xref: /spdk/module/bdev/nvme/bdev_nvme.c (revision db35950a13859f8680bb97c961337d3317a974b2)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "bdev_nvme.h"
37 #include "bdev_ocssd.h"
38 
39 #include "spdk/accel_engine.h"
40 #include "spdk/config.h"
41 #include "spdk/endian.h"
42 #include "spdk/bdev.h"
43 #include "spdk/json.h"
44 #include "spdk/nvme.h"
45 #include "spdk/nvme_ocssd.h"
46 #include "spdk/nvme_zns.h"
47 #include "spdk/thread.h"
48 #include "spdk/string.h"
49 #include "spdk/util.h"
50 
51 #include "spdk/bdev_module.h"
52 #include "spdk/log.h"
53 
54 #define SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT true
55 #define SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS	(10000)
56 
57 static int bdev_nvme_config_json(struct spdk_json_write_ctx *w);
58 
59 struct nvme_bdev_io {
60 	/** array of iovecs to transfer. */
61 	struct iovec *iovs;
62 
63 	/** Number of iovecs in iovs array. */
64 	int iovcnt;
65 
66 	/** Current iovec position. */
67 	int iovpos;
68 
69 	/** Offset in current iovec. */
70 	uint32_t iov_offset;
71 
72 	/** array of iovecs to transfer. */
73 	struct iovec *fused_iovs;
74 
75 	/** Number of iovecs in iovs array. */
76 	int fused_iovcnt;
77 
78 	/** Current iovec position. */
79 	int fused_iovpos;
80 
81 	/** Offset in current iovec. */
82 	uint32_t fused_iov_offset;
83 
84 	/** Saved status for admin passthru completion event, PI error verification, or intermediate compare-and-write status */
85 	struct spdk_nvme_cpl cpl;
86 
87 	/** Originating thread */
88 	struct spdk_thread *orig_thread;
89 
90 	/** Keeps track if first of fused commands was submitted */
91 	bool first_fused_submitted;
92 
93 	/** Temporary pointer to zone report buffer */
94 	struct spdk_nvme_zns_zone_report *zone_report_buf;
95 
96 	/** Keep track of how many zones that have been copied to the spdk_bdev_zone_info struct */
97 	uint64_t handled_zones;
98 };
99 
100 struct nvme_probe_ctx {
101 	size_t count;
102 	struct spdk_nvme_transport_id trids[NVME_MAX_CONTROLLERS];
103 	struct spdk_nvme_host_id hostids[NVME_MAX_CONTROLLERS];
104 	const char *names[NVME_MAX_CONTROLLERS];
105 	uint32_t prchk_flags[NVME_MAX_CONTROLLERS];
106 	const char *hostnqn;
107 };
108 
109 struct nvme_probe_skip_entry {
110 	struct spdk_nvme_transport_id		trid;
111 	TAILQ_ENTRY(nvme_probe_skip_entry)	tailq;
112 };
113 /* All the controllers deleted by users via RPC are skipped by hotplug monitor */
114 static TAILQ_HEAD(, nvme_probe_skip_entry) g_skipped_nvme_ctrlrs = TAILQ_HEAD_INITIALIZER(
115 			g_skipped_nvme_ctrlrs);
116 
117 static struct spdk_bdev_nvme_opts g_opts = {
118 	.action_on_timeout = SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE,
119 	.timeout_us = 0,
120 	.keep_alive_timeout_ms = SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS,
121 	.retry_count = 4,
122 	.arbitration_burst = 0,
123 	.low_priority_weight = 0,
124 	.medium_priority_weight = 0,
125 	.high_priority_weight = 0,
126 	.nvme_adminq_poll_period_us = 10000ULL,
127 	.nvme_ioq_poll_period_us = 0,
128 	.io_queue_requests = 0,
129 	.delay_cmd_submit = SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT,
130 };
131 
132 #define NVME_HOTPLUG_POLL_PERIOD_MAX			10000000ULL
133 #define NVME_HOTPLUG_POLL_PERIOD_DEFAULT		100000ULL
134 
135 static int g_hot_insert_nvme_controller_index = 0;
136 static uint64_t g_nvme_hotplug_poll_period_us = NVME_HOTPLUG_POLL_PERIOD_DEFAULT;
137 static bool g_nvme_hotplug_enabled = false;
138 static struct spdk_thread *g_bdev_nvme_init_thread;
139 static struct spdk_poller *g_hotplug_poller;
140 static struct spdk_poller *g_hotplug_probe_poller;
141 static struct spdk_nvme_probe_ctx *g_hotplug_probe_ctx;
142 
143 static void nvme_ctrlr_populate_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
144 		struct nvme_async_probe_ctx *ctx);
145 static void nvme_ctrlr_populate_namespaces_done(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
146 		struct nvme_async_probe_ctx *ctx);
147 static int bdev_nvme_library_init(void);
148 static void bdev_nvme_library_fini(void);
149 static int bdev_nvme_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
150 			   struct nvme_bdev_io *bio,
151 			   struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba,
152 			   uint32_t flags);
153 static int bdev_nvme_no_pi_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
154 				 struct nvme_bdev_io *bio,
155 				 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba);
156 static int bdev_nvme_writev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
157 			    struct nvme_bdev_io *bio,
158 			    struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba,
159 			    uint32_t flags);
160 static int bdev_nvme_zone_appendv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
161 				  struct nvme_bdev_io *bio,
162 				  struct iovec *iov, int iovcnt, void *md, uint64_t lba_count,
163 				  uint64_t zslba, uint32_t flags);
164 static int bdev_nvme_comparev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
165 			      struct nvme_bdev_io *bio,
166 			      struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba,
167 			      uint32_t flags);
168 static int bdev_nvme_comparev_and_writev(struct spdk_nvme_ns *ns,
169 		struct spdk_nvme_qpair *qpair,
170 		struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt, struct iovec *write_iov,
171 		int write_iovcnt, void *md, uint64_t lba_count, uint64_t lba,
172 		uint32_t flags);
173 static int bdev_nvme_get_zone_info(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
174 				   struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones,
175 				   struct spdk_bdev_zone_info *info);
176 static int bdev_nvme_zone_management(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
177 				     struct nvme_bdev_io *bio, uint64_t zone_id,
178 				     enum spdk_bdev_zone_action action);
179 static int bdev_nvme_admin_passthru(struct nvme_io_path *io_path,
180 				    struct nvme_bdev_io *bio,
181 				    struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes);
182 static int bdev_nvme_io_passthru(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
183 				 struct nvme_bdev_io *bio,
184 				 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes);
185 static int bdev_nvme_io_passthru_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
186 				    struct nvme_bdev_io *bio,
187 				    struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len);
188 static int bdev_nvme_abort(struct nvme_io_path *io_path,
189 			   struct nvme_bdev_io *bio, struct nvme_bdev_io *bio_to_abort);
190 static int bdev_nvme_reset(struct nvme_io_path *io_path, struct spdk_bdev_io *bdev_io);
191 static int bdev_nvme_failover(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool remove);
192 static void remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr);
193 
194 typedef void (*populate_namespace_fn)(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
195 				      struct nvme_bdev_ns *nvme_ns, struct nvme_async_probe_ctx *ctx);
196 static void nvme_ctrlr_populate_standard_namespace(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
197 		struct nvme_bdev_ns *nvme_ns, struct nvme_async_probe_ctx *ctx);
198 
199 static populate_namespace_fn g_populate_namespace_fn[] = {
200 	NULL,
201 	nvme_ctrlr_populate_standard_namespace,
202 	bdev_ocssd_populate_namespace,
203 };
204 
205 typedef void (*depopulate_namespace_fn)(struct nvme_bdev_ns *nvme_ns);
206 static void nvme_ctrlr_depopulate_standard_namespace(struct nvme_bdev_ns *nvme_ns);
207 
208 static depopulate_namespace_fn g_depopulate_namespace_fn[] = {
209 	NULL,
210 	nvme_ctrlr_depopulate_standard_namespace,
211 	bdev_ocssd_depopulate_namespace,
212 };
213 
214 typedef void (*config_json_namespace_fn)(struct spdk_json_write_ctx *w,
215 		struct nvme_bdev_ns *nvme_ns);
216 static void nvme_ctrlr_config_json_standard_namespace(struct spdk_json_write_ctx *w,
217 		struct nvme_bdev_ns *nvme_ns);
218 
219 static config_json_namespace_fn g_config_json_namespace_fn[] = {
220 	NULL,
221 	nvme_ctrlr_config_json_standard_namespace,
222 	bdev_ocssd_namespace_config_json,
223 };
224 
225 struct spdk_nvme_qpair *
226 bdev_nvme_get_io_qpair(struct spdk_io_channel *io_path_ch)
227 {
228 	struct nvme_io_path *io_path;
229 
230 	assert(io_path_ch != NULL);
231 
232 	io_path = spdk_io_channel_get_ctx(io_path_ch);
233 
234 	return io_path->qpair;
235 }
236 
237 static int
238 bdev_nvme_get_ctx_size(void)
239 {
240 	return sizeof(struct nvme_bdev_io);
241 }
242 
243 static struct spdk_bdev_module nvme_if = {
244 	.name = "nvme",
245 	.async_fini = true,
246 	.module_init = bdev_nvme_library_init,
247 	.module_fini = bdev_nvme_library_fini,
248 	.config_json = bdev_nvme_config_json,
249 	.get_ctx_size = bdev_nvme_get_ctx_size,
250 
251 };
252 SPDK_BDEV_MODULE_REGISTER(nvme, &nvme_if)
253 
254 static inline bool
255 bdev_nvme_find_io_path(struct nvme_bdev *nbdev, struct nvme_io_path *io_path,
256 		       struct spdk_nvme_ns **_ns, struct spdk_nvme_qpair **_qpair)
257 {
258 	if (spdk_unlikely(io_path->qpair == NULL)) {
259 		/* The device is currently resetting. */
260 		return false;
261 	}
262 
263 	*_ns = nbdev->nvme_ns->ns;
264 	*_qpair = io_path->qpair;
265 	return true;
266 }
267 
268 static inline bool
269 bdev_nvme_find_admin_path(struct nvme_io_path *io_path,
270 			  struct nvme_bdev_ctrlr **_nvme_bdev_ctrlr)
271 {
272 	*_nvme_bdev_ctrlr = io_path->ctrlr;
273 	return true;
274 }
275 
276 static inline void
277 bdev_nvme_io_complete_nvme_status(struct nvme_bdev_io *bio,
278 				  const struct spdk_nvme_cpl *cpl)
279 {
280 	spdk_bdev_io_complete_nvme_status(spdk_bdev_io_from_ctx(bio), cpl->cdw0,
281 					  cpl->status.sct, cpl->status.sc);
282 }
283 
284 static inline void
285 bdev_nvme_io_complete(struct nvme_bdev_io *bio, int rc)
286 {
287 	enum spdk_bdev_io_status io_status;
288 
289 	if (rc == 0) {
290 		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
291 	} else if (rc == -ENOMEM) {
292 		io_status = SPDK_BDEV_IO_STATUS_NOMEM;
293 	} else {
294 		io_status = SPDK_BDEV_IO_STATUS_FAILED;
295 	}
296 
297 	spdk_bdev_io_complete(spdk_bdev_io_from_ctx(bio), io_status);
298 }
299 
300 static void
301 bdev_nvme_disconnected_qpair_cb(struct spdk_nvme_qpair *qpair, void *poll_group_ctx)
302 {
303 	int rc;
304 
305 	SPDK_DEBUGLOG(bdev_nvme, "qpair %p is disconnected, attempting reconnect.\n", qpair);
306 	/*
307 	 * Currently, just try to reconnect indefinitely. If we are doing a reset, the reset will
308 	 * reconnect a qpair and we will stop getting a callback for this one.
309 	 */
310 	rc = spdk_nvme_ctrlr_reconnect_io_qpair(qpair);
311 	if (rc != 0) {
312 		SPDK_WARNLOG("Failed to reconnect to qpair %p, errno %d\n", qpair, -rc);
313 	}
314 }
315 
316 static int
317 bdev_nvme_poll(void *arg)
318 {
319 	struct nvme_bdev_poll_group *group = arg;
320 	int64_t num_completions;
321 
322 	if (group->collect_spin_stat && group->start_ticks == 0) {
323 		group->start_ticks = spdk_get_ticks();
324 	}
325 
326 	num_completions = spdk_nvme_poll_group_process_completions(group->group, 0,
327 			  bdev_nvme_disconnected_qpair_cb);
328 	if (group->collect_spin_stat) {
329 		if (num_completions > 0) {
330 			if (group->end_ticks != 0) {
331 				group->spin_ticks += (group->end_ticks - group->start_ticks);
332 				group->end_ticks = 0;
333 			}
334 			group->start_ticks = 0;
335 		} else {
336 			group->end_ticks = spdk_get_ticks();
337 		}
338 	}
339 
340 	return num_completions > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
341 }
342 
343 static int
344 bdev_nvme_poll_adminq(void *arg)
345 {
346 	int32_t rc;
347 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = arg;
348 
349 	assert(nvme_bdev_ctrlr != NULL);
350 
351 	rc = spdk_nvme_ctrlr_process_admin_completions(nvme_bdev_ctrlr->ctrlr);
352 	if (rc < 0) {
353 		bdev_nvme_failover(nvme_bdev_ctrlr, false);
354 	}
355 
356 	return rc == 0 ? SPDK_POLLER_IDLE : SPDK_POLLER_BUSY;
357 }
358 
359 static int
360 bdev_nvme_destruct(void *ctx)
361 {
362 	struct nvme_bdev *nvme_disk = ctx;
363 	struct nvme_bdev_ns *nvme_ns = nvme_disk->nvme_ns;
364 
365 	pthread_mutex_lock(&nvme_ns->ctrlr->mutex);
366 
367 	nvme_ns->bdev = NULL;
368 
369 	if (!nvme_ns->populated) {
370 		pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
371 
372 		nvme_bdev_ctrlr_destruct(nvme_ns->ctrlr);
373 	} else {
374 		pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
375 	}
376 
377 	free(nvme_disk->disk.name);
378 	free(nvme_disk);
379 
380 	return 0;
381 }
382 
383 static int
384 bdev_nvme_flush(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
385 		struct nvme_bdev_io *bio, uint64_t offset, uint64_t nbytes)
386 {
387 	bdev_nvme_io_complete(bio, 0);
388 
389 	return 0;
390 }
391 
392 static int
393 bdev_nvme_create_qpair(struct nvme_io_path *io_path)
394 {
395 	struct spdk_nvme_ctrlr *ctrlr = io_path->ctrlr->ctrlr;
396 	struct spdk_nvme_io_qpair_opts opts;
397 	struct spdk_nvme_qpair *qpair;
398 	int rc;
399 
400 	spdk_nvme_ctrlr_get_default_io_qpair_opts(ctrlr, &opts, sizeof(opts));
401 	opts.delay_cmd_submit = g_opts.delay_cmd_submit;
402 	opts.create_only = true;
403 	opts.io_queue_requests = spdk_max(g_opts.io_queue_requests, opts.io_queue_requests);
404 	g_opts.io_queue_requests = opts.io_queue_requests;
405 
406 	qpair = spdk_nvme_ctrlr_alloc_io_qpair(ctrlr, &opts, sizeof(opts));
407 	if (qpair == NULL) {
408 		return -1;
409 	}
410 
411 	assert(io_path->group != NULL);
412 
413 	rc = spdk_nvme_poll_group_add(io_path->group->group, qpair);
414 	if (rc != 0) {
415 		SPDK_ERRLOG("Unable to begin polling on NVMe Channel.\n");
416 		goto err;
417 	}
418 
419 	rc = spdk_nvme_ctrlr_connect_io_qpair(ctrlr, qpair);
420 	if (rc != 0) {
421 		SPDK_ERRLOG("Unable to connect I/O qpair.\n");
422 		goto err;
423 	}
424 
425 	io_path->qpair = qpair;
426 
427 	return 0;
428 
429 err:
430 	spdk_nvme_ctrlr_free_io_qpair(qpair);
431 
432 	return rc;
433 }
434 
435 static int
436 bdev_nvme_destroy_qpair(struct nvme_io_path *io_path)
437 {
438 	int rc;
439 
440 	if (io_path->qpair == NULL) {
441 		return 0;
442 	}
443 
444 	rc = spdk_nvme_ctrlr_free_io_qpair(io_path->qpair);
445 	if (!rc) {
446 		io_path->qpair = NULL;
447 	}
448 	return rc;
449 }
450 
451 static void
452 _bdev_nvme_check_pending_destruct(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr)
453 {
454 	pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
455 	if (nvme_bdev_ctrlr->destruct_after_reset) {
456 		assert(nvme_bdev_ctrlr->ref == 0 && nvme_bdev_ctrlr->destruct);
457 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
458 
459 		spdk_thread_send_msg(nvme_bdev_ctrlr->thread, nvme_bdev_ctrlr_unregister,
460 				     nvme_bdev_ctrlr);
461 	} else {
462 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
463 	}
464 }
465 
466 static void
467 bdev_nvme_check_pending_destruct(struct spdk_io_channel_iter *i, int status)
468 {
469 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = spdk_io_channel_iter_get_ctx(i);
470 
471 	_bdev_nvme_check_pending_destruct(nvme_bdev_ctrlr);
472 }
473 
474 static void
475 _bdev_nvme_complete_pending_resets(struct nvme_io_path *io_path,
476 				   enum spdk_bdev_io_status status)
477 {
478 	struct spdk_bdev_io *bdev_io;
479 
480 	while (!TAILQ_EMPTY(&io_path->pending_resets)) {
481 		bdev_io = TAILQ_FIRST(&io_path->pending_resets);
482 		TAILQ_REMOVE(&io_path->pending_resets, bdev_io, module_link);
483 		spdk_bdev_io_complete(bdev_io, status);
484 	}
485 }
486 
487 static void
488 bdev_nvme_complete_pending_resets(struct spdk_io_channel_iter *i)
489 {
490 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
491 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(_ch);
492 
493 	_bdev_nvme_complete_pending_resets(io_path, SPDK_BDEV_IO_STATUS_SUCCESS);
494 
495 	spdk_for_each_channel_continue(i, 0);
496 }
497 
498 static void
499 bdev_nvme_abort_pending_resets(struct spdk_io_channel_iter *i)
500 {
501 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
502 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(_ch);
503 
504 	_bdev_nvme_complete_pending_resets(io_path, SPDK_BDEV_IO_STATUS_FAILED);
505 
506 	spdk_for_each_channel_continue(i, 0);
507 }
508 
509 static void
510 bdev_nvme_reset_io_complete(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
511 			    struct spdk_bdev_io *bdev_io, int rc)
512 {
513 	enum spdk_bdev_io_status io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
514 
515 	if (rc) {
516 		io_status = SPDK_BDEV_IO_STATUS_FAILED;
517 	}
518 
519 	spdk_bdev_io_complete(bdev_io, io_status);
520 
521 	/* Make sure we clear any pending resets before returning. */
522 	spdk_for_each_channel(nvme_bdev_ctrlr,
523 			      rc == 0 ? bdev_nvme_complete_pending_resets :
524 			      bdev_nvme_abort_pending_resets,
525 			      nvme_bdev_ctrlr,
526 			      bdev_nvme_check_pending_destruct);
527 }
528 
529 static void
530 _bdev_nvme_reset_complete(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, int rc)
531 {
532 	struct nvme_bdev_ctrlr_trid *curr_trid;
533 	struct spdk_bdev_io *bdev_io = nvme_bdev_ctrlr->reset_bdev_io;
534 
535 	nvme_bdev_ctrlr->reset_bdev_io = NULL;
536 
537 	if (rc) {
538 		SPDK_ERRLOG("Resetting controller failed.\n");
539 	} else {
540 		SPDK_NOTICELOG("Resetting controller successful.\n");
541 	}
542 
543 	pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
544 	nvme_bdev_ctrlr->resetting = false;
545 	nvme_bdev_ctrlr->failover_in_progress = false;
546 
547 	curr_trid = TAILQ_FIRST(&nvme_bdev_ctrlr->trids);
548 	assert(curr_trid != NULL);
549 	assert(&curr_trid->trid == nvme_bdev_ctrlr->connected_trid);
550 
551 	curr_trid->is_failed = rc != 0 ? true : false;
552 
553 	if (nvme_bdev_ctrlr->ref == 0 && nvme_bdev_ctrlr->destruct) {
554 		/* Destruct ctrlr after clearing pending resets. */
555 		nvme_bdev_ctrlr->destruct_after_reset = true;
556 	}
557 
558 	pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
559 
560 	if (bdev_io) {
561 		bdev_nvme_reset_io_complete(nvme_bdev_ctrlr, bdev_io, rc);
562 	} else {
563 		/* Make sure we clear any pending resets before returning. */
564 		spdk_for_each_channel(nvme_bdev_ctrlr,
565 				      rc == 0 ? bdev_nvme_complete_pending_resets :
566 				      bdev_nvme_abort_pending_resets,
567 				      nvme_bdev_ctrlr,
568 				      bdev_nvme_check_pending_destruct);
569 	}
570 }
571 
572 static void
573 _bdev_nvme_reset_create_qpairs_done(struct spdk_io_channel_iter *i, int status)
574 {
575 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = spdk_io_channel_iter_get_ctx(i);
576 
577 	_bdev_nvme_reset_complete(nvme_bdev_ctrlr, status);
578 }
579 
580 static void
581 _bdev_nvme_reset_create_qpair(struct spdk_io_channel_iter *i)
582 {
583 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
584 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(_ch);
585 	int rc;
586 
587 	rc = bdev_nvme_create_qpair(io_path);
588 
589 	spdk_for_each_channel_continue(i, rc);
590 }
591 
592 static void
593 _bdev_nvme_reset_ctrlr(struct spdk_io_channel_iter *i, int status)
594 {
595 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = spdk_io_channel_iter_get_ctx(i);
596 	int rc;
597 
598 	if (status) {
599 		rc = status;
600 		goto err;
601 	}
602 
603 	rc = spdk_nvme_ctrlr_reset(nvme_bdev_ctrlr->ctrlr);
604 	if (rc != 0) {
605 		goto err;
606 	}
607 
608 	/* Recreate all of the I/O queue pairs */
609 	spdk_for_each_channel(nvme_bdev_ctrlr,
610 			      _bdev_nvme_reset_create_qpair,
611 			      nvme_bdev_ctrlr,
612 			      _bdev_nvme_reset_create_qpairs_done);
613 	return;
614 
615 err:
616 	_bdev_nvme_reset_complete(nvme_bdev_ctrlr, rc);
617 }
618 
619 static void
620 _bdev_nvme_reset_destroy_qpair(struct spdk_io_channel_iter *i)
621 {
622 	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
623 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(ch);
624 	int rc;
625 
626 	rc = bdev_nvme_destroy_qpair(io_path);
627 
628 	spdk_for_each_channel_continue(i, rc);
629 }
630 
631 static int
632 _bdev_nvme_reset(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr)
633 {
634 	pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
635 	if (nvme_bdev_ctrlr->destruct) {
636 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
637 		return -EBUSY;
638 	}
639 
640 	if (nvme_bdev_ctrlr->resetting) {
641 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
642 		SPDK_NOTICELOG("Unable to perform reset, already in progress.\n");
643 		return -EAGAIN;
644 	}
645 
646 	nvme_bdev_ctrlr->resetting = true;
647 	pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
648 
649 	/* First, delete all NVMe I/O queue pairs. */
650 	spdk_for_each_channel(nvme_bdev_ctrlr,
651 			      _bdev_nvme_reset_destroy_qpair,
652 			      nvme_bdev_ctrlr,
653 			      _bdev_nvme_reset_ctrlr);
654 
655 	return 0;
656 }
657 
658 static int
659 bdev_nvme_reset(struct nvme_io_path *io_path, struct spdk_bdev_io *bdev_io)
660 {
661 	int rc;
662 
663 	rc = _bdev_nvme_reset(io_path->ctrlr);
664 	if (rc == 0) {
665 		assert(io_path->ctrlr->reset_bdev_io == NULL);
666 		io_path->ctrlr->reset_bdev_io = bdev_io;
667 	} else if (rc == -EAGAIN) {
668 		/*
669 		 * Reset call is queued only if it is from the app framework. This is on purpose so that
670 		 * we don't interfere with the app framework reset strategy. i.e. we are deferring to the
671 		 * upper level. If they are in the middle of a reset, we won't try to schedule another one.
672 		 */
673 		TAILQ_INSERT_TAIL(&io_path->pending_resets, bdev_io, module_link);
674 	} else {
675 		return rc;
676 	}
677 
678 	return 0;
679 }
680 
681 static int
682 _bdev_nvme_failover_start(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool remove)
683 {
684 	struct nvme_bdev_ctrlr_trid *curr_trid = NULL, *next_trid = NULL;
685 	int rc;
686 
687 	pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
688 	if (nvme_bdev_ctrlr->destruct) {
689 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
690 		/* Don't bother resetting if the controller is in the process of being destructed. */
691 		return -EBUSY;
692 	}
693 
694 	curr_trid = TAILQ_FIRST(&nvme_bdev_ctrlr->trids);
695 	assert(curr_trid);
696 	assert(&curr_trid->trid == nvme_bdev_ctrlr->connected_trid);
697 	next_trid = TAILQ_NEXT(curr_trid, link);
698 
699 	if (nvme_bdev_ctrlr->resetting) {
700 		if (next_trid && !nvme_bdev_ctrlr->failover_in_progress) {
701 			rc = -EAGAIN;
702 		} else {
703 			rc = -EBUSY;
704 		}
705 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
706 		SPDK_NOTICELOG("Unable to perform reset, already in progress.\n");
707 		return rc;
708 	}
709 
710 	nvme_bdev_ctrlr->resetting = true;
711 	curr_trid->is_failed = true;
712 
713 	if (next_trid) {
714 		assert(curr_trid->trid.trtype != SPDK_NVME_TRANSPORT_PCIE);
715 
716 		SPDK_NOTICELOG("Start failover from %s:%s to %s:%s\n", curr_trid->trid.traddr,
717 			       curr_trid->trid.trsvcid,	next_trid->trid.traddr, next_trid->trid.trsvcid);
718 
719 		nvme_bdev_ctrlr->failover_in_progress = true;
720 		spdk_nvme_ctrlr_fail(nvme_bdev_ctrlr->ctrlr);
721 		nvme_bdev_ctrlr->connected_trid = &next_trid->trid;
722 		rc = spdk_nvme_ctrlr_set_trid(nvme_bdev_ctrlr->ctrlr, &next_trid->trid);
723 		assert(rc == 0);
724 		TAILQ_REMOVE(&nvme_bdev_ctrlr->trids, curr_trid, link);
725 		if (!remove) {
726 			/** Shuffle the old trid to the end of the list and use the new one.
727 			 * Allows for round robin through multiple connections.
728 			 */
729 			TAILQ_INSERT_TAIL(&nvme_bdev_ctrlr->trids, curr_trid, link);
730 		} else {
731 			free(curr_trid);
732 		}
733 	}
734 
735 	pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
736 	return 0;
737 }
738 
739 static int
740 bdev_nvme_failover(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool remove)
741 {
742 	int rc;
743 
744 	rc = _bdev_nvme_failover_start(nvme_bdev_ctrlr, remove);
745 	if (rc == 0) {
746 		/* First, delete all NVMe I/O queue pairs. */
747 		spdk_for_each_channel(nvme_bdev_ctrlr,
748 				      _bdev_nvme_reset_destroy_qpair,
749 				      nvme_bdev_ctrlr,
750 				      _bdev_nvme_reset_ctrlr);
751 	} else if (rc != -EBUSY) {
752 		return rc;
753 	}
754 
755 	return 0;
756 }
757 
758 static int
759 bdev_nvme_unmap(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
760 		struct nvme_bdev_io *bio,
761 		uint64_t offset_blocks,
762 		uint64_t num_blocks);
763 
764 static void
765 bdev_nvme_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
766 		     bool success)
767 {
768 	struct nvme_bdev_io *bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
769 	struct spdk_bdev *bdev = bdev_io->bdev;
770 	struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev->ctxt;
771 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(ch);
772 	struct spdk_nvme_ns *ns;
773 	struct spdk_nvme_qpair *qpair;
774 	int ret;
775 
776 	if (!success) {
777 		ret = -EINVAL;
778 		goto exit;
779 	}
780 
781 	if (spdk_unlikely(!bdev_nvme_find_io_path(nbdev, io_path, &ns, &qpair))) {
782 		ret = -ENXIO;
783 		goto exit;
784 	}
785 
786 	ret = bdev_nvme_readv(ns,
787 			      qpair,
788 			      bio,
789 			      bdev_io->u.bdev.iovs,
790 			      bdev_io->u.bdev.iovcnt,
791 			      bdev_io->u.bdev.md_buf,
792 			      bdev_io->u.bdev.num_blocks,
793 			      bdev_io->u.bdev.offset_blocks,
794 			      bdev->dif_check_flags);
795 
796 exit:
797 	if (spdk_unlikely(ret != 0)) {
798 		bdev_nvme_io_complete(bio, ret);
799 	}
800 }
801 
802 static void
803 bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
804 {
805 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(ch);
806 	struct spdk_bdev *bdev = bdev_io->bdev;
807 	struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev->ctxt;
808 	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
809 	struct nvme_bdev_io *nbdev_io_to_abort;
810 	struct spdk_nvme_ns *ns;
811 	struct spdk_nvme_qpair *qpair;
812 	int rc = 0;
813 
814 	if (spdk_unlikely(!bdev_nvme_find_io_path(nbdev, io_path, &ns, &qpair))) {
815 		rc = -ENXIO;
816 		goto exit;
817 	}
818 
819 	switch (bdev_io->type) {
820 	case SPDK_BDEV_IO_TYPE_READ:
821 		if (bdev_io->u.bdev.iovs && bdev_io->u.bdev.iovs[0].iov_base) {
822 			rc = bdev_nvme_readv(ns,
823 					     qpair,
824 					     nbdev_io,
825 					     bdev_io->u.bdev.iovs,
826 					     bdev_io->u.bdev.iovcnt,
827 					     bdev_io->u.bdev.md_buf,
828 					     bdev_io->u.bdev.num_blocks,
829 					     bdev_io->u.bdev.offset_blocks,
830 					     bdev->dif_check_flags);
831 		} else {
832 			spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb,
833 					     bdev_io->u.bdev.num_blocks * bdev->blocklen);
834 			rc = 0;
835 		}
836 		break;
837 	case SPDK_BDEV_IO_TYPE_WRITE:
838 		rc = bdev_nvme_writev(ns,
839 				      qpair,
840 				      nbdev_io,
841 				      bdev_io->u.bdev.iovs,
842 				      bdev_io->u.bdev.iovcnt,
843 				      bdev_io->u.bdev.md_buf,
844 				      bdev_io->u.bdev.num_blocks,
845 				      bdev_io->u.bdev.offset_blocks,
846 				      bdev->dif_check_flags);
847 		break;
848 	case SPDK_BDEV_IO_TYPE_COMPARE:
849 		rc = bdev_nvme_comparev(ns,
850 					qpair,
851 					nbdev_io,
852 					bdev_io->u.bdev.iovs,
853 					bdev_io->u.bdev.iovcnt,
854 					bdev_io->u.bdev.md_buf,
855 					bdev_io->u.bdev.num_blocks,
856 					bdev_io->u.bdev.offset_blocks,
857 					bdev->dif_check_flags);
858 		break;
859 	case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
860 		rc = bdev_nvme_comparev_and_writev(ns,
861 						   qpair,
862 						   nbdev_io,
863 						   bdev_io->u.bdev.iovs,
864 						   bdev_io->u.bdev.iovcnt,
865 						   bdev_io->u.bdev.fused_iovs,
866 						   bdev_io->u.bdev.fused_iovcnt,
867 						   bdev_io->u.bdev.md_buf,
868 						   bdev_io->u.bdev.num_blocks,
869 						   bdev_io->u.bdev.offset_blocks,
870 						   bdev->dif_check_flags);
871 		break;
872 	case SPDK_BDEV_IO_TYPE_UNMAP:
873 		rc = bdev_nvme_unmap(ns,
874 				     qpair,
875 				     nbdev_io,
876 				     bdev_io->u.bdev.offset_blocks,
877 				     bdev_io->u.bdev.num_blocks);
878 		break;
879 	case SPDK_BDEV_IO_TYPE_RESET:
880 		rc = bdev_nvme_reset(io_path, bdev_io);
881 		break;
882 	case SPDK_BDEV_IO_TYPE_FLUSH:
883 		rc = bdev_nvme_flush(ns,
884 				     qpair,
885 				     nbdev_io,
886 				     bdev_io->u.bdev.offset_blocks,
887 				     bdev_io->u.bdev.num_blocks);
888 		break;
889 	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
890 		rc = bdev_nvme_zone_appendv(ns,
891 					    qpair,
892 					    nbdev_io,
893 					    bdev_io->u.bdev.iovs,
894 					    bdev_io->u.bdev.iovcnt,
895 					    bdev_io->u.bdev.md_buf,
896 					    bdev_io->u.bdev.num_blocks,
897 					    bdev_io->u.bdev.offset_blocks,
898 					    bdev->dif_check_flags);
899 		break;
900 	case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
901 		rc = bdev_nvme_get_zone_info(ns,
902 					     qpair,
903 					     nbdev_io,
904 					     bdev_io->u.zone_mgmt.zone_id,
905 					     bdev_io->u.zone_mgmt.num_zones,
906 					     bdev_io->u.zone_mgmt.buf);
907 		break;
908 	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
909 		rc = bdev_nvme_zone_management(ns,
910 					       qpair,
911 					       nbdev_io,
912 					       bdev_io->u.zone_mgmt.zone_id,
913 					       bdev_io->u.zone_mgmt.zone_action);
914 		break;
915 	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
916 		rc = bdev_nvme_admin_passthru(io_path,
917 					      nbdev_io,
918 					      &bdev_io->u.nvme_passthru.cmd,
919 					      bdev_io->u.nvme_passthru.buf,
920 					      bdev_io->u.nvme_passthru.nbytes);
921 		break;
922 	case SPDK_BDEV_IO_TYPE_NVME_IO:
923 		rc = bdev_nvme_io_passthru(ns,
924 					   qpair,
925 					   nbdev_io,
926 					   &bdev_io->u.nvme_passthru.cmd,
927 					   bdev_io->u.nvme_passthru.buf,
928 					   bdev_io->u.nvme_passthru.nbytes);
929 		break;
930 	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
931 		rc = bdev_nvme_io_passthru_md(ns,
932 					      qpair,
933 					      nbdev_io,
934 					      &bdev_io->u.nvme_passthru.cmd,
935 					      bdev_io->u.nvme_passthru.buf,
936 					      bdev_io->u.nvme_passthru.nbytes,
937 					      bdev_io->u.nvme_passthru.md_buf,
938 					      bdev_io->u.nvme_passthru.md_len);
939 		break;
940 	case SPDK_BDEV_IO_TYPE_ABORT:
941 		nbdev_io_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx;
942 		rc = bdev_nvme_abort(io_path,
943 				     nbdev_io,
944 				     nbdev_io_to_abort);
945 		break;
946 	default:
947 		rc = -EINVAL;
948 		break;
949 	}
950 
951 exit:
952 	if (spdk_unlikely(rc != 0)) {
953 		bdev_nvme_io_complete(nbdev_io, rc);
954 	}
955 }
956 
957 static bool
958 bdev_nvme_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
959 {
960 	struct nvme_bdev *nbdev = ctx;
961 	struct nvme_bdev_ns *nvme_ns;
962 	struct spdk_nvme_ns *ns;
963 	struct spdk_nvme_ctrlr *ctrlr;
964 	const struct spdk_nvme_ctrlr_data *cdata;
965 
966 	nvme_ns = nbdev->nvme_ns;
967 	assert(nvme_ns != NULL);
968 	ns = nvme_ns->ns;
969 	ctrlr = spdk_nvme_ns_get_ctrlr(ns);
970 
971 	switch (io_type) {
972 	case SPDK_BDEV_IO_TYPE_READ:
973 	case SPDK_BDEV_IO_TYPE_WRITE:
974 	case SPDK_BDEV_IO_TYPE_RESET:
975 	case SPDK_BDEV_IO_TYPE_FLUSH:
976 	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
977 	case SPDK_BDEV_IO_TYPE_NVME_IO:
978 	case SPDK_BDEV_IO_TYPE_ABORT:
979 		return true;
980 
981 	case SPDK_BDEV_IO_TYPE_COMPARE:
982 		return spdk_nvme_ns_supports_compare(ns);
983 
984 	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
985 		return spdk_nvme_ns_get_md_size(ns) ? true : false;
986 
987 	case SPDK_BDEV_IO_TYPE_UNMAP:
988 		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
989 		return cdata->oncs.dsm;
990 
991 	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
992 		/*
993 		 * The NVMe controller write_zeroes function is currently not used by our driver.
994 		 * NVMe write zeroes is limited to 16-bit block count, and the bdev layer currently
995 		 * has no mechanism for reporting a max write zeroes block count, nor ability to
996 		 * split a write zeroes request.
997 		 */
998 		return false;
999 
1000 	case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
1001 		if (spdk_nvme_ctrlr_get_flags(ctrlr) &
1002 		    SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED) {
1003 			return true;
1004 		}
1005 		return false;
1006 
1007 	case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
1008 	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
1009 		return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS;
1010 
1011 	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
1012 		return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS &&
1013 		       spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ZONE_APPEND_SUPPORTED;
1014 
1015 	default:
1016 		return false;
1017 	}
1018 }
1019 
1020 static int
1021 bdev_nvme_create_path_cb(void *io_device, void *ctx_buf)
1022 {
1023 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = io_device;
1024 	struct nvme_io_path *io_path = ctx_buf;
1025 	struct spdk_io_channel *pg_ch;
1026 	int rc;
1027 
1028 	pg_ch = spdk_get_io_channel(&g_nvme_bdev_ctrlrs);
1029 	if (!pg_ch) {
1030 		return -1;
1031 	}
1032 
1033 	io_path->group = spdk_io_channel_get_ctx(pg_ch);
1034 
1035 #ifdef SPDK_CONFIG_VTUNE
1036 	io_path->group->collect_spin_stat = true;
1037 #else
1038 	io_path->group->collect_spin_stat = false;
1039 #endif
1040 
1041 	TAILQ_INIT(&io_path->pending_resets);
1042 
1043 	if (spdk_nvme_ctrlr_is_ocssd_supported(nvme_bdev_ctrlr->ctrlr)) {
1044 		rc = bdev_ocssd_create_io_channel(io_path);
1045 		if (rc != 0) {
1046 			goto err_ocssd_ch;
1047 		}
1048 	}
1049 
1050 	io_path->ctrlr = nvme_bdev_ctrlr;
1051 
1052 	rc = bdev_nvme_create_qpair(io_path);
1053 	if (rc != 0) {
1054 		goto err_qpair;
1055 	}
1056 
1057 	return 0;
1058 
1059 err_qpair:
1060 	if (io_path->ocssd_ch) {
1061 		bdev_ocssd_destroy_io_channel(io_path);
1062 	}
1063 err_ocssd_ch:
1064 	spdk_put_io_channel(pg_ch);
1065 
1066 	return rc;
1067 }
1068 
1069 static void
1070 bdev_nvme_destroy_path_cb(void *io_device, void *ctx_buf)
1071 {
1072 	struct nvme_io_path *io_path = ctx_buf;
1073 
1074 	assert(io_path->group != NULL);
1075 
1076 	if (io_path->ocssd_ch != NULL) {
1077 		bdev_ocssd_destroy_io_channel(io_path);
1078 	}
1079 
1080 	bdev_nvme_destroy_qpair(io_path);
1081 
1082 	spdk_put_io_channel(spdk_io_channel_from_ctx(io_path->group));
1083 }
1084 
1085 static void
1086 bdev_nvme_poll_group_submit_accel_crc32c(void *ctx, uint32_t *dst, struct iovec *iov,
1087 		uint32_t iov_cnt, uint32_t seed,
1088 		spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
1089 {
1090 	struct nvme_bdev_poll_group *group = ctx;
1091 	int rc;
1092 
1093 	assert(group->accel_channel != NULL);
1094 	assert(cb_fn != NULL);
1095 
1096 	rc = spdk_accel_submit_crc32cv(group->accel_channel, dst, iov, iov_cnt, seed, cb_fn, cb_arg);
1097 	if (rc) {
1098 		/* For the two cases, spdk_accel_submit_crc32cv does not call the user's cb_fn */
1099 		if (rc == -ENOMEM || rc == -EINVAL) {
1100 			cb_fn(cb_arg, rc);
1101 		}
1102 		SPDK_ERRLOG("Cannot complete the accelerated crc32c operation with iov=%p\n", iov);
1103 	}
1104 }
1105 
1106 static struct spdk_nvme_accel_fn_table g_bdev_nvme_accel_fn_table = {
1107 	.table_size		= sizeof(struct spdk_nvme_accel_fn_table),
1108 	.submit_accel_crc32c	= bdev_nvme_poll_group_submit_accel_crc32c,
1109 };
1110 
1111 static int
1112 bdev_nvme_poll_group_create_cb(void *io_device, void *ctx_buf)
1113 {
1114 	struct nvme_bdev_poll_group *group = ctx_buf;
1115 
1116 	group->group = spdk_nvme_poll_group_create(group, &g_bdev_nvme_accel_fn_table);
1117 	if (group->group == NULL) {
1118 		return -1;
1119 	}
1120 
1121 	group->accel_channel = spdk_accel_engine_get_io_channel();
1122 	if (!group->accel_channel) {
1123 		spdk_nvme_poll_group_destroy(group->group);
1124 		SPDK_ERRLOG("Cannot get the accel_channel for bdev nvme polling group=%p\n",
1125 			    group);
1126 		return -1;
1127 	}
1128 
1129 	group->poller = SPDK_POLLER_REGISTER(bdev_nvme_poll, group, g_opts.nvme_ioq_poll_period_us);
1130 
1131 	if (group->poller == NULL) {
1132 		spdk_put_io_channel(group->accel_channel);
1133 		spdk_nvme_poll_group_destroy(group->group);
1134 		return -1;
1135 	}
1136 
1137 	return 0;
1138 }
1139 
1140 static void
1141 bdev_nvme_poll_group_destroy_cb(void *io_device, void *ctx_buf)
1142 {
1143 	struct nvme_bdev_poll_group *group = ctx_buf;
1144 
1145 	if (group->accel_channel) {
1146 		spdk_put_io_channel(group->accel_channel);
1147 	}
1148 
1149 	spdk_poller_unregister(&group->poller);
1150 	if (spdk_nvme_poll_group_destroy(group->group)) {
1151 		SPDK_ERRLOG("Unable to destroy a poll group for the NVMe bdev module.\n");
1152 		assert(false);
1153 	}
1154 }
1155 
1156 static struct spdk_io_channel *
1157 bdev_nvme_get_io_channel(void *ctx)
1158 {
1159 	struct nvme_bdev *nvme_bdev = ctx;
1160 
1161 	return spdk_get_io_channel(nvme_bdev->nvme_ns->ctrlr);
1162 }
1163 
1164 static void *
1165 bdev_nvme_get_module_ctx(void *ctx)
1166 {
1167 	struct nvme_bdev *nvme_bdev = ctx;
1168 
1169 	return bdev_nvme_get_ctrlr(&nvme_bdev->disk);
1170 }
1171 
1172 static const char *
1173 _nvme_ana_state_str(enum spdk_nvme_ana_state ana_state)
1174 {
1175 	switch (ana_state) {
1176 	case SPDK_NVME_ANA_OPTIMIZED_STATE:
1177 		return "optimized";
1178 	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
1179 		return "non_optimized";
1180 	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
1181 		return "inaccessible";
1182 	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
1183 		return "persistent_loss";
1184 	case SPDK_NVME_ANA_CHANGE_STATE:
1185 		return "change";
1186 	default:
1187 		return NULL;
1188 	}
1189 }
1190 
1191 static int
1192 bdev_nvme_dump_info_json(void *ctx, struct spdk_json_write_ctx *w)
1193 {
1194 	struct nvme_bdev *nvme_bdev = ctx;
1195 	struct nvme_bdev_ns *nvme_ns;
1196 	struct spdk_nvme_ns *ns;
1197 	struct spdk_nvme_ctrlr *ctrlr;
1198 	const struct spdk_nvme_ctrlr_data *cdata;
1199 	const struct spdk_nvme_transport_id *trid;
1200 	union spdk_nvme_vs_register vs;
1201 	union spdk_nvme_csts_register csts;
1202 	char buf[128];
1203 
1204 	nvme_ns = nvme_bdev->nvme_ns;
1205 	assert(nvme_ns != NULL);
1206 	ns = nvme_ns->ns;
1207 	ctrlr = spdk_nvme_ns_get_ctrlr(ns);
1208 
1209 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
1210 	trid = spdk_nvme_ctrlr_get_transport_id(ctrlr);
1211 	vs = spdk_nvme_ctrlr_get_regs_vs(ctrlr);
1212 	csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
1213 
1214 	spdk_json_write_named_object_begin(w, "nvme");
1215 
1216 	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
1217 		spdk_json_write_named_string(w, "pci_address", trid->traddr);
1218 	}
1219 
1220 	spdk_json_write_named_object_begin(w, "trid");
1221 
1222 	nvme_bdev_dump_trid_json(trid, w);
1223 
1224 	spdk_json_write_object_end(w);
1225 
1226 #ifdef SPDK_CONFIG_NVME_CUSE
1227 	size_t cuse_name_size = 128;
1228 	char cuse_name[cuse_name_size];
1229 
1230 	int rc = spdk_nvme_cuse_get_ns_name(ctrlr, spdk_nvme_ns_get_id(ns),
1231 					    cuse_name, &cuse_name_size);
1232 	if (rc == 0) {
1233 		spdk_json_write_named_string(w, "cuse_device", cuse_name);
1234 	}
1235 #endif
1236 
1237 	spdk_json_write_named_object_begin(w, "ctrlr_data");
1238 
1239 	spdk_json_write_named_string_fmt(w, "vendor_id", "0x%04x", cdata->vid);
1240 
1241 	snprintf(buf, sizeof(cdata->mn) + 1, "%s", cdata->mn);
1242 	spdk_str_trim(buf);
1243 	spdk_json_write_named_string(w, "model_number", buf);
1244 
1245 	snprintf(buf, sizeof(cdata->sn) + 1, "%s", cdata->sn);
1246 	spdk_str_trim(buf);
1247 	spdk_json_write_named_string(w, "serial_number", buf);
1248 
1249 	snprintf(buf, sizeof(cdata->fr) + 1, "%s", cdata->fr);
1250 	spdk_str_trim(buf);
1251 	spdk_json_write_named_string(w, "firmware_revision", buf);
1252 
1253 	if (cdata->subnqn[0] != '\0') {
1254 		spdk_json_write_named_string(w, "subnqn", cdata->subnqn);
1255 	}
1256 
1257 	spdk_json_write_named_object_begin(w, "oacs");
1258 
1259 	spdk_json_write_named_uint32(w, "security", cdata->oacs.security);
1260 	spdk_json_write_named_uint32(w, "format", cdata->oacs.format);
1261 	spdk_json_write_named_uint32(w, "firmware", cdata->oacs.firmware);
1262 	spdk_json_write_named_uint32(w, "ns_manage", cdata->oacs.ns_manage);
1263 
1264 	spdk_json_write_object_end(w);
1265 
1266 	spdk_json_write_object_end(w);
1267 
1268 	spdk_json_write_named_object_begin(w, "vs");
1269 
1270 	spdk_json_write_name(w, "nvme_version");
1271 	if (vs.bits.ter) {
1272 		spdk_json_write_string_fmt(w, "%u.%u.%u", vs.bits.mjr, vs.bits.mnr, vs.bits.ter);
1273 	} else {
1274 		spdk_json_write_string_fmt(w, "%u.%u", vs.bits.mjr, vs.bits.mnr);
1275 	}
1276 
1277 	spdk_json_write_object_end(w);
1278 
1279 	spdk_json_write_named_object_begin(w, "csts");
1280 
1281 	spdk_json_write_named_uint32(w, "rdy", csts.bits.rdy);
1282 	spdk_json_write_named_uint32(w, "cfs", csts.bits.cfs);
1283 
1284 	spdk_json_write_object_end(w);
1285 
1286 	spdk_json_write_named_object_begin(w, "ns_data");
1287 
1288 	spdk_json_write_named_uint32(w, "id", spdk_nvme_ns_get_id(ns));
1289 
1290 	if (cdata->cmic.ana_reporting) {
1291 		spdk_json_write_named_string(w, "ana_state",
1292 					     _nvme_ana_state_str(spdk_nvme_ns_get_ana_state(ns)));
1293 	}
1294 
1295 	spdk_json_write_object_end(w);
1296 
1297 	if (cdata->oacs.security) {
1298 		spdk_json_write_named_object_begin(w, "security");
1299 
1300 		spdk_json_write_named_bool(w, "opal", nvme_bdev->opal);
1301 
1302 		spdk_json_write_object_end(w);
1303 	}
1304 
1305 	spdk_json_write_object_end(w);
1306 
1307 	return 0;
1308 }
1309 
1310 static void
1311 bdev_nvme_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1312 {
1313 	/* No config per bdev needed */
1314 }
1315 
1316 static uint64_t
1317 bdev_nvme_get_spin_time(struct spdk_io_channel *ch)
1318 {
1319 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(ch);
1320 	struct nvme_bdev_poll_group *group = io_path->group;
1321 	uint64_t spin_time;
1322 
1323 	if (!group || !group->collect_spin_stat) {
1324 		return 0;
1325 	}
1326 
1327 	if (group->end_ticks != 0) {
1328 		group->spin_ticks += (group->end_ticks - group->start_ticks);
1329 		group->end_ticks = 0;
1330 	}
1331 
1332 	spin_time = (group->spin_ticks * 1000000ULL) / spdk_get_ticks_hz();
1333 	group->start_ticks = 0;
1334 	group->spin_ticks = 0;
1335 
1336 	return spin_time;
1337 }
1338 
1339 static const struct spdk_bdev_fn_table nvmelib_fn_table = {
1340 	.destruct		= bdev_nvme_destruct,
1341 	.submit_request		= bdev_nvme_submit_request,
1342 	.io_type_supported	= bdev_nvme_io_type_supported,
1343 	.get_io_channel		= bdev_nvme_get_io_channel,
1344 	.dump_info_json		= bdev_nvme_dump_info_json,
1345 	.write_config_json	= bdev_nvme_write_config_json,
1346 	.get_spin_time		= bdev_nvme_get_spin_time,
1347 	.get_module_ctx		= bdev_nvme_get_module_ctx,
1348 };
1349 
1350 static int
1351 nvme_disk_create(struct spdk_bdev *disk, const char *base_name,
1352 		 struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns,
1353 		 uint32_t prchk_flags, void *ctx)
1354 {
1355 	const struct spdk_uuid		*uuid;
1356 	const struct spdk_nvme_ctrlr_data *cdata;
1357 	const struct spdk_nvme_ns_data	*nsdata;
1358 	int				rc;
1359 	enum spdk_nvme_csi		csi;
1360 
1361 	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
1362 	csi = spdk_nvme_ns_get_csi(ns);
1363 
1364 	switch (csi) {
1365 	case SPDK_NVME_CSI_NVM:
1366 		disk->product_name = "NVMe disk";
1367 		break;
1368 	case SPDK_NVME_CSI_ZNS:
1369 		disk->product_name = "NVMe ZNS disk";
1370 		disk->zoned = true;
1371 		disk->zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
1372 		disk->max_zone_append_size = spdk_nvme_zns_ctrlr_get_max_zone_append_size(ctrlr) /
1373 					     spdk_nvme_ns_get_extended_sector_size(ns);
1374 		disk->max_open_zones = spdk_nvme_zns_ns_get_max_open_zones(ns);
1375 		disk->max_active_zones = spdk_nvme_zns_ns_get_max_active_zones(ns);
1376 		break;
1377 	default:
1378 		SPDK_ERRLOG("unsupported CSI: %u\n", csi);
1379 		return -ENOTSUP;
1380 	}
1381 
1382 	disk->name = spdk_sprintf_alloc("%sn%d", base_name, spdk_nvme_ns_get_id(ns));
1383 	if (!disk->name) {
1384 		return -ENOMEM;
1385 	}
1386 
1387 	disk->write_cache = 0;
1388 	if (cdata->vwc.present) {
1389 		/* Enable if the Volatile Write Cache exists */
1390 		disk->write_cache = 1;
1391 	}
1392 	disk->blocklen = spdk_nvme_ns_get_extended_sector_size(ns);
1393 	disk->blockcnt = spdk_nvme_ns_get_num_sectors(ns);
1394 	disk->optimal_io_boundary = spdk_nvme_ns_get_optimal_io_boundary(ns);
1395 
1396 	uuid = spdk_nvme_ns_get_uuid(ns);
1397 	if (uuid != NULL) {
1398 		disk->uuid = *uuid;
1399 	}
1400 
1401 	nsdata = spdk_nvme_ns_get_data(ns);
1402 
1403 	disk->md_len = spdk_nvme_ns_get_md_size(ns);
1404 	if (disk->md_len != 0) {
1405 		disk->md_interleave = nsdata->flbas.extended;
1406 		disk->dif_type = (enum spdk_dif_type)spdk_nvme_ns_get_pi_type(ns);
1407 		if (disk->dif_type != SPDK_DIF_DISABLE) {
1408 			disk->dif_is_head_of_md = nsdata->dps.md_start;
1409 			disk->dif_check_flags = prchk_flags;
1410 		}
1411 	}
1412 
1413 	if (!(spdk_nvme_ctrlr_get_flags(ctrlr) &
1414 	      SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED)) {
1415 		disk->acwu = 0;
1416 	} else if (nsdata->nsfeat.ns_atomic_write_unit) {
1417 		disk->acwu = nsdata->nacwu;
1418 	} else {
1419 		disk->acwu = cdata->acwu;
1420 	}
1421 
1422 	disk->ctxt = ctx;
1423 	disk->fn_table = &nvmelib_fn_table;
1424 	disk->module = &nvme_if;
1425 	rc = spdk_bdev_register(disk);
1426 	if (rc) {
1427 		SPDK_ERRLOG("spdk_bdev_register() failed\n");
1428 		free(disk->name);
1429 		return rc;
1430 	}
1431 
1432 	return 0;
1433 }
1434 
1435 static int
1436 nvme_bdev_create(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, struct nvme_bdev_ns *nvme_ns)
1437 {
1438 	struct nvme_bdev *bdev;
1439 	int rc;
1440 
1441 	bdev = calloc(1, sizeof(*bdev));
1442 	if (!bdev) {
1443 		SPDK_ERRLOG("bdev calloc() failed\n");
1444 		return -ENOMEM;
1445 	}
1446 
1447 	bdev->nvme_ns = nvme_ns;
1448 	bdev->opal = nvme_bdev_ctrlr->opal_dev != NULL;
1449 
1450 	rc = nvme_disk_create(&bdev->disk, nvme_bdev_ctrlr->name, nvme_bdev_ctrlr->ctrlr,
1451 			      nvme_ns->ns, nvme_bdev_ctrlr->prchk_flags, bdev);
1452 	if (rc != 0) {
1453 		SPDK_ERRLOG("Failed to create NVMe disk\n");
1454 		free(bdev);
1455 		return rc;
1456 	}
1457 
1458 	nvme_ns->bdev = bdev;
1459 
1460 	return 0;
1461 }
1462 
1463 static bool
1464 bdev_nvme_compare_ns(struct spdk_nvme_ns *ns1, struct spdk_nvme_ns *ns2)
1465 {
1466 	const struct spdk_nvme_ns_data *nsdata1, *nsdata2;
1467 	const struct spdk_uuid *uuid1, *uuid2;
1468 
1469 	nsdata1 = spdk_nvme_ns_get_data(ns1);
1470 	nsdata2 = spdk_nvme_ns_get_data(ns2);
1471 	uuid1 = spdk_nvme_ns_get_uuid(ns1);
1472 	uuid2 = spdk_nvme_ns_get_uuid(ns2);
1473 
1474 	return memcmp(nsdata1->nguid, nsdata2->nguid, sizeof(nsdata1->nguid)) == 0 &&
1475 	       nsdata1->eui64 == nsdata2->eui64 &&
1476 	       uuid1 != NULL && uuid2 != NULL && spdk_uuid_compare(uuid1, uuid2) == 0;
1477 }
1478 
1479 static void
1480 nvme_ctrlr_populate_standard_namespace(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
1481 				       struct nvme_bdev_ns *nvme_ns, struct nvme_async_probe_ctx *ctx)
1482 {
1483 	struct spdk_nvme_ctrlr	*ctrlr = nvme_bdev_ctrlr->ctrlr;
1484 	struct spdk_nvme_ns	*ns;
1485 	int			rc = 0;
1486 
1487 	ns = spdk_nvme_ctrlr_get_ns(ctrlr, nvme_ns->id);
1488 	if (!ns) {
1489 		SPDK_DEBUGLOG(bdev_nvme, "Invalid NS %d\n", nvme_ns->id);
1490 		rc = -EINVAL;
1491 		goto done;
1492 	}
1493 
1494 	nvme_ns->ns = ns;
1495 	nvme_ns->populated = true;
1496 
1497 	rc = nvme_bdev_create(nvme_bdev_ctrlr, nvme_ns);
1498 done:
1499 	nvme_ctrlr_populate_namespace_done(ctx, nvme_ns, rc);
1500 }
1501 
1502 static bool
1503 hotplug_probe_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
1504 		 struct spdk_nvme_ctrlr_opts *opts)
1505 {
1506 	struct nvme_probe_skip_entry *entry;
1507 
1508 	TAILQ_FOREACH(entry, &g_skipped_nvme_ctrlrs, tailq) {
1509 		if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
1510 			return false;
1511 		}
1512 	}
1513 
1514 	opts->arbitration_burst = (uint8_t)g_opts.arbitration_burst;
1515 	opts->low_priority_weight = (uint8_t)g_opts.low_priority_weight;
1516 	opts->medium_priority_weight = (uint8_t)g_opts.medium_priority_weight;
1517 	opts->high_priority_weight = (uint8_t)g_opts.high_priority_weight;
1518 
1519 	SPDK_DEBUGLOG(bdev_nvme, "Attaching to %s\n", trid->traddr);
1520 
1521 	return true;
1522 }
1523 
1524 static void
1525 nvme_abort_cpl(void *ctx, const struct spdk_nvme_cpl *cpl)
1526 {
1527 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = ctx;
1528 
1529 	if (spdk_nvme_cpl_is_error(cpl)) {
1530 		SPDK_WARNLOG("Abort failed. Resetting controller. sc is %u, sct is %u.\n", cpl->status.sc,
1531 			     cpl->status.sct);
1532 		_bdev_nvme_reset(nvme_bdev_ctrlr);
1533 	}
1534 }
1535 
1536 static void
1537 timeout_cb(void *cb_arg, struct spdk_nvme_ctrlr *ctrlr,
1538 	   struct spdk_nvme_qpair *qpair, uint16_t cid)
1539 {
1540 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = cb_arg;
1541 	union spdk_nvme_csts_register csts;
1542 	int rc;
1543 
1544 	assert(nvme_bdev_ctrlr->ctrlr == ctrlr);
1545 
1546 	SPDK_WARNLOG("Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n", ctrlr, qpair, cid);
1547 
1548 	/* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O
1549 	 * queue.  (Note: qpair == NULL when there's an admin cmd timeout.)  Otherwise we
1550 	 * would submit another fabrics cmd on the admin queue to read CSTS and check for its
1551 	 * completion recursively.
1552 	 */
1553 	if (nvme_bdev_ctrlr->connected_trid->trtype == SPDK_NVME_TRANSPORT_PCIE || qpair != NULL) {
1554 		csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
1555 		if (csts.bits.cfs) {
1556 			SPDK_ERRLOG("Controller Fatal Status, reset required\n");
1557 			_bdev_nvme_reset(nvme_bdev_ctrlr);
1558 			return;
1559 		}
1560 	}
1561 
1562 	switch (g_opts.action_on_timeout) {
1563 	case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT:
1564 		if (qpair) {
1565 			/* Don't send abort to ctrlr when reset is running. */
1566 			pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
1567 			if (nvme_bdev_ctrlr->resetting) {
1568 				pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
1569 				SPDK_NOTICELOG("Quit abort. Ctrlr is in the process of reseting.\n");
1570 				return;
1571 			}
1572 			pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
1573 
1574 			rc = spdk_nvme_ctrlr_cmd_abort(ctrlr, qpair, cid,
1575 						       nvme_abort_cpl, nvme_bdev_ctrlr);
1576 			if (rc == 0) {
1577 				return;
1578 			}
1579 
1580 			SPDK_ERRLOG("Unable to send abort. Resetting, rc is %d.\n", rc);
1581 		}
1582 
1583 	/* FALLTHROUGH */
1584 	case SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET:
1585 		_bdev_nvme_reset(nvme_bdev_ctrlr);
1586 		break;
1587 	case SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE:
1588 		SPDK_DEBUGLOG(bdev_nvme, "No action for nvme controller timeout.\n");
1589 		break;
1590 	default:
1591 		SPDK_ERRLOG("An invalid timeout action value is found.\n");
1592 		break;
1593 	}
1594 }
1595 
1596 static void
1597 nvme_ctrlr_depopulate_standard_namespace(struct nvme_bdev_ns *nvme_ns)
1598 {
1599 	struct nvme_bdev *bdev;
1600 
1601 	bdev = nvme_ns->bdev;
1602 	if (bdev != NULL) {
1603 		spdk_bdev_unregister(&bdev->disk, NULL, NULL);
1604 	}
1605 
1606 	nvme_ctrlr_depopulate_namespace_done(nvme_ns);
1607 }
1608 
1609 static void
1610 nvme_ctrlr_populate_namespace(struct nvme_bdev_ctrlr *ctrlr, struct nvme_bdev_ns *nvme_ns,
1611 			      struct nvme_async_probe_ctx *ctx)
1612 {
1613 	g_populate_namespace_fn[nvme_ns->type](ctrlr, nvme_ns, ctx);
1614 }
1615 
1616 static void
1617 nvme_ctrlr_depopulate_namespace(struct nvme_bdev_ctrlr *ctrlr, struct nvme_bdev_ns *nvme_ns)
1618 {
1619 	g_depopulate_namespace_fn[nvme_ns->type](nvme_ns);
1620 }
1621 
1622 void
1623 nvme_ctrlr_populate_namespace_done(struct nvme_async_probe_ctx *ctx,
1624 				   struct nvme_bdev_ns *nvme_ns, int rc)
1625 {
1626 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = nvme_ns->ctrlr;
1627 
1628 	assert(nvme_bdev_ctrlr != NULL);
1629 
1630 	if (rc == 0) {
1631 		pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
1632 		nvme_bdev_ctrlr->ref++;
1633 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
1634 	} else {
1635 		memset(nvme_ns, 0, sizeof(*nvme_ns));
1636 	}
1637 
1638 	if (ctx) {
1639 		ctx->populates_in_progress--;
1640 		if (ctx->populates_in_progress == 0) {
1641 			nvme_ctrlr_populate_namespaces_done(nvme_bdev_ctrlr, ctx);
1642 		}
1643 	}
1644 }
1645 
1646 static void
1647 nvme_ctrlr_populate_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
1648 			       struct nvme_async_probe_ctx *ctx)
1649 {
1650 	struct spdk_nvme_ctrlr	*ctrlr = nvme_bdev_ctrlr->ctrlr;
1651 	struct nvme_bdev_ns	*nvme_ns;
1652 	struct spdk_nvme_ns	*ns;
1653 	struct nvme_bdev	*bdev;
1654 	uint32_t		i;
1655 	int			rc;
1656 	uint64_t		num_sectors;
1657 	bool			ns_is_active;
1658 
1659 	if (ctx) {
1660 		/* Initialize this count to 1 to handle the populate functions
1661 		 * calling nvme_ctrlr_populate_namespace_done() immediately.
1662 		 */
1663 		ctx->populates_in_progress = 1;
1664 	}
1665 
1666 	for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) {
1667 		uint32_t	nsid = i + 1;
1668 
1669 		nvme_ns = nvme_bdev_ctrlr->namespaces[i];
1670 		ns_is_active = spdk_nvme_ctrlr_is_active_ns(ctrlr, nsid);
1671 
1672 		if (nvme_ns->populated && ns_is_active && nvme_ns->type == NVME_BDEV_NS_STANDARD) {
1673 			/* NS is still there but attributes may have changed */
1674 			ns = spdk_nvme_ctrlr_get_ns(ctrlr, nsid);
1675 			num_sectors = spdk_nvme_ns_get_num_sectors(ns);
1676 			bdev = nvme_ns->bdev;
1677 			assert(bdev != NULL);
1678 			if (bdev->disk.blockcnt != num_sectors) {
1679 				SPDK_NOTICELOG("NSID %u is resized: bdev name %s, old size %" PRIu64 ", new size %" PRIu64 "\n",
1680 					       nsid,
1681 					       bdev->disk.name,
1682 					       bdev->disk.blockcnt,
1683 					       num_sectors);
1684 				rc = spdk_bdev_notify_blockcnt_change(&bdev->disk, num_sectors);
1685 				if (rc != 0) {
1686 					SPDK_ERRLOG("Could not change num blocks for nvme bdev: name %s, errno: %d.\n",
1687 						    bdev->disk.name, rc);
1688 				}
1689 			}
1690 		}
1691 
1692 		if (!nvme_ns->populated && ns_is_active) {
1693 			nvme_ns->id = nsid;
1694 			nvme_ns->ctrlr = nvme_bdev_ctrlr;
1695 			if (spdk_nvme_ctrlr_is_ocssd_supported(ctrlr)) {
1696 				nvme_ns->type = NVME_BDEV_NS_OCSSD;
1697 			} else {
1698 				nvme_ns->type = NVME_BDEV_NS_STANDARD;
1699 			}
1700 
1701 			nvme_ns->bdev = NULL;
1702 
1703 			if (ctx) {
1704 				ctx->populates_in_progress++;
1705 			}
1706 			nvme_ctrlr_populate_namespace(nvme_bdev_ctrlr, nvme_ns, ctx);
1707 		}
1708 
1709 		if (nvme_ns->populated && !ns_is_active) {
1710 			nvme_ctrlr_depopulate_namespace(nvme_bdev_ctrlr, nvme_ns);
1711 		}
1712 	}
1713 
1714 	if (ctx) {
1715 		/* Decrement this count now that the loop is over to account
1716 		 * for the one we started with.  If the count is then 0, we
1717 		 * know any populate_namespace functions completed immediately,
1718 		 * so we'll kick the callback here.
1719 		 */
1720 		ctx->populates_in_progress--;
1721 		if (ctx->populates_in_progress == 0) {
1722 			nvme_ctrlr_populate_namespaces_done(nvme_bdev_ctrlr, ctx);
1723 		}
1724 	}
1725 
1726 }
1727 
1728 static void
1729 nvme_ctrlr_depopulate_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr)
1730 {
1731 	uint32_t i;
1732 	struct nvme_bdev_ns *nvme_ns;
1733 
1734 	for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) {
1735 		uint32_t nsid = i + 1;
1736 
1737 		nvme_ns = nvme_bdev_ctrlr->namespaces[nsid - 1];
1738 		if (nvme_ns->populated) {
1739 			assert(nvme_ns->id == nsid);
1740 			nvme_ctrlr_depopulate_namespace(nvme_bdev_ctrlr, nvme_ns);
1741 		}
1742 	}
1743 }
1744 
1745 static void
1746 aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
1747 {
1748 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr		= arg;
1749 	union spdk_nvme_async_event_completion	event;
1750 
1751 	if (spdk_nvme_cpl_is_error(cpl)) {
1752 		SPDK_WARNLOG("AER request execute failed");
1753 		return;
1754 	}
1755 
1756 	event.raw = cpl->cdw0;
1757 	if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
1758 	    (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) {
1759 		nvme_ctrlr_populate_namespaces(nvme_bdev_ctrlr, NULL);
1760 	} else if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_VENDOR) &&
1761 		   (event.bits.log_page_identifier == SPDK_OCSSD_LOG_CHUNK_NOTIFICATION) &&
1762 		   spdk_nvme_ctrlr_is_ocssd_supported(nvme_bdev_ctrlr->ctrlr)) {
1763 		bdev_ocssd_handle_chunk_notification(nvme_bdev_ctrlr);
1764 	}
1765 }
1766 
1767 static void
1768 populate_namespaces_cb(struct nvme_async_probe_ctx *ctx, size_t count, int rc)
1769 {
1770 	if (ctx->cb_fn) {
1771 		ctx->cb_fn(ctx->cb_ctx, count, rc);
1772 	}
1773 
1774 	ctx->namespaces_populated = true;
1775 	if (ctx->probe_done) {
1776 		/* The probe was already completed, so we need to free the context
1777 		 * here.  This can happen for cases like OCSSD, where we need to
1778 		 * send additional commands to the SSD after attach.
1779 		 */
1780 		free(ctx);
1781 	}
1782 }
1783 
1784 static int
1785 _nvme_bdev_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr,
1786 			const char *name,
1787 			const struct spdk_nvme_transport_id *trid,
1788 			uint32_t prchk_flags,
1789 			struct nvme_bdev_ctrlr **_nvme_bdev_ctrlr)
1790 {
1791 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr;
1792 	struct nvme_bdev_ctrlr_trid *trid_entry;
1793 	uint32_t i;
1794 	int rc;
1795 
1796 	nvme_bdev_ctrlr = calloc(1, sizeof(*nvme_bdev_ctrlr));
1797 	if (nvme_bdev_ctrlr == NULL) {
1798 		SPDK_ERRLOG("Failed to allocate device struct\n");
1799 		return -ENOMEM;
1800 	}
1801 
1802 	rc = pthread_mutex_init(&nvme_bdev_ctrlr->mutex, NULL);
1803 	if (rc != 0) {
1804 		goto err_init_mutex;
1805 	}
1806 
1807 	TAILQ_INIT(&nvme_bdev_ctrlr->trids);
1808 	nvme_bdev_ctrlr->num_ns = spdk_nvme_ctrlr_get_num_ns(ctrlr);
1809 	if (nvme_bdev_ctrlr->num_ns != 0) {
1810 		nvme_bdev_ctrlr->namespaces = calloc(nvme_bdev_ctrlr->num_ns, sizeof(struct nvme_bdev_ns *));
1811 		if (!nvme_bdev_ctrlr->namespaces) {
1812 			SPDK_ERRLOG("Failed to allocate block namespaces pointer\n");
1813 			rc = -ENOMEM;
1814 			goto err_alloc_namespaces;
1815 		}
1816 	}
1817 
1818 	trid_entry = calloc(1, sizeof(*trid_entry));
1819 	if (trid_entry == NULL) {
1820 		SPDK_ERRLOG("Failed to allocate trid entry pointer\n");
1821 		rc = -ENOMEM;
1822 		goto err_alloc_trid;
1823 	}
1824 
1825 	trid_entry->trid = *trid;
1826 
1827 	for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) {
1828 		nvme_bdev_ctrlr->namespaces[i] = calloc(1, sizeof(struct nvme_bdev_ns));
1829 		if (nvme_bdev_ctrlr->namespaces[i] == NULL) {
1830 			SPDK_ERRLOG("Failed to allocate block namespace struct\n");
1831 			rc = -ENOMEM;
1832 			goto err_alloc_namespace;
1833 		}
1834 	}
1835 
1836 	nvme_bdev_ctrlr->thread = spdk_get_thread();
1837 	nvme_bdev_ctrlr->adminq_timer_poller = NULL;
1838 	nvme_bdev_ctrlr->ctrlr = ctrlr;
1839 	nvme_bdev_ctrlr->ref = 1;
1840 	nvme_bdev_ctrlr->connected_trid = &trid_entry->trid;
1841 	nvme_bdev_ctrlr->name = strdup(name);
1842 	if (nvme_bdev_ctrlr->name == NULL) {
1843 		rc = -ENOMEM;
1844 		goto err_alloc_name;
1845 	}
1846 
1847 	if (spdk_nvme_ctrlr_is_ocssd_supported(nvme_bdev_ctrlr->ctrlr)) {
1848 		rc = bdev_ocssd_init_ctrlr(nvme_bdev_ctrlr);
1849 		if (spdk_unlikely(rc != 0)) {
1850 			SPDK_ERRLOG("Unable to initialize OCSSD controller\n");
1851 			goto err_init_ocssd;
1852 		}
1853 	}
1854 
1855 	nvme_bdev_ctrlr->prchk_flags = prchk_flags;
1856 
1857 	spdk_io_device_register(nvme_bdev_ctrlr,
1858 				bdev_nvme_create_path_cb,
1859 				bdev_nvme_destroy_path_cb,
1860 				sizeof(struct nvme_io_path),
1861 				name);
1862 
1863 	nvme_bdev_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq, nvme_bdev_ctrlr,
1864 					       g_opts.nvme_adminq_poll_period_us);
1865 
1866 	TAILQ_INSERT_TAIL(&g_nvme_bdev_ctrlrs, nvme_bdev_ctrlr, tailq);
1867 
1868 	if (g_opts.timeout_us > 0) {
1869 		spdk_nvme_ctrlr_register_timeout_callback(ctrlr, g_opts.timeout_us,
1870 				timeout_cb, nvme_bdev_ctrlr);
1871 	}
1872 
1873 	spdk_nvme_ctrlr_register_aer_callback(ctrlr, aer_cb, nvme_bdev_ctrlr);
1874 	spdk_nvme_ctrlr_set_remove_cb(ctrlr, remove_cb, nvme_bdev_ctrlr);
1875 
1876 	if (spdk_nvme_ctrlr_get_flags(nvme_bdev_ctrlr->ctrlr) &
1877 	    SPDK_NVME_CTRLR_SECURITY_SEND_RECV_SUPPORTED) {
1878 		nvme_bdev_ctrlr->opal_dev = spdk_opal_dev_construct(nvme_bdev_ctrlr->ctrlr);
1879 	}
1880 
1881 	TAILQ_INSERT_HEAD(&nvme_bdev_ctrlr->trids, trid_entry, link);
1882 
1883 	if (_nvme_bdev_ctrlr != NULL) {
1884 		*_nvme_bdev_ctrlr = nvme_bdev_ctrlr;
1885 	}
1886 	return 0;
1887 
1888 err_init_ocssd:
1889 	free(nvme_bdev_ctrlr->name);
1890 err_alloc_name:
1891 err_alloc_namespace:
1892 	for (; i > 0; i--) {
1893 		free(nvme_bdev_ctrlr->namespaces[i - 1]);
1894 	}
1895 	free(trid_entry);
1896 err_alloc_trid:
1897 	free(nvme_bdev_ctrlr->namespaces);
1898 err_alloc_namespaces:
1899 	pthread_mutex_destroy(&nvme_bdev_ctrlr->mutex);
1900 err_init_mutex:
1901 	free(nvme_bdev_ctrlr);
1902 	return rc;
1903 }
1904 
1905 static void
1906 nvme_bdev_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr,
1907 		       const char *name,
1908 		       const struct spdk_nvme_transport_id *trid,
1909 		       uint32_t prchk_flags,
1910 		       struct nvme_async_probe_ctx *ctx)
1911 {
1912 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = NULL;
1913 	int rc;
1914 
1915 	rc = _nvme_bdev_ctrlr_create(ctrlr, name, trid, prchk_flags, &nvme_bdev_ctrlr);
1916 	if (rc != 0) {
1917 		SPDK_ERRLOG("Failed to create new NVMe controller\n");
1918 		goto err;
1919 	}
1920 
1921 	nvme_ctrlr_populate_namespaces(nvme_bdev_ctrlr, ctx);
1922 	return;
1923 
1924 err:
1925 	if (ctx != NULL) {
1926 		populate_namespaces_cb(ctx, 0, rc);
1927 	}
1928 }
1929 
1930 static void
1931 attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
1932 	  struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
1933 {
1934 	struct nvme_probe_ctx *ctx = cb_ctx;
1935 	char *name = NULL;
1936 	uint32_t prchk_flags = 0;
1937 	size_t i;
1938 
1939 	if (ctx) {
1940 		for (i = 0; i < ctx->count; i++) {
1941 			if (spdk_nvme_transport_id_compare(trid, &ctx->trids[i]) == 0) {
1942 				prchk_flags = ctx->prchk_flags[i];
1943 				name = strdup(ctx->names[i]);
1944 				break;
1945 			}
1946 		}
1947 	} else {
1948 		name = spdk_sprintf_alloc("HotInNvme%d", g_hot_insert_nvme_controller_index++);
1949 	}
1950 	if (!name) {
1951 		SPDK_ERRLOG("Failed to assign name to NVMe device\n");
1952 		return;
1953 	}
1954 
1955 	SPDK_DEBUGLOG(bdev_nvme, "Attached to %s (%s)\n", trid->traddr, name);
1956 
1957 	nvme_bdev_ctrlr_create(ctrlr, name, trid, prchk_flags, NULL);
1958 
1959 	free(name);
1960 }
1961 
1962 static void
1963 _nvme_bdev_ctrlr_destruct(void *ctx)
1964 {
1965 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = ctx;
1966 
1967 	nvme_ctrlr_depopulate_namespaces(nvme_bdev_ctrlr);
1968 	nvme_bdev_ctrlr_destruct(nvme_bdev_ctrlr);
1969 }
1970 
1971 static int
1972 _bdev_nvme_delete(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool hotplug)
1973 {
1974 	struct nvme_probe_skip_entry *entry;
1975 
1976 	pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
1977 
1978 	/* The controller's destruction was already started */
1979 	if (nvme_bdev_ctrlr->destruct) {
1980 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
1981 		return 0;
1982 	}
1983 
1984 	if (!hotplug &&
1985 	    nvme_bdev_ctrlr->connected_trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
1986 		entry = calloc(1, sizeof(*entry));
1987 		if (!entry) {
1988 			pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
1989 			return -ENOMEM;
1990 		}
1991 		entry->trid = *nvme_bdev_ctrlr->connected_trid;
1992 		TAILQ_INSERT_TAIL(&g_skipped_nvme_ctrlrs, entry, tailq);
1993 	}
1994 
1995 	nvme_bdev_ctrlr->destruct = true;
1996 	pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
1997 
1998 	_nvme_bdev_ctrlr_destruct(nvme_bdev_ctrlr);
1999 
2000 	return 0;
2001 }
2002 
2003 static void
2004 remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr)
2005 {
2006 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = cb_ctx;
2007 
2008 	_bdev_nvme_delete(nvme_bdev_ctrlr, true);
2009 }
2010 
2011 static int
2012 bdev_nvme_hotplug_probe(void *arg)
2013 {
2014 	if (g_hotplug_probe_ctx == NULL) {
2015 		spdk_poller_unregister(&g_hotplug_probe_poller);
2016 		return SPDK_POLLER_IDLE;
2017 	}
2018 
2019 	if (spdk_nvme_probe_poll_async(g_hotplug_probe_ctx) != -EAGAIN) {
2020 		g_hotplug_probe_ctx = NULL;
2021 		spdk_poller_unregister(&g_hotplug_probe_poller);
2022 	}
2023 
2024 	return SPDK_POLLER_BUSY;
2025 }
2026 
2027 static int
2028 bdev_nvme_hotplug(void *arg)
2029 {
2030 	struct spdk_nvme_transport_id trid_pcie;
2031 
2032 	if (g_hotplug_probe_ctx) {
2033 		return SPDK_POLLER_BUSY;
2034 	}
2035 
2036 	memset(&trid_pcie, 0, sizeof(trid_pcie));
2037 	spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);
2038 
2039 	g_hotplug_probe_ctx = spdk_nvme_probe_async(&trid_pcie, NULL,
2040 			      hotplug_probe_cb, attach_cb, NULL);
2041 
2042 	if (g_hotplug_probe_ctx) {
2043 		assert(g_hotplug_probe_poller == NULL);
2044 		g_hotplug_probe_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug_probe, NULL, 1000);
2045 	}
2046 
2047 	return SPDK_POLLER_BUSY;
2048 }
2049 
2050 void
2051 bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts)
2052 {
2053 	*opts = g_opts;
2054 }
2055 
2056 int
2057 bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts)
2058 {
2059 	if (g_bdev_nvme_init_thread != NULL) {
2060 		if (!TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
2061 			return -EPERM;
2062 		}
2063 	}
2064 
2065 	g_opts = *opts;
2066 
2067 	return 0;
2068 }
2069 
2070 struct set_nvme_hotplug_ctx {
2071 	uint64_t period_us;
2072 	bool enabled;
2073 	spdk_msg_fn fn;
2074 	void *fn_ctx;
2075 };
2076 
2077 static void
2078 set_nvme_hotplug_period_cb(void *_ctx)
2079 {
2080 	struct set_nvme_hotplug_ctx *ctx = _ctx;
2081 
2082 	spdk_poller_unregister(&g_hotplug_poller);
2083 	if (ctx->enabled) {
2084 		g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug, NULL, ctx->period_us);
2085 	}
2086 
2087 	g_nvme_hotplug_poll_period_us = ctx->period_us;
2088 	g_nvme_hotplug_enabled = ctx->enabled;
2089 	if (ctx->fn) {
2090 		ctx->fn(ctx->fn_ctx);
2091 	}
2092 
2093 	free(ctx);
2094 }
2095 
2096 int
2097 bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx)
2098 {
2099 	struct set_nvme_hotplug_ctx *ctx;
2100 
2101 	if (enabled == true && !spdk_process_is_primary()) {
2102 		return -EPERM;
2103 	}
2104 
2105 	ctx = calloc(1, sizeof(*ctx));
2106 	if (ctx == NULL) {
2107 		return -ENOMEM;
2108 	}
2109 
2110 	period_us = period_us == 0 ? NVME_HOTPLUG_POLL_PERIOD_DEFAULT : period_us;
2111 	ctx->period_us = spdk_min(period_us, NVME_HOTPLUG_POLL_PERIOD_MAX);
2112 	ctx->enabled = enabled;
2113 	ctx->fn = cb;
2114 	ctx->fn_ctx = cb_ctx;
2115 
2116 	spdk_thread_send_msg(g_bdev_nvme_init_thread, set_nvme_hotplug_period_cb, ctx);
2117 	return 0;
2118 }
2119 
2120 static void
2121 nvme_ctrlr_populate_namespaces_done(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
2122 				    struct nvme_async_probe_ctx *ctx)
2123 {
2124 	struct nvme_bdev_ns	*nvme_ns;
2125 	struct nvme_bdev	*nvme_bdev;
2126 	uint32_t		i, nsid;
2127 	size_t			j;
2128 
2129 	assert(nvme_bdev_ctrlr != NULL);
2130 
2131 	/*
2132 	 * Report the new bdevs that were created in this call.
2133 	 * There can be more than one bdev per NVMe controller.
2134 	 */
2135 	j = 0;
2136 	for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) {
2137 		nsid = i + 1;
2138 		nvme_ns = nvme_bdev_ctrlr->namespaces[nsid - 1];
2139 		if (!nvme_ns->populated) {
2140 			continue;
2141 		}
2142 		assert(nvme_ns->id == nsid);
2143 		nvme_bdev = nvme_ns->bdev;
2144 		if (nvme_bdev == NULL) {
2145 			assert(nvme_ns->type == NVME_BDEV_NS_OCSSD);
2146 			continue;
2147 		}
2148 		if (j < ctx->count) {
2149 			ctx->names[j] = nvme_bdev->disk.name;
2150 			j++;
2151 		} else {
2152 			SPDK_ERRLOG("Maximum number of namespaces supported per NVMe controller is %du. Unable to return all names of created bdevs\n",
2153 				    ctx->count);
2154 			populate_namespaces_cb(ctx, 0, -ERANGE);
2155 			return;
2156 		}
2157 	}
2158 
2159 	populate_namespaces_cb(ctx, j, 0);
2160 }
2161 
2162 static int
2163 bdev_nvme_compare_trids(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
2164 			struct spdk_nvme_ctrlr *new_ctrlr,
2165 			struct spdk_nvme_transport_id *trid)
2166 {
2167 	struct nvme_bdev_ctrlr_trid *tmp_trid;
2168 
2169 	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
2170 		SPDK_ERRLOG("PCIe failover is not supported.\n");
2171 		return -ENOTSUP;
2172 	}
2173 
2174 	/* Currently we only support failover to the same transport type. */
2175 	if (nvme_bdev_ctrlr->connected_trid->trtype != trid->trtype) {
2176 		return -EINVAL;
2177 	}
2178 
2179 	/* Currently we only support failover to the same NQN. */
2180 	if (strncmp(trid->subnqn, nvme_bdev_ctrlr->connected_trid->subnqn, SPDK_NVMF_NQN_MAX_LEN)) {
2181 		return -EINVAL;
2182 	}
2183 
2184 	/* Skip all the other checks if we've already registered this path. */
2185 	TAILQ_FOREACH(tmp_trid, &nvme_bdev_ctrlr->trids, link) {
2186 		if (!spdk_nvme_transport_id_compare(&tmp_trid->trid, trid)) {
2187 			return -EEXIST;
2188 		}
2189 	}
2190 
2191 	return 0;
2192 }
2193 
2194 static int
2195 bdev_nvme_compare_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
2196 			     struct spdk_nvme_ctrlr *new_ctrlr)
2197 {
2198 	uint32_t i, nsid;
2199 	struct nvme_bdev_ns *nvme_ns;
2200 	struct spdk_nvme_ns *new_ns;
2201 
2202 	if (spdk_nvme_ctrlr_get_num_ns(new_ctrlr) != nvme_bdev_ctrlr->num_ns) {
2203 		return -EINVAL;
2204 	}
2205 
2206 	for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) {
2207 		nsid = i + 1;
2208 
2209 		nvme_ns = nvme_bdev_ctrlr->namespaces[i];
2210 		if (!nvme_ns->populated) {
2211 			continue;
2212 		}
2213 
2214 		new_ns = spdk_nvme_ctrlr_get_ns(new_ctrlr, nsid);
2215 		assert(new_ns != NULL);
2216 
2217 		if (!bdev_nvme_compare_ns(nvme_ns->ns, new_ns)) {
2218 			return -EINVAL;
2219 		}
2220 	}
2221 
2222 	return 0;
2223 }
2224 
2225 static int
2226 _bdev_nvme_add_secondary_trid(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
2227 			      struct spdk_nvme_transport_id *trid)
2228 {
2229 	struct nvme_bdev_ctrlr_trid *new_trid, *tmp_trid;
2230 
2231 	new_trid = calloc(1, sizeof(*new_trid));
2232 	if (new_trid == NULL) {
2233 		return -ENOMEM;
2234 	}
2235 	new_trid->trid = *trid;
2236 	new_trid->is_failed = false;
2237 
2238 	TAILQ_FOREACH(tmp_trid, &nvme_bdev_ctrlr->trids, link) {
2239 		if (tmp_trid->is_failed) {
2240 			TAILQ_INSERT_BEFORE(tmp_trid, new_trid, link);
2241 			return 0;
2242 		}
2243 	}
2244 
2245 	TAILQ_INSERT_TAIL(&nvme_bdev_ctrlr->trids, new_trid, link);
2246 	return 0;
2247 }
2248 
2249 /* This is the case that a secondary path is added to an existing
2250  * nvme_bdev_ctrlr for failover. After checking if it can access the same
2251  * namespaces as the primary path, it is disconnected until failover occurs.
2252  */
2253 static void
2254 bdev_nvme_add_secondary_trid(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
2255 			     struct spdk_nvme_ctrlr *new_ctrlr,
2256 			     struct spdk_nvme_transport_id *trid,
2257 			     struct nvme_async_probe_ctx *ctx)
2258 {
2259 	int rc;
2260 
2261 	assert(nvme_bdev_ctrlr != NULL);
2262 
2263 	pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
2264 
2265 	rc = bdev_nvme_compare_trids(nvme_bdev_ctrlr, new_ctrlr, trid);
2266 	if (rc != 0) {
2267 		goto exit;
2268 	}
2269 
2270 	rc = bdev_nvme_compare_namespaces(nvme_bdev_ctrlr, new_ctrlr);
2271 	if (rc != 0) {
2272 		goto exit;
2273 	}
2274 
2275 	rc = _bdev_nvme_add_secondary_trid(nvme_bdev_ctrlr, trid);
2276 
2277 exit:
2278 	pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
2279 
2280 	spdk_nvme_detach(new_ctrlr);
2281 
2282 	if (ctx != NULL) {
2283 		populate_namespaces_cb(ctx, 0, rc);
2284 	}
2285 }
2286 
2287 static void
2288 connect_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
2289 		  struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
2290 {
2291 	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
2292 	struct nvme_bdev_ctrlr	*nvme_bdev_ctrlr;
2293 	struct nvme_async_probe_ctx *ctx;
2294 
2295 	ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, opts);
2296 	ctx->ctrlr_attached = true;
2297 
2298 	nvme_bdev_ctrlr = nvme_bdev_ctrlr_get_by_name(ctx->base_name);
2299 	if (nvme_bdev_ctrlr) {
2300 		bdev_nvme_add_secondary_trid(nvme_bdev_ctrlr, ctrlr, &ctx->trid, ctx);
2301 		return;
2302 	}
2303 
2304 	nvme_bdev_ctrlr_create(ctrlr, ctx->base_name, &ctx->trid, ctx->prchk_flags, ctx);
2305 }
2306 
2307 static int
2308 bdev_nvme_async_poll(void *arg)
2309 {
2310 	struct nvme_async_probe_ctx	*ctx = arg;
2311 	int				rc;
2312 
2313 	rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
2314 	if (spdk_unlikely(rc != -EAGAIN)) {
2315 		ctx->probe_done = true;
2316 		spdk_poller_unregister(&ctx->poller);
2317 		if (!ctx->ctrlr_attached) {
2318 			/* The probe is done, but no controller was attached.
2319 			 * That means we had a failure, so report -EIO back to
2320 			 * the caller (usually the RPC). populate_namespaces_cb()
2321 			 * will take care of freeing the nvme_async_probe_ctx.
2322 			 */
2323 			populate_namespaces_cb(ctx, 0, -EIO);
2324 		} else if (ctx->namespaces_populated) {
2325 			/* The namespaces for the attached controller were all
2326 			 * populated and the response was already sent to the
2327 			 * caller (usually the RPC).  So free the context here.
2328 			 */
2329 			free(ctx);
2330 		}
2331 	}
2332 
2333 	return SPDK_POLLER_BUSY;
2334 }
2335 
2336 int
2337 bdev_nvme_create(struct spdk_nvme_transport_id *trid,
2338 		 struct spdk_nvme_host_id *hostid,
2339 		 const char *base_name,
2340 		 const char **names,
2341 		 uint32_t count,
2342 		 const char *hostnqn,
2343 		 uint32_t prchk_flags,
2344 		 spdk_bdev_create_nvme_fn cb_fn,
2345 		 void *cb_ctx,
2346 		 struct spdk_nvme_ctrlr_opts *opts)
2347 {
2348 	struct nvme_probe_skip_entry	*entry, *tmp;
2349 	struct nvme_async_probe_ctx	*ctx;
2350 
2351 	/* TODO expand this check to include both the host and target TRIDs.
2352 	 * Only if both are the same should we fail.
2353 	 */
2354 	if (nvme_bdev_ctrlr_get(trid) != NULL) {
2355 		SPDK_ERRLOG("A controller with the provided trid (traddr: %s) already exists.\n", trid->traddr);
2356 		return -EEXIST;
2357 	}
2358 
2359 	ctx = calloc(1, sizeof(*ctx));
2360 	if (!ctx) {
2361 		return -ENOMEM;
2362 	}
2363 	ctx->base_name = base_name;
2364 	ctx->names = names;
2365 	ctx->count = count;
2366 	ctx->cb_fn = cb_fn;
2367 	ctx->cb_ctx = cb_ctx;
2368 	ctx->prchk_flags = prchk_flags;
2369 	ctx->trid = *trid;
2370 
2371 	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
2372 		TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, tmp) {
2373 			if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
2374 				TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
2375 				free(entry);
2376 				break;
2377 			}
2378 		}
2379 	}
2380 
2381 	if (opts) {
2382 		memcpy(&ctx->opts, opts, sizeof(*opts));
2383 	} else {
2384 		spdk_nvme_ctrlr_get_default_ctrlr_opts(&ctx->opts, sizeof(ctx->opts));
2385 	}
2386 
2387 	ctx->opts.transport_retry_count = g_opts.retry_count;
2388 	ctx->opts.keep_alive_timeout_ms = g_opts.keep_alive_timeout_ms;
2389 
2390 	if (hostnqn) {
2391 		snprintf(ctx->opts.hostnqn, sizeof(ctx->opts.hostnqn), "%s", hostnqn);
2392 	}
2393 
2394 	if (hostid->hostaddr[0] != '\0') {
2395 		snprintf(ctx->opts.src_addr, sizeof(ctx->opts.src_addr), "%s", hostid->hostaddr);
2396 	}
2397 
2398 	if (hostid->hostsvcid[0] != '\0') {
2399 		snprintf(ctx->opts.src_svcid, sizeof(ctx->opts.src_svcid), "%s", hostid->hostsvcid);
2400 	}
2401 
2402 	ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->opts, connect_attach_cb);
2403 	if (ctx->probe_ctx == NULL) {
2404 		SPDK_ERRLOG("No controller was found with provided trid (traddr: %s)\n", trid->traddr);
2405 		free(ctx);
2406 		return -ENODEV;
2407 	}
2408 	ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_async_poll, ctx, 1000);
2409 
2410 	return 0;
2411 }
2412 
2413 static int
2414 bdev_nvme_delete_secondary_trid(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr,
2415 				const struct spdk_nvme_transport_id *trid)
2416 {
2417 	struct nvme_bdev_ctrlr_trid	*ctrlr_trid, *tmp_trid;
2418 
2419 	if (!spdk_nvme_transport_id_compare(trid, nvme_bdev_ctrlr->connected_trid)) {
2420 		return -EBUSY;
2421 	}
2422 
2423 	TAILQ_FOREACH_SAFE(ctrlr_trid, &nvme_bdev_ctrlr->trids, link, tmp_trid) {
2424 		if (!spdk_nvme_transport_id_compare(&ctrlr_trid->trid, trid)) {
2425 			TAILQ_REMOVE(&nvme_bdev_ctrlr->trids, ctrlr_trid, link);
2426 			free(ctrlr_trid);
2427 			return 0;
2428 		}
2429 	}
2430 
2431 	return -ENXIO;
2432 }
2433 
2434 int
2435 bdev_nvme_delete(const char *name, const struct spdk_nvme_transport_id *trid)
2436 {
2437 	struct nvme_bdev_ctrlr		*nvme_bdev_ctrlr;
2438 	struct nvme_bdev_ctrlr_trid	*ctrlr_trid;
2439 
2440 	if (name == NULL) {
2441 		return -EINVAL;
2442 	}
2443 
2444 	nvme_bdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
2445 	if (nvme_bdev_ctrlr == NULL) {
2446 		SPDK_ERRLOG("Failed to find NVMe controller\n");
2447 		return -ENODEV;
2448 	}
2449 
2450 	/* case 1: remove the controller itself. */
2451 	if (trid == NULL) {
2452 		return _bdev_nvme_delete(nvme_bdev_ctrlr, false);
2453 	}
2454 
2455 	/* case 2: we are currently using the path to be removed. */
2456 	if (!spdk_nvme_transport_id_compare(trid, nvme_bdev_ctrlr->connected_trid)) {
2457 		ctrlr_trid = TAILQ_FIRST(&nvme_bdev_ctrlr->trids);
2458 		assert(nvme_bdev_ctrlr->connected_trid == &ctrlr_trid->trid);
2459 		/* case 2A: the current path is the only path. */
2460 		if (!TAILQ_NEXT(ctrlr_trid, link)) {
2461 			return _bdev_nvme_delete(nvme_bdev_ctrlr, false);
2462 		}
2463 
2464 		/* case 2B: there is an alternative path. */
2465 		return bdev_nvme_failover(nvme_bdev_ctrlr, true);
2466 	}
2467 
2468 	/* case 3: We are not using the specified path. */
2469 	return bdev_nvme_delete_secondary_trid(nvme_bdev_ctrlr, trid);
2470 }
2471 
2472 static int
2473 bdev_nvme_library_init(void)
2474 {
2475 	g_bdev_nvme_init_thread = spdk_get_thread();
2476 
2477 	spdk_io_device_register(&g_nvme_bdev_ctrlrs, bdev_nvme_poll_group_create_cb,
2478 				bdev_nvme_poll_group_destroy_cb,
2479 				sizeof(struct nvme_bdev_poll_group),  "bdev_nvme_poll_groups");
2480 
2481 	return 0;
2482 }
2483 
2484 static void
2485 bdev_nvme_library_fini(void)
2486 {
2487 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, *tmp;
2488 	struct nvme_probe_skip_entry *entry, *entry_tmp;
2489 
2490 	spdk_poller_unregister(&g_hotplug_poller);
2491 	free(g_hotplug_probe_ctx);
2492 	g_hotplug_probe_ctx = NULL;
2493 
2494 	TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, entry_tmp) {
2495 		TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
2496 		free(entry);
2497 	}
2498 
2499 	pthread_mutex_lock(&g_bdev_nvme_mutex);
2500 	TAILQ_FOREACH_SAFE(nvme_bdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq, tmp) {
2501 		pthread_mutex_lock(&nvme_bdev_ctrlr->mutex);
2502 		if (nvme_bdev_ctrlr->destruct) {
2503 			/* This controller's destruction was already started
2504 			 * before the application started shutting down
2505 			 */
2506 			pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
2507 			continue;
2508 		}
2509 		nvme_bdev_ctrlr->destruct = true;
2510 		pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex);
2511 
2512 		spdk_thread_send_msg(nvme_bdev_ctrlr->thread, _nvme_bdev_ctrlr_destruct,
2513 				     nvme_bdev_ctrlr);
2514 	}
2515 
2516 	g_bdev_nvme_module_finish = true;
2517 	if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
2518 		pthread_mutex_unlock(&g_bdev_nvme_mutex);
2519 		spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
2520 		spdk_bdev_module_finish_done();
2521 		return;
2522 	}
2523 
2524 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
2525 }
2526 
2527 static void
2528 bdev_nvme_verify_pi_error(struct nvme_bdev_io *bio)
2529 {
2530 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2531 	struct spdk_bdev *bdev = bdev_io->bdev;
2532 	struct spdk_dif_ctx dif_ctx;
2533 	struct spdk_dif_error err_blk = {};
2534 	int rc;
2535 
2536 	rc = spdk_dif_ctx_init(&dif_ctx,
2537 			       bdev->blocklen, bdev->md_len, bdev->md_interleave,
2538 			       bdev->dif_is_head_of_md, bdev->dif_type, bdev->dif_check_flags,
2539 			       bdev_io->u.bdev.offset_blocks, 0, 0, 0, 0);
2540 	if (rc != 0) {
2541 		SPDK_ERRLOG("Initialization of DIF context failed\n");
2542 		return;
2543 	}
2544 
2545 	if (bdev->md_interleave) {
2546 		rc = spdk_dif_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
2547 				     bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
2548 	} else {
2549 		struct iovec md_iov = {
2550 			.iov_base	= bdev_io->u.bdev.md_buf,
2551 			.iov_len	= bdev_io->u.bdev.num_blocks * bdev->md_len,
2552 		};
2553 
2554 		rc = spdk_dix_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
2555 				     &md_iov, bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
2556 	}
2557 
2558 	if (rc != 0) {
2559 		SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
2560 			    err_blk.err_type, err_blk.err_offset);
2561 	} else {
2562 		SPDK_ERRLOG("Hardware reported PI error but SPDK could not find any.\n");
2563 	}
2564 }
2565 
2566 static void
2567 bdev_nvme_no_pi_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
2568 {
2569 	struct nvme_bdev_io *bio = ref;
2570 
2571 	if (spdk_nvme_cpl_is_success(cpl)) {
2572 		/* Run PI verification for read data buffer. */
2573 		bdev_nvme_verify_pi_error(bio);
2574 	}
2575 
2576 	/* Return original completion status */
2577 	bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
2578 }
2579 
2580 static void
2581 bdev_nvme_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
2582 {
2583 	struct nvme_bdev_io *bio = ref;
2584 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2585 	struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
2586 	struct nvme_io_path *io_path;
2587 	struct spdk_nvme_ns *ns;
2588 	struct spdk_nvme_qpair *qpair;
2589 	int ret;
2590 
2591 	if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
2592 		SPDK_ERRLOG("readv completed with PI error (sct=%d, sc=%d)\n",
2593 			    cpl->status.sct, cpl->status.sc);
2594 
2595 		/* Save completion status to use after verifying PI error. */
2596 		bio->cpl = *cpl;
2597 
2598 		io_path = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));
2599 
2600 		if (spdk_likely(bdev_nvme_find_io_path(nbdev, io_path, &ns, &qpair))) {
2601 			/* Read without PI checking to verify PI error. */
2602 			ret = bdev_nvme_no_pi_readv(ns,
2603 						    qpair,
2604 						    bio,
2605 						    bdev_io->u.bdev.iovs,
2606 						    bdev_io->u.bdev.iovcnt,
2607 						    bdev_io->u.bdev.md_buf,
2608 						    bdev_io->u.bdev.num_blocks,
2609 						    bdev_io->u.bdev.offset_blocks);
2610 			if (ret == 0) {
2611 				return;
2612 			}
2613 		}
2614 	}
2615 
2616 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2617 }
2618 
2619 static void
2620 bdev_nvme_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
2621 {
2622 	struct nvme_bdev_io *bio = ref;
2623 
2624 	if (spdk_nvme_cpl_is_pi_error(cpl)) {
2625 		SPDK_ERRLOG("writev completed with PI error (sct=%d, sc=%d)\n",
2626 			    cpl->status.sct, cpl->status.sc);
2627 		/* Run PI verification for write data buffer if PI error is detected. */
2628 		bdev_nvme_verify_pi_error(bio);
2629 	}
2630 
2631 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2632 }
2633 
2634 static void
2635 bdev_nvme_zone_appendv_done(void *ref, const struct spdk_nvme_cpl *cpl)
2636 {
2637 	struct nvme_bdev_io *bio = ref;
2638 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2639 
2640 	/* spdk_bdev_io_get_append_location() requires that the ALBA is stored in offset_blocks.
2641 	 * Additionally, offset_blocks has to be set before calling bdev_nvme_verify_pi_error().
2642 	 */
2643 	bdev_io->u.bdev.offset_blocks = *(uint64_t *)&cpl->cdw0;
2644 
2645 	if (spdk_nvme_cpl_is_pi_error(cpl)) {
2646 		SPDK_ERRLOG("zone append completed with PI error (sct=%d, sc=%d)\n",
2647 			    cpl->status.sct, cpl->status.sc);
2648 		/* Run PI verification for zone append data buffer if PI error is detected. */
2649 		bdev_nvme_verify_pi_error(bio);
2650 	}
2651 
2652 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2653 }
2654 
2655 static void
2656 bdev_nvme_comparev_done(void *ref, const struct spdk_nvme_cpl *cpl)
2657 {
2658 	struct nvme_bdev_io *bio = ref;
2659 
2660 	if (spdk_nvme_cpl_is_pi_error(cpl)) {
2661 		SPDK_ERRLOG("comparev completed with PI error (sct=%d, sc=%d)\n",
2662 			    cpl->status.sct, cpl->status.sc);
2663 		/* Run PI verification for compare data buffer if PI error is detected. */
2664 		bdev_nvme_verify_pi_error(bio);
2665 	}
2666 
2667 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2668 }
2669 
2670 static void
2671 bdev_nvme_comparev_and_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
2672 {
2673 	struct nvme_bdev_io *bio = ref;
2674 
2675 	/* Compare operation completion */
2676 	if ((cpl->cdw0 & 0xFF) == SPDK_NVME_OPC_COMPARE) {
2677 		/* Save compare result for write callback */
2678 		bio->cpl = *cpl;
2679 		return;
2680 	}
2681 
2682 	/* Write operation completion */
2683 	if (spdk_nvme_cpl_is_error(&bio->cpl)) {
2684 		/* If bio->cpl is already an error, it means the compare operation failed.  In that case,
2685 		 * complete the IO with the compare operation's status.
2686 		 */
2687 		if (!spdk_nvme_cpl_is_error(cpl)) {
2688 			SPDK_ERRLOG("Unexpected write success after compare failure.\n");
2689 		}
2690 
2691 		bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
2692 	} else {
2693 		bdev_nvme_io_complete_nvme_status(bio, cpl);
2694 	}
2695 }
2696 
2697 static void
2698 bdev_nvme_queued_done(void *ref, const struct spdk_nvme_cpl *cpl)
2699 {
2700 	struct nvme_bdev_io *bio = ref;
2701 
2702 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2703 }
2704 
2705 static int
2706 fill_zone_from_report(struct spdk_bdev_zone_info *info, struct spdk_nvme_zns_zone_desc *desc)
2707 {
2708 	switch (desc->zs) {
2709 	case SPDK_NVME_ZONE_STATE_EMPTY:
2710 		info->state = SPDK_BDEV_ZONE_STATE_EMPTY;
2711 		break;
2712 	case SPDK_NVME_ZONE_STATE_IOPEN:
2713 		info->state = SPDK_BDEV_ZONE_STATE_IMP_OPEN;
2714 		break;
2715 	case SPDK_NVME_ZONE_STATE_EOPEN:
2716 		info->state = SPDK_BDEV_ZONE_STATE_EXP_OPEN;
2717 		break;
2718 	case SPDK_NVME_ZONE_STATE_CLOSED:
2719 		info->state = SPDK_BDEV_ZONE_STATE_CLOSED;
2720 		break;
2721 	case SPDK_NVME_ZONE_STATE_RONLY:
2722 		info->state = SPDK_BDEV_ZONE_STATE_READ_ONLY;
2723 		break;
2724 	case SPDK_NVME_ZONE_STATE_FULL:
2725 		info->state = SPDK_BDEV_ZONE_STATE_FULL;
2726 		break;
2727 	case SPDK_NVME_ZONE_STATE_OFFLINE:
2728 		info->state = SPDK_BDEV_ZONE_STATE_OFFLINE;
2729 		break;
2730 	default:
2731 		SPDK_ERRLOG("Invalid zone state: %#x in zone report\n", desc->zs);
2732 		return -EIO;
2733 	}
2734 
2735 	info->zone_id = desc->zslba;
2736 	info->write_pointer = desc->wp;
2737 	info->capacity = desc->zcap;
2738 
2739 	return 0;
2740 }
2741 
2742 static void
2743 bdev_nvme_get_zone_info_done(void *ref, const struct spdk_nvme_cpl *cpl)
2744 {
2745 	struct nvme_bdev_io *bio = ref;
2746 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2747 	struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;
2748 	struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
2749 	struct nvme_io_path *io_path = spdk_io_channel_get_ctx(ch);
2750 	uint64_t zone_id = bdev_io->u.zone_mgmt.zone_id;
2751 	uint32_t zones_to_copy = bdev_io->u.zone_mgmt.num_zones;
2752 	struct spdk_bdev_zone_info *info = bdev_io->u.zone_mgmt.buf;
2753 	uint64_t max_zones_per_buf, i;
2754 	uint32_t zone_report_bufsize;
2755 	struct spdk_nvme_ns *ns;
2756 	struct spdk_nvme_qpair *qpair;
2757 	int ret;
2758 
2759 	if (spdk_nvme_cpl_is_error(cpl)) {
2760 		goto out_complete_io_nvme_cpl;
2761 	}
2762 
2763 	if (!bdev_nvme_find_io_path(nbdev, io_path, &ns, &qpair)) {
2764 		ret = -ENXIO;
2765 		goto out_complete_io_ret;
2766 	}
2767 
2768 	zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
2769 	max_zones_per_buf = (zone_report_bufsize - sizeof(*bio->zone_report_buf)) /
2770 			    sizeof(bio->zone_report_buf->descs[0]);
2771 
2772 	if (bio->zone_report_buf->nr_zones > max_zones_per_buf) {
2773 		ret = -EINVAL;
2774 		goto out_complete_io_ret;
2775 	}
2776 
2777 	if (!bio->zone_report_buf->nr_zones) {
2778 		ret = -EINVAL;
2779 		goto out_complete_io_ret;
2780 	}
2781 
2782 	for (i = 0; i < bio->zone_report_buf->nr_zones && bio->handled_zones < zones_to_copy; i++) {
2783 		ret = fill_zone_from_report(&info[bio->handled_zones],
2784 					    &bio->zone_report_buf->descs[i]);
2785 		if (ret) {
2786 			goto out_complete_io_ret;
2787 		}
2788 		bio->handled_zones++;
2789 	}
2790 
2791 	if (bio->handled_zones < zones_to_copy) {
2792 		uint64_t zone_size_lba = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
2793 		uint64_t slba = zone_id + (zone_size_lba * bio->handled_zones);
2794 
2795 		memset(bio->zone_report_buf, 0, zone_report_bufsize);
2796 		ret = spdk_nvme_zns_report_zones(ns, qpair,
2797 						 bio->zone_report_buf, zone_report_bufsize,
2798 						 slba, SPDK_NVME_ZRA_LIST_ALL, true,
2799 						 bdev_nvme_get_zone_info_done, bio);
2800 		if (!ret) {
2801 			return;
2802 		} else {
2803 			goto out_complete_io_ret;
2804 		}
2805 	}
2806 
2807 out_complete_io_nvme_cpl:
2808 	free(bio->zone_report_buf);
2809 	bio->zone_report_buf = NULL;
2810 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2811 	return;
2812 
2813 out_complete_io_ret:
2814 	free(bio->zone_report_buf);
2815 	bio->zone_report_buf = NULL;
2816 	bdev_nvme_io_complete(bio, ret);
2817 }
2818 
2819 static void
2820 bdev_nvme_zone_management_done(void *ref, const struct spdk_nvme_cpl *cpl)
2821 {
2822 	struct nvme_bdev_io *bio = ref;
2823 
2824 	bdev_nvme_io_complete_nvme_status(bio, cpl);
2825 }
2826 
2827 static void
2828 bdev_nvme_admin_passthru_completion(void *ctx)
2829 {
2830 	struct nvme_bdev_io *bio = ctx;
2831 
2832 	bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
2833 }
2834 
2835 static void
2836 bdev_nvme_abort_completion(void *ctx)
2837 {
2838 	struct nvme_bdev_io *bio = ctx;
2839 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
2840 
2841 	if (spdk_nvme_cpl_is_abort_success(&bio->cpl)) {
2842 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
2843 	} else {
2844 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2845 	}
2846 }
2847 
2848 static void
2849 bdev_nvme_abort_done(void *ref, const struct spdk_nvme_cpl *cpl)
2850 {
2851 	struct nvme_bdev_io *bio = ref;
2852 
2853 	bio->cpl = *cpl;
2854 	spdk_thread_send_msg(bio->orig_thread, bdev_nvme_abort_completion, bio);
2855 }
2856 
2857 static void
2858 bdev_nvme_admin_passthru_done(void *ref, const struct spdk_nvme_cpl *cpl)
2859 {
2860 	struct nvme_bdev_io *bio = ref;
2861 
2862 	bio->cpl = *cpl;
2863 	spdk_thread_send_msg(bio->orig_thread, bdev_nvme_admin_passthru_completion, bio);
2864 }
2865 
2866 static void
2867 bdev_nvme_queued_reset_sgl(void *ref, uint32_t sgl_offset)
2868 {
2869 	struct nvme_bdev_io *bio = ref;
2870 	struct iovec *iov;
2871 
2872 	bio->iov_offset = sgl_offset;
2873 	for (bio->iovpos = 0; bio->iovpos < bio->iovcnt; bio->iovpos++) {
2874 		iov = &bio->iovs[bio->iovpos];
2875 		if (bio->iov_offset < iov->iov_len) {
2876 			break;
2877 		}
2878 
2879 		bio->iov_offset -= iov->iov_len;
2880 	}
2881 }
2882 
2883 static int
2884 bdev_nvme_queued_next_sge(void *ref, void **address, uint32_t *length)
2885 {
2886 	struct nvme_bdev_io *bio = ref;
2887 	struct iovec *iov;
2888 
2889 	assert(bio->iovpos < bio->iovcnt);
2890 
2891 	iov = &bio->iovs[bio->iovpos];
2892 
2893 	*address = iov->iov_base;
2894 	*length = iov->iov_len;
2895 
2896 	if (bio->iov_offset) {
2897 		assert(bio->iov_offset <= iov->iov_len);
2898 		*address += bio->iov_offset;
2899 		*length -= bio->iov_offset;
2900 	}
2901 
2902 	bio->iov_offset += *length;
2903 	if (bio->iov_offset == iov->iov_len) {
2904 		bio->iovpos++;
2905 		bio->iov_offset = 0;
2906 	}
2907 
2908 	return 0;
2909 }
2910 
2911 static void
2912 bdev_nvme_queued_reset_fused_sgl(void *ref, uint32_t sgl_offset)
2913 {
2914 	struct nvme_bdev_io *bio = ref;
2915 	struct iovec *iov;
2916 
2917 	bio->fused_iov_offset = sgl_offset;
2918 	for (bio->fused_iovpos = 0; bio->fused_iovpos < bio->fused_iovcnt; bio->fused_iovpos++) {
2919 		iov = &bio->fused_iovs[bio->fused_iovpos];
2920 		if (bio->fused_iov_offset < iov->iov_len) {
2921 			break;
2922 		}
2923 
2924 		bio->fused_iov_offset -= iov->iov_len;
2925 	}
2926 }
2927 
2928 static int
2929 bdev_nvme_queued_next_fused_sge(void *ref, void **address, uint32_t *length)
2930 {
2931 	struct nvme_bdev_io *bio = ref;
2932 	struct iovec *iov;
2933 
2934 	assert(bio->fused_iovpos < bio->fused_iovcnt);
2935 
2936 	iov = &bio->fused_iovs[bio->fused_iovpos];
2937 
2938 	*address = iov->iov_base;
2939 	*length = iov->iov_len;
2940 
2941 	if (bio->fused_iov_offset) {
2942 		assert(bio->fused_iov_offset <= iov->iov_len);
2943 		*address += bio->fused_iov_offset;
2944 		*length -= bio->fused_iov_offset;
2945 	}
2946 
2947 	bio->fused_iov_offset += *length;
2948 	if (bio->fused_iov_offset == iov->iov_len) {
2949 		bio->fused_iovpos++;
2950 		bio->fused_iov_offset = 0;
2951 	}
2952 
2953 	return 0;
2954 }
2955 
2956 static int
2957 bdev_nvme_no_pi_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
2958 		      struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
2959 		      void *md, uint64_t lba_count, uint64_t lba)
2960 {
2961 	int rc;
2962 
2963 	SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 " without PI check\n",
2964 		      lba_count, lba);
2965 
2966 	bio->iovs = iov;
2967 	bio->iovcnt = iovcnt;
2968 	bio->iovpos = 0;
2969 	bio->iov_offset = 0;
2970 
2971 	rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count,
2972 					    bdev_nvme_no_pi_readv_done, bio, 0,
2973 					    bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
2974 					    md, 0, 0);
2975 
2976 	if (rc != 0 && rc != -ENOMEM) {
2977 		SPDK_ERRLOG("no_pi_readv failed: rc = %d\n", rc);
2978 	}
2979 	return rc;
2980 }
2981 
2982 static int
2983 bdev_nvme_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
2984 		struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
2985 		void *md, uint64_t lba_count, uint64_t lba, uint32_t flags)
2986 {
2987 	int rc;
2988 
2989 	SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 "\n",
2990 		      lba_count, lba);
2991 
2992 	bio->iovs = iov;
2993 	bio->iovcnt = iovcnt;
2994 	bio->iovpos = 0;
2995 	bio->iov_offset = 0;
2996 
2997 	if (iovcnt == 1) {
2998 		rc = spdk_nvme_ns_cmd_read_with_md(ns, qpair, iov[0].iov_base, md, lba,
2999 						   lba_count,
3000 						   bdev_nvme_readv_done, bio,
3001 						   flags,
3002 						   0, 0);
3003 	} else {
3004 		rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count,
3005 						    bdev_nvme_readv_done, bio, flags,
3006 						    bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
3007 						    md, 0, 0);
3008 	}
3009 
3010 	if (rc != 0 && rc != -ENOMEM) {
3011 		SPDK_ERRLOG("readv failed: rc = %d\n", rc);
3012 	}
3013 	return rc;
3014 }
3015 
3016 static int
3017 bdev_nvme_writev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3018 		 struct nvme_bdev_io *bio,
3019 		 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba,
3020 		 uint32_t flags)
3021 {
3022 	int rc;
3023 
3024 	SPDK_DEBUGLOG(bdev_nvme, "write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
3025 		      lba_count, lba);
3026 
3027 	bio->iovs = iov;
3028 	bio->iovcnt = iovcnt;
3029 	bio->iovpos = 0;
3030 	bio->iov_offset = 0;
3031 
3032 	if (iovcnt == 1) {
3033 		rc = spdk_nvme_ns_cmd_write_with_md(ns, qpair, iov[0].iov_base, md, lba,
3034 						    lba_count,
3035 						    bdev_nvme_writev_done, bio,
3036 						    flags,
3037 						    0, 0);
3038 	} else {
3039 		rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
3040 						     bdev_nvme_writev_done, bio, flags,
3041 						     bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
3042 						     md, 0, 0);
3043 	}
3044 
3045 	if (rc != 0 && rc != -ENOMEM) {
3046 		SPDK_ERRLOG("writev failed: rc = %d\n", rc);
3047 	}
3048 	return rc;
3049 }
3050 
3051 static int
3052 bdev_nvme_zone_appendv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3053 		       struct nvme_bdev_io *bio,
3054 		       struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t zslba,
3055 		       uint32_t flags)
3056 {
3057 	int rc;
3058 
3059 	SPDK_DEBUGLOG(bdev_nvme, "zone append %" PRIu64 " blocks to zone start lba %#" PRIx64 "\n",
3060 		      lba_count, zslba);
3061 
3062 	bio->iovs = iov;
3063 	bio->iovcnt = iovcnt;
3064 	bio->iovpos = 0;
3065 	bio->iov_offset = 0;
3066 
3067 	if (iovcnt == 1) {
3068 		rc = spdk_nvme_zns_zone_append_with_md(ns, qpair, iov[0].iov_base, md, zslba,
3069 						       lba_count,
3070 						       bdev_nvme_zone_appendv_done, bio,
3071 						       flags,
3072 						       0, 0);
3073 	} else {
3074 		rc = spdk_nvme_zns_zone_appendv_with_md(ns, qpair, zslba, lba_count,
3075 							bdev_nvme_zone_appendv_done, bio, flags,
3076 							bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
3077 							md, 0, 0);
3078 	}
3079 
3080 	if (rc != 0 && rc != -ENOMEM) {
3081 		SPDK_ERRLOG("zone append failed: rc = %d\n", rc);
3082 	}
3083 	return rc;
3084 }
3085 
3086 static int
3087 bdev_nvme_comparev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3088 		   struct nvme_bdev_io *bio,
3089 		   struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba,
3090 		   uint32_t flags)
3091 {
3092 	int rc;
3093 
3094 	SPDK_DEBUGLOG(bdev_nvme, "compare %" PRIu64 " blocks with offset %#" PRIx64 "\n",
3095 		      lba_count, lba);
3096 
3097 	bio->iovs = iov;
3098 	bio->iovcnt = iovcnt;
3099 	bio->iovpos = 0;
3100 	bio->iov_offset = 0;
3101 
3102 	rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count,
3103 					       bdev_nvme_comparev_done, bio, flags,
3104 					       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
3105 					       md, 0, 0);
3106 
3107 	if (rc != 0 && rc != -ENOMEM) {
3108 		SPDK_ERRLOG("comparev failed: rc = %d\n", rc);
3109 	}
3110 	return rc;
3111 }
3112 
3113 static int
3114 bdev_nvme_comparev_and_writev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3115 			      struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt,
3116 			      struct iovec *write_iov, int write_iovcnt,
3117 			      void *md, uint64_t lba_count, uint64_t lba, uint32_t flags)
3118 {
3119 	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
3120 	int rc;
3121 
3122 	SPDK_DEBUGLOG(bdev_nvme, "compare and write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
3123 		      lba_count, lba);
3124 
3125 	bio->iovs = cmp_iov;
3126 	bio->iovcnt = cmp_iovcnt;
3127 	bio->iovpos = 0;
3128 	bio->iov_offset = 0;
3129 	bio->fused_iovs = write_iov;
3130 	bio->fused_iovcnt = write_iovcnt;
3131 	bio->fused_iovpos = 0;
3132 	bio->fused_iov_offset = 0;
3133 
3134 	if (bdev_io->num_retries == 0) {
3135 		bio->first_fused_submitted = false;
3136 	}
3137 
3138 	if (!bio->first_fused_submitted) {
3139 		flags |= SPDK_NVME_IO_FLAGS_FUSE_FIRST;
3140 		memset(&bio->cpl, 0, sizeof(bio->cpl));
3141 
3142 		rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count,
3143 						       bdev_nvme_comparev_and_writev_done, bio, flags,
3144 						       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, md, 0, 0);
3145 		if (rc == 0) {
3146 			bio->first_fused_submitted = true;
3147 			flags &= ~SPDK_NVME_IO_FLAGS_FUSE_FIRST;
3148 		} else {
3149 			if (rc != -ENOMEM) {
3150 				SPDK_ERRLOG("compare failed: rc = %d\n", rc);
3151 			}
3152 			return rc;
3153 		}
3154 	}
3155 
3156 	flags |= SPDK_NVME_IO_FLAGS_FUSE_SECOND;
3157 
3158 	rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
3159 					     bdev_nvme_comparev_and_writev_done, bio, flags,
3160 					     bdev_nvme_queued_reset_fused_sgl, bdev_nvme_queued_next_fused_sge, md, 0, 0);
3161 	if (rc != 0 && rc != -ENOMEM) {
3162 		SPDK_ERRLOG("write failed: rc = %d\n", rc);
3163 		rc = 0;
3164 	}
3165 
3166 	return rc;
3167 }
3168 
3169 static int
3170 bdev_nvme_unmap(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3171 		struct nvme_bdev_io *bio,
3172 		uint64_t offset_blocks,
3173 		uint64_t num_blocks)
3174 {
3175 	struct spdk_nvme_dsm_range dsm_ranges[SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES];
3176 	struct spdk_nvme_dsm_range *range;
3177 	uint64_t offset, remaining;
3178 	uint64_t num_ranges_u64;
3179 	uint16_t num_ranges;
3180 	int rc;
3181 
3182 	num_ranges_u64 = (num_blocks + SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS - 1) /
3183 			 SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
3184 	if (num_ranges_u64 > SPDK_COUNTOF(dsm_ranges)) {
3185 		SPDK_ERRLOG("Unmap request for %" PRIu64 " blocks is too large\n", num_blocks);
3186 		return -EINVAL;
3187 	}
3188 	num_ranges = (uint16_t)num_ranges_u64;
3189 
3190 	offset = offset_blocks;
3191 	remaining = num_blocks;
3192 	range = &dsm_ranges[0];
3193 
3194 	/* Fill max-size ranges until the remaining blocks fit into one range */
3195 	while (remaining > SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS) {
3196 		range->attributes.raw = 0;
3197 		range->length = SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
3198 		range->starting_lba = offset;
3199 
3200 		offset += SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
3201 		remaining -= SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
3202 		range++;
3203 	}
3204 
3205 	/* Final range describes the remaining blocks */
3206 	range->attributes.raw = 0;
3207 	range->length = remaining;
3208 	range->starting_lba = offset;
3209 
3210 	rc = spdk_nvme_ns_cmd_dataset_management(ns, qpair,
3211 			SPDK_NVME_DSM_ATTR_DEALLOCATE,
3212 			dsm_ranges, num_ranges,
3213 			bdev_nvme_queued_done, bio);
3214 
3215 	return rc;
3216 }
3217 
3218 static int
3219 bdev_nvme_get_zone_info(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3220 			struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones,
3221 			struct spdk_bdev_zone_info *info)
3222 {
3223 	uint32_t zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
3224 	uint64_t zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
3225 	uint64_t total_zones = spdk_nvme_zns_ns_get_num_zones(ns);
3226 
3227 	if (zone_id % zone_size != 0) {
3228 		return -EINVAL;
3229 	}
3230 
3231 	if (num_zones > total_zones || !num_zones) {
3232 		return -EINVAL;
3233 	}
3234 
3235 	assert(!bio->zone_report_buf);
3236 	bio->zone_report_buf = calloc(1, zone_report_bufsize);
3237 	if (!bio->zone_report_buf) {
3238 		return -ENOMEM;
3239 	}
3240 
3241 	bio->handled_zones = 0;
3242 
3243 	return spdk_nvme_zns_report_zones(ns, qpair, bio->zone_report_buf, zone_report_bufsize,
3244 					  zone_id, SPDK_NVME_ZRA_LIST_ALL, true,
3245 					  bdev_nvme_get_zone_info_done, bio);
3246 }
3247 
3248 static int
3249 bdev_nvme_zone_management(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3250 			  struct nvme_bdev_io *bio, uint64_t zone_id,
3251 			  enum spdk_bdev_zone_action action)
3252 {
3253 	switch (action) {
3254 	case SPDK_BDEV_ZONE_CLOSE:
3255 		return spdk_nvme_zns_close_zone(ns, qpair, zone_id, false,
3256 						bdev_nvme_zone_management_done, bio);
3257 	case SPDK_BDEV_ZONE_FINISH:
3258 		return spdk_nvme_zns_finish_zone(ns, qpair, zone_id, false,
3259 						 bdev_nvme_zone_management_done, bio);
3260 	case SPDK_BDEV_ZONE_OPEN:
3261 		return spdk_nvme_zns_open_zone(ns, qpair, zone_id, false,
3262 					       bdev_nvme_zone_management_done, bio);
3263 	case SPDK_BDEV_ZONE_RESET:
3264 		return spdk_nvme_zns_reset_zone(ns, qpair, zone_id, false,
3265 						bdev_nvme_zone_management_done, bio);
3266 	case SPDK_BDEV_ZONE_OFFLINE:
3267 		return spdk_nvme_zns_offline_zone(ns, qpair, zone_id, false,
3268 						  bdev_nvme_zone_management_done, bio);
3269 	default:
3270 		return -EINVAL;
3271 	}
3272 }
3273 
3274 static int
3275 bdev_nvme_admin_passthru(struct nvme_io_path *io_path, struct nvme_bdev_io *bio,
3276 			 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes)
3277 {
3278 	struct nvme_bdev_ctrlr *nvme_bdev_ctrlr;
3279 	uint32_t max_xfer_size;
3280 
3281 	if (!bdev_nvme_find_admin_path(io_path, &nvme_bdev_ctrlr)) {
3282 		return -EINVAL;
3283 	}
3284 
3285 	max_xfer_size = spdk_nvme_ctrlr_get_max_xfer_size(nvme_bdev_ctrlr->ctrlr);
3286 
3287 	if (nbytes > max_xfer_size) {
3288 		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
3289 		return -EINVAL;
3290 	}
3291 
3292 	bio->orig_thread = spdk_get_thread();
3293 
3294 	return spdk_nvme_ctrlr_cmd_admin_raw(nvme_bdev_ctrlr->ctrlr, cmd, buf,
3295 					     (uint32_t)nbytes, bdev_nvme_admin_passthru_done, bio);
3296 }
3297 
3298 static int
3299 bdev_nvme_io_passthru(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3300 		      struct nvme_bdev_io *bio,
3301 		      struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes)
3302 {
3303 	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
3304 	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3305 
3306 	if (nbytes > max_xfer_size) {
3307 		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
3308 		return -EINVAL;
3309 	}
3310 
3311 	/*
3312 	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
3313 	 * so fill it out automatically.
3314 	 */
3315 	cmd->nsid = spdk_nvme_ns_get_id(ns);
3316 
3317 	return spdk_nvme_ctrlr_cmd_io_raw(ctrlr, qpair, cmd, buf,
3318 					  (uint32_t)nbytes, bdev_nvme_queued_done, bio);
3319 }
3320 
3321 static int
3322 bdev_nvme_io_passthru_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair,
3323 			 struct nvme_bdev_io *bio,
3324 			 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len)
3325 {
3326 	size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
3327 	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
3328 	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);
3329 
3330 	if (nbytes > max_xfer_size) {
3331 		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
3332 		return -EINVAL;
3333 	}
3334 
3335 	if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
3336 		SPDK_ERRLOG("invalid meta data buffer size\n");
3337 		return -EINVAL;
3338 	}
3339 
3340 	/*
3341 	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
3342 	 * so fill it out automatically.
3343 	 */
3344 	cmd->nsid = spdk_nvme_ns_get_id(ns);
3345 
3346 	return spdk_nvme_ctrlr_cmd_io_raw_with_md(ctrlr, qpair, cmd, buf,
3347 			(uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio);
3348 }
3349 
3350 static int
3351 bdev_nvme_abort(struct nvme_io_path *io_path, struct nvme_bdev_io *bio,
3352 		struct nvme_bdev_io *bio_to_abort)
3353 {
3354 	int rc;
3355 
3356 	bio->orig_thread = spdk_get_thread();
3357 
3358 	rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->ctrlr->ctrlr,
3359 					   io_path->qpair,
3360 					   bio_to_abort,
3361 					   bdev_nvme_abort_done, bio);
3362 	if (rc == -ENOENT) {
3363 		/* If no command was found in I/O qpair, the target command may be
3364 		 * admin command.
3365 		 */
3366 		rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->ctrlr->ctrlr,
3367 						   NULL,
3368 						   bio_to_abort,
3369 						   bdev_nvme_abort_done, bio);
3370 	}
3371 
3372 	if (rc == -ENOENT) {
3373 		/* If no command was found, complete the abort request with failure. */
3374 		bio->cpl.cdw0 |= 1U;
3375 		bio->cpl.status.sc = SPDK_NVME_SC_SUCCESS;
3376 		bio->cpl.status.sct = SPDK_NVME_SCT_GENERIC;
3377 
3378 		bdev_nvme_abort_completion(bio);
3379 
3380 		rc = 0;
3381 	}
3382 
3383 	return rc;
3384 }
3385 
3386 static void
3387 nvme_ctrlr_config_json_standard_namespace(struct spdk_json_write_ctx *w,
3388 		struct nvme_bdev_ns *nvme_ns)
3389 {
3390 	/* nop */
3391 }
3392 
3393 static void
3394 nvme_namespace_config_json(struct spdk_json_write_ctx *w, struct nvme_bdev_ns *nvme_ns)
3395 {
3396 	g_config_json_namespace_fn[nvme_ns->type](w, nvme_ns);
3397 }
3398 
3399 static void
3400 bdev_nvme_opts_config_json(struct spdk_json_write_ctx *w)
3401 {
3402 	const char	*action;
3403 
3404 	if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET) {
3405 		action = "reset";
3406 	} else if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT) {
3407 		action = "abort";
3408 	} else {
3409 		action = "none";
3410 	}
3411 
3412 	spdk_json_write_object_begin(w);
3413 
3414 	spdk_json_write_named_string(w, "method", "bdev_nvme_set_options");
3415 
3416 	spdk_json_write_named_object_begin(w, "params");
3417 	spdk_json_write_named_string(w, "action_on_timeout", action);
3418 	spdk_json_write_named_uint64(w, "timeout_us", g_opts.timeout_us);
3419 	spdk_json_write_named_uint32(w, "keep_alive_timeout_ms", g_opts.keep_alive_timeout_ms);
3420 	spdk_json_write_named_uint32(w, "retry_count", g_opts.retry_count);
3421 	spdk_json_write_named_uint32(w, "arbitration_burst", g_opts.arbitration_burst);
3422 	spdk_json_write_named_uint32(w, "low_priority_weight", g_opts.low_priority_weight);
3423 	spdk_json_write_named_uint32(w, "medium_priority_weight", g_opts.medium_priority_weight);
3424 	spdk_json_write_named_uint32(w, "high_priority_weight", g_opts.high_priority_weight);
3425 	spdk_json_write_named_uint64(w, "nvme_adminq_poll_period_us", g_opts.nvme_adminq_poll_period_us);
3426 	spdk_json_write_named_uint64(w, "nvme_ioq_poll_period_us", g_opts.nvme_ioq_poll_period_us);
3427 	spdk_json_write_named_uint32(w, "io_queue_requests", g_opts.io_queue_requests);
3428 	spdk_json_write_named_bool(w, "delay_cmd_submit", g_opts.delay_cmd_submit);
3429 	spdk_json_write_object_end(w);
3430 
3431 	spdk_json_write_object_end(w);
3432 }
3433 
3434 static void
3435 nvme_bdev_ctrlr_config_json(struct spdk_json_write_ctx *w,
3436 			    struct nvme_bdev_ctrlr *nvme_bdev_ctrlr)
3437 {
3438 	struct spdk_nvme_transport_id	*trid;
3439 
3440 	trid = nvme_bdev_ctrlr->connected_trid;
3441 
3442 	spdk_json_write_object_begin(w);
3443 
3444 	spdk_json_write_named_string(w, "method", "bdev_nvme_attach_controller");
3445 
3446 	spdk_json_write_named_object_begin(w, "params");
3447 	spdk_json_write_named_string(w, "name", nvme_bdev_ctrlr->name);
3448 	nvme_bdev_dump_trid_json(trid, w);
3449 	spdk_json_write_named_bool(w, "prchk_reftag",
3450 				   (nvme_bdev_ctrlr->prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_REFTAG) != 0);
3451 	spdk_json_write_named_bool(w, "prchk_guard",
3452 				   (nvme_bdev_ctrlr->prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_GUARD) != 0);
3453 
3454 	spdk_json_write_object_end(w);
3455 
3456 	spdk_json_write_object_end(w);
3457 }
3458 
3459 static void
3460 bdev_nvme_hotplug_config_json(struct spdk_json_write_ctx *w)
3461 {
3462 	spdk_json_write_object_begin(w);
3463 	spdk_json_write_named_string(w, "method", "bdev_nvme_set_hotplug");
3464 
3465 	spdk_json_write_named_object_begin(w, "params");
3466 	spdk_json_write_named_uint64(w, "period_us", g_nvme_hotplug_poll_period_us);
3467 	spdk_json_write_named_bool(w, "enable", g_nvme_hotplug_enabled);
3468 	spdk_json_write_object_end(w);
3469 
3470 	spdk_json_write_object_end(w);
3471 }
3472 
3473 static int
3474 bdev_nvme_config_json(struct spdk_json_write_ctx *w)
3475 {
3476 	struct nvme_bdev_ctrlr	*nvme_bdev_ctrlr;
3477 	uint32_t		nsid;
3478 
3479 	bdev_nvme_opts_config_json(w);
3480 
3481 	pthread_mutex_lock(&g_bdev_nvme_mutex);
3482 
3483 	TAILQ_FOREACH(nvme_bdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
3484 		nvme_bdev_ctrlr_config_json(w, nvme_bdev_ctrlr);
3485 
3486 		for (nsid = 0; nsid < nvme_bdev_ctrlr->num_ns; ++nsid) {
3487 			if (!nvme_bdev_ctrlr->namespaces[nsid]->populated) {
3488 				continue;
3489 			}
3490 
3491 			nvme_namespace_config_json(w, nvme_bdev_ctrlr->namespaces[nsid]);
3492 		}
3493 	}
3494 
3495 	/* Dump as last parameter to give all NVMe bdevs chance to be constructed
3496 	 * before enabling hotplug poller.
3497 	 */
3498 	bdev_nvme_hotplug_config_json(w);
3499 
3500 	pthread_mutex_unlock(&g_bdev_nvme_mutex);
3501 	return 0;
3502 }
3503 
3504 struct spdk_nvme_ctrlr *
3505 bdev_nvme_get_ctrlr(struct spdk_bdev *bdev)
3506 {
3507 	if (!bdev || bdev->module != &nvme_if) {
3508 		return NULL;
3509 	}
3510 
3511 	return SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk)->nvme_ns->ctrlr->ctrlr;
3512 }
3513 
3514 SPDK_LOG_REGISTER_COMPONENT(bdev_nvme)
3515