdev.num_blocks * bdev->blocklen);
3148 			rc = 0;
3149 		}
3150 		break;
3151 	case SPDK_BDEV_IO_TYPE_WRITE:
3152 		rc = bdev_nvme_writev(nbdev_io,
3153 				      bdev_io->u.bdev.iovs,
3154 				      bdev_io->u.bdev.iovcnt,
3155 				      bdev_io->u.bdev.md_buf,
3156 				      bdev_io->u.bdev.num_blocks,
3157 				      bdev_io->u.bdev.offset_blocks,
3158 				      bdev_io->u.bdev.dif_check_flags,
3159 				      bdev_io->u.bdev.memory_domain,
3160 				      bdev_io->u.bdev.memory_domain_ctx,
3161 				      bdev_io->u.bdev.accel_sequence,
3162 				      bdev_io->u.bdev.nvme_cdw12,
3163 				      bdev_io->u.bdev.nvme_cdw13);
3164 		break;
3165 	case SPDK_BDEV_IO_TYPE_COMPARE:
3166 		rc = bdev_nvme_comparev(nbdev_io,
3167 					bdev_io->u.bdev.iovs,
3168 					bdev_io->u.bdev.iovcnt,
3169 					bdev_io->u.bdev.md_buf,
3170 					bdev_io->u.bdev.num_blocks,
3171 					bdev_io->u.bdev.offset_blocks,
3172 					bdev_io->u.bdev.dif_check_flags);
3173 		break;
3174 	case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
3175 		rc = bdev_nvme_comparev_and_writev(nbdev_io,
3176 						   bdev_io->u.bdev.iovs,
3177 						   bdev_io->u.bdev.iovcnt,
3178 						   bdev_io->u.bdev.fused_iovs,
3179 						   bdev_io->u.bdev.fused_iovcnt,
3180 						   bdev_io->u.bdev.md_buf,
3181 						   bdev_io->u.bdev.num_blocks,
3182 						   bdev_io->u.bdev.offset_blocks,
3183 						   bdev_io->u.bdev.dif_check_flags);
3184 		break;
3185 	case SPDK_BDEV_IO_TYPE_UNMAP:
3186 		rc = bdev_nvme_unmap(nbdev_io,
3187 				     bdev_io->u.bdev.offset_blocks,
3188 				     bdev_io->u.bdev.num_blocks);
3189 		break;
3190 	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3191 		rc =  bdev_nvme_write_zeroes(nbdev_io,
3192 					     bdev_io->u.bdev.offset_blocks,
3193 					     bdev_io->u.bdev.num_blocks);
3194 		break;
3195 	case SPDK_BDEV_IO_TYPE_RESET:
3196 		nbdev_io->io_path = NULL;
3197 		bdev_nvme_reset_io(bdev->ctxt, nbdev_io);
3198 		return;
3199 
3200 	case SPDK_BDEV_IO_TYPE_FLUSH:
3201 		bdev_nvme_io_complete(nbdev_io, 0);
3202 		return;
3203 
3204 	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
3205 		rc = bdev_nvme_zone_appendv(nbdev_io,
3206 					    bdev_io->u.bdev.iovs,
3207 					    bdev_io->u.bdev.iovcnt,
3208 					    bdev_io->u.bdev.md_buf,
3209 					    bdev_io->u.bdev.num_blocks,
3210 					    bdev_io->u.bdev.offset_blocks,
3211 					    bdev_io->u.bdev.dif_check_flags);
3212 		break;
3213 	case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
3214 		rc = bdev_nvme_get_zone_info(nbdev_io,
3215 					     bdev_io->u.zone_mgmt.zone_id,
3216 					     bdev_io->u.zone_mgmt.num_zones,
3217 					     bdev_io->u.zone_mgmt.buf);
3218 		break;
3219 	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
3220 		rc = bdev_nvme_zone_management(nbdev_io,
3221 					       bdev_io->u.zone_mgmt.zone_id,
3222 					       bdev_io->u.zone_mgmt.zone_action);
3223 		break;
3224 	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3225 		nbdev_io->io_path = NULL;
3226 		bdev_nvme_admin_passthru(nbdev_ch,
3227 					 nbdev_io,
3228 					 &bdev_io->u.nvme_passthru.cmd,
3229 					 bdev_io->u.nvme_passthru.buf,
3230 					 bdev_io->u.nvme_passthru.nbytes);
3231 		return;
3232 
3233 	case SPDK_BDEV_IO_TYPE_NVME_IO:
3234 		rc = bdev_nvme_io_passthru(nbdev_io,
3235 					   &bdev_io->u.nvme_passthru.cmd,
3236 					   bdev_io->u.nvme_passthru.buf,
3237 					   bdev_io->u.nvme_passthru.nbytes);
3238 		break;
3239 	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3240 		rc = bdev_nvme_io_passthru_md(nbdev_io,
3241 					      &bdev_io->u.nvme_passthru.cmd,
3242 					      bdev_io->u.nvme_passthru.buf,
3243 					      bdev_io->u.nvme_passthru.nbytes,
3244 					      bdev_io->u.nvme_passthru.md_buf,
3245 					      bdev_io->u.nvme_passthru.md_len);
3246 		break;
3247 	case SPDK_BDEV_IO_TYPE_NVME_IOV_MD:
3248 		rc = bdev_nvme_iov_passthru_md(nbdev_io,
3249 					       &bdev_io->u.nvme_passthru.cmd,
3250 					       bdev_io->u.nvme_passthru.iovs,
3251 					       bdev_io->u.nvme_passthru.iovcnt,
3252 					       bdev_io->u.nvme_passthru.nbytes,
3253 					       bdev_io->u.nvme_passthru.md_buf,
3254 					       bdev_io->u.nvme_passthru.md_len);
3255 		break;
3256 	case SPDK_BDEV_IO_TYPE_ABORT:
3257 		nbdev_io->io_path = NULL;
3258 		nbdev_io_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx;
3259 		bdev_nvme_abort(nbdev_ch,
3260 				nbdev_io,
3261 				nbdev_io_to_abort);
3262 		return;
3263 
3264 	case SPDK_BDEV_IO_TYPE_COPY:
3265 		rc = bdev_nvme_copy(nbdev_io,
3266 				    bdev_io->u.bdev.offset_blocks,
3267 				    bdev_io->u.bdev.copy.src_offset_blocks,
3268 				    bdev_io->u.bdev.num_blocks);
3269 		break;
3270 	default:
3271 		rc = -EINVAL;
3272 		break;
3273 	}
3274 
3275 	if (spdk_unlikely(rc != 0)) {
3276 		bdev_nvme_io_complete(nbdev_io, rc);
3277 	}
3278 }
3279 
3280 static void
3281 bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
3282 {
3283 	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
3284 	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
3285 
3286 	if (spdk_likely(nbdev_io->submit_tsc == 0)) {
3287 		nbdev_io->submit_tsc = spdk_bdev_io_get_submit_tsc(bdev_io);
3288 	} else {
3289 		/* There are cases where submit_tsc != 0, i.e. retry I/O.
3290 		 * We need to update submit_tsc here.
3291 		 */
3292 		nbdev_io->submit_tsc = spdk_get_ticks();
3293 	}
3294 
3295 	spdk_trace_record(TRACE_BDEV_NVME_IO_START, 0, 0, (uintptr_t)nbdev_io, (uintptr_t)bdev_io);
3296 	nbdev_io->io_path = bdev_nvme_find_io_path(nbdev_ch);
3297 	if (spdk_unlikely(!nbdev_io->io_path)) {
3298 		if (!bdev_nvme_io_type_is_admin(bdev_io->type)) {
3299 			bdev_nvme_io_complete(nbdev_io, -ENXIO);
3300 			return;
3301 		}
3302 
3303 		/* Admin commands do not use the optimal I/O path.
3304 		 * Simply fall through even if it is not found.
3305 		 */
3306 	}
3307 
3308 	_bdev_nvme_submit_request(nbdev_ch, bdev_io);
3309 }
3310 
3311 static bool
3312 bdev_nvme_is_supported_csi(enum spdk_nvme_csi csi)
3313 {
3314 	switch (csi) {
3315 	case SPDK_NVME_CSI_NVM:
3316 		return true;
3317 	case SPDK_NVME_CSI_ZNS:
3318 		return true;
3319 	default:
3320 		return false;
3321 	}
3322 }
3323 
3324 static bool
3325 bdev_nvme_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
3326 {
3327 	struct nvme_bdev *nbdev = ctx;
3328 	struct nvme_ns *nvme_ns;
3329 	struct spdk_nvme_ns *ns;
3330 	struct spdk_nvme_ctrlr *ctrlr;
3331 	const struct spdk_nvme_ctrlr_data *cdata;
3332 
3333 	nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
3334 	assert(nvme_ns != NULL);
3335 	ns = nvme_ns->ns;
3336 	if (ns == NULL) {
3337 		return false;
3338 	}
3339 
3340 	if (!bdev_nvme_is_supported_csi(spdk_nvme_ns_get_csi(ns))) {
3341 		switch (io_type) {
3342 		case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3343 		case SPDK_BDEV_IO_TYPE_NVME_IO:
3344 			return true;
3345 
3346 		case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3347 			return spdk_nvme_ns_get_md_size(ns) ? true : false;
3348 
3349 		default:
3350 			return false;
3351 		}
3352 	}
3353 
3354 	ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3355 
3356 	switch (io_type) {
3357 	case SPDK_BDEV_IO_TYPE_READ:
3358 	case SPDK_BDEV_IO_TYPE_WRITE:
3359 	case SPDK_BDEV_IO_TYPE_RESET:
3360 	case SPDK_BDEV_IO_TYPE_FLUSH:
3361 	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
3362 	case SPDK_BDEV_IO_TYPE_NVME_IO:
3363 	case SPDK_BDEV_IO_TYPE_ABORT:
3364 		return true;
3365 
3366 	case SPDK_BDEV_IO_TYPE_COMPARE:
3367 		return spdk_nvme_ns_supports_compare(ns);
3368 
3369 	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3370 		return spdk_nvme_ns_get_md_size(ns) ? true : false;
3371 
3372 	case SPDK_BDEV_IO_TYPE_UNMAP:
3373 		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3374 		return cdata->oncs.dsm;
3375 
3376 	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3377 		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3378 		return cdata->oncs.write_zeroes;
3379 
3380 	case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
3381 		if (spdk_nvme_ctrlr_get_flags(ctrlr) &
3382 		    SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED) {
3383 			return true;
3384 		}
3385 		return false;
3386 
3387 	case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
3388 	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
3389 		return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS;
3390 
3391 	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
3392 		return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS &&
3393 		       spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ZONE_APPEND_SUPPORTED;
3394 
3395 	case SPDK_BDEV_IO_TYPE_COPY:
3396 		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3397 		return cdata->oncs.copy;
3398 
3399 	default:
3400 		return false;
3401 	}
3402 }
3403 
3404 static int
3405 nvme_qpair_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ctrlr_channel *ctrlr_ch)
3406 {
3407 	struct nvme_qpair *nvme_qpair;
3408 	struct spdk_io_channel *pg_ch;
3409 	int rc;
3410 
3411 	nvme_qpair = calloc(1, sizeof(*nvme_qpair));
3412 	if (!nvme_qpair) {
3413 		SPDK_ERRLOG("Failed to alloc nvme_qpair.\n");
3414 		return -1;
3415 	}
3416 
3417 	TAILQ_INIT(&nvme_qpair->io_path_list);
3418 
3419 	nvme_qpair->ctrlr = nvme_ctrlr;
3420 	nvme_qpair->ctrlr_ch = ctrlr_ch;
3421 
3422 	pg_ch = spdk_get_io_channel(&g_nvme_bdev_ctrlrs);
3423 	if (!pg_ch) {
3424 		free(nvme_qpair);
3425 		return -1;
3426 	}
3427 
3428 	nvme_qpair->group = spdk_io_channel_get_ctx(pg_ch);
3429 
3430 #ifdef SPDK_CONFIG_VTUNE
3431 	nvme_qpair->group->collect_spin_stat = true;
3432 #else
3433 	nvme_qpair->group->collect_spin_stat = false;
3434 #endif
3435 
3436 	if (!nvme_ctrlr->disabled) {
3437 		/* If a nvme_ctrlr is disabled, don't try to create qpair for it. Qpair will
3438 		 * be created when it's enabled.
3439 		 */
3440 		rc = bdev_nvme_create_qpair(nvme_qpair);
3441 		if (rc != 0) {
3442 			/* nvme_ctrlr can't create IO qpair if connection is down.
3443 			 * If reconnect_delay_sec is non-zero, creating IO qpair is retried
3444 			 * after reconnect_delay_sec seconds. If bdev_retry_count is non-zero,
3445 			 * submitted IO will be queued until IO qpair is successfully created.
3446 			 *
3447 			 * Hence, if both are satisfied, ignore the failure.
3448 			 */
3449 			if (nvme_ctrlr->opts.reconnect_delay_sec == 0 || g_opts.bdev_retry_count == 0) {
3450 				spdk_put_io_channel(pg_ch);
3451 				free(nvme_qpair);
3452 				return rc;
3453 			}
3454 		}
3455 	}
3456 
3457 	TAILQ_INSERT_TAIL(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);
3458 
3459 	ctrlr_ch->qpair = nvme_qpair;
3460 
3461 	pthread_mutex_lock(&nvme_qpair->ctrlr->mutex);
3462 	nvme_qpair->ctrlr->ref++;
3463 	pthread_mutex_unlock(&nvme_qpair->ctrlr->mutex);
3464 
3465 	return 0;
3466 }
3467 
3468 static int
3469 bdev_nvme_create_ctrlr_channel_cb(void *io_device, void *ctx_buf)
3470 {
3471 	struct nvme_ctrlr *nvme_ctrlr = io_device;
3472 	struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
3473 
3474 	TAILQ_INIT(&ctrlr_ch->pending_resets);
3475 
3476 	return nvme_qpair_create(nvme_ctrlr, ctrlr_ch);
3477 }
3478 
3479 static void
3480 nvme_qpair_delete(struct nvme_qpair *nvme_qpair)
3481 {
3482 	struct nvme_io_path *io_path, *next;
3483 
3484 	assert(nvme_qpair->group != NULL);
3485 
3486 	TAILQ_FOREACH_SAFE(io_path, &nvme_qpair->io_path_list, tailq, next) {
3487 		TAILQ_REMOVE(&nvme_qpair->io_path_list, io_path, tailq);
3488 		nvme_io_path_free(io_path);
3489 	}
3490 
3491 	TAILQ_REMOVE(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);
3492 
3493 	spdk_put_io_channel(spdk_io_channel_from_ctx(nvme_qpair->group));
3494 
3495 	nvme_ctrlr_release(nvme_qpair->ctrlr);
3496 
3497 	free(nvme_qpair);
3498 }
3499 
3500 static void
3501 bdev_nvme_destroy_ctrlr_channel_cb(void *io_device, void *ctx_buf)
3502 {
3503 	struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
3504 	struct nvme_qpair *nvme_qpair;
3505 
3506 	nvme_qpair = ctrlr_ch->qpair;
3507 	assert(nvme_qpair != NULL);
3508 
3509 	_bdev_nvme_clear_io_path_cache(nvme_qpair);
3510 
3511 	if (nvme_qpair->qpair != NULL) {
3512 		if (ctrlr_ch->reset_iter == NULL) {
3513 			spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);
3514 		} else {
3515 			/* Skip current ctrlr_channel in a full reset sequence because
3516 			 * it is being deleted now. The qpair is already being disconnected.
3517 			 * We do not have to restart disconnecting it.
3518 			 */
3519 			nvme_ctrlr_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
3520 		}
3521 
3522 		/* We cannot release a reference to the poll group now.
3523 		 * The qpair may be disconnected asynchronously later.
3524 		 * We need to poll it until it is actually disconnected.
3525 		 * Just detach the qpair from the deleting ctrlr_channel.
3526 		 */
3527 		nvme_qpair->ctrlr_ch = NULL;
3528 	} else {
3529 		assert(ctrlr_ch->reset_iter == NULL);
3530 
3531 		nvme_qpair_delete(nvme_qpair);
3532 	}
3533 }
3534 
3535 static inline struct spdk_io_channel *
3536 bdev_nvme_get_accel_channel(struct nvme_poll_group *group)
3537 {
3538 	if (spdk_unlikely(!group->accel_channel)) {
3539 		group->accel_channel = spdk_accel_get_io_channel();
3540 		if (!group->accel_channel) {
3541 			SPDK_ERRLOG("Cannot get the accel_channel for bdev nvme polling group=%p\n",
3542 				    group);
3543 			return NULL;
3544 		}
3545 	}
3546 
3547 	return group->accel_channel;
3548 }
3549 
3550 static void
3551 bdev_nvme_finish_sequence(void *seq, spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
3552 {
3553 	spdk_accel_sequence_finish(seq, cb_fn, cb_arg);
3554 }
3555 
3556 static void
3557 bdev_nvme_abort_sequence(void *seq)
3558 {
3559 	spdk_accel_sequence_abort(seq);
3560 }
3561 
3562 static void
3563 bdev_nvme_reverse_sequence(void *seq)
3564 {
3565 	spdk_accel_sequence_reverse(seq);
3566 }
3567 
3568 static int
3569 bdev_nvme_append_crc32c(void *ctx, void **seq, uint32_t *dst, struct iovec *iovs, uint32_t iovcnt,
3570 			struct spdk_memory_domain *domain, void *domain_ctx, uint32_t seed,
3571 			spdk_nvme_accel_step_cb cb_fn, void *cb_arg)
3572 {
3573 	struct spdk_io_channel *ch;
3574 	struct nvme_poll_group *group = ctx;
3575 
3576 	ch = bdev_nvme_get_accel_channel(group);
3577 	if (spdk_unlikely(ch == NULL)) {
3578 		return -ENOMEM;
3579 	}
3580 
3581 	return spdk_accel_append_crc32c((struct spdk_accel_sequence **)seq, ch, dst, iovs, iovcnt,
3582 					domain, domain_ctx, seed, cb_fn, cb_arg);
3583 }
3584 
3585 static int
3586 bdev_nvme_append_copy(void *ctx, void **seq, struct iovec *dst_iovs, uint32_t dst_iovcnt,
3587 		      struct spdk_memory_domain *dst_domain, void *dst_domain_ctx,
3588 		      struct iovec *src_iovs, uint32_t src_iovcnt,
3589 		      struct spdk_memory_domain *src_domain, void *src_domain_ctx,
3590 		      spdk_nvme_accel_step_cb cb_fn, void *cb_arg)
3591 {
3592 	struct spdk_io_channel *ch;
3593 	struct nvme_poll_group *group = ctx;
3594 
3595 	ch = bdev_nvme_get_accel_channel(group);
3596 	if (spdk_unlikely(ch == NULL)) {
3597 		return -ENOMEM;
3598 	}
3599 
3600 	return spdk_accel_append_copy((struct spdk_accel_sequence **)seq, ch,
3601 				      dst_iovs, dst_iovcnt, dst_domain, dst_domain_ctx,
3602 				      src_iovs, src_iovcnt, src_domain, src_domain_ctx,
3603 				      cb_fn, cb_arg);
3604 }
3605 
3606 static struct spdk_nvme_accel_fn_table g_bdev_nvme_accel_fn_table = {
3607 	.table_size		= sizeof(struct spdk_nvme_accel_fn_table),
3608 	.append_crc32c		= bdev_nvme_append_crc32c,
3609 	.append_copy		= bdev_nvme_append_copy,
3610 	.finish_sequence	= bdev_nvme_finish_sequence,
3611 	.reverse_sequence	= bdev_nvme_reverse_sequence,
3612 	.abort_sequence		= bdev_nvme_abort_sequence,
3613 };
3614 
3615 static int
3616 bdev_nvme_create_poll_group_cb(void *io_device, void *ctx_buf)
3617 {
3618 	struct nvme_poll_group *group = ctx_buf;
3619 
3620 	TAILQ_INIT(&group->qpair_list);
3621 
3622 	group->group = spdk_nvme_poll_group_create(group, &g_bdev_nvme_accel_fn_table);
3623 	if (group->group == NULL) {
3624 		return -1;
3625 	}
3626 
3627 	group->poller = SPDK_POLLER_REGISTER(bdev_nvme_poll, group, g_opts.nvme_ioq_poll_period_us);
3628 
3629 	if (group->poller == NULL) {
3630 		spdk_nvme_poll_group_destroy(group->group);
3631 		return -1;
3632 	}
3633 
3634 	return 0;
3635 }
3636 
3637 static void
3638 bdev_nvme_destroy_poll_group_cb(void *io_device, void *ctx_buf)
3639 {
3640 	struct nvme_poll_group *group = ctx_buf;
3641 
3642 	assert(TAILQ_EMPTY(&group->qpair_list));
3643 
3644 	if (group->accel_channel) {
3645 		spdk_put_io_channel(group->accel_channel);
3646 	}
3647 
3648 	spdk_poller_unregister(&group->poller);
3649 	if (spdk_nvme_poll_group_destroy(group->group)) {
3650 		SPDK_ERRLOG("Unable to destroy a poll group for the NVMe bdev module.\n");
3651 		assert(false);
3652 	}
3653 }
3654 
3655 static struct spdk_io_channel *
3656 bdev_nvme_get_io_channel(void *ctx)
3657 {
3658 	struct nvme_bdev *nvme_bdev = ctx;
3659 
3660 	return spdk_get_io_channel(nvme_bdev);
3661 }
3662 
3663 static void *
3664 bdev_nvme_get_module_ctx(void *ctx)
3665 {
3666 	struct nvme_bdev *nvme_bdev = ctx;
3667 	struct nvme_ns *nvme_ns;
3668 
3669 	if (!nvme_bdev || nvme_bdev->disk.module != &nvme_if) {
3670 		return NULL;
3671 	}
3672 
3673 	nvme_ns = TAILQ_FIRST(&nvme_bdev->nvme_ns_list);
3674 	if (!nvme_ns) {
3675 		return NULL;
3676 	}
3677 
3678 	return nvme_ns->ns;
3679 }
3680 
3681 static const char *
3682 _nvme_ana_state_str(enum spdk_nvme_ana_state ana_state)
3683 {
3684 	switch (ana_state) {
3685 	case SPDK_NVME_ANA_OPTIMIZED_STATE:
3686 		return "optimized";
3687 	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
3688 		return "non_optimized";
3689 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
3690 		return "inaccessible";
3691 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
3692 		return "persistent_loss";
3693 	case SPDK_NVME_ANA_CHANGE_STATE:
3694 		return "change";
3695 	default:
3696 		return NULL;
3697 	}
3698 }
3699 
3700 static int
3701 bdev_nvme_get_memory_domains(void *ctx, struct spdk_memory_domain **domains, int array_size)
3702 {
3703 	struct spdk_memory_domain **_domains = NULL;
3704 	struct nvme_bdev *nbdev = ctx;
3705 	struct nvme_ns *nvme_ns;
3706 	int i = 0, _array_size = array_size;
3707 	int rc = 0;
3708 
3709 	TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
3710 		if (domains && array_size >= i) {
3711 			_domains = &domains[i];
3712 		} else {
3713 			_domains = NULL;
3714 		}
3715 		rc = spdk_nvme_ctrlr_get_memory_domains(nvme_ns->ctrlr->ctrlr, _domains, _array_size);
3716 		if (rc > 0) {
3717 			i += rc;
3718 			if (_array_size >= rc) {
3719 				_array_size -= rc;
3720 			} else {
3721 				_array_size = 0;
3722 			}
3723 		} else if (rc < 0) {
3724 			return rc;
3725 		}
3726 	}
3727 
3728 	return i;
3729 }
3730 
3731 static const char *
3732 nvme_ctrlr_get_state_str(struct nvme_ctrlr *nvme_ctrlr)
3733 {
3734 	if (nvme_ctrlr->destruct) {
3735 		return "deleting";
3736 	} else if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
3737 		return "failed";
3738 	} else if (nvme_ctrlr->resetting) {
3739 		return "resetting";
3740 	} else if (nvme_ctrlr->reconnect_is_delayed > 0) {
3741 		return "reconnect_is_delayed";
3742 	} else if (nvme_ctrlr->disabled) {
3743 		return "disabled";
3744 	} else {
3745 		return "enabled";
3746 	}
3747 }
3748 
3749 void
3750 nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr)
3751 {
3752 	struct spdk_nvme_transport_id *trid;
3753 	const struct spdk_nvme_ctrlr_opts *opts;
3754 	const struct spdk_nvme_ctrlr_data *cdata;
3755 	struct nvme_path_id *path_id;
3756 	int32_t numa_id;
3757 
3758 	spdk_json_write_object_begin(w);
3759 
3760 	spdk_json_write_named_string(w, "state", nvme_ctrlr_get_state_str(nvme_ctrlr));
3761 
3762 #ifdef SPDK_CONFIG_NVME_CUSE
3763 	size_t cuse_name_size = 128;
3764 	char cuse_name[cuse_name_size];
3765 
3766 	int rc = spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr, cuse_name, &cuse_name_size);
3767 	if (rc == 0) {
3768 		spdk_json_write_named_string(w, "cuse_device", cuse_name);
3769 	}
3770 #endif
3771 	trid = &nvme_ctrlr->active_path_id->trid;
3772 	spdk_json_write_named_object_begin(w, "trid");
3773 	nvme_bdev_dump_trid_json(trid, w);
3774 	spdk_json_write_object_end(w);
3775 
3776 	path_id = TAILQ_NEXT(nvme_ctrlr->active_path_id, link);
3777 	if (path_id != NULL) {
3778 		spdk_json_write_named_array_begin(w, "alternate_trids");
3779 		do {
3780 			trid = &path_id->trid;
3781 			spdk_json_write_object_begin(w);
3782 			nvme_bdev_dump_trid_json(trid, w);
3783 			spdk_json_write_object_end(w);
3784 
3785 			path_id = TAILQ_NEXT(path_id, link);
3786 		} while (path_id != NULL);
3787 		spdk_json_write_array_end(w);
3788 	}
3789 
3790 	cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
3791 	spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);
3792 
3793 	opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
3794 	spdk_json_write_named_object_begin(w, "host");
3795 	spdk_json_write_named_string(w, "nqn", opts->hostnqn);
3796 	spdk_json_write_named_string(w, "addr", opts->src_addr);
3797 	spdk_json_write_named_string(w, "svcid", opts->src_svcid);
3798 	spdk_json_write_object_end(w);
3799 
3800 	numa_id = spdk_nvme_ctrlr_get_numa_id(nvme_ctrlr->ctrlr);
3801 	if (numa_id != SPDK_ENV_NUMA_ID_ANY) {
3802 		spdk_json_write_named_uint32(w, "numa_id", numa_id);
3803 	}
3804 	spdk_json_write_object_end(w);
3805 }
3806 
3807 static void
3808 nvme_namespace_info_json(struct spdk_json_write_ctx *w,
3809 			 struct nvme_ns *nvme_ns)
3810 {
3811 	struct spdk_nvme_ns *ns;
3812 	struct spdk_nvme_ctrlr *ctrlr;
3813 	const struct spdk_nvme_ctrlr_data *cdata;
3814 	const struct spdk_nvme_transport_id *trid;
3815 	union spdk_nvme_vs_register vs;
3816 	const struct spdk_nvme_ns_data *nsdata;
3817 	char buf[128];
3818 
3819 	ns = nvme_ns->ns;
3820 	if (ns == NULL) {
3821 		return;
3822 	}
3823 
3824 	ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3825 
3826 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
3827 	trid = spdk_nvme_ctrlr_get_transport_id(ctrlr);
3828 	vs = spdk_nvme_ctrlr_get_regs_vs(ctrlr);
3829 
3830 	spdk_json_write_object_begin(w);
3831 
3832 	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
3833 		spdk_json_write_named_string(w, "pci_address", trid->traddr);
3834 	}
3835 
3836 	spdk_json_write_named_object_begin(w, "trid");
3837 
3838 	nvme_bdev_dump_trid_json(trid, w);
3839 
3840 	spdk_json_write_object_end(w);
3841 
3842 #ifdef SPDK_CONFIG_NVME_CUSE
3843 	size_t cuse_name_size = 128;
3844 	char cuse_name[cuse_name_size];
3845 
3846 	int rc = spdk_nvme_cuse_get_ns_name(ctrlr, spdk_nvme_ns_get_id(ns),
3847 					    cuse_name, &cuse_name_size);
3848 	if (rc == 0) {
3849 		spdk_json_write_named_string(w, "cuse_device", cuse_name);
3850 	}
3851 #endif
3852 
3853 	spdk_json_write_named_object_begin(w, "ctrlr_data");
3854 
3855 	spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);
3856 
3857 	spdk_json_write_named_string_fmt(w, "vendor_id", "0x%04x", cdata->vid);
3858 
3859 	snprintf(buf, sizeof(cdata->mn) + 1, "%s", cdata->mn);
3860 	spdk_str_trim(buf);
3861 	spdk_json_write_named_string(w, "model_number", buf);
3862 
3863 	snprintf(buf, sizeof(cdata->sn) + 1, "%s", cdata->sn);
3864 	spdk_str_trim(buf);
3865 	spdk_json_write_named_string(w, "serial_number", buf);
3866 
3867 	snprintf(buf, sizeof(cdata->fr) + 1, "%s", cdata->fr);
3868 	spdk_str_trim(buf);
3869 	spdk_json_write_named_string(w, "firmware_revision", buf);
3870 
3871 	if (cdata->subnqn[0] != '\0') {
3872 		spdk_json_write_named_string(w, "subnqn", cdata->subnqn);
3873 	}
3874 
3875 	spdk_json_write_named_object_begin(w, "oacs");
3876 
3877 	spdk_json_write_named_uint32(w, "security", cdata->oacs.security);
3878 	spdk_json_write_named_uint32(w, "format", cdata->oacs.format);
3879 	spdk_json_write_named_uint32(w, "firmware", cdata->oacs.firmware);
3880 	spdk_json_write_named_uint32(w, "ns_manage", cdata->oacs.ns_manage);
3881 
3882 	spdk_json_write_object_end(w);
3883 
3884 	spdk_json_write_named_bool(w, "multi_ctrlr", cdata->cmic.multi_ctrlr);
3885 	spdk_json_write_named_bool(w, "ana_reporting", cdata->cmic.ana_reporting);
3886 
3887 	spdk_json_write_object_end(w);
3888 
3889 	spdk_json_write_named_object_begin(w, "vs");
3890 
3891 	spdk_json_write_name(w, "nvme_version");
3892 	if (vs.bits.ter) {
3893 		spdk_json_write_string_fmt(w, "%u.%u.%u", vs.bits.mjr, vs.bits.mnr, vs.bits.ter);
3894 	} else {
3895 		spdk_json_write_string_fmt(w, "%u.%u", vs.bits.mjr, vs.bits.mnr);
3896 	}
3897 
3898 	spdk_json_write_object_end(w);
3899 
3900 	nsdata = spdk_nvme_ns_get_data(ns);
3901 
3902 	spdk_json_write_named_object_begin(w, "ns_data");
3903 
3904 	spdk_json_write_named_uint32(w, "id", spdk_nvme_ns_get_id(ns));
3905 
3906 	if (cdata->cmic.ana_reporting) {
3907 		spdk_json_write_named_string(w, "ana_state",
3908 					     _nvme_ana_state_str(nvme_ns->ana_state));
3909 	}
3910 
3911 	spdk_json_write_named_bool(w, "can_share", nsdata->nmic.can_share);
3912 
3913 	spdk_json_write_object_end(w);
3914 
3915 	if (cdata->oacs.security) {
3916 		spdk_json_write_named_object_begin(w, "security");
3917 
3918 		spdk_json_write_named_bool(w, "opal", nvme_ns->bdev->opal);
3919 
3920 		spdk_json_write_object_end(w);
3921 	}
3922 
3923 	spdk_json_write_object_end(w);
3924 }
3925 
3926 static const char *
3927 nvme_bdev_get_mp_policy_str(struct nvme_bdev *nbdev)
3928 {
3929 	switch (nbdev->mp_policy) {
3930 	case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
3931 		return "active_passive";
3932 	case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
3933 		return "active_active";
3934 	default:
3935 		assert(false);
3936 		return "invalid";
3937 	}
3938 }
3939 
3940 static const char *
3941 nvme_bdev_get_mp_selector_str(struct nvme_bdev *nbdev)
3942 {
3943 	switch (nbdev->mp_selector) {
3944 	case BDEV_NVME_MP_SELECTOR_ROUND_ROBIN:
3945 		return "round_robin";
3946 	case BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH:
3947 		return "queue_depth";
3948 	default:
3949 		assert(false);
3950 		return "invalid";
3951 	}
3952 }
3953 
3954 static int
3955 bdev_nvme_dump_info_json(void *ctx, struct spdk_json_write_ctx *w)
3956 {
3957 	struct nvme_bdev *nvme_bdev = ctx;
3958 	struct nvme_ns *nvme_ns;
3959 
3960 	pthread_mutex_lock(&nvme_bdev->mutex);
3961 	spdk_json_write_named_array_begin(w, "nvme");
3962 	TAILQ_FOREACH(nvme_ns, &nvme_bdev->nvme_ns_list, tailq) {
3963 		nvme_namespace_info_json(w, nvme_ns);
3964 	}
3965 	spdk_json_write_array_end(w);
3966 	spdk_json_write_named_string(w, "mp_policy", nvme_bdev_get_mp_policy_str(nvme_bdev));
3967 	if (nvme_bdev->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
3968 		spdk_json_write_named_string(w, "selector", nvme_bdev_get_mp_selector_str(nvme_bdev));
3969 		if (nvme_bdev->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
3970 			spdk_json_write_named_uint32(w, "rr_min_io", nvme_bdev->rr_min_io);
3971 		}
3972 	}
3973 	pthread_mutex_unlock(&nvme_bdev->mutex);
3974 
3975 	return 0;
3976 }
3977 
3978 static void
3979 bdev_nvme_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
3980 {
3981 	/* No config per bdev needed */
3982 }
3983 
3984 static uint64_t
3985 bdev_nvme_get_spin_time(struct spdk_io_channel *ch)
3986 {
3987 	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
3988 	struct nvme_io_path *io_path;
3989 	struct nvme_poll_group *group;
3990 	uint64_t spin_time = 0;
3991 
3992 	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
3993 		group = io_path->qpair->group;
3994 
3995 		if (!group || !group->collect_spin_stat) {
3996 			continue;
3997 		}
3998 
3999 		if (group->end_ticks != 0) {
4000 			group->spin_ticks += (group->end_ticks - group->start_ticks);
4001 			group->end_ticks = 0;
4002 		}
4003 
4004 		spin_time += group->spin_ticks;
4005 		group->start_ticks = 0;
4006 		group->spin_ticks = 0;
4007 	}
4008 
4009 	return (spin_time * 1000000ULL) / spdk_get_ticks_hz();
4010 }
4011 
4012 static void
4013 bdev_nvme_reset_device_stat(void *ctx)
4014 {
4015 	struct nvme_bdev *nbdev = ctx;
4016 
4017 	if (nbdev->err_stat != NULL) {
4018 		memset(nbdev->err_stat, 0, sizeof(struct nvme_error_stat));
4019 	}
4020 }
4021 
4022 /* JSON string should be lowercases and underscore delimited string. */
4023 static void
4024 bdev_nvme_format_nvme_status(char *dst, const char *src)
4025 {
4026 	char tmp[256];
4027 
4028 	spdk_strcpy_replace(dst, 256, src, " - ", "_");
4029 	spdk_strcpy_replace(tmp, 256, dst, "-", "_");
4030 	spdk_strcpy_replace(dst, 256, tmp, " ", "_");
4031 	spdk_strlwr(dst);
4032 }
4033 
4034 static void
4035 bdev_nvme_dump_device_stat_json(void *ctx, struct spdk_json_write_ctx *w)
4036 {
4037 	struct nvme_bdev *nbdev = ctx;
4038 	struct spdk_nvme_status status = {};
4039 	uint16_t sct, sc;
4040 	char status_json[256];
4041 	const char *status_str;
4042 
4043 	if (nbdev->err_stat == NULL) {
4044 		return;
4045 	}
4046 
4047 	spdk_json_write_named_object_begin(w, "nvme_error");
4048 
4049 	spdk_json_write_named_object_begin(w, "status_type");
4050 	for (sct = 0; sct < 8; sct++) {
4051 		if (nbdev->err_stat->status_type[sct] == 0) {
4052 			continue;
4053 		}
4054 		status.sct = sct;
4055 
4056 		status_str = spdk_nvme_cpl_get_status_type_string(&status);
4057 		assert(status_str != NULL);
4058 		bdev_nvme_format_nvme_status(status_json, status_str);
4059 
4060 		spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status_type[sct]);
4061 	}
4062 	spdk_json_write_object_end(w);
4063 
4064 	spdk_json_write_named_object_begin(w, "status_code");
4065 	for (sct = 0; sct < 4; sct++) {
4066 		status.sct = sct;
4067 		for (sc = 0; sc < 256; sc++) {
4068 			if (nbdev->err_stat->status[sct][sc] == 0) {
4069 				continue;
4070 			}
4071 			status.sc = sc;
4072 
4073 			status_str = spdk_nvme_cpl_get_status_string(&status);
4074 			assert(status_str != NULL);
4075 			bdev_nvme_format_nvme_status(status_json, status_str);
4076 
4077 			spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status[sct][sc]);
4078 		}
4079 	}
4080 	spdk_json_write_object_end(w);
4081 
4082 	spdk_json_write_object_end(w);
4083 }
4084 
4085 static bool
4086 bdev_nvme_accel_sequence_supported(void *ctx, enum spdk_bdev_io_type type)
4087 {
4088 	struct nvme_bdev *nbdev = ctx;
4089 	struct spdk_nvme_ctrlr *ctrlr;
4090 
4091 	if (!g_opts.allow_accel_sequence) {
4092 		return false;
4093 	}
4094 
4095 	switch (type) {
4096 	case SPDK_BDEV_IO_TYPE_WRITE:
4097 	case SPDK_BDEV_IO_TYPE_READ:
4098 		break;
4099 	default:
4100 		return false;
4101 	}
4102 
4103 	ctrlr = bdev_nvme_get_ctrlr(&nbdev->disk);
4104 	assert(ctrlr != NULL);
4105 
4106 	return spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ACCEL_SEQUENCE_SUPPORTED;
4107 }
4108 
4109 static const struct spdk_bdev_fn_table nvmelib_fn_table = {
4110 	.destruct			= bdev_nvme_destruct,
4111 	.submit_request			= bdev_nvme_submit_request,
4112 	.io_type_supported		= bdev_nvme_io_type_supported,
4113 	.get_io_channel			= bdev_nvme_get_io_channel,
4114 	.dump_info_json			= bdev_nvme_dump_info_json,
4115 	.write_config_json		= bdev_nvme_write_config_json,
4116 	.get_spin_time			= bdev_nvme_get_spin_time,
4117 	.get_module_ctx			= bdev_nvme_get_module_ctx,
4118 	.get_memory_domains		= bdev_nvme_get_memory_domains,
4119 	.accel_sequence_supported	= bdev_nvme_accel_sequence_supported,
4120 	.reset_device_stat		= bdev_nvme_reset_device_stat,
4121 	.dump_device_stat_json		= bdev_nvme_dump_device_stat_json,
4122 };
4123 
4124 typedef int (*bdev_nvme_parse_ana_log_page_cb)(
4125 	const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg);
4126 
4127 static int
4128 bdev_nvme_parse_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
4129 			     bdev_nvme_parse_ana_log_page_cb cb_fn, void *cb_arg)
4130 {
4131 	struct spdk_nvme_ana_group_descriptor *copied_desc;
4132 	uint8_t *orig_desc;
4133 	uint32_t i, desc_size, copy_len;
4134 	int rc = 0;
4135 
4136 	if (nvme_ctrlr->ana_log_page == NULL) {
4137 		return -EINVAL;
4138 	}
4139 
4140 	copied_desc = nvme_ctrlr->copied_ana_desc;
4141 
4142 	orig_desc = (uint8_t *)nvme_ctrlr->ana_log_page + sizeof(struct spdk_nvme_ana_page);
4143 	copy_len = nvme_ctrlr->max_ana_log_page_size - sizeof(struct spdk_nvme_ana_page);
4144 
4145 	for (i = 0; i < nvme_ctrlr->ana_log_page->num_ana_group_desc; i++) {
4146 		memcpy(copied_desc, orig_desc, copy_len);
4147 
4148 		rc = cb_fn(copied_desc, cb_arg);
4149 		if (rc != 0) {
4150 			break;
4151 		}
4152 
4153 		desc_size = sizeof(struct spdk_nvme_ana_group_descriptor) +
4154 			    copied_desc->num_of_nsid * sizeof(uint32_t);
4155 		orig_desc += desc_size;
4156 		copy_len -= desc_size;
4157 	}
4158 
4159 	return rc;
4160 }
4161 
4162 static int
4163 nvme_ns_ana_transition_timedout(void *ctx)
4164 {
4165 	struct nvme_ns *nvme_ns = ctx;
4166 
4167 	spdk_poller_unregister(&nvme_ns->anatt_timer);
4168 	nvme_ns->ana_transition_timedout = true;
4169 
4170 	return SPDK_POLLER_BUSY;
4171 }
4172 
4173 static void
4174 _nvme_ns_set_ana_state(struct nvme_ns *nvme_ns,
4175 		       const struct spdk_nvme_ana_group_descriptor *desc)
4176 {
4177 	const struct spdk_nvme_ctrlr_data *cdata;
4178 
4179 	nvme_ns->ana_group_id = desc->ana_group_id;
4180 	nvme_ns->ana_state = desc->ana_state;
4181 	nvme_ns->ana_state_updating = false;
4182 
4183 	switch (nvme_ns->ana_state) {
4184 	case SPDK_NVME_ANA_OPTIMIZED_STATE:
4185 	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
4186 		nvme_ns->ana_transition_timedout = false;
4187 		spdk_poller_unregister(&nvme_ns->anatt_timer);
4188 		break;
4189 
4190 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
4191 	case SPDK_NVME_ANA_CHANGE_STATE:
4192 		if (nvme_ns->anatt_timer != NULL) {
4193 			break;
4194 		}
4195 
4196 		cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
4197 		nvme_ns->anatt_timer = SPDK_POLLER_REGISTER(nvme_ns_ana_transition_timedout,
4198 				       nvme_ns,
4199 				       cdata->anatt * SPDK_SEC_TO_USEC);
4200 		break;
4201 	default:
4202 		break;
4203 	}
4204 }
4205 
4206 static int
4207 nvme_ns_set_ana_state(const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg)
4208 {
4209 	struct nvme_ns *nvme_ns = cb_arg;
4210 	uint32_t i;
4211 
4212 	assert(nvme_ns->ns != NULL);
4213 
4214 	for (i = 0; i < desc->num_of_nsid; i++) {
4215 		if (desc->nsid[i] != spdk_nvme_ns_get_id(nvme_ns->ns)) {
4216 			continue;
4217 		}
4218 
4219 		_nvme_ns_set_ana_state(nvme_ns, desc);
4220 		return 1;
4221 	}
4222 
4223 	return 0;
4224 }
4225 
4226 static int
4227 nvme_generate_uuid(const char *sn, uint32_t nsid, struct spdk_uuid *uuid)
4228 {
4229 	int rc = 0;
4230 	struct spdk_uuid new_uuid, namespace_uuid;
4231 	char merged_str[SPDK_NVME_CTRLR_SN_LEN + NSID_STR_LEN + 1] = {'\0'};
4232 	/* This namespace UUID was generated using uuid_generate() method. */
4233 	const char *namespace_str = {"edaed2de-24bc-4b07-b559-f47ecbe730fd"};
4234 	int size;
4235 
4236 	assert(strlen(sn) <= SPDK_NVME_CTRLR_SN_LEN);
4237 
4238 	spdk_uuid_set_null(&new_uuid);
4239 	spdk_uuid_set_null(&namespace_uuid);
4240 
4241 	size = snprintf(merged_str, sizeof(merged_str), "%s%"PRIu32, sn, nsid);
4242 	if (size <= 0 || (unsigned long)size >= sizeof(merged_str)) {
4243 		return -EINVAL;
4244 	}
4245 
4246 	spdk_uuid_parse(&namespace_uuid, namespace_str);
4247 
4248 	rc = spdk_uuid_generate_sha1(&new_uuid, &namespace_uuid, merged_str, size);
4249 	if (rc == 0) {
4250 		memcpy(uuid, &new_uuid, sizeof(struct spdk_uuid));
4251 	}
4252 
4253 	return rc;
4254 }
4255 
4256 static int
4257 nvme_disk_create(struct spdk_bdev *disk, const char *base_name,
4258 		 struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns,
4259 		 struct spdk_bdev_nvme_ctrlr_opts *bdev_opts, void *ctx)
4260 {
4261 	const struct spdk_uuid		*uuid;
4262 	const uint8_t *nguid;
4263 	const struct spdk_nvme_ctrlr_data *cdata;
4264 	const struct spdk_nvme_ns_data	*nsdata;
4265 	const struct spdk_nvme_ctrlr_opts *opts;
4266 	enum spdk_nvme_csi		csi;
4267 	uint32_t atomic_bs, phys_bs, bs;
4268 	char sn_tmp[SPDK_NVME_CTRLR_SN_LEN + 1] = {'\0'};
4269 	int rc;
4270 
4271 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
4272 	csi = spdk_nvme_ns_get_csi(ns);
4273 	opts = spdk_nvme_ctrlr_get_opts(ctrlr);
4274 
4275 	switch (csi) {
4276 	case SPDK_NVME_CSI_NVM:
4277 		disk->product_name = "NVMe disk";
4278 		break;
4279 	case SPDK_NVME_CSI_ZNS:
4280 		disk->product_name = "NVMe ZNS disk";
4281 		disk->zoned = true;
4282 		disk->zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
4283 		disk->max_zone_append_size = spdk_nvme_zns_ctrlr_get_max_zone_append_size(ctrlr) /
4284 					     spdk_nvme_ns_get_extended_sector_size(ns);
4285 		disk->max_open_zones = spdk_nvme_zns_ns_get_max_open_zones(ns);
4286 		disk->max_active_zones = spdk_nvme_zns_ns_get_max_active_zones(ns);
4287 		break;
4288 	default:
4289 		if (bdev_opts->allow_unrecognized_csi) {
4290 			disk->product_name = "NVMe Passthrough disk";
4291 			break;
4292 		}
4293 		SPDK_ERRLOG("unsupported CSI: %u\n", csi);
4294 		return -ENOTSUP;
4295 	}
4296 
4297 	nguid = spdk_nvme_ns_get_nguid(ns);
4298 	if (!nguid) {
4299 		uuid = spdk_nvme_ns_get_uuid(ns);
4300 		if (uuid) {
4301 			disk->uuid = *uuid;
4302 		} else if (g_opts.generate_uuids) {
4303 			spdk_strcpy_pad(sn_tmp, cdata->sn, SPDK_NVME_CTRLR_SN_LEN, '\0');
4304 			rc = nvme_generate_uuid(sn_tmp, spdk_nvme_ns_get_id(ns), &disk->uuid);
4305 			if (rc < 0) {
4306 				SPDK_ERRLOG("UUID generation failed (%s)\n", spdk_strerror(-rc));
4307 				return rc;
4308 			}
4309 		}
4310 	} else {
4311 		memcpy(&disk->uuid, nguid, sizeof(disk->uuid));
4312 	}
4313 
4314 	disk->name = spdk_sprintf_alloc("%sn%d", base_name, spdk_nvme_ns_get_id(ns));
4315 	if (!disk->name) {
4316 		return -ENOMEM;
4317 	}
4318 
4319 	disk->write_cache = 0;
4320 	if (cdata->vwc.present) {
4321 		/* Enable if the Volatile Write Cache exists */
4322 		disk->write_cache = 1;
4323 	}
4324 	if (cdata->oncs.write_zeroes) {
4325 		disk->max_write_zeroes = UINT16_MAX + 1;
4326 	}
4327 	disk->blocklen = spdk_nvme_ns_get_extended_sector_size(ns);
4328 	disk->blockcnt = spdk_nvme_ns_get_num_sectors(ns);
4329 	disk->max_segment_size = spdk_nvme_ctrlr_get_max_xfer_size(ctrlr);
4330 	disk->ctratt.raw = cdata->ctratt.raw;
4331 	/* NVMe driver will split one request into multiple requests
4332 	 * based on MDTS and stripe boundary, the bdev layer will use
4333 	 * max_segment_size and max_num_segments to split one big IO
4334 	 * into multiple requests, then small request can't run out
4335 	 * of NVMe internal requests data structure.
4336 	 */
4337 	if (opts && opts->io_queue_requests) {
4338 		disk->max_num_segments = opts->io_queue_requests / 2;
4339 	}
4340 	if (spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_SGL_SUPPORTED) {
4341 		/* The nvme driver will try to split I/O that have too many
4342 		 * SGEs, but it doesn't work if that last SGE doesn't end on
4343 		 * an aggregate total that is block aligned. The bdev layer has
4344 		 * a more robust splitting framework, so use that instead for
4345 		 * this case. (See issue #3269.)
4346 		 */
4347 		uint16_t max_sges = spdk_nvme_ctrlr_get_max_sges(ctrlr);
4348 
4349 		if (disk->max_num_segments == 0) {
4350 			disk->max_num_segments = max_sges;
4351 		} else {
4352 			disk->max_num_segments = spdk_min(disk->max_num_segments, max_sges);
4353 		}
4354 	}
4355 	disk->optimal_io_boundary = spdk_nvme_ns_get_optimal_io_boundary(ns);
4356 
4357 	nsdata = spdk_nvme_ns_get_data(ns);
4358 	bs = spdk_nvme_ns_get_sector_size(ns);
4359 	atomic_bs = bs;
4360 	phys_bs = bs;
4361 	if (nsdata->nabo == 0) {
4362 		if (nsdata->nsfeat.ns_atomic_write_unit && nsdata->nawupf) {
4363 			atomic_bs = bs * (1 + nsdata->nawupf);
4364 		} else {
4365 			atomic_bs = bs * (1 + cdata->awupf);
4366 		}
4367 	}
4368 	if (nsdata->nsfeat.optperf) {
4369 		phys_bs = bs * (1 + nsdata->npwg);
4370 	}
4371 	disk->phys_blocklen = spdk_min(phys_bs, atomic_bs);
4372 
4373 	disk->md_len = spdk_nvme_ns_get_md_size(ns);
4374 	if (disk->md_len != 0) {
4375 		disk->md_interleave = nsdata->flbas.extended;
4376 		disk->dif_type = (enum spdk_dif_type)spdk_nvme_ns_get_pi_type(ns);
4377 		if (disk->dif_type != SPDK_DIF_DISABLE) {
4378 			disk->dif_is_head_of_md = nsdata->dps.md_start;
4379 			disk->dif_check_flags = bdev_opts->prchk_flags;
4380 			disk->dif_pi_format = (enum spdk_dif_pi_format)spdk_nvme_ns_get_pi_format(ns);
4381 		}
4382 	}
4383 
4384 	if (!(spdk_nvme_ctrlr_get_flags(ctrlr) &
4385 	      SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED)) {
4386 		disk->acwu = 0;
4387 	} else if (nsdata->nsfeat.ns_atomic_write_unit) {
4388 		disk->acwu = nsdata->nacwu + 1; /* 0-based */
4389 	} else {
4390 		disk->acwu = cdata->acwu + 1; /* 0-based */
4391 	}
4392 
4393 	if (cdata->oncs.copy) {
4394 		/* For now bdev interface allows only single segment copy */
4395 		disk->max_copy = nsdata->mssrl;
4396 	}
4397 
4398 	disk->ctxt = ctx;
4399 	disk->fn_table = &nvmelib_fn_table;
4400 	disk->module = &nvme_if;
4401 
4402 	disk->numa.id_valid = 1;
4403 	disk->numa.id = spdk_nvme_ctrlr_get_numa_id(ctrlr);
4404 
4405 	return 0;
4406 }
4407 
4408 static struct nvme_bdev *
4409 nvme_bdev_alloc(void)
4410 {
4411 	struct nvme_bdev *bdev;
4412 	int rc;
4413 
4414 	bdev = calloc(1, sizeof(*bdev));
4415 	if (!bdev) {
4416 		SPDK_ERRLOG("bdev calloc() failed\n");
4417 		return NULL;
4418 	}
4419 
4420 	if (g_opts.nvme_error_stat) {
4421 		bdev->err_stat = calloc(1, sizeof(struct nvme_error_stat));
4422 		if (!bdev->err_stat) {
4423 			SPDK_ERRLOG("err_stat calloc() failed\n");
4424 			free(bdev);
4425 			return NULL;
4426 		}
4427 	}
4428 
4429 	rc = pthread_mutex_init(&bdev->mutex, NULL);
4430 	if (rc != 0) {
4431 		free(bdev->err_stat);
4432 		free(bdev);
4433 		return NULL;
4434 	}
4435 
4436 	bdev->ref = 1;
4437 	bdev->mp_policy = BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE;
4438 	bdev->mp_selector = BDEV_NVME_MP_SELECTOR_ROUND_ROBIN;
4439 	bdev->rr_min_io = UINT32_MAX;
4440 	TAILQ_INIT(&bdev->nvme_ns_list);
4441 
4442 	return bdev;
4443 }
4444 
4445 static int
4446 nvme_bdev_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4447 {
4448 	struct nvme_bdev *bdev;
4449 	struct nvme_bdev_ctrlr *nbdev_ctrlr = nvme_ctrlr->nbdev_ctrlr;
4450 	int rc;
4451 
4452 	bdev = nvme_bdev_alloc();
4453 	if (bdev == NULL) {
4454 		SPDK_ERRLOG("Failed to allocate NVMe bdev\n");
4455 		return -ENOMEM;
4456 	}
4457 
4458 	bdev->opal = nvme_ctrlr->opal_dev != NULL;
4459 
4460 	rc = nvme_disk_create(&bdev->disk, nbdev_ctrlr->name, nvme_ctrlr->ctrlr,
4461 			      nvme_ns->ns, &nvme_ctrlr->opts, bdev);
4462 	if (rc != 0) {
4463 		SPDK_ERRLOG("Failed to create NVMe disk\n");
4464 		nvme_bdev_free(bdev);
4465 		return rc;
4466 	}
4467 
4468 	spdk_io_device_register(bdev,
4469 				bdev_nvme_create_bdev_channel_cb,
4470 				bdev_nvme_destroy_bdev_channel_cb,
4471 				sizeof(struct nvme_bdev_channel),
4472 				bdev->disk.name);
4473 
4474 	nvme_ns->bdev = bdev;
4475 	bdev->nsid = nvme_ns->id;
4476 	TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
4477 
4478 	bdev->nbdev_ctrlr = nbdev_ctrlr;
4479 	TAILQ_INSERT_TAIL(&nbdev_ctrlr->bdevs, bdev, tailq);
4480 
4481 	rc = spdk_bdev_register(&bdev->disk);
4482 	if (rc != 0) {
4483 		SPDK_ERRLOG("spdk_bdev_register() failed\n");
4484 		spdk_io_device_unregister(bdev, NULL);
4485 		nvme_ns->bdev = NULL;
4486 		TAILQ_REMOVE(&nbdev_ctrlr->bdevs, bdev, tailq);
4487 		nvme_bdev_free(bdev);
4488 		return rc;
4489 	}
4490 
4491 	return 0;
4492 }
4493 
4494 static bool
4495 bdev_nvme_compare_ns(struct spdk_nvme_ns *ns1, struct spdk_nvme_ns *ns2)
4496 {
4497 	const struct spdk_nvme_ns_data *nsdata1, *nsdata2;
4498 	const struct spdk_uuid *uuid1, *uuid2;
4499 
4500 	nsdata1 = spdk_nvme_ns_get_data(ns1);
4501 	nsdata2 = spdk_nvme_ns_get_data(ns2);
4502 	uuid1 = spdk_nvme_ns_get_uuid(ns1);
4503 	uuid2 = spdk_nvme_ns_get_uuid(ns2);
4504 
4505 	return memcmp(nsdata1->nguid, nsdata2->nguid, sizeof(nsdata1->nguid)) == 0 &&
4506 	       nsdata1->eui64 == nsdata2->eui64 &&
4507 	       ((uuid1 == NULL && uuid2 == NULL) ||
4508 		(uuid1 != NULL && uuid2 != NULL && spdk_uuid_compare(uuid1, uuid2) == 0)) &&
4509 	       spdk_nvme_ns_get_csi(ns1) == spdk_nvme_ns_get_csi(ns2);
4510 }
4511 
4512 static bool
4513 hotplug_probe_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
4514 		 struct spdk_nvme_ctrlr_opts *opts)
4515 {
4516 	struct nvme_probe_skip_entry *entry;
4517 
4518 	TAILQ_FOREACH(entry, &g_skipped_nvme_ctrlrs, tailq) {
4519 		if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
4520 			return false;
4521 		}
4522 	}
4523 
4524 	opts->arbitration_burst = (uint8_t)g_opts.arbitration_burst;
4525 	opts->low_priority_weight = (uint8_t)g_opts.low_priority_weight;
4526 	opts->medium_priority_weight = (uint8_t)g_opts.medium_priority_weight;
4527 	opts->high_priority_weight = (uint8_t)g_opts.high_priority_weight;
4528 	opts->disable_read_ana_log_page = true;
4529 
4530 	SPDK_DEBUGLOG(bdev_nvme, "Attaching to %s\n", trid->traddr);
4531 
4532 	return true;
4533 }
4534 
4535 static void
4536 nvme_abort_cpl(void *ctx, const struct spdk_nvme_cpl *cpl)
4537 {
4538 	struct nvme_ctrlr *nvme_ctrlr = ctx;
4539 
4540 	if (spdk_nvme_cpl_is_error(cpl)) {
4541 		SPDK_WARNLOG("Abort failed. Resetting controller. sc is %u, sct is %u.\n", cpl->status.sc,
4542 			     cpl->status.sct);
4543 		bdev_nvme_reset_ctrlr(nvme_ctrlr);
4544 	} else if (cpl->cdw0 & 0x1) {
4545 		SPDK_WARNLOG("Specified command could not be aborted.\n");
4546 		bdev_nvme_reset_ctrlr(nvme_ctrlr);
4547 	}
4548 }
4549 
4550 static void
4551 timeout_cb(void *cb_arg, struct spdk_nvme_ctrlr *ctrlr,
4552 	   struct spdk_nvme_qpair *qpair, uint16_t cid)
4553 {
4554 	struct nvme_ctrlr *nvme_ctrlr = cb_arg;
4555 	union spdk_nvme_csts_register csts;
4556 	int rc;
4557 
4558 	assert(nvme_ctrlr->ctrlr == ctrlr);
4559 
4560 	SPDK_WARNLOG("Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n", ctrlr, qpair, cid);
4561 
4562 	/* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O
4563 	 * queue.  (Note: qpair == NULL when there's an admin cmd timeout.)  Otherwise we
4564 	 * would submit another fabrics cmd on the admin queue to read CSTS and check for its
4565 	 * completion recursively.
4566 	 */
4567 	if (nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE || qpair != NULL) {
4568 		csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
4569 		if (csts.bits.cfs) {
4570 			SPDK_ERRLOG("Controller Fatal Status, reset required\n");
4571 			bdev_nvme_reset_ctrlr(nvme_ctrlr);
4572 			return;
4573 		}
4574 	}
4575 
4576 	switch (g_opts.action_on_timeout) {
4577 	case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT:
4578 		if (qpair) {
4579 			/* Don't send abort to ctrlr when ctrlr is not available. */
4580 			pthread_mutex_lock(&nvme_ctrlr->mutex);
4581 			if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
4582 				pthread_mutex_unlock(&nvme_ctrlr->mutex);
4583 				SPDK_NOTICELOG("Quit abort. Ctrlr is not available.\n");
4584 				return;
4585 			}
4586 			pthread_mutex_unlock(&nvme_ctrlr->mutex);
4587 
4588 			rc = spdk_nvme_ctrlr_cmd_abort(ctrlr, qpair, cid,
4589 						       nvme_abort_cpl, nvme_ctrlr);
4590 			if (rc == 0) {
4591 				return;
4592 			}
4593 
4594 			SPDK_ERRLOG("Unable to send abort. Resetting, rc is %d.\n", rc);
4595 		}
4596 
4597 	/* FALLTHROUGH */
4598 	case SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET:
4599 		bdev_nvme_reset_ctrlr(nvme_ctrlr);
4600 		break;
4601 	case SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE:
4602 		SPDK_DEBUGLOG(bdev_nvme, "No action for nvme controller timeout.\n");
4603 		break;
4604 	default:
4605 		SPDK_ERRLOG("An invalid timeout action value is found.\n");
4606 		break;
4607 	}
4608 }
4609 
4610 static struct nvme_ns *
4611 nvme_ns_alloc(void)
4612 {
4613 	struct nvme_ns *nvme_ns;
4614 
4615 	nvme_ns = calloc(1, sizeof(struct nvme_ns));
4616 	if (nvme_ns == NULL) {
4617 		return NULL;
4618 	}
4619 
4620 	if (g_opts.io_path_stat) {
4621 		nvme_ns->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
4622 		if (nvme_ns->stat == NULL) {
4623 			free(nvme_ns);
4624 			return NULL;
4625 		}
4626 		spdk_bdev_reset_io_stat(nvme_ns->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
4627 	}
4628 
4629 	return nvme_ns;
4630 }
4631 
4632 static void
4633 nvme_ns_free(struct nvme_ns *nvme_ns)
4634 {
4635 	free(nvme_ns->stat);
4636 	free(nvme_ns);
4637 }
4638 
4639 static void
4640 nvme_ctrlr_populate_namespace_done(struct nvme_ns *nvme_ns, int rc)
4641 {
4642 	struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
4643 	struct nvme_async_probe_ctx *ctx = nvme_ns->probe_ctx;
4644 
4645 	if (rc == 0) {
4646 		nvme_ns->probe_ctx = NULL;
4647 		pthread_mutex_lock(&nvme_ctrlr->mutex);
4648 		nvme_ctrlr->ref++;
4649 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
4650 	} else {
4651 		RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4652 		nvme_ns_free(nvme_ns);
4653 	}
4654 
4655 	if (ctx) {
4656 		ctx->populates_in_progress--;
4657 		if (ctx->populates_in_progress == 0) {
4658 			nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
4659 		}
4660 	}
4661 }
4662 
4663 static void
4664 bdev_nvme_add_io_path(struct nvme_bdev_channel_iter *i,
4665 		      struct nvme_bdev *nbdev,
4666 		      struct nvme_bdev_channel *nbdev_ch, void *ctx)
4667 {
4668 	struct nvme_ns *nvme_ns = ctx;
4669 	int rc;
4670 
4671 	rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
4672 	if (rc != 0) {
4673 		SPDK_ERRLOG("Failed to add I/O path to bdev_channel dynamically.\n");
4674 	}
4675 
4676 	nvme_bdev_for_each_channel_continue(i, rc);
4677 }
4678 
4679 static void
4680 bdev_nvme_delete_io_path(struct nvme_bdev_channel_iter *i,
4681 			 struct nvme_bdev *nbdev,
4682 			 struct nvme_bdev_channel *nbdev_ch, void *ctx)
4683 {
4684 	struct nvme_ns *nvme_ns = ctx;
4685 	struct nvme_io_path *io_path;
4686 
4687 	io_path = _bdev_nvme_get_io_path(nbdev_ch, nvme_ns);
4688 	if (io_path != NULL) {
4689 		_bdev_nvme_delete_io_path(nbdev_ch, io_path);
4690 	}
4691 
4692 	nvme_bdev_for_each_channel_continue(i, 0);
4693 }
4694 
4695 static void
4696 bdev_nvme_add_io_path_failed(struct nvme_bdev *nbdev, void *ctx, int status)
4697 {
4698 	struct nvme_ns *nvme_ns = ctx;
4699 
4700 	nvme_ctrlr_populate_namespace_done(nvme_ns, -1);
4701 }
4702 
4703 static void
4704 bdev_nvme_add_io_path_done(struct nvme_bdev *nbdev, void *ctx, int status)
4705 {
4706 	struct nvme_ns *nvme_ns = ctx;
4707 
4708 	if (status == 0) {
4709 		nvme_ctrlr_populate_namespace_done(nvme_ns, 0);
4710 	} else {
4711 		/* Delete the added io_paths and fail populating the namespace. */
4712 		nvme_bdev_for_each_channel(nbdev,
4713 					   bdev_nvme_delete_io_path,
4714 					   nvme_ns,
4715 					   bdev_nvme_add_io_path_failed);
4716 	}
4717 }
4718 
4719 static int
4720 nvme_bdev_add_ns(struct nvme_bdev *bdev, struct nvme_ns *nvme_ns)
4721 {
4722 	struct nvme_ns *tmp_ns;
4723 	const struct spdk_nvme_ns_data *nsdata;
4724 
4725 	nsdata = spdk_nvme_ns_get_data(nvme_ns->ns);
4726 	if (!nsdata->nmic.can_share) {
4727 		SPDK_ERRLOG("Namespace cannot be shared.\n");
4728 		return -EINVAL;
4729 	}
4730 
4731 	pthread_mutex_lock(&bdev->mutex);
4732 
4733 	tmp_ns = TAILQ_FIRST(&bdev->nvme_ns_list);
4734 	assert(tmp_ns != NULL);
4735 
4736 	if (tmp_ns->ns != NULL && !bdev_nvme_compare_ns(nvme_ns->ns, tmp_ns->ns)) {
4737 		pthread_mutex_unlock(&bdev->mutex);
4738 		SPDK_ERRLOG("Namespaces are not identical.\n");
4739 		return -EINVAL;
4740 	}
4741 
4742 	bdev->ref++;
4743 	TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
4744 	nvme_ns->bdev = bdev;
4745 
4746 	pthread_mutex_unlock(&bdev->mutex);
4747 
4748 	/* Add nvme_io_path to nvme_bdev_channels dynamically. */
4749 	nvme_bdev_for_each_channel(bdev,
4750 				   bdev_nvme_add_io_path,
4751 				   nvme_ns,
4752 				   bdev_nvme_add_io_path_done);
4753 
4754 	return 0;
4755 }
4756 
4757 static void
4758 nvme_ctrlr_populate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4759 {
4760 	struct spdk_nvme_ns	*ns;
4761 	struct nvme_bdev	*bdev;
4762 	int			rc = 0;
4763 
4764 	ns = spdk_nvme_ctrlr_get_ns(nvme_ctrlr->ctrlr, nvme_ns->id);
4765 	if (!ns) {
4766 		SPDK_DEBUGLOG(bdev_nvme, "Invalid NS %d\n", nvme_ns->id);
4767 		rc = -EINVAL;
4768 		goto done;
4769 	}
4770 
4771 	nvme_ns->ns = ns;
4772 	nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
4773 
4774 	if (nvme_ctrlr->ana_log_page != NULL) {
4775 		bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ns_set_ana_state, nvme_ns);
4776 	}
4777 
4778 	bdev = nvme_bdev_ctrlr_get_bdev(nvme_ctrlr->nbdev_ctrlr, nvme_ns->id);
4779 	if (bdev == NULL) {
4780 		rc = nvme_bdev_create(nvme_ctrlr, nvme_ns);
4781 	} else {
4782 		rc = nvme_bdev_add_ns(bdev, nvme_ns);
4783 		if (rc == 0) {
4784 			return;
4785 		}
4786 	}
4787 done:
4788 	nvme_ctrlr_populate_namespace_done(nvme_ns, rc);
4789 }
4790 
4791 static void
4792 nvme_ctrlr_depopulate_namespace_done(struct nvme_ns *nvme_ns)
4793 {
4794 	struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
4795 
4796 	assert(nvme_ctrlr != NULL);
4797 
4798 	pthread_mutex_lock(&nvme_ctrlr->mutex);
4799 
4800 	RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4801 
4802 	if (nvme_ns->bdev != NULL) {
4803 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
4804 		return;
4805 	}
4806 
4807 	nvme_ns_free(nvme_ns);
4808 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
4809 
4810 	nvme_ctrlr_release(nvme_ctrlr);
4811 }
4812 
4813 static void
4814 bdev_nvme_delete_io_path_done(struct nvme_bdev *nbdev, void *ctx, int status)
4815 {
4816 	struct nvme_ns *nvme_ns = ctx;
4817 
4818 	nvme_ctrlr_depopulate_namespace_done(nvme_ns);
4819 }
4820 
4821 static void
4822 nvme_ctrlr_depopulate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
4823 {
4824 	struct nvme_bdev *bdev;
4825 
4826 	spdk_poller_unregister(&nvme_ns->anatt_timer);
4827 
4828 	bdev = nvme_ns->bdev;
4829 	if (bdev != NULL) {
4830 		pthread_mutex_lock(&bdev->mutex);
4831 
4832 		assert(bdev->ref > 0);
4833 		bdev->ref--;
4834 		if (bdev->ref == 0) {
4835 			pthread_mutex_unlock(&bdev->mutex);
4836 
4837 			spdk_bdev_unregister(&bdev->disk, NULL, NULL);
4838 		} else {
4839 			/* spdk_bdev_unregister() is not called until the last nvme_ns is
4840 			 * depopulated. Hence we need to remove nvme_ns from bdev->nvme_ns_list
4841 			 * and clear nvme_ns->bdev here.
4842 			 */
4843 			TAILQ_REMOVE(&bdev->nvme_ns_list, nvme_ns, tailq);
4844 			nvme_ns->bdev = NULL;
4845 
4846 			pthread_mutex_unlock(&bdev->mutex);
4847 
4848 			/* Delete nvme_io_paths from nvme_bdev_channels dynamically. After that,
4849 			 * we call depopulate_namespace_done() to avoid use-after-free.
4850 			 */
4851 			nvme_bdev_for_each_channel(bdev,
4852 						   bdev_nvme_delete_io_path,
4853 						   nvme_ns,
4854 						   bdev_nvme_delete_io_path_done);
4855 			return;
4856 		}
4857 	}
4858 
4859 	nvme_ctrlr_depopulate_namespace_done(nvme_ns);
4860 }
4861 
4862 static void
4863 nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
4864 			       struct nvme_async_probe_ctx *ctx)
4865 {
4866 	struct spdk_nvme_ctrlr	*ctrlr = nvme_ctrlr->ctrlr;
4867 	struct nvme_ns	*nvme_ns, *next;
4868 	struct spdk_nvme_ns	*ns;
4869 	struct nvme_bdev	*bdev;
4870 	uint32_t		nsid;
4871 	int			rc;
4872 	uint64_t		num_sectors;
4873 
4874 	if (ctx) {
4875 		/* Initialize this count to 1 to handle the populate functions
4876 		 * calling nvme_ctrlr_populate_namespace_done() immediately.
4877 		 */
4878 		ctx->populates_in_progress = 1;
4879 	}
4880 
4881 	/* First loop over our existing namespaces and see if they have been
4882 	 * removed. */
4883 	nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
4884 	while (nvme_ns != NULL) {
4885 		next = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
4886 
4887 		if (spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
4888 			/* NS is still there or added again. Its attributes may have changed. */
4889 			ns = spdk_nvme_ctrlr_get_ns(ctrlr, nvme_ns->id);
4890 			if (nvme_ns->ns != ns) {
4891 				assert(nvme_ns->ns == NULL);
4892 				nvme_ns->ns = ns;
4893 				SPDK_DEBUGLOG(bdev_nvme, "NSID %u was added\n", nvme_ns->id);
4894 			}
4895 
4896 			num_sectors = spdk_nvme_ns_get_num_sectors(ns);
4897 			bdev = nvme_ns->bdev;
4898 			assert(bdev != NULL);
4899 			if (bdev->disk.blockcnt != num_sectors) {
4900 				SPDK_NOTICELOG("NSID %u is resized: bdev name %s, old size %" PRIu64 ", new size %" PRIu64 "\n",
4901 					       nvme_ns->id,
4902 					       bdev->disk.name,
4903 					       bdev->disk.blockcnt,
4904 					       num_sectors);
4905 				rc = spdk_bdev_notify_blockcnt_change(&bdev->disk, num_sectors);
4906 				if (rc != 0) {
4907 					SPDK_ERRLOG("Could not change num blocks for nvme bdev: name %s, errno: %d.\n",
4908 						    bdev->disk.name, rc);
4909 				}
4910 			}
4911 		} else {
4912 			/* Namespace was removed */
4913 			nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
4914 		}
4915 
4916 		nvme_ns = next;
4917 	}
4918 
4919 	/* Loop through all of the namespaces at the nvme level and see if any of them are new */
4920 	nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
4921 	while (nsid != 0) {
4922 		nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);
4923 
4924 		if (nvme_ns == NULL) {
4925 			/* Found a new one */
4926 			nvme_ns = nvme_ns_alloc();
4927 			if (nvme_ns == NULL) {
4928 				SPDK_ERRLOG("Failed to allocate namespace\n");
4929 				/* This just fails to attach the namespace. It may work on a future attempt. */
4930 				continue;
4931 			}
4932 
4933 			nvme_ns->id = nsid;
4934 			nvme_ns->ctrlr = nvme_ctrlr;
4935 
4936 			nvme_ns->bdev = NULL;
4937 
4938 			if (ctx) {
4939 				ctx->populates_in_progress++;
4940 			}
4941 			nvme_ns->probe_ctx = ctx;
4942 
4943 			RB_INSERT(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
4944 
4945 			nvme_ctrlr_populate_namespace(nvme_ctrlr, nvme_ns);
4946 		}
4947 
4948 		nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid);
4949 	}
4950 
4951 	if (ctx) {
4952 		/* Decrement this count now that the loop is over to account
4953 		 * for the one we started with.  If the count is then 0, we
4954 		 * know any populate_namespace functions completed immediately,
4955 		 * so we'll kick the callback here.
4956 		 */
4957 		ctx->populates_in_progress--;
4958 		if (ctx->populates_in_progress == 0) {
4959 			nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
4960 		}
4961 	}
4962 
4963 }
4964 
4965 static void
4966 nvme_ctrlr_depopulate_namespaces(struct nvme_ctrlr *nvme_ctrlr)
4967 {
4968 	struct nvme_ns *nvme_ns, *tmp;
4969 
4970 	RB_FOREACH_SAFE(nvme_ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp) {
4971 		nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
4972 	}
4973 }
4974 
4975 static uint32_t
4976 nvme_ctrlr_get_ana_log_page_size(struct nvme_ctrlr *nvme_ctrlr)
4977 {
4978 	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
4979 	const struct spdk_nvme_ctrlr_data *cdata;
4980 	uint32_t nsid, ns_count = 0;
4981 
4982 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
4983 
4984 	for (nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
4985 	     nsid != 0; nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid)) {
4986 		ns_count++;
4987 	}
4988 
4989 	return sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
4990 	       sizeof(struct spdk_nvme_ana_group_descriptor) + ns_count *
4991 	       sizeof(uint32_t);
4992 }
4993 
4994 static int
4995 nvme_ctrlr_set_ana_states(const struct spdk_nvme_ana_group_descriptor *desc,
4996 			  void *cb_arg)
4997 {
4998 	struct nvme_ctrlr *nvme_ctrlr = cb_arg;
4999 	struct nvme_ns *nvme_ns;
5000 	uint32_t i, nsid;
5001 
5002 	for (i = 0; i < desc->num_of_nsid; i++) {
5003 		nsid = desc->nsid[i];
5004 		if (nsid == 0) {
5005 			continue;
5006 		}
5007 
5008 		nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);
5009 
5010 		if (nvme_ns == NULL) {
5011 			/* Target told us that an inactive namespace had an ANA change */
5012 			continue;
5013 		}
5014 
5015 		_nvme_ns_set_ana_state(nvme_ns, desc);
5016 	}
5017 
5018 	return 0;
5019 }
5020 
5021 static void
5022 bdev_nvme_disable_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
5023 {
5024 	struct nvme_ns *nvme_ns;
5025 
5026 	spdk_free(nvme_ctrlr->ana_log_page);
5027 	nvme_ctrlr->ana_log_page = NULL;
5028 
5029 	for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
5030 	     nvme_ns != NULL;
5031 	     nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
5032 		nvme_ns->ana_state_updating = false;
5033 		nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
5034 	}
5035 }
5036 
5037 static void
5038 nvme_ctrlr_read_ana_log_page_done(void *ctx, const struct spdk_nvme_cpl *cpl)
5039 {
5040 	struct nvme_ctrlr *nvme_ctrlr = ctx;
5041 
5042 	if (cpl != NULL && spdk_nvme_cpl_is_success(cpl)) {
5043 		bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ctrlr_set_ana_states,
5044 					     nvme_ctrlr);
5045 	} else {
5046 		bdev_nvme_disable_read_ana_log_page(nvme_ctrlr);
5047 	}
5048 
5049 	pthread_mutex_lock(&nvme_ctrlr->mutex);
5050 
5051 	assert(nvme_ctrlr->ana_log_page_updating == true);
5052 	nvme_ctrlr->ana_log_page_updating = false;
5053 
5054 	if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
5055 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
5056 
5057 		nvme_ctrlr_unregister(nvme_ctrlr);
5058 	} else {
5059 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
5060 
5061 		bdev_nvme_clear_io_path_caches(nvme_ctrlr);
5062 	}
5063 }
5064 
5065 static int
5066 nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
5067 {
5068 	uint32_t ana_log_page_size;
5069 	int rc;
5070 
5071 	if (nvme_ctrlr->ana_log_page == NULL) {
5072 		return -EINVAL;
5073 	}
5074 
5075 	ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);
5076 
5077 	if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
5078 		SPDK_ERRLOG("ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
5079 			    ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
5080 		return -EINVAL;
5081 	}
5082 
5083 	pthread_mutex_lock(&nvme_ctrlr->mutex);
5084 	if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
5085 	    nvme_ctrlr->ana_log_page_updating) {
5086 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
5087 		return -EBUSY;
5088 	}
5089 
5090 	nvme_ctrlr->ana_log_page_updating = true;
5091 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
5092 
5093 	rc = spdk_nvme_ctrlr_cmd_get_log_page(nvme_ctrlr->ctrlr,
5094 					      SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
5095 					      SPDK_NVME_GLOBAL_NS_TAG,
5096 					      nvme_ctrlr->ana_log_page,
5097 					      ana_log_page_size, 0,
5098 					      nvme_ctrlr_read_ana_log_page_done,
5099 					      nvme_ctrlr);
5100 	if (rc != 0) {
5101 		nvme_ctrlr_read_ana_log_page_done(nvme_ctrlr, NULL);
5102 	}
5103 
5104 	return rc;
5105 }
5106 
5107 static void
5108 dummy_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5109 {
5110 }
5111 
5112 struct bdev_nvme_set_preferred_path_ctx {
5113 	struct spdk_bdev_desc *desc;
5114 	struct nvme_ns *nvme_ns;
5115 	bdev_nvme_set_preferred_path_cb cb_fn;
5116 	void *cb_arg;
5117 };
5118 
5119 static void
5120 bdev_nvme_set_preferred_path_done(struct nvme_bdev *nbdev, void *_ctx, int status)
5121 {
5122 	struct bdev_nvme_set_preferred_path_ctx *ctx = _ctx;
5123 
5124 	assert(ctx != NULL);
5125 	assert(ctx->desc != NULL);
5126 	assert(ctx->cb_fn != NULL);
5127 
5128 	spdk_bdev_close(ctx->desc);
5129 
5130 	ctx->cb_fn(ctx->cb_arg, status);
5131 
5132 	free(ctx);
5133 }
5134 
5135 static void
5136 _bdev_nvme_set_preferred_path(struct nvme_bdev_channel_iter *i,
5137 			      struct nvme_bdev *nbdev,
5138 			      struct nvme_bdev_channel *nbdev_ch, void *_ctx)
5139 {
5140 	struct bdev_nvme_set_preferred_path_ctx *ctx = _ctx;
5141 	struct nvme_io_path *io_path, *prev;
5142 
5143 	prev = NULL;
5144 	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
5145 		if (io_path->nvme_ns == ctx->nvme_ns) {
5146 			break;
5147 		}
5148 		prev = io_path;
5149 	}
5150 
5151 	if (io_path != NULL) {
5152 		if (prev != NULL) {
5153 			STAILQ_REMOVE_AFTER(&nbdev_ch->io_path_list, prev, stailq);
5154 			STAILQ_INSERT_HEAD(&nbdev_ch->io_path_list, io_path, stailq);
5155 		}
5156 
5157 		/* We can set io_path to nbdev_ch->current_io_path directly here.
5158 		 * However, it needs to be conditional. To simplify the code,
5159 		 * just clear nbdev_ch->current_io_path and let find_io_path()
5160 		 * fill it.
5161 		 *
5162 		 * Automatic failback may be disabled. Hence even if the io_path is
5163 		 * already at the head, clear nbdev_ch->current_io_path.
5164 		 */
5165 		bdev_nvme_clear_current_io_path(nbdev_ch);
5166 	}
5167 
5168 	nvme_bdev_for_each_channel_continue(i, 0);
5169 }
5170 
5171 static struct nvme_ns *
5172 bdev_nvme_set_preferred_ns(struct nvme_bdev *nbdev, uint16_t cntlid)
5173 {
5174 	struct nvme_ns *nvme_ns, *prev;
5175 	const struct spdk_nvme_ctrlr_data *cdata;
5176 
5177 	prev = NULL;
5178 	TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
5179 		cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
5180 
5181 		if (cdata->cntlid == cntlid) {
5182 			break;
5183 		}
5184 		prev = nvme_ns;
5185 	}
5186 
5187 	if (nvme_ns != NULL && prev != NULL) {
5188 		TAILQ_REMOVE(&nbdev->nvme_ns_list, nvme_ns, tailq);
5189 		TAILQ_INSERT_HEAD(&nbdev->nvme_ns_list, nvme_ns, tailq);
5190 	}
5191 
5192 	return nvme_ns;
5193 }
5194 
5195 /* This function supports only multipath mode. There is only a single I/O path
5196  * for each NVMe-oF controller. Hence, just move the matched I/O path to the
5197  * head of the I/O path list for each NVMe bdev channel.
5198  *
5199  * NVMe bdev channel may be acquired after completing this function. move the
5200  * matched namespace to the head of the namespace list for the NVMe bdev too.
5201  */
5202 void
5203 bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid,
5204 			     bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg)
5205 {
5206 	struct bdev_nvme_set_preferred_path_ctx *ctx;
5207 	struct spdk_bdev *bdev;
5208 	struct nvme_bdev *nbdev;
5209 	int rc = 0;
5210 
5211 	assert(cb_fn != NULL);
5212 
5213 	ctx = calloc(1, sizeof(*ctx));
5214 	if (ctx == NULL) {
5215 		SPDK_ERRLOG("Failed to alloc context.\n");
5216 		rc = -ENOMEM;
5217 		goto err_alloc;
5218 	}
5219 
5220 	ctx->cb_fn = cb_fn;
5221 	ctx->cb_arg = cb_arg;
5222 
5223 	rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
5224 	if (rc != 0) {
5225 		SPDK_ERRLOG("Failed to open bdev %s.\n", name);
5226 		goto err_open;
5227 	}
5228 
5229 	bdev = spdk_bdev_desc_get_bdev(ctx->desc);
5230 
5231 	if (bdev->module != &nvme_if) {
5232 		SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
5233 		rc = -ENODEV;
5234 		goto err_bdev;
5235 	}
5236 
5237 	nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
5238 
5239 	pthread_mutex_lock(&nbdev->mutex);
5240 
5241 	ctx->nvme_ns = bdev_nvme_set_preferred_ns(nbdev, cntlid);
5242 	if (ctx->nvme_ns == NULL) {
5243 		pthread_mutex_unlock(&nbdev->mutex);
5244 
5245 		SPDK_ERRLOG("bdev %s does not have namespace to controller %u.\n", name, cntlid);
5246 		rc = -ENODEV;
5247 		goto err_bdev;
5248 	}
5249 
5250 	pthread_mutex_unlock(&nbdev->mutex);
5251 
5252 	nvme_bdev_for_each_channel(nbdev,
5253 				   _bdev_nvme_set_preferred_path,
5254 				   ctx,
5255 				   bdev_nvme_set_preferred_path_done);
5256 	return;
5257 
5258 err_bdev:
5259 	spdk_bdev_close(ctx->desc);
5260 err_open:
5261 	free(ctx);
5262 err_alloc:
5263 	cb_fn(cb_arg, rc);
5264 }
5265 
5266 struct bdev_nvme_set_multipath_policy_ctx {
5267 	struct spdk_bdev_desc *desc;
5268 	spdk_bdev_nvme_set_multipath_policy_cb cb_fn;
5269 	void *cb_arg;
5270 };
5271 
5272 static void
5273 bdev_nvme_set_multipath_policy_done(struct nvme_bdev *nbdev, void *_ctx, int status)
5274 {
5275 	struct bdev_nvme_set_multipath_policy_ctx *ctx = _ctx;
5276 
5277 	assert(ctx != NULL);
5278 	assert(ctx->desc != NULL);
5279 	assert(ctx->cb_fn != NULL);
5280 
5281 	spdk_bdev_close(ctx->desc);
5282 
5283 	ctx->cb_fn(ctx->cb_arg, status);
5284 
5285 	free(ctx);
5286 }
5287 
5288 static void
5289 _bdev_nvme_set_multipath_policy(struct nvme_bdev_channel_iter *i,
5290 				struct nvme_bdev *nbdev,
5291 				struct nvme_bdev_channel *nbdev_ch, void *ctx)
5292 {
5293 	nbdev_ch->mp_policy = nbdev->mp_policy;
5294 	nbdev_ch->mp_selector = nbdev->mp_selector;
5295 	nbdev_ch->rr_min_io = nbdev->rr_min_io;
5296 	bdev_nvme_clear_current_io_path(nbdev_ch);
5297 
5298 	nvme_bdev_for_each_channel_continue(i, 0);
5299 }
5300 
5301 void
5302 spdk_bdev_nvme_set_multipath_policy(const char *name, enum spdk_bdev_nvme_multipath_policy policy,
5303 				    enum spdk_bdev_nvme_multipath_selector selector, uint32_t rr_min_io,
5304 				    spdk_bdev_nvme_set_multipath_policy_cb cb_fn, void *cb_arg)
5305 {
5306 	struct bdev_nvme_set_multipath_policy_ctx *ctx;
5307 	struct spdk_bdev *bdev;
5308 	struct nvme_bdev *nbdev;
5309 	int rc;
5310 
5311 	assert(cb_fn != NULL);
5312 
5313 	switch (policy) {
5314 	case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
5315 		break;
5316 	case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
5317 		switch (selector) {
5318 		case BDEV_NVME_MP_SELECTOR_ROUND_ROBIN:
5319 			if (rr_min_io == UINT32_MAX) {
5320 				rr_min_io = 1;
5321 			} else if (rr_min_io == 0) {
5322 				rc = -EINVAL;
5323 				goto exit;
5324 			}
5325 			break;
5326 		case BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH:
5327 			break;
5328 		default:
5329 			rc = -EINVAL;
5330 			goto exit;
5331 		}
5332 		break;
5333 	default:
5334 		rc = -EINVAL;
5335 		goto exit;
5336 	}
5337 
5338 	ctx = calloc(1, sizeof(*ctx));
5339 	if (ctx == NULL) {
5340 		SPDK_ERRLOG("Failed to alloc context.\n");
5341 		rc = -ENOMEM;
5342 		goto exit;
5343 	}
5344 
5345 	ctx->cb_fn = cb_fn;
5346 	ctx->cb_arg = cb_arg;
5347 
5348 	rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
5349 	if (rc != 0) {
5350 		SPDK_ERRLOG("Failed to open bdev %s.\n", name);
5351 		rc = -ENODEV;
5352 		goto err_open;
5353 	}
5354 
5355 	bdev = spdk_bdev_desc_get_bdev(ctx->desc);
5356 	if (bdev->module != &nvme_if) {
5357 		SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
5358 		rc = -ENODEV;
5359 		goto err_module;
5360 	}
5361 	nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
5362 
5363 	pthread_mutex_lock(&nbdev->mutex);
5364 	nbdev->mp_policy = policy;
5365 	nbdev->mp_selector = selector;
5366 	nbdev->rr_min_io = rr_min_io;
5367 	pthread_mutex_unlock(&nbdev->mutex);
5368 
5369 	nvme_bdev_for_each_channel(nbdev,
5370 				   _bdev_nvme_set_multipath_policy,
5371 				   ctx,
5372 				   bdev_nvme_set_multipath_policy_done);
5373 	return;
5374 
5375 err_module:
5376 	spdk_bdev_close(ctx->desc);
5377 err_open:
5378 	free(ctx);
5379 exit:
5380 	cb_fn(cb_arg, rc);
5381 }
5382 
5383 static void
5384 aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
5385 {
5386 	struct nvme_ctrlr *nvme_ctrlr		= arg;
5387 	union spdk_nvme_async_event_completion	event;
5388 
5389 	if (spdk_nvme_cpl_is_error(cpl)) {
5390 		SPDK_WARNLOG("AER request execute failed\n");
5391 		return;
5392 	}
5393 
5394 	event.raw = cpl->cdw0;
5395 	if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
5396 	    (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) {
5397 		nvme_ctrlr_populate_namespaces(nvme_ctrlr, NULL);
5398 	} else if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
5399 		   (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_ANA_CHANGE)) {
5400 		nvme_ctrlr_read_ana_log_page(nvme_ctrlr);
5401 	}
5402 }
5403 
5404 static void
5405 free_nvme_async_probe_ctx(struct nvme_async_probe_ctx *ctx)
5406 {
5407 	spdk_keyring_put_key(ctx->drv_opts.tls_psk);
5408 	spdk_keyring_put_key(ctx->drv_opts.dhchap_key);
5409 	spdk_keyring_put_key(ctx->drv_opts.dhchap_ctrlr_key);
5410 	free(ctx);
5411 }
5412 
5413 static void
5414 populate_namespaces_cb(struct nvme_async_probe_ctx *ctx, int rc)
5415 {
5416 	if (ctx->cb_fn) {
5417 		ctx->cb_fn(ctx->cb_ctx, ctx->reported_bdevs, rc);
5418 	}
5419 
5420 	ctx->namespaces_populated = true;
5421 	if (ctx->probe_done) {
5422 		/* The probe was already completed, so we need to free the context
5423 		 * here.  This can happen for cases like OCSSD, where we need to
5424 		 * send additional commands to the SSD after attach.
5425 		 */
5426 		free_nvme_async_probe_ctx(ctx);
5427 	}
5428 }
5429 
5430 static int
5431 bdev_nvme_remove_poller(void *ctx)
5432 {
5433 	struct spdk_nvme_transport_id trid_pcie;
5434 
5435 	if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
5436 		spdk_poller_unregister(&g_hotplug_poller);
5437 		return SPDK_POLLER_IDLE;
5438 	}
5439 
5440 	memset(&trid_pcie, 0, sizeof(trid_pcie));
5441 	spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);
5442 
5443 	if (spdk_nvme_scan_attached(&trid_pcie)) {
5444 		SPDK_ERRLOG_RATELIMIT("spdk_nvme_scan_attached() failed\n");
5445 	}
5446 
5447 	return SPDK_POLLER_BUSY;
5448 }
5449 
5450 static void
5451 nvme_ctrlr_create_done(struct nvme_ctrlr *nvme_ctrlr,
5452 		       struct nvme_async_probe_ctx *ctx)
5453 {
5454 	spdk_io_device_register(nvme_ctrlr,
5455 				bdev_nvme_create_ctrlr_channel_cb,
5456 				bdev_nvme_destroy_ctrlr_channel_cb,
5457 				sizeof(struct nvme_ctrlr_channel),
5458 				nvme_ctrlr->nbdev_ctrlr->name);
5459 
5460 	nvme_ctrlr_populate_namespaces(nvme_ctrlr, ctx);
5461 
5462 	if (g_hotplug_poller == NULL) {
5463 		g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_remove_poller, NULL,
5464 							NVME_HOTPLUG_POLL_PERIOD_DEFAULT);
5465 	}
5466 }
5467 
5468 static void
5469 nvme_ctrlr_init_ana_log_page_done(void *_ctx, const struct spdk_nvme_cpl *cpl)
5470 {
5471 	struct nvme_ctrlr *nvme_ctrlr = _ctx;
5472 	struct nvme_async_probe_ctx *ctx = nvme_ctrlr->probe_ctx;
5473 
5474 	nvme_ctrlr->probe_ctx = NULL;
5475 
5476 	if (spdk_nvme_cpl_is_error(cpl)) {
5477 		nvme_ctrlr_delete(nvme_ctrlr);
5478 
5479 		if (ctx != NULL) {
5480 			ctx->reported_bdevs = 0;
5481 			populate_namespaces_cb(ctx, -1);
5482 		}
5483 		return;
5484 	}
5485 
5486 	nvme_ctrlr_create_done(nvme_ctrlr, ctx);
5487 }
5488 
5489 static int
5490 nvme_ctrlr_init_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
5491 			     struct nvme_async_probe_ctx *ctx)
5492 {
5493 	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5494 	const struct spdk_nvme_ctrlr_data *cdata;
5495 	uint32_t ana_log_page_size;
5496 
5497 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5498 
5499 	/* Set buffer size enough to include maximum number of allowed namespaces. */
5500 	ana_log_page_size = sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
5501 			    sizeof(struct spdk_nvme_ana_group_descriptor) + cdata->mnan *
5502 			    sizeof(uint32_t);
5503 
5504 	nvme_ctrlr->ana_log_page = spdk_zmalloc(ana_log_page_size, 64, NULL,
5505 						SPDK_ENV_NUMA_ID_ANY, SPDK_MALLOC_DMA);
5506 	if (nvme_ctrlr->ana_log_page == NULL) {
5507 		SPDK_ERRLOG("could not allocate ANA log page buffer\n");
5508 		return -ENXIO;
5509 	}
5510 
5511 	/* Each descriptor in a ANA log page is not ensured to be 8-bytes aligned.
5512 	 * Hence copy each descriptor to a temporary area when parsing it.
5513 	 *
5514 	 * Allocate a buffer whose size is as large as ANA log page buffer because
5515 	 * we do not know the size of a descriptor until actually reading it.
5516 	 */
5517 	nvme_ctrlr->copied_ana_desc = calloc(1, ana_log_page_size);
5518 	if (nvme_ctrlr->copied_ana_desc == NULL) {
5519 		SPDK_ERRLOG("could not allocate a buffer to parse ANA descriptor\n");
5520 		return -ENOMEM;
5521 	}
5522 
5523 	nvme_ctrlr->max_ana_log_page_size = ana_log_page_size;
5524 
5525 	nvme_ctrlr->probe_ctx = ctx;
5526 
5527 	/* Then, set the read size only to include the current active namespaces. */
5528 	ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);
5529 
5530 	if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
5531 		SPDK_ERRLOG("ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
5532 			    ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
5533 		return -EINVAL;
5534 	}
5535 
5536 	return spdk_nvme_ctrlr_cmd_get_log_page(ctrlr,
5537 						SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
5538 						SPDK_NVME_GLOBAL_NS_TAG,
5539 						nvme_ctrlr->ana_log_page,
5540 						ana_log_page_size, 0,
5541 						nvme_ctrlr_init_ana_log_page_done,
5542 						nvme_ctrlr);
5543 }
5544 
5545 /* hostnqn and subnqn were already verified before attaching a controller.
5546  * Hence check only the multipath capability and cntlid here.
5547  */
5548 static bool
5549 bdev_nvme_check_multipath(struct nvme_bdev_ctrlr *nbdev_ctrlr, struct spdk_nvme_ctrlr *ctrlr)
5550 {
5551 	struct nvme_ctrlr *tmp;
5552 	const struct spdk_nvme_ctrlr_data *cdata, *tmp_cdata;
5553 
5554 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5555 
5556 	if (!cdata->cmic.multi_ctrlr) {
5557 		SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
5558 		return false;
5559 	}
5560 
5561 	TAILQ_FOREACH(tmp, &nbdev_ctrlr->ctrlrs, tailq) {
5562 		tmp_cdata = spdk_nvme_ctrlr_get_data(tmp->ctrlr);
5563 
5564 		if (!tmp_cdata->cmic.multi_ctrlr) {
5565 			SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
5566 			return false;
5567 		}
5568 		if (cdata->cntlid == tmp_cdata->cntlid) {
5569 			SPDK_ERRLOG("cntlid %u are duplicated.\n", tmp_cdata->cntlid);
5570 			return false;
5571 		}
5572 	}
5573 
5574 	return true;
5575 }
5576 
5577 
5578 static int
5579 nvme_bdev_ctrlr_create(const char *name, struct nvme_ctrlr *nvme_ctrlr)
5580 {
5581 	struct nvme_bdev_ctrlr *nbdev_ctrlr;
5582 	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
5583 	struct nvme_ctrlr      *nctrlr;
5584 	int rc = 0;
5585 
5586 	pthread_mutex_lock(&g_bdev_nvme_mutex);
5587 
5588 	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
5589 	if (nbdev_ctrlr != NULL) {
5590 		if (!bdev_nvme_check_multipath(nbdev_ctrlr, ctrlr)) {
5591 			rc = -EINVAL;
5592 			goto exit;
5593 		}
5594 		TAILQ_FOREACH(nctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
5595 			if (nctrlr->opts.multipath != nvme_ctrlr->opts.multipath) {
5596 				/* All controllers with the same name must be configured the same
5597 				 * way, either for multipath or failover. If the configuration doesn't
5598 				 * match - report error.
5599 				 */
5600 				rc = -EINVAL;
5601 				goto exit;
5602 			}
5603 		}
5604 	} else {
5605 		nbdev_ctrlr = calloc(1, sizeof(*nbdev_ctrlr));
5606 		if (nbdev_ctrlr == NULL) {
5607 			SPDK_ERRLOG("Failed to allocate nvme_bdev_ctrlr.\n");
5608 			rc = -ENOMEM;
5609 			goto exit;
5610 		}
5611 		nbdev_ctrlr->name = strdup(name);
5612 		if (nbdev_ctrlr->name == NULL) {
5613 			SPDK_ERRLOG("Failed to allocate name of nvme_bdev_ctrlr.\n");
5614 			free(nbdev_ctrlr);
5615 			goto exit;
5616 		}
5617 		TAILQ_INIT(&nbdev_ctrlr->ctrlrs);
5618 		TAILQ_INIT(&nbdev_ctrlr->bdevs);
5619 		TAILQ_INSERT_TAIL(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
5620 	}
5621 	nvme_ctrlr->nbdev_ctrlr = nbdev_ctrlr;
5622 	TAILQ_INSERT_TAIL(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
5623 exit:
5624 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
5625 	return rc;
5626 }
5627 
5628 static int
5629 nvme_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr,
5630 		  const char *name,
5631 		  const struct spdk_nvme_transport_id *trid,
5632 		  struct nvme_async_probe_ctx *ctx)
5633 {
5634 	struct nvme_ctrlr *nvme_ctrlr;
5635 	struct nvme_path_id *path_id;
5636 	const struct spdk_nvme_ctrlr_data *cdata;
5637 	int rc;
5638 
5639 	nvme_ctrlr = calloc(1, sizeof(*nvme_ctrlr));
5640 	if (nvme_ctrlr == NULL) {
5641 		SPDK_ERRLOG("Failed to allocate device struct\n");
5642 		return -ENOMEM;
5643 	}
5644 
5645 	rc = pthread_mutex_init(&nvme_ctrlr->mutex, NULL);
5646 	if (rc != 0) {
5647 		free(nvme_ctrlr);
5648 		return rc;
5649 	}
5650 
5651 	TAILQ_INIT(&nvme_ctrlr->trids);
5652 	RB_INIT(&nvme_ctrlr->namespaces);
5653 
5654 	/* Get another reference to the key, so the first one can be released from probe_ctx */
5655 	if (ctx != NULL) {
5656 		if (ctx->drv_opts.tls_psk != NULL) {
5657 			nvme_ctrlr->psk = spdk_keyring_get_key(
5658 						  spdk_key_get_name(ctx->drv_opts.tls_psk));
5659 			if (nvme_ctrlr->psk == NULL) {
5660 				/* Could only happen if the key was removed in the meantime */
5661 				SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5662 					    spdk_key_get_name(ctx->drv_opts.tls_psk));
5663 				rc = -ENOKEY;
5664 				goto err;
5665 			}
5666 		}
5667 
5668 		if (ctx->drv_opts.dhchap_key != NULL) {
5669 			nvme_ctrlr->dhchap_key = spdk_keyring_get_key(
5670 							 spdk_key_get_name(ctx->drv_opts.dhchap_key));
5671 			if (nvme_ctrlr->dhchap_key == NULL) {
5672 				SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5673 					    spdk_key_get_name(ctx->drv_opts.dhchap_key));
5674 				rc = -ENOKEY;
5675 				goto err;
5676 			}
5677 		}
5678 
5679 		if (ctx->drv_opts.dhchap_ctrlr_key != NULL) {
5680 			nvme_ctrlr->dhchap_ctrlr_key =
5681 				spdk_keyring_get_key(
5682 					spdk_key_get_name(ctx->drv_opts.dhchap_ctrlr_key));
5683 			if (nvme_ctrlr->dhchap_ctrlr_key == NULL) {
5684 				SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
5685 					    spdk_key_get_name(ctx->drv_opts.dhchap_ctrlr_key));
5686 				rc = -ENOKEY;
5687 				goto err;
5688 			}
5689 		}
5690 	}
5691 
5692 	path_id = calloc(1, sizeof(*path_id));
5693 	if (path_id == NULL) {
5694 		SPDK_ERRLOG("Failed to allocate trid entry pointer\n");
5695 		rc = -ENOMEM;
5696 		goto err;
5697 	}
5698 
5699 	path_id->trid = *trid;
5700 	if (ctx != NULL) {
5701 		memcpy(path_id->hostid.hostaddr, ctx->drv_opts.src_addr, sizeof(path_id->hostid.hostaddr));
5702 		memcpy(path_id->hostid.hostsvcid, ctx->drv_opts.src_svcid, sizeof(path_id->hostid.hostsvcid));
5703 	}
5704 	nvme_ctrlr->active_path_id = path_id;
5705 	TAILQ_INSERT_HEAD(&nvme_ctrlr->trids, path_id, link);
5706 
5707 	nvme_ctrlr->thread = spdk_get_thread();
5708 	nvme_ctrlr->ctrlr = ctrlr;
5709 	nvme_ctrlr->ref = 1;
5710 
5711 	if (spdk_nvme_ctrlr_is_ocssd_supported(ctrlr)) {
5712 		SPDK_ERRLOG("OCSSDs are not supported");
5713 		rc = -ENOTSUP;
5714 		goto err;
5715 	}
5716 
5717 	if (ctx != NULL) {
5718 		memcpy(&nvme_ctrlr->opts, &ctx->bdev_opts, sizeof(ctx->bdev_opts));
5719 	} else {
5720 		spdk_bdev_nvme_get_default_ctrlr_opts(&nvme_ctrlr->opts);
5721 	}
5722 
5723 	nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq, nvme_ctrlr,
5724 					  g_opts.nvme_adminq_poll_period_us);
5725 
5726 	if (g_opts.timeout_us > 0) {
5727 		/* Register timeout callback. Timeout values for IO vs. admin reqs can be different. */
5728 		/* If timeout_admin_us is 0 (not specified), admin uses same timeout as IO. */
5729 		uint64_t adm_timeout_us = (g_opts.timeout_admin_us == 0) ?
5730 					  g_opts.timeout_us : g_opts.timeout_admin_us;
5731 		spdk_nvme_ctrlr_register_timeout_callback(ctrlr, g_opts.timeout_us,
5732 				adm_timeout_us, timeout_cb, nvme_ctrlr);
5733 	}
5734 
5735 	spdk_nvme_ctrlr_register_aer_callback(ctrlr, aer_cb, nvme_ctrlr);
5736 	spdk_nvme_ctrlr_set_remove_cb(ctrlr, remove_cb, nvme_ctrlr);
5737 
5738 	if (spdk_nvme_ctrlr_get_flags(ctrlr) &
5739 	    SPDK_NVME_CTRLR_SECURITY_SEND_RECV_SUPPORTED) {
5740 		nvme_ctrlr->opal_dev = spdk_opal_dev_construct(ctrlr);
5741 	}
5742 
5743 	rc = nvme_bdev_ctrlr_create(name, nvme_ctrlr);
5744 	if (rc != 0) {
5745 		goto err;
5746 	}
5747 
5748 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
5749 
5750 	if (cdata->cmic.ana_reporting) {
5751 		rc = nvme_ctrlr_init_ana_log_page(nvme_ctrlr, ctx);
5752 		if (rc == 0) {
5753 			return 0;
5754 		}
5755 	} else {
5756 		nvme_ctrlr_create_done(nvme_ctrlr, ctx);
5757 		return 0;
5758 	}
5759 
5760 err:
5761 	nvme_ctrlr_delete(nvme_ctrlr);
5762 	return rc;
5763 }
5764 
5765 void
5766 spdk_bdev_nvme_get_default_ctrlr_opts(struct spdk_bdev_nvme_ctrlr_opts *opts)
5767 {
5768 	opts->prchk_flags = 0;
5769 	opts->ctrlr_loss_timeout_sec = g_opts.ctrlr_loss_timeout_sec;
5770 	opts->reconnect_delay_sec = g_opts.reconnect_delay_sec;
5771 	opts->fast_io_fail_timeout_sec = g_opts.fast_io_fail_timeout_sec;
5772 	opts->multipath = true;
5773 }
5774 
5775 static void
5776 attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
5777 	  struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *drv_opts)
5778 {
5779 	char *name;
5780 
5781 	name = spdk_sprintf_alloc("HotInNvme%d", g_hot_insert_nvme_controller_index++);
5782 	if (!name) {
5783 		SPDK_ERRLOG("Failed to assign name to NVMe device\n");
5784 		return;
5785 	}
5786 
5787 	if (nvme_ctrlr_create(ctrlr, name, trid, NULL) == 0) {
5788 		SPDK_DEBUGLOG(bdev_nvme, "Attached to %s (%s)\n", trid->traddr, name);
5789 	} else {
5790 		SPDK_ERRLOG("Failed to attach to %s (%s)\n", trid->traddr, name);
5791 	}
5792 
5793 	free(name);
5794 }
5795 
5796 static void
5797 _nvme_ctrlr_destruct(void *ctx)
5798 {
5799 	struct nvme_ctrlr *nvme_ctrlr = ctx;
5800 
5801 	nvme_ctrlr_depopulate_namespaces(nvme_ctrlr);
5802 	nvme_ctrlr_release(nvme_ctrlr);
5803 }
5804 
5805 static int
5806 bdev_nvme_delete_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
5807 {
5808 	struct nvme_probe_skip_entry *entry;
5809 
5810 	/* The controller's destruction was already started */
5811 	if (nvme_ctrlr->destruct) {
5812 		return -EALREADY;
5813 	}
5814 
5815 	if (!hotplug &&
5816 	    nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
5817 		entry = calloc(1, sizeof(*entry));
5818 		if (!entry) {
5819 			return -ENOMEM;
5820 		}
5821 		entry->trid = nvme_ctrlr->active_path_id->trid;
5822 		TAILQ_INSERT_TAIL(&g_skipped_nvme_ctrlrs, entry, tailq);
5823 	}
5824 
5825 	nvme_ctrlr->destruct = true;
5826 	return 0;
5827 }
5828 
5829 static int
5830 bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
5831 {
5832 	int rc;
5833 
5834 	pthread_mutex_lock(&nvme_ctrlr->mutex);
5835 	rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, hotplug);
5836 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
5837 
5838 	if (rc == 0) {
5839 		_nvme_ctrlr_destruct(nvme_ctrlr);
5840 	} else if (rc == -EALREADY) {
5841 		rc = 0;
5842 	}
5843 
5844 	return rc;
5845 }
5846 
5847 static void
5848 remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr)
5849 {
5850 	struct nvme_ctrlr *nvme_ctrlr = cb_ctx;
5851 
5852 	bdev_nvme_delete_ctrlr(nvme_ctrlr, true);
5853 }
5854 
5855 static int
5856 bdev_nvme_hotplug_probe(void *arg)
5857 {
5858 	if (g_hotplug_probe_ctx == NULL) {
5859 		spdk_poller_unregister(&g_hotplug_probe_poller);
5860 		return SPDK_POLLER_IDLE;
5861 	}
5862 
5863 	if (spdk_nvme_probe_poll_async(g_hotplug_probe_ctx) != -EAGAIN) {
5864 		g_hotplug_probe_ctx = NULL;
5865 		spdk_poller_unregister(&g_hotplug_probe_poller);
5866 	}
5867 
5868 	return SPDK_POLLER_BUSY;
5869 }
5870 
5871 static int
5872 bdev_nvme_hotplug(void *arg)
5873 {
5874 	struct spdk_nvme_transport_id trid_pcie;
5875 
5876 	if (g_hotplug_probe_ctx) {
5877 		return SPDK_POLLER_BUSY;
5878 	}
5879 
5880 	memset(&trid_pcie, 0, sizeof(trid_pcie));
5881 	spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);
5882 
5883 	g_hotplug_probe_ctx = spdk_nvme_probe_async(&trid_pcie, NULL,
5884 			      hotplug_probe_cb, attach_cb, NULL);
5885 
5886 	if (g_hotplug_probe_ctx) {
5887 		assert(g_hotplug_probe_poller == NULL);
5888 		g_hotplug_probe_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug_probe, NULL, 1000);
5889 	}
5890 
5891 	return SPDK_POLLER_BUSY;
5892 }
5893 
5894 void
5895 bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts)
5896 {
5897 	*opts = g_opts;
5898 }
5899 
5900 static bool bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
5901 		uint32_t reconnect_delay_sec,
5902 		uint32_t fast_io_fail_timeout_sec);
5903 
5904 static int
5905 bdev_nvme_validate_opts(const struct spdk_bdev_nvme_opts *opts)
5906 {
5907 	if ((opts->timeout_us == 0) && (opts->timeout_admin_us != 0)) {
5908 		/* Can't set timeout_admin_us without also setting timeout_us */
5909 		SPDK_WARNLOG("Invalid options: Can't have (timeout_us == 0) with (timeout_admin_us > 0)\n");
5910 		return -EINVAL;
5911 	}
5912 
5913 	if (opts->bdev_retry_count < -1) {
5914 		SPDK_WARNLOG("Invalid option: bdev_retry_count can't be less than -1.\n");
5915 		return -EINVAL;
5916 	}
5917 
5918 	if (!bdev_nvme_check_io_error_resiliency_params(opts->ctrlr_loss_timeout_sec,
5919 			opts->reconnect_delay_sec,
5920 			opts->fast_io_fail_timeout_sec)) {
5921 		return -EINVAL;
5922 	}
5923 
5924 	return 0;
5925 }
5926 
5927 int
5928 bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts)
5929 {
5930 	int ret;
5931 
5932 	ret = bdev_nvme_validate_opts(opts);
5933 	if (ret) {
5934 		SPDK_WARNLOG("Failed to set nvme opts.\n");
5935 		return ret;
5936 	}
5937 
5938 	if (g_bdev_nvme_init_thread != NULL) {
5939 		if (!TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
5940 			return -EPERM;
5941 		}
5942 	}
5943 
5944 	if (opts->rdma_srq_size != 0 ||
5945 	    opts->rdma_max_cq_size != 0 ||
5946 	    opts->rdma_cm_event_timeout_ms != 0) {
5947 		struct spdk_nvme_transport_opts drv_opts;
5948 
5949 		spdk_nvme_transport_get_opts(&drv_opts, sizeof(drv_opts));
5950 		if (opts->rdma_srq_size != 0) {
5951 			drv_opts.rdma_srq_size = opts->rdma_srq_size;
5952 		}
5953 		if (opts->rdma_max_cq_size != 0) {
5954 			drv_opts.rdma_max_cq_size = opts->rdma_max_cq_size;
5955 		}
5956 		if (opts->rdma_cm_event_timeout_ms != 0) {
5957 			drv_opts.rdma_cm_event_timeout_ms = opts->rdma_cm_event_timeout_ms;
5958 		}
5959 
5960 		ret = spdk_nvme_transport_set_opts(&drv_opts, sizeof(drv_opts));
5961 		if (ret) {
5962 			SPDK_ERRLOG("Failed to set NVMe transport opts.\n");
5963 			return ret;
5964 		}
5965 	}
5966 
5967 	g_opts = *opts;
5968 
5969 	return 0;
5970 }
5971 
5972 struct set_nvme_hotplug_ctx {
5973 	uint64_t period_us;
5974 	bool enabled;
5975 	spdk_msg_fn fn;
5976 	void *fn_ctx;
5977 };
5978 
5979 static void
5980 set_nvme_hotplug_period_cb(void *_ctx)
5981 {
5982 	struct set_nvme_hotplug_ctx *ctx = _ctx;
5983 
5984 	spdk_poller_unregister(&g_hotplug_poller);
5985 	if (ctx->enabled) {
5986 		g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug, NULL, ctx->period_us);
5987 	} else {
5988 		g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_remove_poller, NULL,
5989 							NVME_HOTPLUG_POLL_PERIOD_DEFAULT);
5990 	}
5991 
5992 	g_nvme_hotplug_poll_period_us = ctx->period_us;
5993 	g_nvme_hotplug_enabled = ctx->enabled;
5994 	if (ctx->fn) {
5995 		ctx->fn(ctx->fn_ctx);
5996 	}
5997 
5998 	free(ctx);
5999 }
6000 
6001 int
6002 bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx)
6003 {
6004 	struct set_nvme_hotplug_ctx *ctx;
6005 
6006 	if (enabled == true && !spdk_process_is_primary()) {
6007 		return -EPERM;
6008 	}
6009 
6010 	ctx = calloc(1, sizeof(*ctx));
6011 	if (ctx == NULL) {
6012 		return -ENOMEM;
6013 	}
6014 
6015 	period_us = period_us == 0 ? NVME_HOTPLUG_POLL_PERIOD_DEFAULT : period_us;
6016 	ctx->period_us = spdk_min(period_us, NVME_HOTPLUG_POLL_PERIOD_MAX);
6017 	ctx->enabled = enabled;
6018 	ctx->fn = cb;
6019 	ctx->fn_ctx = cb_ctx;
6020 
6021 	spdk_thread_send_msg(g_bdev_nvme_init_thread, set_nvme_hotplug_period_cb, ctx);
6022 	return 0;
6023 }
6024 
6025 static void
6026 nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
6027 				    struct nvme_async_probe_ctx *ctx)
6028 {
6029 	struct nvme_ns	*nvme_ns;
6030 	struct nvme_bdev	*nvme_bdev;
6031 	size_t			j;
6032 
6033 	assert(nvme_ctrlr != NULL);
6034 
6035 	if (ctx->names == NULL) {
6036 		ctx->reported_bdevs = 0;
6037 		populate_namespaces_cb(ctx, 0);
6038 		return;
6039 	}
6040 
6041 	/*
6042 	 * Report the new bdevs that were created in this call.
6043 	 * There can be more than one bdev per NVMe controller.
6044 	 */
6045 	j = 0;
6046 	nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
6047 	while (nvme_ns != NULL) {
6048 		nvme_bdev = nvme_ns->bdev;
6049 		if (j < ctx->max_bdevs) {
6050 			ctx->names[j] = nvme_bdev->disk.name;
6051 			j++;
6052 		} else {
6053 			SPDK_ERRLOG("Maximum number of namespaces supported per NVMe controller is %du. Unable to return all names of created bdevs\n",
6054 				    ctx->max_bdevs);
6055 			ctx->reported_bdevs = 0;
6056 			populate_namespaces_cb(ctx, -ERANGE);
6057 			return;
6058 		}
6059 
6060 		nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
6061 	}
6062 
6063 	ctx->reported_bdevs = j;
6064 	populate_namespaces_cb(ctx, 0);
6065 }
6066 
6067 static int
6068 bdev_nvme_check_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
6069 			       struct spdk_nvme_ctrlr *new_ctrlr,
6070 			       struct spdk_nvme_transport_id *trid)
6071 {
6072 	struct nvme_path_id *tmp_trid;
6073 
6074 	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
6075 		SPDK_ERRLOG("PCIe failover is not supported.\n");
6076 		return -ENOTSUP;
6077 	}
6078 
6079 	/* Currently we only support failover to the same transport type. */
6080 	if (nvme_ctrlr->active_path_id->trid.trtype != trid->trtype) {
6081 		SPDK_WARNLOG("Failover from trtype: %s to a different trtype: %s is not supported currently\n",
6082 			     spdk_nvme_transport_id_trtype_str(nvme_ctrlr->active_path_id->trid.trtype),
6083 			     spdk_nvme_transport_id_trtype_str(trid->trtype));
6084 		return -EINVAL;
6085 	}
6086 
6087 
6088 	/* Currently we only support failover to the same NQN. */
6089 	if (strncmp(trid->subnqn, nvme_ctrlr->active_path_id->trid.subnqn, SPDK_NVMF_NQN_MAX_LEN)) {
6090 		SPDK_WARNLOG("Failover from subnqn: %s to a different subnqn: %s is not supported currently\n",
6091 			     nvme_ctrlr->active_path_id->trid.subnqn, trid->subnqn);
6092 		return -EINVAL;
6093 	}
6094 
6095 	/* Skip all the other checks if we've already registered this path. */
6096 	TAILQ_FOREACH(tmp_trid, &nvme_ctrlr->trids, link) {
6097 		if (!spdk_nvme_transport_id_compare(&tmp_trid->trid, trid)) {
6098 			SPDK_WARNLOG("This path (traddr: %s subnqn: %s) is already registered\n", trid->traddr,
6099 				     trid->subnqn);
6100 			return -EALREADY;
6101 		}
6102 	}
6103 
6104 	return 0;
6105 }
6106 
6107 static int
6108 bdev_nvme_check_secondary_namespace(struct nvme_ctrlr *nvme_ctrlr,
6109 				    struct spdk_nvme_ctrlr *new_ctrlr)
6110 {
6111 	struct nvme_ns *nvme_ns;
6112 	struct spdk_nvme_ns *new_ns;
6113 
6114 	nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
6115 	while (nvme_ns != NULL) {
6116 		new_ns = spdk_nvme_ctrlr_get_ns(new_ctrlr, nvme_ns->id);
6117 		assert(new_ns != NULL);
6118 
6119 		if (!bdev_nvme_compare_ns(nvme_ns->ns, new_ns)) {
6120 			return -EINVAL;
6121 		}
6122 
6123 		nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
6124 	}
6125 
6126 	return 0;
6127 }
6128 
6129 static int
6130 _bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
6131 			      struct spdk_nvme_transport_id *trid)
6132 {
6133 	struct nvme_path_id *active_id, *new_trid, *tmp_trid;
6134 
6135 	new_trid = calloc(1, sizeof(*new_trid));
6136 	if (new_trid == NULL) {
6137 		return -ENOMEM;
6138 	}
6139 	new_trid->trid = *trid;
6140 
6141 	active_id = nvme_ctrlr->active_path_id;
6142 	assert(active_id != NULL);
6143 	assert(active_id == TAILQ_FIRST(&nvme_ctrlr->trids));
6144 
6145 	/* Skip the active trid not to replace it until it is failed. */
6146 	tmp_trid = TAILQ_NEXT(active_id, link);
6147 	if (tmp_trid == NULL) {
6148 		goto add_tail;
6149 	}
6150 
6151 	/* It means the trid is faled if its last failed time is non-zero.
6152 	 * Insert the new alternate trid before any failed trid.
6153 	 */
6154 	TAILQ_FOREACH_FROM(tmp_trid, &nvme_ctrlr->trids, link) {
6155 		if (tmp_trid->last_failed_tsc != 0) {
6156 			TAILQ_INSERT_BEFORE(tmp_trid, new_trid, link);
6157 			return 0;
6158 		}
6159 	}
6160 
6161 add_tail:
6162 	TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, new_trid, link);
6163 	return 0;
6164 }
6165 
6166 /* This is the case that a secondary path is added to an existing
6167  * nvme_ctrlr for failover. After checking if it can access the same
6168  * namespaces as the primary path, it is disconnected until failover occurs.
6169  */
6170 static int
6171 bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
6172 			     struct spdk_nvme_ctrlr *new_ctrlr,
6173 			     struct spdk_nvme_transport_id *trid)
6174 {
6175 	int rc;
6176 
6177 	assert(nvme_ctrlr != NULL);
6178 
6179 	pthread_mutex_lock(&nvme_ctrlr->mutex);
6180 
6181 	rc = bdev_nvme_check_secondary_trid(nvme_ctrlr, new_ctrlr, trid);
6182 	if (rc != 0) {
6183 		goto exit;
6184 	}
6185 
6186 	rc = bdev_nvme_check_secondary_namespace(nvme_ctrlr, new_ctrlr);
6187 	if (rc != 0) {
6188 		goto exit;
6189 	}
6190 
6191 	rc = _bdev_nvme_add_secondary_trid(nvme_ctrlr, trid);
6192 
6193 exit:
6194 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
6195 
6196 	spdk_nvme_detach(new_ctrlr);
6197 
6198 	return rc;
6199 }
6200 
6201 static void
6202 connect_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
6203 		  struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
6204 {
6205 	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6206 	struct nvme_async_probe_ctx *ctx;
6207 	int rc;
6208 
6209 	ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
6210 	ctx->ctrlr_attached = true;
6211 
6212 	rc = nvme_ctrlr_create(ctrlr, ctx->base_name, &ctx->trid, ctx);
6213 	if (rc != 0) {
6214 		ctx->reported_bdevs = 0;
6215 		populate_namespaces_cb(ctx, rc);
6216 	}
6217 }
6218 
6219 
6220 static void
6221 connect_set_failover_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
6222 			struct spdk_nvme_ctrlr *ctrlr,
6223 			const struct spdk_nvme_ctrlr_opts *opts)
6224 {
6225 	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
6226 	struct nvme_ctrlr *nvme_ctrlr;
6227 	struct nvme_async_probe_ctx *ctx;
6228 	int rc;
6229 
6230 	ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
6231 	ctx->ctrlr_attached = true;
6232 
6233 	nvme_ctrlr = nvme_ctrlr_get_by_name(ctx->base_name);
6234 	if (nvme_ctrlr) {
6235 		rc = bdev_nvme_add_secondary_trid(nvme_ctrlr, ctrlr, &ctx->trid);
6236 	} else {
6237 		rc = -ENODEV;
6238 	}
6239 
6240 	ctx->reported_bdevs = 0;
6241 	populate_namespaces_cb(ctx, rc);
6242 }
6243 
6244 static int
6245 bdev_nvme_async_poll(void *arg)
6246 {
6247 	struct nvme_async_probe_ctx	*ctx = arg;
6248 	int				rc;
6249 
6250 	rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
6251 	if (spdk_unlikely(rc != -EAGAIN)) {
6252 		ctx->probe_done = true;
6253 		spdk_poller_unregister(&ctx->poller);
6254 		if (!ctx->ctrlr_attached) {
6255 			/* The probe is done, but no controller was attached.
6256 			 * That means we had a failure, so report -EIO back to
6257 			 * the caller (usually the RPC). populate_namespaces_cb()
6258 			 * will take care of freeing the nvme_async_probe_ctx.
6259 			 */
6260 			ctx->reported_bdevs = 0;
6261 			populate_namespaces_cb(ctx, -EIO);
6262 		} else if (ctx->namespaces_populated) {
6263 			/* The namespaces for the attached controller were all
6264 			 * populated and the response was already sent to the
6265 			 * caller (usually the RPC).  So free the context here.
6266 			 */
6267 			free_nvme_async_probe_ctx(ctx);
6268 		}
6269 	}
6270 
6271 	return SPDK_POLLER_BUSY;
6272 }
6273 
6274 static bool
6275 bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
6276 		uint32_t reconnect_delay_sec,
6277 		uint32_t fast_io_fail_timeout_sec)
6278 {
6279 	if (ctrlr_loss_timeout_sec < -1) {
6280 		SPDK_ERRLOG("ctrlr_loss_timeout_sec can't be less than -1.\n");
6281 		return false;
6282 	} else if (ctrlr_loss_timeout_sec == -1) {
6283 		if (reconnect_delay_sec == 0) {
6284 			SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
6285 			return false;
6286 		} else if (fast_io_fail_timeout_sec != 0 &&
6287 			   fast_io_fail_timeout_sec < reconnect_delay_sec) {
6288 			SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io-fail_timeout_sec.\n");
6289 			return false;
6290 		}
6291 	} else if (ctrlr_loss_timeout_sec != 0) {
6292 		if (reconnect_delay_sec == 0) {
6293 			SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
6294 			return false;
6295 		} else if (reconnect_delay_sec > (uint32_t)ctrlr_loss_timeout_sec) {
6296 			SPDK_ERRLOG("reconnect_delay_sec can't be more than ctrlr_loss_timeout_sec.\n");
6297 			return false;
6298 		} else if (fast_io_fail_timeout_sec != 0) {
6299 			if (fast_io_fail_timeout_sec < reconnect_delay_sec) {
6300 				SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io_fail_timeout_sec.\n");
6301 				return false;
6302 			} else if (fast_io_fail_timeout_sec > (uint32_t)ctrlr_loss_timeout_sec) {
6303 				SPDK_ERRLOG("fast_io_fail_timeout_sec can't be more than ctrlr_loss_timeout_sec.\n");
6304 				return false;
6305 			}
6306 		}
6307 	} else if (reconnect_delay_sec != 0 || fast_io_fail_timeout_sec != 0) {
6308 		SPDK_ERRLOG("Both reconnect_delay_sec and fast_io_fail_timeout_sec must be 0 if ctrlr_loss_timeout_sec is 0.\n");
6309 		return false;
6310 	}
6311 
6312 	return true;
6313 }
6314 
6315 int
6316 spdk_bdev_nvme_create(struct spdk_nvme_transport_id *trid,
6317 		      const char *base_name,
6318 		      const char **names,
6319 		      uint32_t count,
6320 		      spdk_bdev_nvme_create_cb cb_fn,
6321 		      void *cb_ctx,
6322 		      struct spdk_nvme_ctrlr_opts *drv_opts,
6323 		      struct spdk_bdev_nvme_ctrlr_opts *bdev_opts)
6324 {
6325 	struct nvme_probe_skip_entry *entry, *tmp;
6326 	struct nvme_async_probe_ctx *ctx;
6327 	spdk_nvme_attach_cb attach_cb;
6328 	struct nvme_ctrlr *nvme_ctrlr;
6329 	int len;
6330 
6331 	/* TODO expand this check to include both the host and target TRIDs.
6332 	 * Only if both are the same should we fail.
6333 	 */
6334 	if (nvme_ctrlr_get(trid, drv_opts->hostnqn) != NULL) {
6335 		SPDK_ERRLOG("A controller with the provided trid (traddr: %s, hostnqn: %s) "
6336 			    "already exists.\n", trid->traddr, drv_opts->hostnqn);
6337 		return -EEXIST;
6338 	}
6339 
6340 	len = strnlen(base_name, SPDK_CONTROLLER_NAME_MAX);
6341 
6342 	if (len == 0 || len == SPDK_CONTROLLER_NAME_MAX) {
6343 		SPDK_ERRLOG("controller name must be between 1 and %d characters\n", SPDK_CONTROLLER_NAME_MAX - 1);
6344 		return -EINVAL;
6345 	}
6346 
6347 	if (bdev_opts != NULL &&
6348 	    !bdev_nvme_check_io_error_resiliency_params(bdev_opts->ctrlr_loss_timeout_sec,
6349 			    bdev_opts->reconnect_delay_sec,
6350 			    bdev_opts->fast_io_fail_timeout_sec)) {
6351 		return -EINVAL;
6352 	}
6353 
6354 	ctx = calloc(1, sizeof(*ctx));
6355 	if (!ctx) {
6356 		return -ENOMEM;
6357 	}
6358 	ctx->base_name = base_name;
6359 	ctx->names = names;
6360 	ctx->max_bdevs = count;
6361 	ctx->cb_fn = cb_fn;
6362 	ctx->cb_ctx = cb_ctx;
6363 	ctx->trid = *trid;
6364 
6365 	if (bdev_opts) {
6366 		memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
6367 	} else {
6368 		spdk_bdev_nvme_get_default_ctrlr_opts(&ctx->bdev_opts);
6369 	}
6370 
6371 	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
6372 		TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, tmp) {
6373 			if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
6374 				TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
6375 				free(entry);
6376 				break;
6377 			}
6378 		}
6379 	}
6380 
6381 	memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
6382 	ctx->drv_opts.transport_retry_count = g_opts.transport_retry_count;
6383 	ctx->drv_opts.transport_ack_timeout = g_opts.transport_ack_timeout;
6384 	ctx->drv_opts.keep_alive_timeout_ms = g_opts.keep_alive_timeout_ms;
6385 	ctx->drv_opts.disable_read_ana_log_page = true;
6386 	ctx->drv_opts.transport_tos = g_opts.transport_tos;
6387 
6388 	if (ctx->bdev_opts.psk != NULL) {
6389 		ctx->drv_opts.tls_psk = spdk_keyring_get_key(ctx->bdev_opts.psk);
6390 		if (ctx->drv_opts.tls_psk == NULL) {
6391 			SPDK_ERRLOG("Could not load PSK: %s\n", ctx->bdev_opts.psk);
6392 			free_nvme_async_probe_ctx(ctx);
6393 			return -ENOKEY;
6394 		}
6395 	}
6396 
6397 	if (ctx->bdev_opts.dhchap_key != NULL) {
6398 		ctx->drv_opts.dhchap_key = spdk_keyring_get_key(ctx->bdev_opts.dhchap_key);
6399 		if (ctx->drv_opts.dhchap_key == NULL) {
6400 			SPDK_ERRLOG("Could not load DH-HMAC-CHAP key: %s\n",
6401 				    ctx->bdev_opts.dhchap_key);
6402 			free_nvme_async_probe_ctx(ctx);
6403 			return -ENOKEY;
6404 		}
6405 
6406 		ctx->drv_opts.dhchap_digests = g_opts.dhchap_digests;
6407 		ctx->drv_opts.dhchap_dhgroups = g_opts.dhchap_dhgroups;
6408 	}
6409 	if (ctx->bdev_opts.dhchap_ctrlr_key != NULL) {
6410 		ctx->drv_opts.dhchap_ctrlr_key =
6411 			spdk_keyring_get_key(ctx->bdev_opts.dhchap_ctrlr_key);
6412 		if (ctx->drv_opts.dhchap_ctrlr_key == NULL) {
6413 			SPDK_ERRLOG("Could not load DH-HMAC-CHAP controller key: %s\n",
6414 				    ctx->bdev_opts.dhchap_ctrlr_key);
6415 			free_nvme_async_probe_ctx(ctx);
6416 			return -ENOKEY;
6417 		}
6418 	}
6419 
6420 	if (nvme_bdev_ctrlr_get_by_name(base_name) == NULL || ctx->bdev_opts.multipath) {
6421 		attach_cb = connect_attach_cb;
6422 	} else {
6423 		attach_cb = connect_set_failover_cb;
6424 	}
6425 
6426 	nvme_ctrlr = nvme_ctrlr_get_by_name(ctx->base_name);
6427 	if (nvme_ctrlr  && nvme_ctrlr->opts.multipath != ctx->bdev_opts.multipath) {
6428 		/* All controllers with the same name must be configured the same
6429 		 * way, either for multipath or failover. If the configuration doesn't
6430 		 * match - report error.
6431 		 */
6432 		free_nvme_async_probe_ctx(ctx);
6433 		return -EINVAL;
6434 	}
6435 
6436 	ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, attach_cb);
6437 	if (ctx->probe_ctx == NULL) {
6438 		SPDK_ERRLOG("No controller was found with provided trid (traddr: %s)\n", trid->traddr);
6439 		free_nvme_async_probe_ctx(ctx);
6440 		return -ENODEV;
6441 	}
6442 	ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_async_poll, ctx, 1000);
6443 
6444 	return 0;
6445 }
6446 
6447 struct bdev_nvme_delete_ctx {
6448 	char                        *name;
6449 	struct nvme_path_id         path_id;
6450 	bdev_nvme_delete_done_fn    delete_done;
6451 	void                        *delete_done_ctx;
6452 	uint64_t                    timeout_ticks;
6453 	struct spdk_poller          *poller;
6454 };
6455 
6456 static void
6457 free_bdev_nvme_delete_ctx(struct bdev_nvme_delete_ctx *ctx)
6458 {
6459 	if (ctx != NULL) {
6460 		free(ctx->name);
6461 		free(ctx);
6462 	}
6463 }
6464 
6465 static bool
6466 nvme_path_id_compare(struct nvme_path_id *p, const struct nvme_path_id *path_id)
6467 {
6468 	if (path_id->trid.trtype != 0) {
6469 		if (path_id->trid.trtype == SPDK_NVME_TRANSPORT_CUSTOM) {
6470 			if (strcasecmp(path_id->trid.trstring, p->trid.trstring) != 0) {
6471 				return false;
6472 			}
6473 		} else {
6474 			if (path_id->trid.trtype != p->trid.trtype) {
6475 				return false;
6476 			}
6477 		}
6478 	}
6479 
6480 	if (!spdk_mem_all_zero(path_id->trid.traddr, sizeof(path_id->trid.traddr))) {
6481 		if (strcasecmp(path_id->trid.traddr, p->trid.traddr) != 0) {
6482 			return false;
6483 		}
6484 	}
6485 
6486 	if (path_id->trid.adrfam != 0) {
6487 		if (path_id->trid.adrfam != p->trid.adrfam) {
6488 			return false;
6489 		}
6490 	}
6491 
6492 	if (!spdk_mem_all_zero(path_id->trid.trsvcid, sizeof(path_id->trid.trsvcid))) {
6493 		if (strcasecmp(path_id->trid.trsvcid, p->trid.trsvcid) != 0) {
6494 			return false;
6495 		}
6496 	}
6497 
6498 	if (!spdk_mem_all_zero(path_id->trid.subnqn, sizeof(path_id->trid.subnqn))) {
6499 		if (strcmp(path_id->trid.subnqn, p->trid.subnqn) != 0) {
6500 			return false;
6501 		}
6502 	}
6503 
6504 	if (!spdk_mem_all_zero(path_id->hostid.hostaddr, sizeof(path_id->hostid.hostaddr))) {
6505 		if (strcmp(path_id->hostid.hostaddr, p->hostid.hostaddr) != 0) {
6506 			return false;
6507 		}
6508 	}
6509 
6510 	if (!spdk_mem_all_zero(path_id->hostid.hostsvcid, sizeof(path_id->hostid.hostsvcid))) {
6511 		if (strcmp(path_id->hostid.hostsvcid, p->hostid.hostsvcid) != 0) {
6512 			return false;
6513 		}
6514 	}
6515 
6516 	return true;
6517 }
6518 
6519 static bool
6520 nvme_path_id_exists(const char *name, const struct nvme_path_id *path_id)
6521 {
6522 	struct nvme_bdev_ctrlr  *nbdev_ctrlr;
6523 	struct nvme_ctrlr       *ctrlr;
6524 	struct nvme_path_id     *p;
6525 
6526 	pthread_mutex_lock(&g_bdev_nvme_mutex);
6527 	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
6528 	if (!nbdev_ctrlr) {
6529 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
6530 		return false;
6531 	}
6532 
6533 	TAILQ_FOREACH(ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
6534 		pthread_mutex_lock(&ctrlr->mutex);
6535 		TAILQ_FOREACH(p, &ctrlr->trids, link) {
6536 			if (nvme_path_id_compare(p, path_id)) {
6537 				pthread_mutex_unlock(&ctrlr->mutex);
6538 				pthread_mutex_unlock(&g_bdev_nvme_mutex);
6539 				return true;
6540 			}
6541 		}
6542 		pthread_mutex_unlock(&ctrlr->mutex);
6543 	}
6544 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
6545 
6546 	return false;
6547 }
6548 
6549 static int
6550 bdev_nvme_delete_complete_poll(void *arg)
6551 {
6552 	struct bdev_nvme_delete_ctx     *ctx = arg;
6553 	int                             rc = 0;
6554 
6555 	if (nvme_path_id_exists(ctx->name, &ctx->path_id)) {
6556 		if (ctx->timeout_ticks > spdk_get_ticks()) {
6557 			return SPDK_POLLER_BUSY;
6558 		}
6559 
6560 		SPDK_ERRLOG("NVMe path '%s' still exists after delete\n", ctx->name);
6561 		rc = -ETIMEDOUT;
6562 	}
6563 
6564 	spdk_poller_unregister(&ctx->poller);
6565 
6566 	ctx->delete_done(ctx->delete_done_ctx, rc);
6567 	free_bdev_nvme_delete_ctx(ctx);
6568 
6569 	return SPDK_POLLER_BUSY;
6570 }
6571 
6572 static int
6573 _bdev_nvme_delete(struct nvme_ctrlr *nvme_ctrlr, const struct nvme_path_id *path_id)
6574 {
6575 	struct nvme_path_id	*p, *t;
6576 	spdk_msg_fn		msg_fn;
6577 	int			rc = -ENXIO;
6578 
6579 	pthread_mutex_lock(&nvme_ctrlr->mutex);
6580 
6581 	TAILQ_FOREACH_REVERSE_SAFE(p, &nvme_ctrlr->trids, nvme_paths, link, t) {
6582 		if (p == TAILQ_FIRST(&nvme_ctrlr->trids)) {
6583 			break;
6584 		}
6585 
6586 		if (!nvme_path_id_compare(p, path_id)) {
6587 			continue;
6588 		}
6589 
6590 		/* We are not using the specified path. */
6591 		TAILQ_REMOVE(&nvme_ctrlr->trids, p, link);
6592 		free(p);
6593 		rc = 0;
6594 	}
6595 
6596 	if (p == NULL || !nvme_path_id_compare(p, path_id)) {
6597 		pthread_mutex_unlock(&nvme_ctrlr->mutex);
6598 		return rc;
6599 	}
6600 
6601 	/* If we made it here, then this path is a match! Now we need to remove it. */
6602 
6603 	/* This is the active path in use right now. The active path is always the first in the list. */
6604 	assert(p == nvme_ctrlr->active_path_id);
6605 
6606 	if (!TAILQ_NEXT(p, link)) {
6607 		/* The current path is the only path. */
6608 		msg_fn = _nvme_ctrlr_destruct;
6609 		rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, false);
6610 	} else {
6611 		/* There is an alternative path. */
6612 		msg_fn = _bdev_nvme_reset_ctrlr;
6613 		rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, true);
6614 	}
6615 
6616 	pthread_mutex_unlock(&nvme_ctrlr->mutex);
6617 
6618 	if (rc == 0) {
6619 		spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
6620 	} else if (rc == -EALREADY) {
6621 		rc = 0;
6622 	}
6623 
6624 	return rc;
6625 }
6626 
6627 int
6628 bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id,
6629 		 bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx)
6630 {
6631 	struct nvme_bdev_ctrlr		*nbdev_ctrlr;
6632 	struct nvme_ctrlr		*nvme_ctrlr, *tmp_nvme_ctrlr;
6633 	struct bdev_nvme_delete_ctx     *ctx = NULL;
6634 	int				rc = -ENXIO, _rc;
6635 
6636 	if (name == NULL || path_id == NULL) {
6637 		rc = -EINVAL;
6638 		goto exit;
6639 	}
6640 
6641 	pthread_mutex_lock(&g_bdev_nvme_mutex);
6642 
6643 	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
6644 	if (nbdev_ctrlr == NULL) {
6645 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
6646 
6647 		SPDK_ERRLOG("Failed to find NVMe bdev controller\n");
6648 		rc = -ENODEV;
6649 		goto exit;
6650 	}
6651 
6652 	TAILQ_FOREACH_SAFE(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq, tmp_nvme_ctrlr) {
6653 		_rc = _bdev_nvme_delete(nvme_ctrlr, path_id);
6654 		if (_rc < 0 && _rc != -ENXIO) {
6655 			pthread_mutex_unlock(&g_bdev_nvme_mutex);
6656 			rc = _rc;
6657 			goto exit;
6658 		} else if (_rc == 0) {
6659 			/* We traverse all remaining nvme_ctrlrs even if one nvme_ctrlr
6660 			 * was deleted successfully. To remember the successful deletion,
6661 			 * overwrite rc only if _rc is zero.
6662 			 */
6663 			rc = 0;
6664 		}
6665 	}
6666 
6667 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
6668 
6669 	if (rc != 0 || delete_done == NULL) {
6670 		goto exit;
6671 	}
6672 
6673 	ctx = calloc(1, sizeof(*ctx));
6674 	if (ctx == NULL) {
6675 		SPDK_ERRLOG("Failed to allocate context for bdev_nvme_delete\n");
6676 		rc = -ENOMEM;
6677 		goto exit;
6678 	}
6679 
6680 	ctx->name = strdup(name);
6681 	if (ctx->name == NULL) {
6682 		SPDK_ERRLOG("Failed to copy controller name for deletion\n");
6683 		rc = -ENOMEM;
6684 		goto exit;
6685 	}
6686 
6687 	ctx->delete_done = delete_done;
6688 	ctx->delete_done_ctx = delete_done_ctx;
6689 	ctx->path_id = *path_id;
6690 	ctx->timeout_ticks = spdk_get_ticks() + 10 * spdk_get_ticks_hz();
6691 	ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_delete_complete_poll, ctx, 1000);
6692 	if (ctx->poller == NULL) {
6693 		SPDK_ERRLOG("Failed to register bdev_nvme_delete poller\n");
6694 		rc = -ENOMEM;
6695 		goto exit;
6696 	}
6697 
6698 exit:
6699 	if (rc != 0) {
6700 		free_bdev_nvme_delete_ctx(ctx);
6701 	}
6702 
6703 	return rc;
6704 }
6705 
6706 #define DISCOVERY_INFOLOG(ctx, format, ...) \
6707 	SPDK_INFOLOG(bdev_nvme, "Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);
6708 
6709 #define DISCOVERY_ERRLOG(ctx, format, ...) \
6710 	SPDK_ERRLOG("Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);
6711 
6712 struct discovery_entry_ctx {
6713 	char						name[128];
6714 	struct spdk_nvme_transport_id			trid;
6715 	struct spdk_nvme_ctrlr_opts			drv_opts;
6716 	struct spdk_nvmf_discovery_log_page_entry	entry;
6717 	TAILQ_ENTRY(discovery_entry_ctx)		tailq;
6718 	struct discovery_ctx				*ctx;
6719 };
6720 
6721 struct discovery_ctx {
6722 	char					*name;
6723 	spdk_bdev_nvme_start_discovery_fn	start_cb_fn;
6724 	spdk_bdev_nvme_stop_discovery_fn	stop_cb_fn;
6725 	void					*cb_ctx;
6726 	struct spdk_nvme_probe_ctx		*probe_ctx;
6727 	struct spdk_nvme_detach_ctx		*detach_ctx;
6728 	struct spdk_nvme_ctrlr			*ctrlr;
6729 	struct spdk_nvme_transport_id		trid;
6730 	struct discovery_entry_ctx		*entry_ctx_in_use;
6731 	struct spdk_poller			*poller;
6732 	struct spdk_nvme_ctrlr_opts		drv_opts;
6733 	struct spdk_bdev_nvme_ctrlr_opts	bdev_opts;
6734 	struct spdk_nvmf_discovery_log_page	*log_page;
6735 	TAILQ_ENTRY(discovery_ctx)		tailq;
6736 	TAILQ_HEAD(, discovery_entry_ctx)	nvm_entry_ctxs;
6737 	TAILQ_HEAD(, discovery_entry_ctx)	discovery_entry_ctxs;
6738 	int					rc;
6739 	bool					wait_for_attach;
6740 	uint64_t				timeout_ticks;
6741 	/* Denotes that the discovery service is being started. We're waiting
6742 	 * for the initial connection to the discovery controller to be
6743 	 * established and attach discovered NVM ctrlrs.
6744 	 */
6745 	bool					initializing;
6746 	/* Denotes if a discovery is currently in progress for this context.
6747 	 * That includes connecting to newly discovered subsystems.  Used to
6748 	 * ensure we do not start a new discovery until an existing one is
6749 	 * complete.
6750 	 */
6751 	bool					in_progress;
6752 
6753 	/* Denotes if another discovery is needed after the one in progress
6754 	 * completes.  Set when we receive an AER completion while a discovery
6755 	 * is already in progress.
6756 	 */
6757 	bool					pending;
6758 
6759 	/* Signal to the discovery context poller that it should stop the
6760 	 * discovery service, including detaching from the current discovery
6761 	 * controller.
6762 	 */
6763 	bool					stop;
6764 
6765 	struct spdk_thread			*calling_thread;
6766 	uint32_t				index;
6767 	uint32_t				attach_in_progress;
6768 	char					*hostnqn;
6769 
6770 	/* Denotes if the discovery service was started by the mdns discovery.
6771 	 */
6772 	bool					from_mdns_discovery_service;
6773 };
6774 
6775 TAILQ_HEAD(discovery_ctxs, discovery_ctx);
6776 static struct discovery_ctxs g_discovery_ctxs = TAILQ_HEAD_INITIALIZER(g_discovery_ctxs);
6777 
6778 static void get_discovery_log_page(struct discovery_ctx *ctx);
6779 
6780 static void
6781 free_discovery_ctx(struct discovery_ctx *ctx)
6782 {
6783 	free(ctx->log_page);
6784 	free(ctx->hostnqn);
6785 	free(ctx->name);
6786 	free(ctx);
6787 }
6788 
6789 static void
6790 discovery_complete(struct discovery_ctx *ctx)
6791 {
6792 	ctx->initializing = false;
6793 	ctx->in_progress = false;
6794 	if (ctx->pending) {
6795 		ctx->pending = false;
6796 		get_discovery_log_page(ctx);
6797 	}
6798 }
6799 
6800 static void
6801 build_trid_from_log_page_entry(struct spdk_nvme_transport_id *trid,
6802 			       struct spdk_nvmf_discovery_log_page_entry *entry)
6803 {
6804 	char *space;
6805 
6806 	trid->trtype = entry->trtype;
6807 	trid->adrfam = entry->adrfam;
6808 	memcpy(trid->traddr, entry->traddr, sizeof(entry->traddr));
6809 	memcpy(trid->trsvcid, entry->trsvcid, sizeof(entry->trsvcid));
6810 	/* Because the source buffer (entry->subnqn) is longer than trid->subnqn, and
6811 	 * before call to this function trid->subnqn is zeroed out, we need
6812 	 * to copy sizeof(trid->subnqn) minus one byte to make sure the last character
6813 	 * remains 0. Then we can shorten the string (replace ' ' with 0) if required
6814 	 */
6815 	memcpy(trid->subnqn, entry->subnqn, sizeof(trid->subnqn) - 1);
6816 
6817 	/* We want the traddr, trsvcid and subnqn fields to be NULL-terminated.
6818 	 * But the log page entries typically pad them with spaces, not zeroes.
6819 	 * So add a NULL terminator to each of these fields at the appropriate
6820 	 * location.
6821 	 */
6822 	space = strchr(trid->traddr, ' ');
6823 	if (space) {
6824 		*space = 0;
6825 	}
6826 	space = strchr(trid->trsvcid, ' ');
6827 	if (space) {
6828 		*space = 0;
6829 	}
6830 	space = strchr(trid->subnqn, ' ');
6831 	if (space) {
6832 		*space = 0;
6833 	}
6834 }
6835 
6836 static void
6837 _stop_discovery(void *_ctx)
6838 {
6839 	struct discovery_ctx *ctx = _ctx;
6840 
6841 	if (ctx->attach_in_progress > 0) {
6842 		spdk_thread_send_msg(spdk_get_thread(), _stop_discovery, ctx);
6843 		return;
6844 	}
6845 
6846 	ctx->stop = true;
6847 
6848 	while (!TAILQ_EMPTY(&ctx->nvm_entry_ctxs)) {
6849 		struct discovery_entry_ctx *entry_ctx;
6850 		struct nvme_path_id path = {};
6851 
6852 		entry_ctx = TAILQ_FIRST(&ctx->nvm_entry_ctxs);
6853 		path.trid = entry_ctx->trid;
6854 		bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
6855 		TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
6856 		free(entry_ctx);
6857 	}
6858 
6859 	while (!TAILQ_EMPTY(&ctx->discovery_entry_ctxs)) {
6860 		struct discovery_entry_ctx *entry_ctx;
6861 
6862 		entry_ctx = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
6863 		TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
6864 		free(entry_ctx);
6865 	}
6866 
6867 	free(ctx->entry_ctx_in_use);
6868 	ctx->entry_ctx_in_use = NULL;
6869 }
6870 
6871 static void
6872 stop_discovery(struct discovery_ctx *ctx, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
6873 {
6874 	ctx->stop_cb_fn = cb_fn;
6875 	ctx->cb_ctx = cb_ctx;
6876 
6877 	if (ctx->attach_in_progress > 0) {
6878 		DISCOVERY_INFOLOG(ctx, "stopping discovery with attach_in_progress: %"PRIu32"\n",
6879 				  ctx->attach_in_progress);
6880 	}
6881 
6882 	_stop_discovery(ctx);
6883 }
6884 
6885 static void
6886 remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr)
6887 {
6888 	struct discovery_ctx *d_ctx;
6889 	struct nvme_path_id *path_id;
6890 	struct spdk_nvme_transport_id trid = {};
6891 	struct discovery_entry_ctx *entry_ctx, *tmp;
6892 
6893 	path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
6894 
6895 	TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
6896 		TAILQ_FOREACH_SAFE(entry_ctx, &d_ctx->nvm_entry_ctxs, tailq, tmp) {
6897 			build_trid_from_log_page_entry(&trid, &entry_ctx->entry);
6898 			if (spdk_nvme_transport_id_compare(&trid, &path_id->trid) != 0) {
6899 				continue;
6900 			}
6901 
6902 			TAILQ_REMOVE(&d_ctx->nvm_entry_ctxs, entry_ctx, tailq);
6903 			free(entry_ctx);
6904 			DISCOVERY_INFOLOG(d_ctx, "Remove discovery entry: %s:%s:%s\n",
6905 					  trid.subnqn, trid.traddr, trid.trsvcid);
6906 
6907 			/* Fail discovery ctrlr to force reattach attempt */
6908 			spdk_nvme_ctrlr_fail(d_ctx->ctrlr);
6909 		}
6910 	}
6911 }
6912 
6913 static void
6914 discovery_remove_controllers(struct discovery_ctx *ctx)
6915 {
6916 	struct spdk_nvmf_discovery_log_page *log_page = ctx->log_page;
6917 	struct discovery_entry_ctx *entry_ctx, *tmp;
6918 	struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
6919 	struct spdk_nvme_transport_id old_trid = {};
6920 	uint64_t numrec, i;
6921 	bool found;
6922 
6923 	numrec = from_le64(&log_page->numrec);
6924 	TAILQ_FOREACH_SAFE(entry_ctx, &ctx->nvm_entry_ctxs, tailq, tmp) {
6925 		found = false;
6926 		old_entry = &entry_ctx->entry;
6927 		build_trid_from_log_page_entry(&old_trid, old_entry);
6928 		for (i = 0; i < numrec; i++) {
6929 			new_entry = &log_page->entries[i];
6930 			if (!memcmp(old_entry, new_entry, sizeof(*old_entry))) {
6931 				DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s found again\n",
6932 						  old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
6933 				found = true;
6934 				break;
6935 			}
6936 		}
6937 		if (!found) {
6938 			struct nvme_path_id path = {};
6939 
6940 			DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s not found\n",
6941 					  old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
6942 
6943 			path.trid = entry_ctx->trid;
6944 			bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
6945 			TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
6946 			free(entry_ctx);
6947 		}
6948 	}
6949 	free(log_page);
6950 	ctx->log_page = NULL;
6951 	discovery_complete(ctx);
6952 }
6953 
6954 static void
6955 complete_discovery_start(struct discovery_ctx *ctx, int status)
6956 {
6957 	ctx->timeout_ticks = 0;
6958 	ctx->rc = status;
6959 	if (ctx->start_cb_fn) {
6960 		ctx->start_cb_fn(ctx->cb_ctx, status);
6961 		ctx->start_cb_fn = NULL;
6962 		ctx->cb_ctx = NULL;
6963 	}
6964 }
6965 
6966 static void
6967 discovery_attach_controller_done(void *cb_ctx, size_t bdev_count, int rc)
6968 {
6969 	struct discovery_entry_ctx *entry_ctx = cb_ctx;
6970 	struct discovery_ctx *ctx = entry_ctx->ctx;
6971 
6972 	DISCOVERY_INFOLOG(ctx, "attach %s done\n", entry_ctx->name);
6973 	ctx->attach_in_progress--;
6974 	if (ctx->attach_in_progress == 0) {
6975 		complete_discovery_start(ctx, ctx->rc);
6976 		if (ctx->initializing && ctx->rc != 0) {
6977 			DISCOVERY_ERRLOG(ctx, "stopping discovery due to errors: %d\n", ctx->rc);
6978 			stop_discovery(ctx, NULL, ctx->cb_ctx);
6979 		} else {
6980 			discovery_remove_controllers(ctx);
6981 		}
6982 	}
6983 }
6984 
6985 static struct discovery_entry_ctx *
6986 create_discovery_entry_ctx(struct discovery_ctx *ctx, struct spdk_nvme_transport_id *trid)
6987 {
6988 	struct discovery_entry_ctx *new_ctx;
6989 
6990 	new_ctx = calloc(1, sizeof(*new_ctx));
6991 	if (new_ctx == NULL) {
6992 		DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
6993 		return NULL;
6994 	}
6995 
6996 	new_ctx->ctx = ctx;
6997 	memcpy(&new_ctx->trid, trid, sizeof(*trid));
6998 	spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
6999 	snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
7000 	return new_ctx;
7001 }
7002 
7003 static void
7004 discovery_log_page_cb(void *cb_arg, int rc, const struct spdk_nvme_cpl *cpl,
7005 		      struct spdk_nvmf_discovery_log_page *log_page)
7006 {
7007 	struct discovery_ctx *ctx = cb_arg;
7008 	struct discovery_entry_ctx *entry_ctx, *tmp;
7009 	struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
7010 	uint64_t numrec, i;
7011 	bool found;
7012 
7013 	if (rc || spdk_nvme_cpl_is_error(cpl)) {
7014 		DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
7015 		return;
7016 	}
7017 
7018 	ctx->log_page = log_page;
7019 	assert(ctx->attach_in_progress == 0);
7020 	numrec = from_le64(&log_page->numrec);
7021 	TAILQ_FOREACH_SAFE(entry_ctx, &ctx->discovery_entry_ctxs, tailq, tmp) {
7022 		TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
7023 		free(entry_ctx);
7024 	}
7025 	for (i = 0; i < numrec; i++) {
7026 		found = false;
7027 		new_entry = &log_page->entries[i];
7028 		if (new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
7029 		    new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
7030 			struct discovery_entry_ctx *new_ctx;
7031 			struct spdk_nvme_transport_id trid = {};
7032 
7033 			build_trid_from_log_page_entry(&trid, new_entry);
7034 			new_ctx = create_discovery_entry_ctx(ctx, &trid);
7035 			if (new_ctx == NULL) {
7036 				DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7037 				break;
7038 			}
7039 
7040 			TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, new_ctx, tailq);
7041 			continue;
7042 		}
7043 		TAILQ_FOREACH(entry_ctx, &ctx->nvm_entry_ctxs, tailq) {
7044 			old_entry = &entry_ctx->entry;
7045 			if (!memcmp(new_entry, old_entry, sizeof(*new_entry))) {
7046 				found = true;
7047 				break;
7048 			}
7049 		}
7050 		if (!found) {
7051 			struct discovery_entry_ctx *subnqn_ctx = NULL, *new_ctx;
7052 			struct discovery_ctx *d_ctx;
7053 
7054 			TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
7055 				TAILQ_FOREACH(subnqn_ctx, &d_ctx->nvm_entry_ctxs, tailq) {
7056 					if (!memcmp(subnqn_ctx->entry.subnqn, new_entry->subnqn,
7057 						    sizeof(new_entry->subnqn))) {
7058 						break;
7059 					}
7060 				}
7061 				if (subnqn_ctx) {
7062 					break;
7063 				}
7064 			}
7065 
7066 			new_ctx = calloc(1, sizeof(*new_ctx));
7067 			if (new_ctx == NULL) {
7068 				DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7069 				break;
7070 			}
7071 
7072 			new_ctx->ctx = ctx;
7073 			memcpy(&new_ctx->entry, new_entry, sizeof(*new_entry));
7074 			build_trid_from_log_page_entry(&new_ctx->trid, new_entry);
7075 			if (subnqn_ctx) {
7076 				snprintf(new_ctx->name, sizeof(new_ctx->name), "%s", subnqn_ctx->name);
7077 				DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new path for %s\n",
7078 						  new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
7079 						  new_ctx->name);
7080 			} else {
7081 				snprintf(new_ctx->name, sizeof(new_ctx->name), "%s%d", ctx->name, ctx->index++);
7082 				DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new subsystem %s\n",
7083 						  new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
7084 						  new_ctx->name);
7085 			}
7086 			spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
7087 			snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
7088 			rc = spdk_bdev_nvme_create(&new_ctx->trid, new_ctx->name, NULL, 0,
7089 						   discovery_attach_controller_done, new_ctx,
7090 						   &new_ctx->drv_opts, &ctx->bdev_opts);
7091 			if (rc == 0) {
7092 				TAILQ_INSERT_TAIL(&ctx->nvm_entry_ctxs, new_ctx, tailq);
7093 				ctx->attach_in_progress++;
7094 			} else {
7095 				DISCOVERY_ERRLOG(ctx, "spdk_bdev_nvme_create failed (%s)\n", spdk_strerror(-rc));
7096 			}
7097 		}
7098 	}
7099 
7100 	if (ctx->attach_in_progress == 0) {
7101 		discovery_remove_controllers(ctx);
7102 	}
7103 }
7104 
7105 static void
7106 get_discovery_log_page(struct discovery_ctx *ctx)
7107 {
7108 	int rc;
7109 
7110 	assert(ctx->in_progress == false);
7111 	ctx->in_progress = true;
7112 	rc = spdk_nvme_ctrlr_get_discovery_log_page(ctx->ctrlr, discovery_log_page_cb, ctx);
7113 	if (rc != 0) {
7114 		DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
7115 	}
7116 	DISCOVERY_INFOLOG(ctx, "sent discovery log page command\n");
7117 }
7118 
7119 static void
7120 discovery_aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
7121 {
7122 	struct discovery_ctx *ctx = arg;
7123 	uint32_t log_page_id = (cpl->cdw0 & 0xFF0000) >> 16;
7124 
7125 	if (spdk_nvme_cpl_is_error(cpl)) {
7126 		DISCOVERY_ERRLOG(ctx, "aer failed\n");
7127 		return;
7128 	}
7129 
7130 	if (log_page_id != SPDK_NVME_LOG_DISCOVERY) {
7131 		DISCOVERY_ERRLOG(ctx, "unexpected log page 0x%x\n", log_page_id);
7132 		return;
7133 	}
7134 
7135 	DISCOVERY_INFOLOG(ctx, "got aer\n");
7136 	if (ctx->in_progress) {
7137 		ctx->pending = true;
7138 		return;
7139 	}
7140 
7141 	get_discovery_log_page(ctx);
7142 }
7143 
7144 static void
7145 discovery_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
7146 		    struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
7147 {
7148 	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
7149 	struct discovery_ctx *ctx;
7150 
7151 	ctx = SPDK_CONTAINEROF(user_opts, struct discovery_ctx, drv_opts);
7152 
7153 	DISCOVERY_INFOLOG(ctx, "discovery ctrlr attached\n");
7154 	ctx->probe_ctx = NULL;
7155 	ctx->ctrlr = ctrlr;
7156 
7157 	if (ctx->rc != 0) {
7158 		DISCOVERY_ERRLOG(ctx, "encountered error while attaching discovery ctrlr: %d\n",
7159 				 ctx->rc);
7160 		return;
7161 	}
7162 
7163 	spdk_nvme_ctrlr_register_aer_callback(ctx->ctrlr, discovery_aer_cb, ctx);
7164 }
7165 
7166 static int
7167 discovery_poller(void *arg)
7168 {
7169 	struct discovery_ctx *ctx = arg;
7170 	struct spdk_nvme_transport_id *trid;
7171 	int rc;
7172 
7173 	if (ctx->detach_ctx) {
7174 		rc = spdk_nvme_detach_poll_async(ctx->detach_ctx);
7175 		if (rc != -EAGAIN) {
7176 			ctx->detach_ctx = NULL;
7177 			ctx->ctrlr = NULL;
7178 		}
7179 	} else if (ctx->stop) {
7180 		if (ctx->ctrlr != NULL) {
7181 			rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
7182 			if (rc == 0) {
7183 				return SPDK_POLLER_BUSY;
7184 			}
7185 			DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
7186 		}
7187 		spdk_poller_unregister(&ctx->poller);
7188 		TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
7189 		assert(ctx->start_cb_fn == NULL);
7190 		if (ctx->stop_cb_fn != NULL) {
7191 			ctx->stop_cb_fn(ctx->cb_ctx);
7192 		}
7193 		free_discovery_ctx(ctx);
7194 	} else if (ctx->probe_ctx == NULL && ctx->ctrlr == NULL) {
7195 		if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7196 			DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
7197 			assert(ctx->initializing);
7198 			spdk_poller_unregister(&ctx->poller);
7199 			TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
7200 			complete_discovery_start(ctx, -ETIMEDOUT);
7201 			stop_discovery(ctx, NULL, NULL);
7202 			free_discovery_ctx(ctx);
7203 			return SPDK_POLLER_BUSY;
7204 		}
7205 
7206 		assert(ctx->entry_ctx_in_use == NULL);
7207 		ctx->entry_ctx_in_use = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
7208 		TAILQ_REMOVE(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7209 		trid = &ctx->entry_ctx_in_use->trid;
7210 
7211 		/* All controllers must be configured explicitely either for multipath or failover.
7212 		 * While discovery use multipath mode, we need to set this in bdev options as well.
7213 		 */
7214 		ctx->bdev_opts.multipath = true;
7215 
7216 		ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, discovery_attach_cb);
7217 		if (ctx->probe_ctx) {
7218 			spdk_poller_unregister(&ctx->poller);
7219 			ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000);
7220 		} else {
7221 			DISCOVERY_ERRLOG(ctx, "could not start discovery connect\n");
7222 			TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7223 			ctx->entry_ctx_in_use = NULL;
7224 		}
7225 	} else if (ctx->probe_ctx) {
7226 		if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7227 			DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
7228 			complete_discovery_start(ctx, -ETIMEDOUT);
7229 			return SPDK_POLLER_BUSY;
7230 		}
7231 
7232 		rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
7233 		if (rc != -EAGAIN) {
7234 			if (ctx->rc != 0) {
7235 				assert(ctx->initializing);
7236 				stop_discovery(ctx, NULL, ctx->cb_ctx);
7237 			} else {
7238 				assert(rc == 0);
7239 				DISCOVERY_INFOLOG(ctx, "discovery ctrlr connected\n");
7240 				ctx->rc = rc;
7241 				get_discovery_log_page(ctx);
7242 			}
7243 		}
7244 	} else {
7245 		if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
7246 			DISCOVERY_ERRLOG(ctx, "timed out while attaching NVM ctrlrs\n");
7247 			complete_discovery_start(ctx, -ETIMEDOUT);
7248 			/* We need to wait until all NVM ctrlrs are attached before we stop the
7249 			 * discovery service to make sure we don't detach a ctrlr that is still
7250 			 * being attached.
7251 			 */
7252 			if (ctx->attach_in_progress == 0) {
7253 				stop_discovery(ctx, NULL, ctx->cb_ctx);
7254 				return SPDK_POLLER_BUSY;
7255 			}
7256 		}
7257 
7258 		rc = spdk_nvme_ctrlr_process_admin_completions(ctx->ctrlr);
7259 		if (rc < 0) {
7260 			spdk_poller_unregister(&ctx->poller);
7261 			ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
7262 			TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
7263 			ctx->entry_ctx_in_use = NULL;
7264 
7265 			rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
7266 			if (rc != 0) {
7267 				DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
7268 				ctx->ctrlr = NULL;
7269 			}
7270 		}
7271 	}
7272 
7273 	return SPDK_POLLER_BUSY;
7274 }
7275 
7276 static void
7277 start_discovery_poller(void *arg)
7278 {
7279 	struct discovery_ctx *ctx = arg;
7280 
7281 	TAILQ_INSERT_TAIL(&g_discovery_ctxs, ctx, tailq);
7282 	ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
7283 }
7284 
7285 int
7286 bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid,
7287 			  const char *base_name,
7288 			  struct spdk_nvme_ctrlr_opts *drv_opts,
7289 			  struct spdk_bdev_nvme_ctrlr_opts *bdev_opts,
7290 			  uint64_t attach_timeout,
7291 			  bool from_mdns,
7292 			  spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx)
7293 {
7294 	struct discovery_ctx *ctx;
7295 	struct discovery_entry_ctx *discovery_entry_ctx;
7296 
7297 	snprintf(trid->subnqn, sizeof(trid->subnqn), "%s", SPDK_NVMF_DISCOVERY_NQN);
7298 	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7299 		if (strcmp(ctx->name, base_name) == 0) {
7300 			return -EEXIST;
7301 		}
7302 
7303 		if (ctx->entry_ctx_in_use != NULL) {
7304 			if (!spdk_nvme_transport_id_compare(trid, &ctx->entry_ctx_in_use->trid)) {
7305 				return -EEXIST;
7306 			}
7307 		}
7308 
7309 		TAILQ_FOREACH(discovery_entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
7310 			if (!spdk_nvme_transport_id_compare(trid, &discovery_entry_ctx->trid)) {
7311 				return -EEXIST;
7312 			}
7313 		}
7314 	}
7315 
7316 	ctx = calloc(1, sizeof(*ctx));
7317 	if (ctx == NULL) {
7318 		return -ENOMEM;
7319 	}
7320 
7321 	ctx->name = strdup(base_name);
7322 	if (ctx->name == NULL) {
7323 		free_discovery_ctx(ctx);
7324 		return -ENOMEM;
7325 	}
7326 	memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
7327 	memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
7328 	ctx->from_mdns_discovery_service = from_mdns;
7329 	ctx->bdev_opts.from_discovery_service = true;
7330 	ctx->calling_thread = spdk_get_thread();
7331 	ctx->start_cb_fn = cb_fn;
7332 	ctx->cb_ctx = cb_ctx;
7333 	ctx->initializing = true;
7334 	if (ctx->start_cb_fn) {
7335 		/* We can use this when dumping json to denote if this RPC parameter
7336 		 * was specified or not.
7337 		 */
7338 		ctx->wait_for_attach = true;
7339 	}
7340 	if (attach_timeout != 0) {
7341 		ctx->timeout_ticks = spdk_get_ticks() + attach_timeout *
7342 				     spdk_get_ticks_hz() / 1000ull;
7343 	}
7344 	TAILQ_INIT(&ctx->nvm_entry_ctxs);
7345 	TAILQ_INIT(&ctx->discovery_entry_ctxs);
7346 	memcpy(&ctx->trid, trid, sizeof(*trid));
7347 	/* Even if user did not specify hostnqn, we can still strdup("\0"); */
7348 	ctx->hostnqn = strdup(ctx->drv_opts.hostnqn);
7349 	if (ctx->hostnqn == NULL) {
7350 		free_discovery_ctx(ctx);
7351 		return -ENOMEM;
7352 	}
7353 	discovery_entry_ctx = create_discovery_entry_ctx(ctx, trid);
7354 	if (discovery_entry_ctx == NULL) {
7355 		DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
7356 		free_discovery_ctx(ctx);
7357 		return -ENOMEM;
7358 	}
7359 
7360 	TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, discovery_entry_ctx, tailq);
7361 	spdk_thread_send_msg(g_bdev_nvme_init_thread, start_discovery_poller, ctx);
7362 	return 0;
7363 }
7364 
7365 int
7366 bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
7367 {
7368 	struct discovery_ctx *ctx;
7369 
7370 	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7371 		if (strcmp(name, ctx->name) == 0) {
7372 			if (ctx->stop) {
7373 				return -EALREADY;
7374 			}
7375 			/* If we're still starting the discovery service and ->rc is non-zero, we're
7376 			 * going to stop it as soon as we can
7377 			 */
7378 			if (ctx->initializing && ctx->rc != 0) {
7379 				return -EALREADY;
7380 			}
7381 			stop_discovery(ctx, cb_fn, cb_ctx);
7382 			return 0;
7383 		}
7384 	}
7385 
7386 	return -ENOENT;
7387 }
7388 
7389 static int
7390 bdev_nvme_library_init(void)
7391 {
7392 	g_bdev_nvme_init_thread = spdk_get_thread();
7393 
7394 	spdk_io_device_register(&g_nvme_bdev_ctrlrs, bdev_nvme_create_poll_group_cb,
7395 				bdev_nvme_destroy_poll_group_cb,
7396 				sizeof(struct nvme_poll_group),  "nvme_poll_groups");
7397 
7398 	return 0;
7399 }
7400 
7401 static void
7402 bdev_nvme_fini_destruct_ctrlrs(void)
7403 {
7404 	struct nvme_bdev_ctrlr *nbdev_ctrlr;
7405 	struct nvme_ctrlr *nvme_ctrlr;
7406 
7407 	pthread_mutex_lock(&g_bdev_nvme_mutex);
7408 	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
7409 		TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
7410 			pthread_mutex_lock(&nvme_ctrlr->mutex);
7411 			if (nvme_ctrlr->destruct) {
7412 				/* This controller's destruction was already started
7413 				 * before the application started shutting down
7414 				 */
7415 				pthread_mutex_unlock(&nvme_ctrlr->mutex);
7416 				continue;
7417 			}
7418 			nvme_ctrlr->destruct = true;
7419 			pthread_mutex_unlock(&nvme_ctrlr->mutex);
7420 
7421 			spdk_thread_send_msg(nvme_ctrlr->thread, _nvme_ctrlr_destruct,
7422 					     nvme_ctrlr);
7423 		}
7424 	}
7425 
7426 	g_bdev_nvme_module_finish = true;
7427 	if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
7428 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
7429 		spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
7430 		spdk_bdev_module_fini_done();
7431 		return;
7432 	}
7433 
7434 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
7435 }
7436 
7437 static void
7438 check_discovery_fini(void *arg)
7439 {
7440 	if (TAILQ_EMPTY(&g_discovery_ctxs)) {
7441 		bdev_nvme_fini_destruct_ctrlrs();
7442 	}
7443 }
7444 
7445 static void
7446 bdev_nvme_library_fini(void)
7447 {
7448 	struct nvme_probe_skip_entry *entry, *entry_tmp;
7449 	struct discovery_ctx *ctx;
7450 
7451 	spdk_poller_unregister(&g_hotplug_poller);
7452 	free(g_hotplug_probe_ctx);
7453 	g_hotplug_probe_ctx = NULL;
7454 
7455 	TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, entry_tmp) {
7456 		TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
7457 		free(entry);
7458 	}
7459 
7460 	assert(spdk_get_thread() == g_bdev_nvme_init_thread);
7461 	if (TAILQ_EMPTY(&g_discovery_ctxs)) {
7462 		bdev_nvme_fini_destruct_ctrlrs();
7463 	} else {
7464 		TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
7465 			stop_discovery(ctx, check_discovery_fini, NULL);
7466 		}
7467 	}
7468 }
7469 
7470 static void
7471 bdev_nvme_verify_pi_error(struct nvme_bdev_io *bio)
7472 {
7473 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7474 	struct spdk_bdev *bdev = bdev_io->bdev;
7475 	struct spdk_dif_ctx dif_ctx;
7476 	struct spdk_dif_error err_blk = {};
7477 	int rc;
7478 	struct spdk_dif_ctx_init_ext_opts dif_opts;
7479 
7480 	dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
7481 	dif_opts.dif_pi_format = bdev->dif_pi_format;
7482 	rc = spdk_dif_ctx_init(&dif_ctx,
7483 			       bdev->blocklen, bdev->md_len, bdev->md_interleave,
7484 			       bdev->dif_is_head_of_md, bdev->dif_type,
7485 			       bdev_io->u.bdev.dif_check_flags,
7486 			       bdev_io->u.bdev.offset_blocks, 0, 0, 0, 0, &dif_opts);
7487 	if (rc != 0) {
7488 		SPDK_ERRLOG("Initialization of DIF context failed\n");
7489 		return;
7490 	}
7491 
7492 	if (bdev->md_interleave) {
7493 		rc = spdk_dif_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
7494 				     bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
7495 	} else {
7496 		struct iovec md_iov = {
7497 			.iov_base	= bdev_io->u.bdev.md_buf,
7498 			.iov_len	= bdev_io->u.bdev.num_blocks * bdev->md_len,
7499 		};
7500 
7501 		rc = spdk_dix_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
7502 				     &md_iov, bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
7503 	}
7504 
7505 	if (rc != 0) {
7506 		SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
7507 			    err_blk.err_type, err_blk.err_offset);
7508 	} else {
7509 		SPDK_ERRLOG("Hardware reported PI error but SPDK could not find any.\n");
7510 	}
7511 }
7512 
7513 static void
7514 bdev_nvme_no_pi_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7515 {
7516 	struct nvme_bdev_io *bio = ref;
7517 
7518 	if (spdk_nvme_cpl_is_success(cpl)) {
7519 		/* Run PI verification for read data buffer. */
7520 		bdev_nvme_verify_pi_error(bio);
7521 	}
7522 
7523 	/* Return original completion status */
7524 	bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
7525 }
7526 
7527 static void
7528 bdev_nvme_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7529 {
7530 	struct nvme_bdev_io *bio = ref;
7531 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7532 	int ret;
7533 
7534 	if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
7535 		SPDK_ERRLOG("readv completed with PI error (sct=%d, sc=%d)\n",
7536 			    cpl->status.sct, cpl->status.sc);
7537 
7538 		/* Save completion status to use after verifying PI error. */
7539 		bio->cpl = *cpl;
7540 
7541 		if (spdk_likely(nvme_io_path_is_available(bio->io_path))) {
7542 			/* Read without PI checking to verify PI error. */
7543 			ret = bdev_nvme_no_pi_readv(bio,
7544 						    bdev_io->u.bdev.iovs,
7545 						    bdev_io->u.bdev.iovcnt,
7546 						    bdev_io->u.bdev.md_buf,
7547 						    bdev_io->u.bdev.num_blocks,
7548 						    bdev_io->u.bdev.offset_blocks);
7549 			if (ret == 0) {
7550 				return;
7551 			}
7552 		}
7553 	}
7554 
7555 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7556 }
7557 
7558 static void
7559 bdev_nvme_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7560 {
7561 	struct nvme_bdev_io *bio = ref;
7562 
7563 	if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
7564 		SPDK_ERRLOG("writev completed with PI error (sct=%d, sc=%d)\n",
7565 			    cpl->status.sct, cpl->status.sc);
7566 		/* Run PI verification for write data buffer if PI error is detected. */
7567 		bdev_nvme_verify_pi_error(bio);
7568 	}
7569 
7570 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7571 }
7572 
7573 static void
7574 bdev_nvme_zone_appendv_done(void *ref, const struct spdk_nvme_cpl *cpl)
7575 {
7576 	struct nvme_bdev_io *bio = ref;
7577 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7578 
7579 	/* spdk_bdev_io_get_append_location() requires that the ALBA is stored in offset_blocks.
7580 	 * Additionally, offset_blocks has to be set before calling bdev_nvme_verify_pi_error().
7581 	 */
7582 	bdev_io->u.bdev.offset_blocks = *(uint64_t *)&cpl->cdw0;
7583 
7584 	if (spdk_nvme_cpl_is_pi_error(cpl)) {
7585 		SPDK_ERRLOG("zone append completed with PI error (sct=%d, sc=%d)\n",
7586 			    cpl->status.sct, cpl->status.sc);
7587 		/* Run PI verification for zone append data buffer if PI error is detected. */
7588 		bdev_nvme_verify_pi_error(bio);
7589 	}
7590 
7591 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7592 }
7593 
7594 static void
7595 bdev_nvme_comparev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7596 {
7597 	struct nvme_bdev_io *bio = ref;
7598 
7599 	if (spdk_nvme_cpl_is_pi_error(cpl)) {
7600 		SPDK_ERRLOG("comparev completed with PI error (sct=%d, sc=%d)\n",
7601 			    cpl->status.sct, cpl->status.sc);
7602 		/* Run PI verification for compare data buffer if PI error is detected. */
7603 		bdev_nvme_verify_pi_error(bio);
7604 	}
7605 
7606 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7607 }
7608 
7609 static void
7610 bdev_nvme_comparev_and_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
7611 {
7612 	struct nvme_bdev_io *bio = ref;
7613 
7614 	/* Compare operation completion */
7615 	if (!bio->first_fused_completed) {
7616 		/* Save compare result for write callback */
7617 		bio->cpl = *cpl;
7618 		bio->first_fused_completed = true;
7619 		return;
7620 	}
7621 
7622 	/* Write operation completion */
7623 	if (spdk_nvme_cpl_is_error(&bio->cpl)) {
7624 		/* If bio->cpl is already an error, it means the compare operation failed.  In that case,
7625 		 * complete the IO with the compare operation's status.
7626 		 */
7627 		if (!spdk_nvme_cpl_is_error(cpl)) {
7628 			SPDK_ERRLOG("Unexpected write success after compare failure.\n");
7629 		}
7630 
7631 		bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
7632 	} else {
7633 		bdev_nvme_io_complete_nvme_status(bio, cpl);
7634 	}
7635 }
7636 
7637 static void
7638 bdev_nvme_queued_done(void *ref, const struct spdk_nvme_cpl *cpl)
7639 {
7640 	struct nvme_bdev_io *bio = ref;
7641 
7642 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7643 }
7644 
7645 static int
7646 fill_zone_from_report(struct spdk_bdev_zone_info *info, struct spdk_nvme_zns_zone_desc *desc)
7647 {
7648 	switch (desc->zt) {
7649 	case SPDK_NVME_ZONE_TYPE_SEQWR:
7650 		info->type = SPDK_BDEV_ZONE_TYPE_SEQWR;
7651 		break;
7652 	default:
7653 		SPDK_ERRLOG("Invalid zone type: %#x in zone report\n", desc->zt);
7654 		return -EIO;
7655 	}
7656 
7657 	switch (desc->zs) {
7658 	case SPDK_NVME_ZONE_STATE_EMPTY:
7659 		info->state = SPDK_BDEV_ZONE_STATE_EMPTY;
7660 		break;
7661 	case SPDK_NVME_ZONE_STATE_IOPEN:
7662 		info->state = SPDK_BDEV_ZONE_STATE_IMP_OPEN;
7663 		break;
7664 	case SPDK_NVME_ZONE_STATE_EOPEN:
7665 		info->state = SPDK_BDEV_ZONE_STATE_EXP_OPEN;
7666 		break;
7667 	case SPDK_NVME_ZONE_STATE_CLOSED:
7668 		info->state = SPDK_BDEV_ZONE_STATE_CLOSED;
7669 		break;
7670 	case SPDK_NVME_ZONE_STATE_RONLY:
7671 		info->state = SPDK_BDEV_ZONE_STATE_READ_ONLY;
7672 		break;
7673 	case SPDK_NVME_ZONE_STATE_FULL:
7674 		info->state = SPDK_BDEV_ZONE_STATE_FULL;
7675 		break;
7676 	case SPDK_NVME_ZONE_STATE_OFFLINE:
7677 		info->state = SPDK_BDEV_ZONE_STATE_OFFLINE;
7678 		break;
7679 	default:
7680 		SPDK_ERRLOG("Invalid zone state: %#x in zone report\n", desc->zs);
7681 		return -EIO;
7682 	}
7683 
7684 	info->zone_id = desc->zslba;
7685 	info->write_pointer = desc->wp;
7686 	info->capacity = desc->zcap;
7687 
7688 	return 0;
7689 }
7690 
7691 static void
7692 bdev_nvme_get_zone_info_done(void *ref, const struct spdk_nvme_cpl *cpl)
7693 {
7694 	struct nvme_bdev_io *bio = ref;
7695 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7696 	uint64_t zone_id = bdev_io->u.zone_mgmt.zone_id;
7697 	uint32_t zones_to_copy = bdev_io->u.zone_mgmt.num_zones;
7698 	struct spdk_bdev_zone_info *info = bdev_io->u.zone_mgmt.buf;
7699 	uint64_t max_zones_per_buf, i;
7700 	uint32_t zone_report_bufsize;
7701 	struct spdk_nvme_ns *ns;
7702 	struct spdk_nvme_qpair *qpair;
7703 	int ret;
7704 
7705 	if (spdk_nvme_cpl_is_error(cpl)) {
7706 		goto out_complete_io_nvme_cpl;
7707 	}
7708 
7709 	if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
7710 		ret = -ENXIO;
7711 		goto out_complete_io_ret;
7712 	}
7713 
7714 	ns = bio->io_path->nvme_ns->ns;
7715 	qpair = bio->io_path->qpair->qpair;
7716 
7717 	zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
7718 	max_zones_per_buf = (zone_report_bufsize - sizeof(*bio->zone_report_buf)) /
7719 			    sizeof(bio->zone_report_buf->descs[0]);
7720 
7721 	if (bio->zone_report_buf->nr_zones > max_zones_per_buf) {
7722 		ret = -EINVAL;
7723 		goto out_complete_io_ret;
7724 	}
7725 
7726 	if (!bio->zone_report_buf->nr_zones) {
7727 		ret = -EINVAL;
7728 		goto out_complete_io_ret;
7729 	}
7730 
7731 	for (i = 0; i < bio->zone_report_buf->nr_zones && bio->handled_zones < zones_to_copy; i++) {
7732 		ret = fill_zone_from_report(&info[bio->handled_zones],
7733 					    &bio->zone_report_buf->descs[i]);
7734 		if (ret) {
7735 			goto out_complete_io_ret;
7736 		}
7737 		bio->handled_zones++;
7738 	}
7739 
7740 	if (bio->handled_zones < zones_to_copy) {
7741 		uint64_t zone_size_lba = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
7742 		uint64_t slba = zone_id + (zone_size_lba * bio->handled_zones);
7743 
7744 		memset(bio->zone_report_buf, 0, zone_report_bufsize);
7745 		ret = spdk_nvme_zns_report_zones(ns, qpair,
7746 						 bio->zone_report_buf, zone_report_bufsize,
7747 						 slba, SPDK_NVME_ZRA_LIST_ALL, true,
7748 						 bdev_nvme_get_zone_info_done, bio);
7749 		if (!ret) {
7750 			return;
7751 		} else {
7752 			goto out_complete_io_ret;
7753 		}
7754 	}
7755 
7756 out_complete_io_nvme_cpl:
7757 	free(bio->zone_report_buf);
7758 	bio->zone_report_buf = NULL;
7759 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7760 	return;
7761 
7762 out_complete_io_ret:
7763 	free(bio->zone_report_buf);
7764 	bio->zone_report_buf = NULL;
7765 	bdev_nvme_io_complete(bio, ret);
7766 }
7767 
7768 static void
7769 bdev_nvme_zone_management_done(void *ref, const struct spdk_nvme_cpl *cpl)
7770 {
7771 	struct nvme_bdev_io *bio = ref;
7772 
7773 	bdev_nvme_io_complete_nvme_status(bio, cpl);
7774 }
7775 
7776 static void
7777 bdev_nvme_admin_passthru_complete_nvme_status(void *ctx)
7778 {
7779 	struct nvme_bdev_io *bio = ctx;
7780 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7781 	const struct spdk_nvme_cpl *cpl = &bio->cpl;
7782 
7783 	assert(bdev_nvme_io_type_is_admin(bdev_io->type));
7784 
7785 	__bdev_nvme_io_complete(bdev_io, 0, cpl);
7786 }
7787 
7788 static void
7789 bdev_nvme_abort_complete(void *ctx)
7790 {
7791 	struct nvme_bdev_io *bio = ctx;
7792 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7793 
7794 	if (spdk_nvme_cpl_is_abort_success(&bio->cpl)) {
7795 		__bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS, NULL);
7796 	} else {
7797 		__bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED, NULL);
7798 	}
7799 }
7800 
7801 static void
7802 bdev_nvme_abort_done(void *ref, const struct spdk_nvme_cpl *cpl)
7803 {
7804 	struct nvme_bdev_io *bio = ref;
7805 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7806 
7807 	bio->cpl = *cpl;
7808 	spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), bdev_nvme_abort_complete, bio);
7809 }
7810 
7811 static void
7812 bdev_nvme_admin_passthru_done(void *ref, const struct spdk_nvme_cpl *cpl)
7813 {
7814 	struct nvme_bdev_io *bio = ref;
7815 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
7816 
7817 	bio->cpl = *cpl;
7818 	spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7819 			     bdev_nvme_admin_passthru_complete_nvme_status, bio);
7820 }
7821 
7822 static void
7823 bdev_nvme_queued_reset_sgl(void *ref, uint32_t sgl_offset)
7824 {
7825 	struct nvme_bdev_io *bio = ref;
7826 	struct iovec *iov;
7827 
7828 	bio->iov_offset = sgl_offset;
7829 	for (bio->iovpos = 0; bio->iovpos < bio->iovcnt; bio->iovpos++) {
7830 		iov = &bio->iovs[bio->iovpos];
7831 		if (bio->iov_offset < iov->iov_len) {
7832 			break;
7833 		}
7834 
7835 		bio->iov_offset -= iov->iov_len;
7836 	}
7837 }
7838 
7839 static int
7840 bdev_nvme_queued_next_sge(void *ref, void **address, uint32_t *length)
7841 {
7842 	struct nvme_bdev_io *bio = ref;
7843 	struct iovec *iov;
7844 
7845 	assert(bio->iovpos < bio->iovcnt);
7846 
7847 	iov = &bio->iovs[bio->iovpos];
7848 
7849 	*address = iov->iov_base;
7850 	*length = iov->iov_len;
7851 
7852 	if (bio->iov_offset) {
7853 		assert(bio->iov_offset <= iov->iov_len);
7854 		*address += bio->iov_offset;
7855 		*length -= bio->iov_offset;
7856 	}
7857 
7858 	bio->iov_offset += *length;
7859 	if (bio->iov_offset == iov->iov_len) {
7860 		bio->iovpos++;
7861 		bio->iov_offset = 0;
7862 	}
7863 
7864 	return 0;
7865 }
7866 
7867 static void
7868 bdev_nvme_queued_reset_fused_sgl(void *ref, uint32_t sgl_offset)
7869 {
7870 	struct nvme_bdev_io *bio = ref;
7871 	struct iovec *iov;
7872 
7873 	bio->fused_iov_offset = sgl_offset;
7874 	for (bio->fused_iovpos = 0; bio->fused_iovpos < bio->fused_iovcnt; bio->fused_iovpos++) {
7875 		iov = &bio->fused_iovs[bio->fused_iovpos];
7876 		if (bio->fused_iov_offset < iov->iov_len) {
7877 			break;
7878 		}
7879 
7880 		bio->fused_iov_offset -= iov->iov_len;
7881 	}
7882 }
7883 
7884 static int
7885 bdev_nvme_queued_next_fused_sge(void *ref, void **address, uint32_t *length)
7886 {
7887 	struct nvme_bdev_io *bio = ref;
7888 	struct iovec *iov;
7889 
7890 	assert(bio->fused_iovpos < bio->fused_iovcnt);
7891 
7892 	iov = &bio->fused_iovs[bio->fused_iovpos];
7893 
7894 	*address = iov->iov_base;
7895 	*length = iov->iov_len;
7896 
7897 	if (bio->fused_iov_offset) {
7898 		assert(bio->fused_iov_offset <= iov->iov_len);
7899 		*address += bio->fused_iov_offset;
7900 		*length -= bio->fused_iov_offset;
7901 	}
7902 
7903 	bio->fused_iov_offset += *length;
7904 	if (bio->fused_iov_offset == iov->iov_len) {
7905 		bio->fused_iovpos++;
7906 		bio->fused_iov_offset = 0;
7907 	}
7908 
7909 	return 0;
7910 }
7911 
7912 static int
7913 bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7914 		      void *md, uint64_t lba_count, uint64_t lba)
7915 {
7916 	int rc;
7917 
7918 	SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 " without PI check\n",
7919 		      lba_count, lba);
7920 
7921 	bio->iovs = iov;
7922 	bio->iovcnt = iovcnt;
7923 	bio->iovpos = 0;
7924 	bio->iov_offset = 0;
7925 
7926 	rc = spdk_nvme_ns_cmd_readv_with_md(bio->io_path->nvme_ns->ns,
7927 					    bio->io_path->qpair->qpair,
7928 					    lba, lba_count,
7929 					    bdev_nvme_no_pi_readv_done, bio, 0,
7930 					    bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
7931 					    md, 0, 0);
7932 
7933 	if (rc != 0 && rc != -ENOMEM) {
7934 		SPDK_ERRLOG("no_pi_readv failed: rc = %d\n", rc);
7935 	}
7936 	return rc;
7937 }
7938 
7939 static int
7940 bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7941 		void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
7942 		struct spdk_memory_domain *domain, void *domain_ctx,
7943 		struct spdk_accel_sequence *seq)
7944 {
7945 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
7946 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
7947 	int rc;
7948 
7949 	SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 "\n",
7950 		      lba_count, lba);
7951 
7952 	bio->iovs = iov;
7953 	bio->iovcnt = iovcnt;
7954 	bio->iovpos = 0;
7955 	bio->iov_offset = 0;
7956 
7957 	if (domain != NULL || seq != NULL) {
7958 		bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
7959 		bio->ext_opts.memory_domain = domain;
7960 		bio->ext_opts.memory_domain_ctx = domain_ctx;
7961 		bio->ext_opts.io_flags = flags;
7962 		bio->ext_opts.metadata = md;
7963 		bio->ext_opts.accel_sequence = seq;
7964 
7965 		if (iovcnt == 1) {
7966 			rc = spdk_nvme_ns_cmd_read_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_readv_done,
7967 						       bio, &bio->ext_opts);
7968 		} else {
7969 			rc = spdk_nvme_ns_cmd_readv_ext(ns, qpair, lba, lba_count,
7970 							bdev_nvme_readv_done, bio,
7971 							bdev_nvme_queued_reset_sgl,
7972 							bdev_nvme_queued_next_sge,
7973 							&bio->ext_opts);
7974 		}
7975 	} else if (iovcnt == 1) {
7976 		rc = spdk_nvme_ns_cmd_read_with_md(ns, qpair, iov[0].iov_base,
7977 						   md, lba, lba_count, bdev_nvme_readv_done,
7978 						   bio, flags, 0, 0);
7979 	} else {
7980 		rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count,
7981 						    bdev_nvme_readv_done, bio, flags,
7982 						    bdev_nvme_queued_reset_sgl,
7983 						    bdev_nvme_queued_next_sge, md, 0, 0);
7984 	}
7985 
7986 	if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
7987 		SPDK_ERRLOG("readv failed: rc = %d\n", rc);
7988 	}
7989 	return rc;
7990 }
7991 
7992 static int
7993 bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
7994 		 void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
7995 		 struct spdk_memory_domain *domain, void *domain_ctx,
7996 		 struct spdk_accel_sequence *seq,
7997 		 union spdk_bdev_nvme_cdw12 cdw12, union spdk_bdev_nvme_cdw13 cdw13)
7998 {
7999 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8000 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8001 	int rc;
8002 
8003 	SPDK_DEBUGLOG(bdev_nvme, "write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8004 		      lba_count, lba);
8005 
8006 	bio->iovs = iov;
8007 	bio->iovcnt = iovcnt;
8008 	bio->iovpos = 0;
8009 	bio->iov_offset = 0;
8010 
8011 	if (domain != NULL || seq != NULL) {
8012 		bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
8013 		bio->ext_opts.memory_domain = domain;
8014 		bio->ext_opts.memory_domain_ctx = domain_ctx;
8015 		bio->ext_opts.io_flags = flags | SPDK_NVME_IO_FLAGS_DIRECTIVE(cdw12.write.dtype);
8016 		bio->ext_opts.cdw13 = cdw13.raw;
8017 		bio->ext_opts.metadata = md;
8018 		bio->ext_opts.accel_sequence = seq;
8019 
8020 		if (iovcnt == 1) {
8021 			rc = spdk_nvme_ns_cmd_write_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_writev_done,
8022 							bio, &bio->ext_opts);
8023 		} else {
8024 			rc = spdk_nvme_ns_cmd_writev_ext(ns, qpair, lba, lba_count,
8025 							 bdev_nvme_writev_done, bio,
8026 							 bdev_nvme_queued_reset_sgl,
8027 							 bdev_nvme_queued_next_sge,
8028 							 &bio->ext_opts);
8029 		}
8030 	} else if (iovcnt == 1) {
8031 		rc = spdk_nvme_ns_cmd_write_with_md(ns, qpair, iov[0].iov_base,
8032 						    md, lba, lba_count, bdev_nvme_writev_done,
8033 						    bio, flags, 0, 0);
8034 	} else {
8035 		rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
8036 						     bdev_nvme_writev_done, bio, flags,
8037 						     bdev_nvme_queued_reset_sgl,
8038 						     bdev_nvme_queued_next_sge, md, 0, 0);
8039 	}
8040 
8041 	if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
8042 		SPDK_ERRLOG("writev failed: rc = %d\n", rc);
8043 	}
8044 	return rc;
8045 }
8046 
8047 static int
8048 bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8049 		       void *md, uint64_t lba_count, uint64_t zslba,
8050 		       uint32_t flags)
8051 {
8052 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8053 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8054 	int rc;
8055 
8056 	SPDK_DEBUGLOG(bdev_nvme, "zone append %" PRIu64 " blocks to zone start lba %#" PRIx64 "\n",
8057 		      lba_count, zslba);
8058 
8059 	bio->iovs = iov;
8060 	bio->iovcnt = iovcnt;
8061 	bio->iovpos = 0;
8062 	bio->iov_offset = 0;
8063 
8064 	if (iovcnt == 1) {
8065 		rc = spdk_nvme_zns_zone_append_with_md(ns, qpair, iov[0].iov_base, md, zslba,
8066 						       lba_count,
8067 						       bdev_nvme_zone_appendv_done, bio,
8068 						       flags,
8069 						       0, 0);
8070 	} else {
8071 		rc = spdk_nvme_zns_zone_appendv_with_md(ns, qpair, zslba, lba_count,
8072 							bdev_nvme_zone_appendv_done, bio, flags,
8073 							bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
8074 							md, 0, 0);
8075 	}
8076 
8077 	if (rc != 0 && rc != -ENOMEM) {
8078 		SPDK_ERRLOG("zone append failed: rc = %d\n", rc);
8079 	}
8080 	return rc;
8081 }
8082 
8083 static int
8084 bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
8085 		   void *md, uint64_t lba_count, uint64_t lba,
8086 		   uint32_t flags)
8087 {
8088 	int rc;
8089 
8090 	SPDK_DEBUGLOG(bdev_nvme, "compare %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8091 		      lba_count, lba);
8092 
8093 	bio->iovs = iov;
8094 	bio->iovcnt = iovcnt;
8095 	bio->iovpos = 0;
8096 	bio->iov_offset = 0;
8097 
8098 	rc = spdk_nvme_ns_cmd_comparev_with_md(bio->io_path->nvme_ns->ns,
8099 					       bio->io_path->qpair->qpair,
8100 					       lba, lba_count,
8101 					       bdev_nvme_comparev_done, bio, flags,
8102 					       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
8103 					       md, 0, 0);
8104 
8105 	if (rc != 0 && rc != -ENOMEM) {
8106 		SPDK_ERRLOG("comparev failed: rc = %d\n", rc);
8107 	}
8108 	return rc;
8109 }
8110 
8111 static int
8112 bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt,
8113 			      struct iovec *write_iov, int write_iovcnt,
8114 			      void *md, uint64_t lba_count, uint64_t lba, uint32_t flags)
8115 {
8116 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8117 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8118 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
8119 	int rc;
8120 
8121 	SPDK_DEBUGLOG(bdev_nvme, "compare and write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
8122 		      lba_count, lba);
8123 
8124 	bio->iovs = cmp_iov;
8125 	bio->iovcnt = cmp_iovcnt;
8126 	bio->iovpos = 0;
8127 	bio->iov_offset = 0;
8128 	bio->fused_iovs = write_iov;
8129 	bio->fused_iovcnt = write_iovcnt;
8130 	bio->fused_iovpos = 0;
8131 	bio->fused_iov_offset = 0;
8132 
8133 	if (bdev_io->num_retries == 0) {
8134 		bio->first_fused_submitted = false;
8135 		bio->first_fused_completed = false;
8136 	}
8137 
8138 	if (!bio->first_fused_submitted) {
8139 		flags |= SPDK_NVME_IO_FLAGS_FUSE_FIRST;
8140 		memset(&bio->cpl, 0, sizeof(bio->cpl));
8141 
8142 		rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count,
8143 						       bdev_nvme_comparev_and_writev_done, bio, flags,
8144 						       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, md, 0, 0);
8145 		if (rc == 0) {
8146 			bio->first_fused_submitted = true;
8147 			flags &= ~SPDK_NVME_IO_FLAGS_FUSE_FIRST;
8148 		} else {
8149 			if (rc != -ENOMEM) {
8150 				SPDK_ERRLOG("compare failed: rc = %d\n", rc);
8151 			}
8152 			return rc;
8153 		}
8154 	}
8155 
8156 	flags |= SPDK_NVME_IO_FLAGS_FUSE_SECOND;
8157 
8158 	rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
8159 					     bdev_nvme_comparev_and_writev_done, bio, flags,
8160 					     bdev_nvme_queued_reset_fused_sgl, bdev_nvme_queued_next_fused_sge, md, 0, 0);
8161 	if (rc != 0 && rc != -ENOMEM) {
8162 		SPDK_ERRLOG("write failed: rc = %d\n", rc);
8163 		rc = 0;
8164 	}
8165 
8166 	return rc;
8167 }
8168 
8169 static int
8170 bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
8171 {
8172 	struct spdk_nvme_dsm_range dsm_ranges[SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES];
8173 	struct spdk_nvme_dsm_range *range;
8174 	uint64_t offset, remaining;
8175 	uint64_t num_ranges_u64;
8176 	uint16_t num_ranges;
8177 	int rc;
8178 
8179 	num_ranges_u64 = (num_blocks + SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS - 1) /
8180 			 SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8181 	if (num_ranges_u64 > SPDK_COUNTOF(dsm_ranges)) {
8182 		SPDK_ERRLOG("Unmap request for %" PRIu64 " blocks is too large\n", num_blocks);
8183 		return -EINVAL;
8184 	}
8185 	num_ranges = (uint16_t)num_ranges_u64;
8186 
8187 	offset = offset_blocks;
8188 	remaining = num_blocks;
8189 	range = &dsm_ranges[0];
8190 
8191 	/* Fill max-size ranges until the remaining blocks fit into one range */
8192 	while (remaining > SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS) {
8193 		range->attributes.raw = 0;
8194 		range->length = SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8195 		range->starting_lba = offset;
8196 
8197 		offset += SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8198 		remaining -= SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
8199 		range++;
8200 	}
8201 
8202 	/* Final range describes the remaining blocks */
8203 	range->attributes.raw = 0;
8204 	range->length = remaining;
8205 	range->starting_lba = offset;
8206 
8207 	rc = spdk_nvme_ns_cmd_dataset_management(bio->io_path->nvme_ns->ns,
8208 			bio->io_path->qpair->qpair,
8209 			SPDK_NVME_DSM_ATTR_DEALLOCATE,
8210 			dsm_ranges, num_ranges,
8211 			bdev_nvme_queued_done, bio);
8212 
8213 	return rc;
8214 }
8215 
8216 static int
8217 bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
8218 {
8219 	if (num_blocks > UINT16_MAX + 1) {
8220 		SPDK_ERRLOG("NVMe write zeroes is limited to 16-bit block count\n");
8221 		return -EINVAL;
8222 	}
8223 
8224 	return spdk_nvme_ns_cmd_write_zeroes(bio->io_path->nvme_ns->ns,
8225 					     bio->io_path->qpair->qpair,
8226 					     offset_blocks, num_blocks,
8227 					     bdev_nvme_queued_done, bio,
8228 					     0);
8229 }
8230 
8231 static int
8232 bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones,
8233 			struct spdk_bdev_zone_info *info)
8234 {
8235 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8236 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8237 	uint32_t zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
8238 	uint64_t zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
8239 	uint64_t total_zones = spdk_nvme_zns_ns_get_num_zones(ns);
8240 
8241 	if (zone_id % zone_size != 0) {
8242 		return -EINVAL;
8243 	}
8244 
8245 	if (num_zones > total_zones || !num_zones) {
8246 		return -EINVAL;
8247 	}
8248 
8249 	assert(!bio->zone_report_buf);
8250 	bio->zone_report_buf = calloc(1, zone_report_bufsize);
8251 	if (!bio->zone_report_buf) {
8252 		return -ENOMEM;
8253 	}
8254 
8255 	bio->handled_zones = 0;
8256 
8257 	return spdk_nvme_zns_report_zones(ns, qpair, bio->zone_report_buf, zone_report_bufsize,
8258 					  zone_id, SPDK_NVME_ZRA_LIST_ALL, true,
8259 					  bdev_nvme_get_zone_info_done, bio);
8260 }
8261 
8262 static int
8263 bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
8264 			  enum spdk_bdev_zone_action action)
8265 {
8266 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8267 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8268 
8269 	switch (action) {
8270 	case SPDK_BDEV_ZONE_CLOSE:
8271 		return spdk_nvme_zns_close_zone(ns, qpair, zone_id, false,
8272 						bdev_nvme_zone_management_done, bio);
8273 	case SPDK_BDEV_ZONE_FINISH:
8274 		return spdk_nvme_zns_finish_zone(ns, qpair, zone_id, false,
8275 						 bdev_nvme_zone_management_done, bio);
8276 	case SPDK_BDEV_ZONE_OPEN:
8277 		return spdk_nvme_zns_open_zone(ns, qpair, zone_id, false,
8278 					       bdev_nvme_zone_management_done, bio);
8279 	case SPDK_BDEV_ZONE_RESET:
8280 		return spdk_nvme_zns_reset_zone(ns, qpair, zone_id, false,
8281 						bdev_nvme_zone_management_done, bio);
8282 	case SPDK_BDEV_ZONE_OFFLINE:
8283 		return spdk_nvme_zns_offline_zone(ns, qpair, zone_id, false,
8284 						  bdev_nvme_zone_management_done, bio);
8285 	default:
8286 		return -EINVAL;
8287 	}
8288 }
8289 
8290 static void
8291 bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
8292 			 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes)
8293 {
8294 	struct nvme_io_path *io_path;
8295 	struct nvme_ctrlr *nvme_ctrlr;
8296 	uint32_t max_xfer_size;
8297 	int rc = -ENXIO;
8298 
8299 	/* Choose the first ctrlr which is not failed. */
8300 	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
8301 		nvme_ctrlr = io_path->qpair->ctrlr;
8302 
8303 		/* We should skip any unavailable nvme_ctrlr rather than checking
8304 		 * if the return value of spdk_nvme_ctrlr_cmd_admin_raw() is -ENXIO.
8305 		 */
8306 		if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
8307 			continue;
8308 		}
8309 
8310 		max_xfer_size = spdk_nvme_ctrlr_get_max_xfer_size(nvme_ctrlr->ctrlr);
8311 
8312 		if (nbytes > max_xfer_size) {
8313 			SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8314 			rc = -EINVAL;
8315 			goto err;
8316 		}
8317 
8318 		rc = spdk_nvme_ctrlr_cmd_admin_raw(nvme_ctrlr->ctrlr, cmd, buf, (uint32_t)nbytes,
8319 						   bdev_nvme_admin_passthru_done, bio);
8320 		if (rc == 0) {
8321 			return;
8322 		}
8323 	}
8324 
8325 err:
8326 	bdev_nvme_admin_complete(bio, rc);
8327 }
8328 
8329 static int
8330 bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
8331 		      void *buf, size_t nbytes)
8332 {
8333 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8334 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8335 	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8336 	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8337 
8338 	if (nbytes > max_xfer_size) {
8339 		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8340 		return -EINVAL;
8341 	}
8342 
8343 	/*
8344 	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
8345 	 * so fill it out automatically.
8346 	 */
8347 	cmd->nsid = spdk_nvme_ns_get_id(ns);
8348 
8349 	return spdk_nvme_ctrlr_cmd_io_raw(ctrlr, qpair, cmd, buf,
8350 					  (uint32_t)nbytes, bdev_nvme_queued_done, bio);
8351 }
8352 
8353 static int
8354 bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
8355 			 void *buf, size_t nbytes, void *md_buf, size_t md_len)
8356 {
8357 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8358 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8359 	size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
8360 	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8361 	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8362 
8363 	if (nbytes > max_xfer_size) {
8364 		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8365 		return -EINVAL;
8366 	}
8367 
8368 	if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
8369 		SPDK_ERRLOG("invalid meta data buffer size\n");
8370 		return -EINVAL;
8371 	}
8372 
8373 	/*
8374 	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
8375 	 * so fill it out automatically.
8376 	 */
8377 	cmd->nsid = spdk_nvme_ns_get_id(ns);
8378 
8379 	return spdk_nvme_ctrlr_cmd_io_raw_with_md(ctrlr, qpair, cmd, buf,
8380 			(uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio);
8381 }
8382 
8383 static int
8384 bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio,
8385 			  struct spdk_nvme_cmd *cmd, struct iovec *iov, int iovcnt,
8386 			  size_t nbytes, void *md_buf, size_t md_len)
8387 {
8388 	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
8389 	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
8390 	size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
8391 	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
8392 	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
8393 
8394 	bio->iovs = iov;
8395 	bio->iovcnt = iovcnt;
8396 	bio->iovpos = 0;
8397 	bio->iov_offset = 0;
8398 
8399 	if (nbytes > max_xfer_size) {
8400 		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
8401 		return -EINVAL;
8402 	}
8403 
8404 	if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
8405 		SPDK_ERRLOG("invalid meta data buffer size\n");
8406 		return -EINVAL;
8407 	}
8408 
8409 	/*
8410 	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands
8411 	 * require a nsid, so fill it out automatically.
8412 	 */
8413 	cmd->nsid = spdk_nvme_ns_get_id(ns);
8414 
8415 	return spdk_nvme_ctrlr_cmd_iov_raw_with_md(
8416 		       ctrlr, qpair, cmd, (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio,
8417 		       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge);
8418 }
8419 
8420 static void
8421 bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
8422 		struct nvme_bdev_io *bio_to_abort)
8423 {
8424 	struct nvme_io_path *io_path;
8425 	int rc = 0;
8426 
8427 	rc = bdev_nvme_abort_retry_io(nbdev_ch, bio_to_abort);
8428 	if (rc == 0) {
8429 		bdev_nvme_admin_complete(bio, 0);
8430 		return;
8431 	}
8432 
8433 	io_path = bio_to_abort->io_path;
8434 	if (io_path != NULL) {
8435 		rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
8436 						   io_path->qpair->qpair,
8437 						   bio_to_abort,
8438 						   bdev_nvme_abort_done, bio);
8439 	} else {
8440 		STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
8441 			rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
8442 							   NULL,
8443 							   bio_to_abort,
8444 							   bdev_nvme_abort_done, bio);
8445 
8446 			if (rc != -ENOENT) {
8447 				break;
8448 			}
8449 		}
8450 	}
8451 
8452 	if (rc != 0) {
8453 		/* If no command was found or there was any error, complete the abort
8454 		 * request with failure.
8455 		 */
8456 		bdev_nvme_admin_complete(bio, rc);
8457 	}
8458 }
8459 
8460 static int
8461 bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks, uint64_t src_offset_blocks,
8462 	       uint64_t num_blocks)
8463 {
8464 	struct spdk_nvme_scc_source_range range = {
8465 		.slba = src_offset_blocks,
8466 		.nlb = num_blocks - 1
8467 	};
8468 
8469 	return spdk_nvme_ns_cmd_copy(bio->io_path->nvme_ns->ns,
8470 				     bio->io_path->qpair->qpair,
8471 				     &range, 1, dst_offset_blocks,
8472 				     bdev_nvme_queued_done, bio);
8473 }
8474 
8475 static void
8476 bdev_nvme_opts_config_json(struct spdk_json_write_ctx *w)
8477 {
8478 	const char *action;
8479 	uint32_t i;
8480 
8481 	if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET) {
8482 		action = "reset";
8483 	} else if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT) {
8484 		action = "abort";
8485 	} else {
8486 		action = "none";
8487 	}
8488 
8489 	spdk_json_write_object_begin(w);
8490 
8491 	spdk_json_write_named_string(w, "method", "bdev_nvme_set_options");
8492 
8493 	spdk_json_write_named_object_begin(w, "params");
8494 	spdk_json_write_named_string(w, "action_on_timeout", action);
8495 	spdk_json_write_named_uint64(w, "timeout_us", g_opts.timeout_us);
8496 	spdk_json_write_named_uint64(w, "timeout_admin_us", g_opts.timeout_admin_us);
8497 	spdk_json_write_named_uint32(w, "keep_alive_timeout_ms", g_opts.keep_alive_timeout_ms);
8498 	spdk_json_write_named_uint32(w, "arbitration_burst", g_opts.arbitration_burst);
8499 	spdk_json_write_named_uint32(w, "low_priority_weight", g_opts.low_priority_weight);
8500 	spdk_json_write_named_uint32(w, "medium_priority_weight", g_opts.medium_priority_weight);
8501 	spdk_json_write_named_uint32(w, "high_priority_weight", g_opts.high_priority_weight);
8502 	spdk_json_write_named_uint64(w, "nvme_adminq_poll_period_us", g_opts.nvme_adminq_poll_period_us);
8503 	spdk_json_write_named_uint64(w, "nvme_ioq_poll_period_us", g_opts.nvme_ioq_poll_period_us);
8504 	spdk_json_write_named_uint32(w, "io_queue_requests", g_opts.io_queue_requests);
8505 	spdk_json_write_named_bool(w, "delay_cmd_submit", g_opts.delay_cmd_submit);
8506 	spdk_json_write_named_uint32(w, "transport_retry_count", g_opts.transport_retry_count);
8507 	spdk_json_write_named_int32(w, "bdev_retry_count", g_opts.bdev_retry_count);
8508 	spdk_json_write_named_uint8(w, "transport_ack_timeout", g_opts.transport_ack_timeout);
8509 	spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", g_opts.ctrlr_loss_timeout_sec);
8510 	spdk_json_write_named_uint32(w, "reconnect_delay_sec", g_opts.reconnect_delay_sec);
8511 	spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec", g_opts.fast_io_fail_timeout_sec);
8512 	spdk_json_write_named_bool(w, "disable_auto_failback", g_opts.disable_auto_failback);
8513 	spdk_json_write_named_bool(w, "generate_uuids", g_opts.generate_uuids);
8514 	spdk_json_write_named_uint8(w, "transport_tos", g_opts.transport_tos);
8515 	spdk_json_write_named_bool(w, "nvme_error_stat", g_opts.nvme_error_stat);
8516 	spdk_json_write_named_uint32(w, "rdma_srq_size", g_opts.rdma_srq_size);
8517 	spdk_json_write_named_bool(w, "io_path_stat", g_opts.io_path_stat);
8518 	spdk_json_write_named_bool(w, "allow_accel_sequence", g_opts.allow_accel_sequence);
8519 	spdk_json_write_named_uint32(w, "rdma_max_cq_size", g_opts.rdma_max_cq_size);
8520 	spdk_json_write_named_uint16(w, "rdma_cm_event_timeout_ms", g_opts.rdma_cm_event_timeout_ms);
8521 	spdk_json_write_named_array_begin(w, "dhchap_digests");
8522 	for (i = 0; i < 32; ++i) {
8523 		if (g_opts.dhchap_digests & SPDK_BIT(i)) {
8524 			spdk_json_write_string(w, spdk_nvme_dhchap_get_digest_name(i));
8525 		}
8526 	}
8527 	spdk_json_write_array_end(w);
8528 	spdk_json_write_named_array_begin(w, "dhchap_dhgroups");
8529 	for (i = 0; i < 32; ++i) {
8530 		if (g_opts.dhchap_dhgroups & SPDK_BIT(i)) {
8531 			spdk_json_write_string(w, spdk_nvme_dhchap_get_dhgroup_name(i));
8532 		}
8533 	}
8534 
8535 	spdk_json_write_array_end(w);
8536 	spdk_json_write_object_end(w);
8537 
8538 	spdk_json_write_object_end(w);
8539 }
8540 
8541 static void
8542 bdev_nvme_discovery_config_json(struct spdk_json_write_ctx *w, struct discovery_ctx *ctx)
8543 {
8544 	struct spdk_nvme_transport_id trid;
8545 
8546 	spdk_json_write_object_begin(w);
8547 
8548 	spdk_json_write_named_string(w, "method", "bdev_nvme_start_discovery");
8549 
8550 	spdk_json_write_named_object_begin(w, "params");
8551 	spdk_json_write_named_string(w, "name", ctx->name);
8552 	spdk_json_write_named_string(w, "hostnqn", ctx->hostnqn);
8553 
8554 	trid = ctx->trid;
8555 	memset(trid.subnqn, 0, sizeof(trid.subnqn));
8556 	nvme_bdev_dump_trid_json(&trid, w);
8557 
8558 	spdk_json_write_named_bool(w, "wait_for_attach", ctx->wait_for_attach);
8559 	spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", ctx->bdev_opts.ctrlr_loss_timeout_sec);
8560 	spdk_json_write_named_uint32(w, "reconnect_delay_sec", ctx->bdev_opts.reconnect_delay_sec);
8561 	spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
8562 				     ctx->bdev_opts.fast_io_fail_timeout_sec);
8563 	spdk_json_write_object_end(w);
8564 
8565 	spdk_json_write_object_end(w);
8566 }
8567 
8568 #ifdef SPDK_CONFIG_NVME_CUSE
8569 static void
8570 nvme_ctrlr_cuse_config_json(struct spdk_json_write_ctx *w,
8571 			    struct nvme_ctrlr *nvme_ctrlr)
8572 {
8573 	size_t cuse_name_size = 128;
8574 	char cuse_name[cuse_name_size];
8575 
8576 	if (spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr,
8577 					  cuse_name, &cuse_name_size) != 0) {
8578 		return;
8579 	}
8580 
8581 	spdk_json_write_object_begin(w);
8582 
8583 	spdk_json_write_named_string(w, "method", "bdev_nvme_cuse_register");
8584 
8585 	spdk_json_write_named_object_begin(w, "params");
8586 	spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
8587 	spdk_json_write_object_end(w);
8588 
8589 	spdk_json_write_object_end(w);
8590 }
8591 #endif
8592 
8593 static void
8594 nvme_ctrlr_config_json(struct spdk_json_write_ctx *w,
8595 		       struct nvme_ctrlr *nvme_ctrlr,
8596 		       struct nvme_path_id *path_id)
8597 {
8598 	struct spdk_nvme_transport_id	*trid;
8599 	const struct spdk_nvme_ctrlr_opts *opts;
8600 
8601 	if (nvme_ctrlr->opts.from_discovery_service) {
8602 		/* Do not emit an RPC for this - it will be implicitly
8603 		 * covered by a separate bdev_nvme_start_discovery or
8604 		 * bdev_nvme_start_mdns_discovery RPC.
8605 		 */
8606 		return;
8607 	}
8608 
8609 	trid = &path_id->trid;
8610 
8611 	spdk_json_write_object_begin(w);
8612 
8613 	spdk_json_write_named_string(w, "method", "bdev_nvme_attach_controller");
8614 
8615 	spdk_json_write_named_object_begin(w, "params");
8616 	spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
8617 	nvme_bdev_dump_trid_json(trid, w);
8618 	spdk_json_write_named_bool(w, "prchk_reftag",
8619 				   (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_REFTAG) != 0);
8620 	spdk_json_write_named_bool(w, "prchk_guard",
8621 				   (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_GUARD) != 0);
8622 	spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", nvme_ctrlr->opts.ctrlr_loss_timeout_sec);
8623 	spdk_json_write_named_uint32(w, "reconnect_delay_sec", nvme_ctrlr->opts.reconnect_delay_sec);
8624 	spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
8625 				     nvme_ctrlr->opts.fast_io_fail_timeout_sec);
8626 	if (nvme_ctrlr->psk != NULL) {
8627 		spdk_json_write_named_string(w, "psk", spdk_key_get_name(nvme_ctrlr->psk));
8628 	}
8629 	if (nvme_ctrlr->dhchap_key != NULL) {
8630 		spdk_json_write_named_string(w, "dhchap_key",
8631 					     spdk_key_get_name(nvme_ctrlr->dhchap_key));
8632 	}
8633 	if (nvme_ctrlr->dhchap_ctrlr_key != NULL) {
8634 		spdk_json_write_named_string(w, "dhchap_ctrlr_key",
8635 					     spdk_key_get_name(nvme_ctrlr->dhchap_ctrlr_key));
8636 	}
8637 	opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
8638 	spdk_json_write_named_string(w, "hostnqn", opts->hostnqn);
8639 	spdk_json_write_named_bool(w, "hdgst", opts->header_digest);
8640 	spdk_json_write_named_bool(w, "ddgst", opts->data_digest);
8641 	if (opts->src_addr[0] != '\0') {
8642 		spdk_json_write_named_string(w, "hostaddr", opts->src_addr);
8643 	}
8644 	if (opts->src_svcid[0] != '\0') {
8645 		spdk_json_write_named_string(w, "hostsvcid", opts->src_svcid);
8646 	}
8647 
8648 	if (nvme_ctrlr->opts.multipath) {
8649 		spdk_json_write_named_string(w, "multipath", "multipath");
8650 	}
8651 	spdk_json_write_object_end(w);
8652 
8653 	spdk_json_write_object_end(w);
8654 }
8655 
8656 static void
8657 bdev_nvme_hotplug_config_json(struct spdk_json_write_ctx *w)
8658 {
8659 	spdk_json_write_object_begin(w);
8660 	spdk_json_write_named_string(w, "method", "bdev_nvme_set_hotplug");
8661 
8662 	spdk_json_write_named_object_begin(w, "params");
8663 	spdk_json_write_named_uint64(w, "period_us", g_nvme_hotplug_poll_period_us);
8664 	spdk_json_write_named_bool(w, "enable", g_nvme_hotplug_enabled);
8665 	spdk_json_write_object_end(w);
8666 
8667 	spdk_json_write_object_end(w);
8668 }
8669 
8670 static int
8671 bdev_nvme_config_json(struct spdk_json_write_ctx *w)
8672 {
8673 	struct nvme_bdev_ctrlr	*nbdev_ctrlr;
8674 	struct nvme_ctrlr	*nvme_ctrlr;
8675 	struct discovery_ctx	*ctx;
8676 	struct nvme_path_id	*path_id;
8677 
8678 	bdev_nvme_opts_config_json(w);
8679 
8680 	pthread_mutex_lock(&g_bdev_nvme_mutex);
8681 
8682 	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
8683 		TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
8684 			path_id = nvme_ctrlr->active_path_id;
8685 			assert(path_id == TAILQ_FIRST(&nvme_ctrlr->trids));
8686 			nvme_ctrlr_config_json(w, nvme_ctrlr, path_id);
8687 
8688 			path_id = TAILQ_NEXT(path_id, link);
8689 			while (path_id != NULL) {
8690 				nvme_ctrlr_config_json(w, nvme_ctrlr, path_id);
8691 				path_id = TAILQ_NEXT(path_id, link);
8692 			}
8693 
8694 #ifdef SPDK_CONFIG_NVME_CUSE
8695 			nvme_ctrlr_cuse_config_json(w, nvme_ctrlr);
8696 #endif
8697 		}
8698 	}
8699 
8700 	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
8701 		if (!ctx->from_mdns_discovery_service) {
8702 			bdev_nvme_discovery_config_json(w, ctx);
8703 		}
8704 	}
8705 
8706 	bdev_nvme_mdns_discovery_config_json(w);
8707 
8708 	/* Dump as last parameter to give all NVMe bdevs chance to be constructed
8709 	 * before enabling hotplug poller.
8710 	 */
8711 	bdev_nvme_hotplug_config_json(w);
8712 
8713 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
8714 	return 0;
8715 }
8716 
8717 struct spdk_nvme_ctrlr *
8718 bdev_nvme_get_ctrlr(struct spdk_bdev *bdev)
8719 {
8720 	struct nvme_bdev *nbdev;
8721 	struct nvme_ns *nvme_ns;
8722 
8723 	if (!bdev || bdev->module != &nvme_if) {
8724 		return NULL;
8725 	}
8726 
8727 	nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
8728 	nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
8729 	assert(nvme_ns != NULL);
8730 
8731 	return nvme_ns->ctrlr->ctrlr;
8732 }
8733 
8734 static bool
8735 nvme_io_path_is_current(struct nvme_io_path *io_path)
8736 {
8737 	const struct nvme_bdev_channel *nbdev_ch;
8738 	bool current;
8739 
8740 	if (!nvme_io_path_is_available(io_path)) {
8741 		return false;
8742 	}
8743 
8744 	nbdev_ch = io_path->nbdev_ch;
8745 	if (nbdev_ch == NULL) {
8746 		current = false;
8747 	} else if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
8748 		struct nvme_io_path *optimized_io_path = NULL;
8749 
8750 		STAILQ_FOREACH(optimized_io_path, &nbdev_ch->io_path_list, stailq) {
8751 			if (optimized_io_path->nvme_ns->ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE) {
8752 				break;
8753 			}
8754 		}
8755 
8756 		/* A non-optimized path is only current if there are no optimized paths. */
8757 		current = (io_path->nvme_ns->ana_state == SPDK_NVME_ANA_OPTIMIZED_STATE) ||
8758 			  (optimized_io_path == NULL);
8759 	} else {
8760 		if (nbdev_ch->current_io_path) {
8761 			current = (io_path == nbdev_ch->current_io_path);
8762 		} else {
8763 			struct nvme_io_path *first_path;
8764 
8765 			/* We arrived here as there are no optimized paths for active-passive
8766 			 * mode. Check if this io_path is the first one available on the list.
8767 			 */
8768 			current = false;
8769 			STAILQ_FOREACH(first_path, &nbdev_ch->io_path_list, stailq) {
8770 				if (nvme_io_path_is_available(first_path)) {
8771 					current = (io_path == first_path);
8772 					break;
8773 				}
8774 			}
8775 		}
8776 	}
8777 
8778 	return current;
8779 }
8780 
8781 static struct nvme_ctrlr *
8782 bdev_nvme_next_ctrlr_unsafe(struct nvme_bdev_ctrlr *nbdev_ctrlr, struct nvme_ctrlr *prev)
8783 {
8784 	struct nvme_ctrlr *next;
8785 
8786 	/* Must be called under g_bdev_nvme_mutex */
8787 	next = prev != NULL ? TAILQ_NEXT(prev, tailq) : TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
8788 	while (next != NULL) {
8789 		/* ref can be 0 when the ctrlr was released, but hasn't been detached yet */
8790 		pthread_mutex_lock(&next->mutex);
8791 		if (next->ref > 0) {
8792 			next->ref++;
8793 			pthread_mutex_unlock(&next->mutex);
8794 			return next;
8795 		}
8796 
8797 		pthread_mutex_unlock(&next->mutex);
8798 		next = TAILQ_NEXT(next, tailq);
8799 	}
8800 
8801 	return NULL;
8802 }
8803 
8804 struct bdev_nvme_set_keys_ctx {
8805 	struct nvme_ctrlr	*nctrlr;
8806 	struct spdk_key		*dhchap_key;
8807 	struct spdk_key		*dhchap_ctrlr_key;
8808 	struct spdk_thread	*thread;
8809 	bdev_nvme_set_keys_cb	cb_fn;
8810 	void			*cb_ctx;
8811 	int			status;
8812 };
8813 
8814 static void
8815 bdev_nvme_free_set_keys_ctx(struct bdev_nvme_set_keys_ctx *ctx)
8816 {
8817 	if (ctx == NULL) {
8818 		return;
8819 	}
8820 
8821 	spdk_keyring_put_key(ctx->dhchap_key);
8822 	spdk_keyring_put_key(ctx->dhchap_ctrlr_key);
8823 	free(ctx);
8824 }
8825 
8826 static void
8827 _bdev_nvme_set_keys_done(void *_ctx)
8828 {
8829 	struct bdev_nvme_set_keys_ctx *ctx = _ctx;
8830 
8831 	ctx->cb_fn(ctx->cb_ctx, ctx->status);
8832 
8833 	if (ctx->nctrlr != NULL) {
8834 		nvme_ctrlr_release(ctx->nctrlr);
8835 	}
8836 	bdev_nvme_free_set_keys_ctx(ctx);
8837 }
8838 
8839 static void
8840 bdev_nvme_set_keys_done(struct bdev_nvme_set_keys_ctx *ctx, int status)
8841 {
8842 	ctx->status = status;
8843 	spdk_thread_exec_msg(ctx->thread, _bdev_nvme_set_keys_done, ctx);
8844 }
8845 
8846 static void bdev_nvme_authenticate_ctrlr(struct bdev_nvme_set_keys_ctx *ctx);
8847 
8848 static void
8849 bdev_nvme_authenticate_ctrlr_continue(struct bdev_nvme_set_keys_ctx *ctx)
8850 {
8851 	struct nvme_ctrlr *next;
8852 
8853 	pthread_mutex_lock(&g_bdev_nvme_mutex);
8854 	next = bdev_nvme_next_ctrlr_unsafe(NULL, ctx->nctrlr);
8855 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
8856 
8857 	nvme_ctrlr_release(ctx->nctrlr);
8858 	ctx->nctrlr = next;
8859 
8860 	if (next == NULL) {
8861 		bdev_nvme_set_keys_done(ctx, 0);
8862 	} else {
8863 		bdev_nvme_authenticate_ctrlr(ctx);
8864 	}
8865 }
8866 
8867 static void
8868 bdev_nvme_authenticate_qpairs_done(struct spdk_io_channel_iter *i, int status)
8869 {
8870 	struct bdev_nvme_set_keys_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
8871 
8872 	if (status != 0) {
8873 		bdev_nvme_set_keys_done(ctx, status);
8874 		return;
8875 	}
8876 	bdev_nvme_authenticate_ctrlr_continue(ctx);
8877 }
8878 
8879 static void
8880 bdev_nvme_authenticate_qpair_done(void *ctx, int status)
8881 {
8882 	spdk_for_each_channel_continue(ctx, status);
8883 }
8884 
8885 static void
8886 bdev_nvme_authenticate_qpair(struct spdk_io_channel_iter *i)
8887 {
8888 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
8889 	struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(ch);
8890 	struct nvme_qpair *qpair = ctrlr_ch->qpair;
8891 	int rc;
8892 
8893 	if (!nvme_qpair_is_connected(qpair)) {
8894 		spdk_for_each_channel_continue(i, 0);
8895 		return;
8896 	}
8897 
8898 	rc = spdk_nvme_qpair_authenticate(qpair->qpair, bdev_nvme_authenticate_qpair_done, i);
8899 	if (rc != 0) {
8900 		spdk_for_each_channel_continue(i, rc);
8901 	}
8902 }
8903 
8904 static void
8905 bdev_nvme_authenticate_ctrlr_done(void *_ctx, int status)
8906 {
8907 	struct bdev_nvme_set_keys_ctx *ctx = _ctx;
8908 
8909 	if (status != 0) {
8910 		bdev_nvme_set_keys_done(ctx, status);
8911 		return;
8912 	}
8913 
8914 	spdk_for_each_channel(ctx->nctrlr, bdev_nvme_authenticate_qpair, ctx,
8915 			      bdev_nvme_authenticate_qpairs_done);
8916 }
8917 
8918 static void
8919 bdev_nvme_authenticate_ctrlr(struct bdev_nvme_set_keys_ctx *ctx)
8920 {
8921 	struct spdk_nvme_ctrlr_key_opts opts = {};
8922 	struct nvme_ctrlr *nctrlr = ctx->nctrlr;
8923 	int rc;
8924 
8925 	opts.size = SPDK_SIZEOF(&opts, dhchap_ctrlr_key);
8926 	opts.dhchap_key = ctx->dhchap_key;
8927 	opts.dhchap_ctrlr_key = ctx->dhchap_ctrlr_key;
8928 	rc = spdk_nvme_ctrlr_set_keys(nctrlr->ctrlr, &opts);
8929 	if (rc != 0) {
8930 		bdev_nvme_set_keys_done(ctx, rc);
8931 		return;
8932 	}
8933 
8934 	if (ctx->dhchap_key != NULL) {
8935 		rc = spdk_nvme_ctrlr_authenticate(nctrlr->ctrlr,
8936 						  bdev_nvme_authenticate_ctrlr_done, ctx);
8937 		if (rc != 0) {
8938 			bdev_nvme_set_keys_done(ctx, rc);
8939 		}
8940 	} else {
8941 		bdev_nvme_authenticate_ctrlr_continue(ctx);
8942 	}
8943 }
8944 
8945 int
8946 bdev_nvme_set_keys(const char *name, const char *dhchap_key, const char *dhchap_ctrlr_key,
8947 		   bdev_nvme_set_keys_cb cb_fn, void *cb_ctx)
8948 {
8949 	struct bdev_nvme_set_keys_ctx *ctx;
8950 	struct nvme_bdev_ctrlr *nbdev_ctrlr;
8951 	struct nvme_ctrlr *nctrlr;
8952 
8953 	ctx = calloc(1, sizeof(*ctx));
8954 	if (ctx == NULL) {
8955 		return -ENOMEM;
8956 	}
8957 
8958 	if (dhchap_key != NULL) {
8959 		ctx->dhchap_key = spdk_keyring_get_key(dhchap_key);
8960 		if (ctx->dhchap_key == NULL) {
8961 			SPDK_ERRLOG("Could not find key %s for bdev %s\n", dhchap_key, name);
8962 			bdev_nvme_free_set_keys_ctx(ctx);
8963 			return -ENOKEY;
8964 		}
8965 	}
8966 	if (dhchap_ctrlr_key != NULL) {
8967 		ctx->dhchap_ctrlr_key = spdk_keyring_get_key(dhchap_ctrlr_key);
8968 		if (ctx->dhchap_ctrlr_key == NULL) {
8969 			SPDK_ERRLOG("Could not find key %s for bdev %s\n", dhchap_ctrlr_key, name);
8970 			bdev_nvme_free_set_keys_ctx(ctx);
8971 			return -ENOKEY;
8972 		}
8973 	}
8974 
8975 	pthread_mutex_lock(&g_bdev_nvme_mutex);
8976 	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
8977 	if (nbdev_ctrlr == NULL) {
8978 		SPDK_ERRLOG("Could not find bdev_ctrlr %s\n", name);
8979 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
8980 		bdev_nvme_free_set_keys_ctx(ctx);
8981 		return -ENODEV;
8982 	}
8983 	nctrlr = bdev_nvme_next_ctrlr_unsafe(nbdev_ctrlr, NULL);
8984 	if (nctrlr == NULL) {
8985 		SPDK_ERRLOG("Could not find any nvme_ctrlrs on bdev_ctrlr %s\n", name);
8986 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
8987 		bdev_nvme_free_set_keys_ctx(ctx);
8988 		return -ENODEV;
8989 	}
8990 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
8991 
8992 	ctx->nctrlr = nctrlr;
8993 	ctx->cb_fn = cb_fn;
8994 	ctx->cb_ctx = cb_ctx;
8995 	ctx->thread = spdk_get_thread();
8996 
8997 	bdev_nvme_authenticate_ctrlr(ctx);
8998 
8999 	return 0;
9000 }
9001 
9002 void
9003 nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path)
9004 {
9005 	struct nvme_ns *nvme_ns = io_path->nvme_ns;
9006 	struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
9007 	const struct spdk_nvme_ctrlr_data *cdata;
9008 	const struct spdk_nvme_transport_id *trid;
9009 	const char *adrfam_str;
9010 
9011 	spdk_json_write_object_begin(w);
9012 
9013 	spdk_json_write_named_string(w, "bdev_name", nvme_ns->bdev->disk.name);
9014 
9015 	cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
9016 	trid = spdk_nvme_ctrlr_get_transport_id(nvme_ctrlr->ctrlr);
9017 
9018 	spdk_json_write_named_uint32(w, "cntlid", cdata->cntlid);
9019 	spdk_json_write_named_bool(w, "current", nvme_io_path_is_current(io_path));
9020 	spdk_json_write_named_bool(w, "connected", nvme_qpair_is_connected(io_path->qpair));
9021 	spdk_json_write_named_bool(w, "accessible", nvme_ns_is_accessible(nvme_ns));
9022 
9023 	spdk_json_write_named_object_begin(w, "transport");
9024 	spdk_json_write_named_string(w, "trtype", trid->trstring);
9025 	spdk_json_write_named_string(w, "traddr", trid->traddr);
9026 	if (trid->trsvcid[0] != '\0') {
9027 		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
9028 	}
9029 	adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
9030 	if (adrfam_str) {
9031 		spdk_json_write_named_string(w, "adrfam", adrfam_str);
9032 	}
9033 	spdk_json_write_object_end(w);
9034 
9035 	spdk_json_write_object_end(w);
9036 }
9037 
9038 void
9039 bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w)
9040 {
9041 	struct discovery_ctx *ctx;
9042 	struct discovery_entry_ctx *entry_ctx;
9043 
9044 	spdk_json_write_array_begin(w);
9045 	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
9046 		spdk_json_write_object_begin(w);
9047 		spdk_json_write_named_string(w, "name", ctx->name);
9048 
9049 		spdk_json_write_named_object_begin(w, "trid");
9050 		nvme_bdev_dump_trid_json(&ctx->trid, w);
9051 		spdk_json_write_object_end(w);
9052 
9053 		spdk_json_write_named_array_begin(w, "referrals");
9054 		TAILQ_FOREACH(entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
9055 			spdk_json_write_object_begin(w);
9056 			spdk_json_write_named_object_begin(w, "trid");
9057 			nvme_bdev_dump_trid_json(&entry_ctx->trid, w);
9058 			spdk_json_write_object_end(w);
9059 			spdk_json_write_object_end(w);
9060 		}
9061 		spdk_json_write_array_end(w);
9062 
9063 		spdk_json_write_object_end(w);
9064 	}
9065 	spdk_json_write_array_end(w);
9066 }
9067 
9068 SPDK_LOG_REGISTER_COMPONENT(bdev_nvme)
9069 
9070 static void
9071 bdev_nvme_trace(void)
9072 {
9073 	struct spdk_trace_tpoint_opts opts[] = {
9074 		{
9075 			"BDEV_NVME_IO_START", TRACE_BDEV_NVME_IO_START,
9076 			OWNER_TYPE_NONE, OBJECT_BDEV_NVME_IO, 1,
9077 			{{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
9078 		},
9079 		{
9080 			"BDEV_NVME_IO_DONE", TRACE_BDEV_NVME_IO_DONE,
9081 			OWNER_TYPE_NONE, OBJECT_BDEV_NVME_IO, 0,
9082 			{{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
9083 		}
9084 	};
9085 
9086 
9087 	spdk_trace_register_object(OBJECT_BDEV_NVME_IO, 'N');
9088 	spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
9089 	spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
9090 	spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
9091 	spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
9092 	spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
9093 }
9094 SPDK_TRACE_REGISTER_FN(bdev_nvme_trace, "bdev_nvme", TRACE_GROUP_BDEV_NVME)
9095