xref: /spdk/module/bdev/nvme/bdev_nvme.c (revision 7025ceb9c119a6da0b6ee2013b6ae94b51fac2df)

/*   SPDX-License-Identifier: BSD-3-Clause
 *   Copyright (C) 2016 Intel Corporation. All rights reserved.
 *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
 *   Copyright (c) 2021-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 *   Copyright (c) 2022 Dell Inc, or its subsidiaries. All rights reserved.
 */

#include "spdk/stdinc.h"

#include "bdev_nvme.h"

#include "spdk/accel.h"
#include "spdk/config.h"
#include "spdk/endian.h"
#include "spdk/bdev.h"
#include "spdk/json.h"
#include "spdk/keyring.h"
#include "spdk/likely.h"
#include "spdk/nvme.h"
#include "spdk/nvme_ocssd.h"
#include "spdk/nvme_zns.h"
#include "spdk/opal.h"
#include "spdk/thread.h"
#include "spdk/trace.h"
#include "spdk/string.h"
#include "spdk/util.h"
#include "spdk/uuid.h"

#include "spdk/bdev_module.h"
#include "spdk/log.h"

#include "spdk_internal/usdt.h"
#include "spdk_internal/trace_defs.h"

#define SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT true
#define SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS	(10000)

#define NSID_STR_LEN 10

#define SPDK_CONTROLLER_NAME_MAX 512

static int bdev_nvme_config_json(struct spdk_json_write_ctx *w);

struct nvme_bdev_io {
	/** array of iovecs to transfer. */
	struct iovec *iovs;

	/** Number of iovecs in iovs array. */
	int iovcnt;

	/** Current iovec position. */
	int iovpos;

	/** Offset in current iovec. */
	uint32_t iov_offset;

	/** I/O path the current I/O or admin passthrough is submitted on, or the I/O path
	 *  being reset in a reset I/O.
	 */
	struct nvme_io_path *io_path;

	/** array of iovecs to transfer. */
	struct iovec *fused_iovs;

	/** Number of iovecs in iovs array. */
	int fused_iovcnt;

	/** Current iovec position. */
	int fused_iovpos;

	/** Offset in current iovec. */
	uint32_t fused_iov_offset;

	/** Saved status for admin passthru completion event, PI error verification, or intermediate compare-and-write status */
	struct spdk_nvme_cpl cpl;

	/** Extended IO opts passed by the user to bdev layer and mapped to NVME format */
	struct spdk_nvme_ns_cmd_ext_io_opts ext_opts;

	/** Keeps track if first of fused commands was submitted */
	bool first_fused_submitted;

	/** Keeps track if first of fused commands was completed */
	bool first_fused_completed;

	/** Temporary pointer to zone report buffer */
	struct spdk_nvme_zns_zone_report *zone_report_buf;

	/** Keep track of how many zones that have been copied to the spdk_bdev_zone_info struct */
	uint64_t handled_zones;

	/** Expiration value in ticks to retry the current I/O. */
	uint64_t retry_ticks;

	/* How many times the current I/O was retried. */
	int32_t retry_count;

	/* Current tsc at submit time. */
	uint64_t submit_tsc;

	/* Used to put nvme_bdev_io into the list */
	TAILQ_ENTRY(nvme_bdev_io) retry_link;
};

struct nvme_probe_skip_entry {
	struct spdk_nvme_transport_id		trid;
	TAILQ_ENTRY(nvme_probe_skip_entry)	tailq;
};
/* All the controllers deleted by users via RPC are skipped by hotplug monitor */
static TAILQ_HEAD(, nvme_probe_skip_entry) g_skipped_nvme_ctrlrs = TAILQ_HEAD_INITIALIZER(
			g_skipped_nvme_ctrlrs);

#define BDEV_NVME_DEFAULT_DIGESTS (SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA256) | \
				   SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA384) | \
				   SPDK_BIT(SPDK_NVMF_DHCHAP_HASH_SHA512))

#define BDEV_NVME_DEFAULT_DHGROUPS (SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_NULL) | \
				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_2048) | \
				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_3072) | \
				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_4096) | \
				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_6144) | \
				    SPDK_BIT(SPDK_NVMF_DHCHAP_DHGROUP_8192))

static struct spdk_bdev_nvme_opts g_opts = {
	.action_on_timeout = SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE,
	.timeout_us = 0,
	.timeout_admin_us = 0,
	.keep_alive_timeout_ms = SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS,
	.transport_retry_count = 4,
	.arbitration_burst = 0,
	.low_priority_weight = 0,
	.medium_priority_weight = 0,
	.high_priority_weight = 0,
	.nvme_adminq_poll_period_us = 10000ULL,
	.nvme_ioq_poll_period_us = 0,
	.io_queue_requests = 0,
	.delay_cmd_submit = SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT,
	.bdev_retry_count = 3,
	.transport_ack_timeout = 0,
	.ctrlr_loss_timeout_sec = 0,
	.reconnect_delay_sec = 0,
	.fast_io_fail_timeout_sec = 0,
	.disable_auto_failback = false,
	.generate_uuids = false,
	.transport_tos = 0,
	.nvme_error_stat = false,
	.io_path_stat = false,
	.allow_accel_sequence = false,
	.dhchap_digests = BDEV_NVME_DEFAULT_DIGESTS,
	.dhchap_dhgroups = BDEV_NVME_DEFAULT_DHGROUPS,
};

#define NVME_HOTPLUG_POLL_PERIOD_MAX			10000000ULL
#define NVME_HOTPLUG_POLL_PERIOD_DEFAULT		100000ULL

static int g_hot_insert_nvme_controller_index = 0;
static uint64_t g_nvme_hotplug_poll_period_us = NVME_HOTPLUG_POLL_PERIOD_DEFAULT;
static bool g_nvme_hotplug_enabled = false;
struct spdk_thread *g_bdev_nvme_init_thread;
static struct spdk_poller *g_hotplug_poller;
static struct spdk_poller *g_hotplug_probe_poller;
static struct spdk_nvme_probe_ctx *g_hotplug_probe_ctx;

static void nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
		struct nvme_async_probe_ctx *ctx);
static void nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
		struct nvme_async_probe_ctx *ctx);
static int bdev_nvme_library_init(void);
static void bdev_nvme_library_fini(void);
static void _bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch,
				      struct spdk_bdev_io *bdev_io);
static void bdev_nvme_submit_request(struct spdk_io_channel *ch,
				     struct spdk_bdev_io *bdev_io);
static int bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
			   void *md, uint64_t lba_count, uint64_t lba,
			   uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
			   struct spdk_accel_sequence *seq);
static int bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
				 void *md, uint64_t lba_count, uint64_t lba);
static int bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
			    void *md, uint64_t lba_count, uint64_t lba,
			    uint32_t flags, struct spdk_memory_domain *domain, void *domain_ctx,
			    struct spdk_accel_sequence *seq);
static int bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
				  void *md, uint64_t lba_count,
				  uint64_t zslba, uint32_t flags);
static int bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
			      void *md, uint64_t lba_count, uint64_t lba,
			      uint32_t flags);
static int bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio,
		struct iovec *cmp_iov, int cmp_iovcnt, struct iovec *write_iov,
		int write_iovcnt, void *md, uint64_t lba_count, uint64_t lba,
		uint32_t flags);
static int bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id,
				   uint32_t num_zones, struct spdk_bdev_zone_info *info);
static int bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
				     enum spdk_bdev_zone_action action);
static void bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch,
				     struct nvme_bdev_io *bio,
				     struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes);
static int bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
				 void *buf, size_t nbytes);
static int bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
				    void *buf, size_t nbytes, void *md_buf, size_t md_len);
static int bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
				     struct iovec *iov, int iovcnt, size_t nbytes,
				     void *md_buf, size_t md_len);
static void bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch,
			    struct nvme_bdev_io *bio, struct nvme_bdev_io *bio_to_abort);
static void bdev_nvme_reset_io(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio);
static int bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
static int bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr);
static void remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr);
static int nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr);

static struct nvme_ns *nvme_ns_alloc(void);
static void nvme_ns_free(struct nvme_ns *ns);

static int
nvme_ns_cmp(struct nvme_ns *ns1, struct nvme_ns *ns2)
{
	return ns1->id < ns2->id ? -1 : ns1->id > ns2->id;
}

RB_GENERATE_STATIC(nvme_ns_tree, nvme_ns, node, nvme_ns_cmp);

struct spdk_nvme_qpair *
bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch)
{
	struct nvme_ctrlr_channel *ctrlr_ch;

	assert(ctrlr_io_ch != NULL);

	ctrlr_ch = spdk_io_channel_get_ctx(ctrlr_io_ch);

	return ctrlr_ch->qpair->qpair;
}

static int
bdev_nvme_get_ctx_size(void)
{
	return sizeof(struct nvme_bdev_io);
}

static struct spdk_bdev_module nvme_if = {
	.name = "nvme",
	.async_fini = true,
	.module_init = bdev_nvme_library_init,
	.module_fini = bdev_nvme_library_fini,
	.config_json = bdev_nvme_config_json,
	.get_ctx_size = bdev_nvme_get_ctx_size,

};
SPDK_BDEV_MODULE_REGISTER(nvme, &nvme_if)

struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs = TAILQ_HEAD_INITIALIZER(g_nvme_bdev_ctrlrs);
pthread_mutex_t g_bdev_nvme_mutex = PTHREAD_MUTEX_INITIALIZER;
bool g_bdev_nvme_module_finish;

struct nvme_bdev_ctrlr *
nvme_bdev_ctrlr_get_by_name(const char *name)
{
	struct nvme_bdev_ctrlr *nbdev_ctrlr;

	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
		if (strcmp(name, nbdev_ctrlr->name) == 0) {
			break;
		}
	}

	return nbdev_ctrlr;
}

static struct nvme_ctrlr *
nvme_bdev_ctrlr_get_ctrlr(struct nvme_bdev_ctrlr *nbdev_ctrlr,
			  const struct spdk_nvme_transport_id *trid)
{
	struct nvme_ctrlr *nvme_ctrlr;

	TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
		if (spdk_nvme_transport_id_compare(trid, &nvme_ctrlr->active_path_id->trid) == 0) {
			break;
		}
	}

	return nvme_ctrlr;
}

struct nvme_ctrlr *
nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr,
				uint16_t cntlid)
{
	struct nvme_ctrlr *nvme_ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;

	TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
		cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
		if (cdata->cntlid == cntlid) {
			break;
		}
	}

	return nvme_ctrlr;
}

static struct nvme_bdev *
nvme_bdev_ctrlr_get_bdev(struct nvme_bdev_ctrlr *nbdev_ctrlr, uint32_t nsid)
{
	struct nvme_bdev *bdev;

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	TAILQ_FOREACH(bdev, &nbdev_ctrlr->bdevs, tailq) {
		if (bdev->nsid == nsid) {
			break;
		}
	}
	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	return bdev;
}

struct nvme_ns *
nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid)
{
	struct nvme_ns ns;

	assert(nsid > 0);

	ns.id = nsid;
	return RB_FIND(nvme_ns_tree, &nvme_ctrlr->namespaces, &ns);
}

struct nvme_ns *
nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr)
{
	return RB_MIN(nvme_ns_tree, &nvme_ctrlr->namespaces);
}

struct nvme_ns *
nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns)
{
	if (ns == NULL) {
		return NULL;
	}

	return RB_NEXT(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
}

static struct nvme_ctrlr *
nvme_ctrlr_get(const struct spdk_nvme_transport_id *trid)
{
	struct nvme_bdev_ctrlr	*nbdev_ctrlr;
	struct nvme_ctrlr	*nvme_ctrlr = NULL;

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
		nvme_ctrlr = nvme_bdev_ctrlr_get_ctrlr(nbdev_ctrlr, trid);
		if (nvme_ctrlr != NULL) {
			break;
		}
	}
	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	return nvme_ctrlr;
}

struct nvme_ctrlr *
nvme_ctrlr_get_by_name(const char *name)
{
	struct nvme_bdev_ctrlr *nbdev_ctrlr;
	struct nvme_ctrlr *nvme_ctrlr = NULL;

	if (name == NULL) {
		return NULL;
	}

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
	if (nbdev_ctrlr != NULL) {
		nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
	}
	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	return nvme_ctrlr;
}

void
nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx)
{
	struct nvme_bdev_ctrlr *nbdev_ctrlr;

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
		fn(nbdev_ctrlr, ctx);
	}
	pthread_mutex_unlock(&g_bdev_nvme_mutex);
}

void
nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid, struct spdk_json_write_ctx *w)
{
	const char *trtype_str;
	const char *adrfam_str;

	trtype_str = spdk_nvme_transport_id_trtype_str(trid->trtype);
	if (trtype_str) {
		spdk_json_write_named_string(w, "trtype", trtype_str);
	}

	adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
	if (adrfam_str) {
		spdk_json_write_named_string(w, "adrfam", adrfam_str);
	}

	if (trid->traddr[0] != '\0') {
		spdk_json_write_named_string(w, "traddr", trid->traddr);
	}

	if (trid->trsvcid[0] != '\0') {
		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
	}

	if (trid->subnqn[0] != '\0') {
		spdk_json_write_named_string(w, "subnqn", trid->subnqn);
	}
}

static void
nvme_bdev_ctrlr_delete(struct nvme_bdev_ctrlr *nbdev_ctrlr,
		       struct nvme_ctrlr *nvme_ctrlr)
{
	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_delete, nvme_ctrlr->nbdev_ctrlr->name);
	pthread_mutex_lock(&g_bdev_nvme_mutex);

	TAILQ_REMOVE(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
	if (!TAILQ_EMPTY(&nbdev_ctrlr->ctrlrs)) {
		pthread_mutex_unlock(&g_bdev_nvme_mutex);

		return;
	}
	TAILQ_REMOVE(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);

	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	assert(TAILQ_EMPTY(&nbdev_ctrlr->bdevs));

	free(nbdev_ctrlr->name);
	free(nbdev_ctrlr);
}

static void
_nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
{
	struct nvme_path_id *path_id, *tmp_path;
	struct nvme_ns *ns, *tmp_ns;

	free(nvme_ctrlr->copied_ana_desc);
	spdk_free(nvme_ctrlr->ana_log_page);

	if (nvme_ctrlr->opal_dev) {
		spdk_opal_dev_destruct(nvme_ctrlr->opal_dev);
		nvme_ctrlr->opal_dev = NULL;
	}

	if (nvme_ctrlr->nbdev_ctrlr) {
		nvme_bdev_ctrlr_delete(nvme_ctrlr->nbdev_ctrlr, nvme_ctrlr);
	}

	RB_FOREACH_SAFE(ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp_ns) {
		RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, ns);
		nvme_ns_free(ns);
	}

	TAILQ_FOREACH_SAFE(path_id, &nvme_ctrlr->trids, link, tmp_path) {
		TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
		free(path_id);
	}

	pthread_mutex_destroy(&nvme_ctrlr->mutex);
	spdk_keyring_put_key(nvme_ctrlr->psk);
	spdk_keyring_put_key(nvme_ctrlr->dhchap_key);
	free(nvme_ctrlr);

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	if (g_bdev_nvme_module_finish && TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
		pthread_mutex_unlock(&g_bdev_nvme_mutex);
		spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
		spdk_bdev_module_fini_done();
		return;
	}
	pthread_mutex_unlock(&g_bdev_nvme_mutex);
}

static int
nvme_detach_poller(void *arg)
{
	struct nvme_ctrlr *nvme_ctrlr = arg;
	int rc;

	rc = spdk_nvme_detach_poll_async(nvme_ctrlr->detach_ctx);
	if (rc != -EAGAIN) {
		spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
		_nvme_ctrlr_delete(nvme_ctrlr);
	}

	return SPDK_POLLER_BUSY;
}

static void
nvme_ctrlr_delete(struct nvme_ctrlr *nvme_ctrlr)
{
	int rc;

	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);

	/* First, unregister the adminq poller, as the driver will poll adminq if necessary */
	spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);

	/* If we got here, the reset/detach poller cannot be active */
	assert(nvme_ctrlr->reset_detach_poller == NULL);
	nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(nvme_detach_poller,
					  nvme_ctrlr, 1000);
	if (nvme_ctrlr->reset_detach_poller == NULL) {
		SPDK_ERRLOG("Failed to register detach poller\n");
		goto error;
	}

	rc = spdk_nvme_detach_async(nvme_ctrlr->ctrlr, &nvme_ctrlr->detach_ctx);
	if (rc != 0) {
		SPDK_ERRLOG("Failed to detach the NVMe controller\n");
		goto error;
	}

	return;
error:
	/* We don't have a good way to handle errors here, so just do what we can and delete the
	 * controller without detaching the underlying NVMe device.
	 */
	spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
	_nvme_ctrlr_delete(nvme_ctrlr);
}

static void
nvme_ctrlr_unregister_cb(void *io_device)
{
	struct nvme_ctrlr *nvme_ctrlr = io_device;

	nvme_ctrlr_delete(nvme_ctrlr);
}

static void
nvme_ctrlr_unregister(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	spdk_io_device_unregister(nvme_ctrlr, nvme_ctrlr_unregister_cb);
}

static bool
nvme_ctrlr_can_be_unregistered(struct nvme_ctrlr *nvme_ctrlr)
{
	if (!nvme_ctrlr->destruct) {
		return false;
	}

	if (nvme_ctrlr->ref > 0) {
		return false;
	}

	if (nvme_ctrlr->resetting) {
		return false;
	}

	if (nvme_ctrlr->ana_log_page_updating) {
		return false;
	}

	if (nvme_ctrlr->io_path_cache_clearing) {
		return false;
	}

	return true;
}

static void
nvme_ctrlr_release(struct nvme_ctrlr *nvme_ctrlr)
{
	pthread_mutex_lock(&nvme_ctrlr->mutex);
	SPDK_DTRACE_PROBE2(bdev_nvme_ctrlr_release, nvme_ctrlr->nbdev_ctrlr->name, nvme_ctrlr->ref);

	assert(nvme_ctrlr->ref > 0);
	nvme_ctrlr->ref--;

	if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return;
	}

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_thread_exec_msg(nvme_ctrlr->thread, nvme_ctrlr_unregister, nvme_ctrlr);
}

static void
bdev_nvme_clear_current_io_path(struct nvme_bdev_channel *nbdev_ch)
{
	nbdev_ch->current_io_path = NULL;
	nbdev_ch->rr_counter = 0;
}

static struct nvme_io_path *
_bdev_nvme_get_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
{
	struct nvme_io_path *io_path;

	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
		if (io_path->nvme_ns == nvme_ns) {
			break;
		}
	}

	return io_path;
}

static struct nvme_io_path *
nvme_io_path_alloc(void)
{
	struct nvme_io_path *io_path;

	io_path = calloc(1, sizeof(*io_path));
	if (io_path == NULL) {
		SPDK_ERRLOG("Failed to alloc io_path.\n");
		return NULL;
	}

	if (g_opts.io_path_stat) {
		io_path->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
		if (io_path->stat == NULL) {
			free(io_path);
			SPDK_ERRLOG("Failed to alloc io_path stat.\n");
			return NULL;
		}
		spdk_bdev_reset_io_stat(io_path->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
	}

	return io_path;
}

static void
nvme_io_path_free(struct nvme_io_path *io_path)
{
	free(io_path->stat);
	free(io_path);
}

static int
_bdev_nvme_add_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_ns *nvme_ns)
{
	struct nvme_io_path *io_path;
	struct spdk_io_channel *ch;
	struct nvme_ctrlr_channel *ctrlr_ch;
	struct nvme_qpair *nvme_qpair;

	io_path = nvme_io_path_alloc();
	if (io_path == NULL) {
		return -ENOMEM;
	}

	io_path->nvme_ns = nvme_ns;

	ch = spdk_get_io_channel(nvme_ns->ctrlr);
	if (ch == NULL) {
		nvme_io_path_free(io_path);
		SPDK_ERRLOG("Failed to alloc io_channel.\n");
		return -ENOMEM;
	}

	ctrlr_ch = spdk_io_channel_get_ctx(ch);

	nvme_qpair = ctrlr_ch->qpair;
	assert(nvme_qpair != NULL);

	io_path->qpair = nvme_qpair;
	TAILQ_INSERT_TAIL(&nvme_qpair->io_path_list, io_path, tailq);

	io_path->nbdev_ch = nbdev_ch;
	STAILQ_INSERT_TAIL(&nbdev_ch->io_path_list, io_path, stailq);

	bdev_nvme_clear_current_io_path(nbdev_ch);

	return 0;
}

static void
bdev_nvme_clear_retry_io_path(struct nvme_bdev_channel *nbdev_ch,
			      struct nvme_io_path *io_path)
{
	struct nvme_bdev_io *bio;

	TAILQ_FOREACH(bio, &nbdev_ch->retry_io_list, retry_link) {
		if (bio->io_path == io_path) {
			bio->io_path = NULL;
		}
	}
}

static void
_bdev_nvme_delete_io_path(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *io_path)
{
	struct spdk_io_channel *ch;
	struct nvme_qpair *nvme_qpair;
	struct nvme_ctrlr_channel *ctrlr_ch;
	struct nvme_bdev *nbdev;

	nbdev = spdk_io_channel_get_io_device(spdk_io_channel_from_ctx(nbdev_ch));

	/* Add the statistics to nvme_ns before this path is destroyed. */
	pthread_mutex_lock(&nbdev->mutex);
	if (nbdev->ref != 0 && io_path->nvme_ns->stat != NULL && io_path->stat != NULL) {
		spdk_bdev_add_io_stat(io_path->nvme_ns->stat, io_path->stat);
	}
	pthread_mutex_unlock(&nbdev->mutex);

	bdev_nvme_clear_current_io_path(nbdev_ch);
	bdev_nvme_clear_retry_io_path(nbdev_ch, io_path);

	STAILQ_REMOVE(&nbdev_ch->io_path_list, io_path, nvme_io_path, stailq);
	io_path->nbdev_ch = NULL;

	nvme_qpair = io_path->qpair;
	assert(nvme_qpair != NULL);

	ctrlr_ch = nvme_qpair->ctrlr_ch;
	assert(ctrlr_ch != NULL);

	ch = spdk_io_channel_from_ctx(ctrlr_ch);
	spdk_put_io_channel(ch);

	/* After an io_path is removed, I/Os submitted to it may complete and update statistics
	 * of the io_path. To avoid heap-use-after-free error from this case, do not free the
	 * io_path here but free the io_path when the associated qpair is freed. It is ensured
	 * that all I/Os submitted to the io_path are completed when the associated qpair is freed.
	 */
}

static void
_bdev_nvme_delete_io_paths(struct nvme_bdev_channel *nbdev_ch)
{
	struct nvme_io_path *io_path, *tmp_io_path;

	STAILQ_FOREACH_SAFE(io_path, &nbdev_ch->io_path_list, stailq, tmp_io_path) {
		_bdev_nvme_delete_io_path(nbdev_ch, io_path);
	}
}

static int
bdev_nvme_create_bdev_channel_cb(void *io_device, void *ctx_buf)
{
	struct nvme_bdev_channel *nbdev_ch = ctx_buf;
	struct nvme_bdev *nbdev = io_device;
	struct nvme_ns *nvme_ns;
	int rc;

	STAILQ_INIT(&nbdev_ch->io_path_list);
	TAILQ_INIT(&nbdev_ch->retry_io_list);

	pthread_mutex_lock(&nbdev->mutex);

	nbdev_ch->mp_policy = nbdev->mp_policy;
	nbdev_ch->mp_selector = nbdev->mp_selector;
	nbdev_ch->rr_min_io = nbdev->rr_min_io;

	TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
		rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
		if (rc != 0) {
			pthread_mutex_unlock(&nbdev->mutex);

			_bdev_nvme_delete_io_paths(nbdev_ch);
			return rc;
		}
	}
	pthread_mutex_unlock(&nbdev->mutex);

	return 0;
}

/* If cpl != NULL, complete the bdev_io with nvme status based on 'cpl'.
 * If cpl == NULL, complete the bdev_io with bdev status based on 'status'.
 */
static inline void
__bdev_nvme_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status,
			const struct spdk_nvme_cpl *cpl)
{
	spdk_trace_record(TRACE_BDEV_NVME_IO_DONE, 0, 0, (uintptr_t)bdev_io->driver_ctx,
			  (uintptr_t)bdev_io);
	if (cpl) {
		spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc);
	} else {
		spdk_bdev_io_complete(bdev_io, status);
	}
}

static void bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch);

static void
bdev_nvme_destroy_bdev_channel_cb(void *io_device, void *ctx_buf)
{
	struct nvme_bdev_channel *nbdev_ch = ctx_buf;

	bdev_nvme_abort_retry_ios(nbdev_ch);
	_bdev_nvme_delete_io_paths(nbdev_ch);
}

static inline bool
bdev_nvme_io_type_is_admin(enum spdk_bdev_io_type io_type)
{
	switch (io_type) {
	case SPDK_BDEV_IO_TYPE_RESET:
	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
	case SPDK_BDEV_IO_TYPE_ABORT:
		return true;
	default:
		break;
	}

	return false;
}

static inline bool
nvme_ns_is_active(struct nvme_ns *nvme_ns)
{
	if (spdk_unlikely(nvme_ns->ana_state_updating)) {
		return false;
	}

	if (spdk_unlikely(nvme_ns->ns == NULL)) {
		return false;
	}

	return true;
}

static inline bool
nvme_ns_is_accessible(struct nvme_ns *nvme_ns)
{
	if (spdk_unlikely(!nvme_ns_is_active(nvme_ns))) {
		return false;
	}

	switch (nvme_ns->ana_state) {
	case SPDK_NVME_ANA_OPTIMIZED_STATE:
	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
		return true;
	default:
		break;
	}

	return false;
}

static inline bool
nvme_qpair_is_connected(struct nvme_qpair *nvme_qpair)
{
	if (spdk_unlikely(nvme_qpair->qpair == NULL)) {
		return false;
	}

	if (spdk_unlikely(spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
			  SPDK_NVME_QPAIR_FAILURE_NONE)) {
		return false;
	}

	if (spdk_unlikely(nvme_qpair->ctrlr_ch->reset_iter != NULL)) {
		return false;
	}

	return true;
}

static inline bool
nvme_io_path_is_available(struct nvme_io_path *io_path)
{
	if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
		return false;
	}

	if (spdk_unlikely(!nvme_ns_is_accessible(io_path->nvme_ns))) {
		return false;
	}

	return true;
}

static inline bool
nvme_ctrlr_is_failed(struct nvme_ctrlr *nvme_ctrlr)
{
	if (nvme_ctrlr->destruct) {
		return true;
	}

	if (nvme_ctrlr->fast_io_fail_timedout) {
		return true;
	}

	if (nvme_ctrlr->resetting) {
		if (nvme_ctrlr->opts.reconnect_delay_sec != 0) {
			return false;
		} else {
			return true;
		}
	}

	if (nvme_ctrlr->reconnect_is_delayed) {
		return false;
	}

	if (nvme_ctrlr->disabled) {
		return true;
	}

	if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
		return true;
	} else {
		return false;
	}
}

static bool
nvme_ctrlr_is_available(struct nvme_ctrlr *nvme_ctrlr)
{
	if (nvme_ctrlr->destruct) {
		return false;
	}

	if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
		return false;
	}

	if (nvme_ctrlr->resetting || nvme_ctrlr->reconnect_is_delayed) {
		return false;
	}

	if (nvme_ctrlr->disabled) {
		return false;
	}

	return true;
}

/* Simulate circular linked list. */
static inline struct nvme_io_path *
nvme_io_path_get_next(struct nvme_bdev_channel *nbdev_ch, struct nvme_io_path *prev_path)
{
	struct nvme_io_path *next_path;

	if (prev_path != NULL) {
		next_path = STAILQ_NEXT(prev_path, stailq);
		if (next_path != NULL) {
			return next_path;
		}
	}

	return STAILQ_FIRST(&nbdev_ch->io_path_list);
}

static struct nvme_io_path *
_bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
{
	struct nvme_io_path *io_path, *start, *non_optimized = NULL;

	start = nvme_io_path_get_next(nbdev_ch, nbdev_ch->current_io_path);

	io_path = start;
	do {
		if (spdk_likely(nvme_qpair_is_connected(io_path->qpair) &&
				nvme_ns_is_active(io_path->nvme_ns))) {
			switch (io_path->nvme_ns->ana_state) {
			case SPDK_NVME_ANA_OPTIMIZED_STATE:
				nbdev_ch->current_io_path = io_path;
				return io_path;
			case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
				if (non_optimized == NULL) {
					non_optimized = io_path;
				}
				break;
			default:
				break;
			}
		}
		io_path = nvme_io_path_get_next(nbdev_ch, io_path);
	} while (io_path != start);

	if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE) {
		/* We come here only if there is no optimized path. Cache even non_optimized
		 * path for load balance across multiple non_optimized paths.
		 */
		nbdev_ch->current_io_path = non_optimized;
	}

	return non_optimized;
}

static struct nvme_io_path *
_bdev_nvme_find_io_path_min_qd(struct nvme_bdev_channel *nbdev_ch)
{
	struct nvme_io_path *io_path;
	struct nvme_io_path *optimized = NULL, *non_optimized = NULL;
	uint32_t opt_min_qd = UINT32_MAX, non_opt_min_qd = UINT32_MAX;
	uint32_t num_outstanding_reqs;

	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
		if (spdk_unlikely(!nvme_qpair_is_connected(io_path->qpair))) {
			/* The device is currently resetting. */
			continue;
		}

		if (spdk_unlikely(!nvme_ns_is_active(io_path->nvme_ns))) {
			continue;
		}

		num_outstanding_reqs = spdk_nvme_qpair_get_num_outstanding_reqs(io_path->qpair->qpair);
		switch (io_path->nvme_ns->ana_state) {
		case SPDK_NVME_ANA_OPTIMIZED_STATE:
			if (num_outstanding_reqs < opt_min_qd) {
				opt_min_qd = num_outstanding_reqs;
				optimized = io_path;
			}
			break;
		case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
			if (num_outstanding_reqs < non_opt_min_qd) {
				non_opt_min_qd = num_outstanding_reqs;
				non_optimized = io_path;
			}
			break;
		default:
			break;
		}
	}

	/* don't cache io path for BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH selector */
	if (optimized != NULL) {
		return optimized;
	}

	return non_optimized;
}

static inline struct nvme_io_path *
bdev_nvme_find_io_path(struct nvme_bdev_channel *nbdev_ch)
{
	if (spdk_likely(nbdev_ch->current_io_path != NULL)) {
		if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE) {
			return nbdev_ch->current_io_path;
		} else if (nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
			if (++nbdev_ch->rr_counter < nbdev_ch->rr_min_io) {
				return nbdev_ch->current_io_path;
			}
			nbdev_ch->rr_counter = 0;
		}
	}

	if (nbdev_ch->mp_policy == BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE ||
	    nbdev_ch->mp_selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
		return _bdev_nvme_find_io_path(nbdev_ch);
	} else {
		return _bdev_nvme_find_io_path_min_qd(nbdev_ch);
	}
}

/* Return true if there is any io_path whose qpair is active or ctrlr is not failed,
 * or false otherwise.
 *
 * If any io_path has an active qpair but find_io_path() returned NULL, its namespace
 * is likely to be non-accessible now but may become accessible.
 *
 * If any io_path has an unfailed ctrlr but find_io_path() returned NULL, the ctrlr
 * is likely to be resetting now but the reset may succeed. A ctrlr is set to unfailed
 * when starting to reset it but it is set to failed when the reset failed. Hence, if
 * a ctrlr is unfailed, it is likely that it works fine or is resetting.
 */
static bool
any_io_path_may_become_available(struct nvme_bdev_channel *nbdev_ch)
{
	struct nvme_io_path *io_path;

	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
		if (io_path->nvme_ns->ana_transition_timedout) {
			continue;
		}

		if (nvme_qpair_is_connected(io_path->qpair) ||
		    !nvme_ctrlr_is_failed(io_path->qpair->ctrlr)) {
			return true;
		}
	}

	return false;
}

static void
bdev_nvme_retry_io(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
{
	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
	struct spdk_io_channel *ch;

	if (nbdev_io->io_path != NULL && nvme_io_path_is_available(nbdev_io->io_path)) {
		_bdev_nvme_submit_request(nbdev_ch, bdev_io);
	} else {
		ch = spdk_io_channel_from_ctx(nbdev_ch);
		bdev_nvme_submit_request(ch, bdev_io);
	}
}

static int
bdev_nvme_retry_ios(void *arg)
{
	struct nvme_bdev_channel *nbdev_ch = arg;
	struct nvme_bdev_io *bio, *tmp_bio;
	uint64_t now, delay_us;

	now = spdk_get_ticks();

	TAILQ_FOREACH_SAFE(bio, &nbdev_ch->retry_io_list, retry_link, tmp_bio) {
		if (bio->retry_ticks > now) {
			break;
		}

		TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);

		bdev_nvme_retry_io(nbdev_ch, spdk_bdev_io_from_ctx(bio));
	}

	spdk_poller_unregister(&nbdev_ch->retry_io_poller);

	bio = TAILQ_FIRST(&nbdev_ch->retry_io_list);
	if (bio != NULL) {
		delay_us = (bio->retry_ticks - now) * SPDK_SEC_TO_USEC / spdk_get_ticks_hz();

		nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
					    delay_us);
	}

	return SPDK_POLLER_BUSY;
}

static void
bdev_nvme_queue_retry_io(struct nvme_bdev_channel *nbdev_ch,
			 struct nvme_bdev_io *bio, uint64_t delay_ms)
{
	struct nvme_bdev_io *tmp_bio;

	bio->retry_ticks = spdk_get_ticks() + delay_ms * spdk_get_ticks_hz() / 1000ULL;

	TAILQ_FOREACH_REVERSE(tmp_bio, &nbdev_ch->retry_io_list, retry_io_head, retry_link) {
		if (tmp_bio->retry_ticks <= bio->retry_ticks) {
			TAILQ_INSERT_AFTER(&nbdev_ch->retry_io_list, tmp_bio, bio,
					   retry_link);
			return;
		}
	}

	/* No earlier I/Os were found. This I/O must be the new head. */
	TAILQ_INSERT_HEAD(&nbdev_ch->retry_io_list, bio, retry_link);

	spdk_poller_unregister(&nbdev_ch->retry_io_poller);

	nbdev_ch->retry_io_poller = SPDK_POLLER_REGISTER(bdev_nvme_retry_ios, nbdev_ch,
				    delay_ms * 1000ULL);
}

static void
bdev_nvme_abort_retry_ios(struct nvme_bdev_channel *nbdev_ch)
{
	struct nvme_bdev_io *bio, *tmp_bio;

	TAILQ_FOREACH_SAFE(bio, &nbdev_ch->retry_io_list, retry_link, tmp_bio) {
		TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
		__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_ABORTED, NULL);
	}

	spdk_poller_unregister(&nbdev_ch->retry_io_poller);
}

static int
bdev_nvme_abort_retry_io(struct nvme_bdev_channel *nbdev_ch,
			 struct nvme_bdev_io *bio_to_abort)
{
	struct nvme_bdev_io *bio;

	TAILQ_FOREACH(bio, &nbdev_ch->retry_io_list, retry_link) {
		if (bio == bio_to_abort) {
			TAILQ_REMOVE(&nbdev_ch->retry_io_list, bio, retry_link);
			__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_ABORTED, NULL);
			return 0;
		}
	}

	return -ENOENT;
}

static void
bdev_nvme_update_nvme_error_stat(struct spdk_bdev_io *bdev_io, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev *nbdev;
	uint16_t sct, sc;

	assert(spdk_nvme_cpl_is_error(cpl));

	nbdev = bdev_io->bdev->ctxt;

	if (nbdev->err_stat == NULL) {
		return;
	}

	sct = cpl->status.sct;
	sc = cpl->status.sc;

	pthread_mutex_lock(&nbdev->mutex);

	nbdev->err_stat->status_type[sct]++;
	switch (sct) {
	case SPDK_NVME_SCT_GENERIC:
	case SPDK_NVME_SCT_COMMAND_SPECIFIC:
	case SPDK_NVME_SCT_MEDIA_ERROR:
	case SPDK_NVME_SCT_PATH:
		nbdev->err_stat->status[sct][sc]++;
		break;
	default:
		break;
	}

	pthread_mutex_unlock(&nbdev->mutex);
}

static inline void
bdev_nvme_update_io_path_stat(struct nvme_bdev_io *bio)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
	uint32_t blocklen = bdev_io->bdev->blocklen;
	struct spdk_bdev_io_stat *stat;
	uint64_t tsc_diff;

	if (bio->io_path->stat == NULL) {
		return;
	}

	tsc_diff = spdk_get_ticks() - bio->submit_tsc;
	stat = bio->io_path->stat;

	switch (bdev_io->type) {
	case SPDK_BDEV_IO_TYPE_READ:
		stat->bytes_read += num_blocks * blocklen;
		stat->num_read_ops++;
		stat->read_latency_ticks += tsc_diff;
		if (stat->max_read_latency_ticks < tsc_diff) {
			stat->max_read_latency_ticks = tsc_diff;
		}
		if (stat->min_read_latency_ticks > tsc_diff) {
			stat->min_read_latency_ticks = tsc_diff;
		}
		break;
	case SPDK_BDEV_IO_TYPE_WRITE:
		stat->bytes_written += num_blocks * blocklen;
		stat->num_write_ops++;
		stat->write_latency_ticks += tsc_diff;
		if (stat->max_write_latency_ticks < tsc_diff) {
			stat->max_write_latency_ticks = tsc_diff;
		}
		if (stat->min_write_latency_ticks > tsc_diff) {
			stat->min_write_latency_ticks = tsc_diff;
		}
		break;
	case SPDK_BDEV_IO_TYPE_UNMAP:
		stat->bytes_unmapped += num_blocks * blocklen;
		stat->num_unmap_ops++;
		stat->unmap_latency_ticks += tsc_diff;
		if (stat->max_unmap_latency_ticks < tsc_diff) {
			stat->max_unmap_latency_ticks = tsc_diff;
		}
		if (stat->min_unmap_latency_ticks > tsc_diff) {
			stat->min_unmap_latency_ticks = tsc_diff;
		}
		break;
	case SPDK_BDEV_IO_TYPE_ZCOPY:
		/* Track the data in the start phase only */
		if (!bdev_io->u.bdev.zcopy.start) {
			break;
		}
		if (bdev_io->u.bdev.zcopy.populate) {
			stat->bytes_read += num_blocks * blocklen;
			stat->num_read_ops++;
			stat->read_latency_ticks += tsc_diff;
			if (stat->max_read_latency_ticks < tsc_diff) {
				stat->max_read_latency_ticks = tsc_diff;
			}
			if (stat->min_read_latency_ticks > tsc_diff) {
				stat->min_read_latency_ticks = tsc_diff;
			}
		} else {
			stat->bytes_written += num_blocks * blocklen;
			stat->num_write_ops++;
			stat->write_latency_ticks += tsc_diff;
			if (stat->max_write_latency_ticks < tsc_diff) {
				stat->max_write_latency_ticks = tsc_diff;
			}
			if (stat->min_write_latency_ticks > tsc_diff) {
				stat->min_write_latency_ticks = tsc_diff;
			}
		}
		break;
	case SPDK_BDEV_IO_TYPE_COPY:
		stat->bytes_copied += num_blocks * blocklen;
		stat->num_copy_ops++;
		stat->copy_latency_ticks += tsc_diff;
		if (stat->max_copy_latency_ticks < tsc_diff) {
			stat->max_copy_latency_ticks = tsc_diff;
		}
		if (stat->min_copy_latency_ticks > tsc_diff) {
			stat->min_copy_latency_ticks = tsc_diff;
		}
		break;
	default:
		break;
	}
}

static bool
bdev_nvme_check_retry_io(struct nvme_bdev_io *bio,
			 const struct spdk_nvme_cpl *cpl,
			 struct nvme_bdev_channel *nbdev_ch,
			 uint64_t *_delay_ms)
{
	struct nvme_io_path *io_path = bio->io_path;
	struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;

	if (spdk_nvme_cpl_is_path_error(cpl) ||
	    spdk_nvme_cpl_is_aborted_sq_deletion(cpl) ||
	    !nvme_io_path_is_available(io_path) ||
	    !nvme_ctrlr_is_available(nvme_ctrlr)) {
		bdev_nvme_clear_current_io_path(nbdev_ch);
		bio->io_path = NULL;
		if (spdk_nvme_cpl_is_ana_error(cpl)) {
			if (nvme_ctrlr_read_ana_log_page(nvme_ctrlr) == 0) {
				io_path->nvme_ns->ana_state_updating = true;
			}
		}
		if (!any_io_path_may_become_available(nbdev_ch)) {
			return false;
		}
		*_delay_ms = 0;
	} else {
		bio->retry_count++;

		cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);

		if (cpl->status.crd != 0) {
			*_delay_ms = cdata->crdt[cpl->status.crd] * 100;
		} else {
			*_delay_ms = 0;
		}
	}

	return true;
}

static inline void
bdev_nvme_io_complete_nvme_status(struct nvme_bdev_io *bio,
				  const struct spdk_nvme_cpl *cpl)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	struct nvme_bdev_channel *nbdev_ch;
	uint64_t delay_ms;

	assert(!bdev_nvme_io_type_is_admin(bdev_io->type));

	if (spdk_likely(spdk_nvme_cpl_is_success(cpl))) {
		bdev_nvme_update_io_path_stat(bio);
		goto complete;
	}

	/* Update error counts before deciding if retry is needed.
	 * Hence, error counts may be more than the number of I/O errors.
	 */
	bdev_nvme_update_nvme_error_stat(bdev_io, cpl);

	if (cpl->status.dnr != 0 || spdk_nvme_cpl_is_aborted_by_request(cpl) ||
	    (g_opts.bdev_retry_count != -1 && bio->retry_count >= g_opts.bdev_retry_count)) {
		goto complete;
	}

	/* At this point we don't know whether the sequence was successfully executed or not, so we
	 * cannot retry the IO */
	if (bdev_io->u.bdev.accel_sequence != NULL) {
		goto complete;
	}

	nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));

	if (bdev_nvme_check_retry_io(bio, cpl, nbdev_ch, &delay_ms)) {
		bdev_nvme_queue_retry_io(nbdev_ch, bio, delay_ms);
		return;
	}

complete:
	bio->retry_count = 0;
	bio->submit_tsc = 0;
	bdev_io->u.bdev.accel_sequence = NULL;
	__bdev_nvme_io_complete(bdev_io, 0, cpl);
}

static inline void
bdev_nvme_io_complete(struct nvme_bdev_io *bio, int rc)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	struct nvme_bdev_channel *nbdev_ch;
	enum spdk_bdev_io_status io_status;

	assert(!bdev_nvme_io_type_is_admin(bdev_io->type));

	switch (rc) {
	case 0:
		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
		break;
	case -ENOMEM:
		io_status = SPDK_BDEV_IO_STATUS_NOMEM;
		break;
	case -ENXIO:
		if (g_opts.bdev_retry_count == -1 || bio->retry_count < g_opts.bdev_retry_count) {
			nbdev_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io));

			bdev_nvme_clear_current_io_path(nbdev_ch);
			bio->io_path = NULL;

			if (any_io_path_may_become_available(nbdev_ch)) {
				bdev_nvme_queue_retry_io(nbdev_ch, bio, 1000ULL);
				return;
			}
		}

	/* fallthrough */
	default:
		spdk_accel_sequence_abort(bdev_io->u.bdev.accel_sequence);
		bdev_io->u.bdev.accel_sequence = NULL;
		io_status = SPDK_BDEV_IO_STATUS_FAILED;
		break;
	}

	bio->retry_count = 0;
	bio->submit_tsc = 0;
	__bdev_nvme_io_complete(bdev_io, io_status, NULL);
}

static inline void
bdev_nvme_admin_complete(struct nvme_bdev_io *bio, int rc)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	enum spdk_bdev_io_status io_status;

	switch (rc) {
	case 0:
		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
		break;
	case -ENOMEM:
		io_status = SPDK_BDEV_IO_STATUS_NOMEM;
		break;
	case -ENXIO:
	/* fallthrough */
	default:
		io_status = SPDK_BDEV_IO_STATUS_FAILED;
		break;
	}

	__bdev_nvme_io_complete(bdev_io, io_status, NULL);
}

static void
bdev_nvme_clear_io_path_caches_done(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);

	pthread_mutex_lock(&nvme_ctrlr->mutex);

	assert(nvme_ctrlr->io_path_cache_clearing == true);
	nvme_ctrlr->io_path_cache_clearing = false;

	if (!nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return;
	}

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	nvme_ctrlr_unregister(nvme_ctrlr);
}

static void
_bdev_nvme_clear_io_path_cache(struct nvme_qpair *nvme_qpair)
{
	struct nvme_io_path *io_path;

	TAILQ_FOREACH(io_path, &nvme_qpair->io_path_list, tailq) {
		if (io_path->nbdev_ch == NULL) {
			continue;
		}
		bdev_nvme_clear_current_io_path(io_path->nbdev_ch);
	}
}

static void
bdev_nvme_clear_io_path_cache(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(_ch);

	assert(ctrlr_ch->qpair != NULL);

	_bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);

	spdk_for_each_channel_continue(i, 0);
}

static void
bdev_nvme_clear_io_path_caches(struct nvme_ctrlr *nvme_ctrlr)
{
	pthread_mutex_lock(&nvme_ctrlr->mutex);
	if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
	    nvme_ctrlr->io_path_cache_clearing) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return;
	}

	nvme_ctrlr->io_path_cache_clearing = true;
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_for_each_channel(nvme_ctrlr,
			      bdev_nvme_clear_io_path_cache,
			      NULL,
			      bdev_nvme_clear_io_path_caches_done);
}

static struct nvme_qpair *
nvme_poll_group_get_qpair(struct nvme_poll_group *group, struct spdk_nvme_qpair *qpair)
{
	struct nvme_qpair *nvme_qpair;

	TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
		if (nvme_qpair->qpair == qpair) {
			break;
		}
	}

	return nvme_qpair;
}

static void nvme_qpair_delete(struct nvme_qpair *nvme_qpair);

static void
bdev_nvme_disconnected_qpair_cb(struct spdk_nvme_qpair *qpair, void *poll_group_ctx)
{
	struct nvme_poll_group *group = poll_group_ctx;
	struct nvme_qpair *nvme_qpair;
	struct nvme_ctrlr_channel *ctrlr_ch;
	int status;

	nvme_qpair = nvme_poll_group_get_qpair(group, qpair);
	if (nvme_qpair == NULL) {
		return;
	}

	if (nvme_qpair->qpair != NULL) {
		spdk_nvme_ctrlr_free_io_qpair(nvme_qpair->qpair);
		nvme_qpair->qpair = NULL;
	}

	_bdev_nvme_clear_io_path_cache(nvme_qpair);

	ctrlr_ch = nvme_qpair->ctrlr_ch;

	if (ctrlr_ch != NULL) {
		if (ctrlr_ch->reset_iter != NULL) {
			/* We are in a full reset sequence. */
			if (ctrlr_ch->connect_poller != NULL) {
				/* qpair was failed to connect. Abort the reset sequence. */
				SPDK_DEBUGLOG(bdev_nvme, "qpair %p was failed to connect. abort the reset ctrlr sequence.\n",
					      qpair);
				spdk_poller_unregister(&ctrlr_ch->connect_poller);
				status = -1;
			} else {
				/* qpair was completed to disconnect. Just move to the next ctrlr_channel. */
				SPDK_DEBUGLOG(bdev_nvme, "qpair %p was disconnected and freed in a reset ctrlr sequence.\n",
					      qpair);
				status = 0;
			}
			spdk_for_each_channel_continue(ctrlr_ch->reset_iter, status);
			ctrlr_ch->reset_iter = NULL;
		} else {
			/* qpair was disconnected unexpectedly. Reset controller for recovery. */
			SPDK_NOTICELOG("qpair %p was disconnected and freed. reset controller.\n", qpair);
			bdev_nvme_failover_ctrlr(nvme_qpair->ctrlr);
		}
	} else {
		/* In this case, ctrlr_channel is already deleted. */
		SPDK_DEBUGLOG(bdev_nvme, "qpair %p was disconnected and freed. delete nvme_qpair.\n", qpair);
		nvme_qpair_delete(nvme_qpair);
	}
}

static void
bdev_nvme_check_io_qpairs(struct nvme_poll_group *group)
{
	struct nvme_qpair *nvme_qpair;

	TAILQ_FOREACH(nvme_qpair, &group->qpair_list, tailq) {
		if (nvme_qpair->qpair == NULL || nvme_qpair->ctrlr_ch == NULL) {
			continue;
		}

		if (spdk_nvme_qpair_get_failure_reason(nvme_qpair->qpair) !=
		    SPDK_NVME_QPAIR_FAILURE_NONE) {
			_bdev_nvme_clear_io_path_cache(nvme_qpair);
		}
	}
}

static int
bdev_nvme_poll(void *arg)
{
	struct nvme_poll_group *group = arg;
	int64_t num_completions;

	if (group->collect_spin_stat && group->start_ticks == 0) {
		group->start_ticks = spdk_get_ticks();
	}

	num_completions = spdk_nvme_poll_group_process_completions(group->group, 0,
			  bdev_nvme_disconnected_qpair_cb);
	if (group->collect_spin_stat) {
		if (num_completions > 0) {
			if (group->end_ticks != 0) {
				group->spin_ticks += (group->end_ticks - group->start_ticks);
				group->end_ticks = 0;
			}
			group->start_ticks = 0;
		} else {
			group->end_ticks = spdk_get_ticks();
		}
	}

	if (spdk_unlikely(num_completions < 0)) {
		bdev_nvme_check_io_qpairs(group);
	}

	return num_completions > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE;
}

static int bdev_nvme_poll_adminq(void *arg);

static void
bdev_nvme_change_adminq_poll_period(struct nvme_ctrlr *nvme_ctrlr, uint64_t new_period_us)
{
	spdk_poller_unregister(&nvme_ctrlr->adminq_timer_poller);

	nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq,
					  nvme_ctrlr, new_period_us);
}

static int
bdev_nvme_poll_adminq(void *arg)
{
	int32_t rc;
	struct nvme_ctrlr *nvme_ctrlr = arg;
	nvme_ctrlr_disconnected_cb disconnected_cb;

	assert(nvme_ctrlr != NULL);

	rc = spdk_nvme_ctrlr_process_admin_completions(nvme_ctrlr->ctrlr);
	if (rc < 0) {
		disconnected_cb = nvme_ctrlr->disconnected_cb;
		nvme_ctrlr->disconnected_cb = NULL;

		if (disconnected_cb != NULL) {
			bdev_nvme_change_adminq_poll_period(nvme_ctrlr,
							    g_opts.nvme_adminq_poll_period_us);
			disconnected_cb(nvme_ctrlr);
		} else {
			bdev_nvme_failover_ctrlr(nvme_ctrlr);
		}
	} else if (spdk_nvme_ctrlr_get_admin_qp_failure_reason(nvme_ctrlr->ctrlr) !=
		   SPDK_NVME_QPAIR_FAILURE_NONE) {
		bdev_nvme_clear_io_path_caches(nvme_ctrlr);
	}

	return rc == 0 ? SPDK_POLLER_IDLE : SPDK_POLLER_BUSY;
}

static void
nvme_bdev_free(void *io_device)
{
	struct nvme_bdev *nvme_disk = io_device;

	pthread_mutex_destroy(&nvme_disk->mutex);
	free(nvme_disk->disk.name);
	free(nvme_disk->err_stat);
	free(nvme_disk);
}

static int
bdev_nvme_destruct(void *ctx)
{
	struct nvme_bdev *nvme_disk = ctx;
	struct nvme_ns *nvme_ns, *tmp_nvme_ns;

	SPDK_DTRACE_PROBE2(bdev_nvme_destruct, nvme_disk->nbdev_ctrlr->name, nvme_disk->nsid);

	TAILQ_FOREACH_SAFE(nvme_ns, &nvme_disk->nvme_ns_list, tailq, tmp_nvme_ns) {
		pthread_mutex_lock(&nvme_ns->ctrlr->mutex);

		nvme_ns->bdev = NULL;

		assert(nvme_ns->id > 0);

		if (nvme_ctrlr_get_ns(nvme_ns->ctrlr, nvme_ns->id) == NULL) {
			pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);

			nvme_ctrlr_release(nvme_ns->ctrlr);
			nvme_ns_free(nvme_ns);
		} else {
			pthread_mutex_unlock(&nvme_ns->ctrlr->mutex);
		}
	}

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	TAILQ_REMOVE(&nvme_disk->nbdev_ctrlr->bdevs, nvme_disk, tailq);
	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	spdk_io_device_unregister(nvme_disk, nvme_bdev_free);

	return 0;
}

static int
bdev_nvme_create_qpair(struct nvme_qpair *nvme_qpair)
{
	struct nvme_ctrlr *nvme_ctrlr;
	struct spdk_nvme_io_qpair_opts opts;
	struct spdk_nvme_qpair *qpair;
	int rc;

	nvme_ctrlr = nvme_qpair->ctrlr;

	spdk_nvme_ctrlr_get_default_io_qpair_opts(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
	opts.delay_cmd_submit = g_opts.delay_cmd_submit;
	opts.create_only = true;
	opts.async_mode = true;
	opts.io_queue_requests = spdk_max(g_opts.io_queue_requests, opts.io_queue_requests);
	g_opts.io_queue_requests = opts.io_queue_requests;

	qpair = spdk_nvme_ctrlr_alloc_io_qpair(nvme_ctrlr->ctrlr, &opts, sizeof(opts));
	if (qpair == NULL) {
		return -1;
	}

	SPDK_DTRACE_PROBE3(bdev_nvme_create_qpair, nvme_ctrlr->nbdev_ctrlr->name,
			   spdk_nvme_qpair_get_id(qpair), spdk_thread_get_id(nvme_ctrlr->thread));

	assert(nvme_qpair->group != NULL);

	rc = spdk_nvme_poll_group_add(nvme_qpair->group->group, qpair);
	if (rc != 0) {
		SPDK_ERRLOG("Unable to begin polling on NVMe Channel.\n");
		goto err;
	}

	rc = spdk_nvme_ctrlr_connect_io_qpair(nvme_ctrlr->ctrlr, qpair);
	if (rc != 0) {
		SPDK_ERRLOG("Unable to connect I/O qpair.\n");
		goto err;
	}

	nvme_qpair->qpair = qpair;

	if (!g_opts.disable_auto_failback) {
		_bdev_nvme_clear_io_path_cache(nvme_qpair);
	}

	return 0;

err:
	spdk_nvme_ctrlr_free_io_qpair(qpair);

	return rc;
}

static void
bdev_nvme_complete_pending_resets(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(_ch);
	enum spdk_bdev_io_status status = SPDK_BDEV_IO_STATUS_SUCCESS;
	struct nvme_bdev_io *bio;

	if (spdk_io_channel_iter_get_ctx(i) != NULL) {
		status = SPDK_BDEV_IO_STATUS_FAILED;
	}

	while (!TAILQ_EMPTY(&ctrlr_ch->pending_resets)) {
		bio = TAILQ_FIRST(&ctrlr_ch->pending_resets);
		TAILQ_REMOVE(&ctrlr_ch->pending_resets, bio, retry_link);
		__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), status, NULL);
	}

	spdk_for_each_channel_continue(i, 0);
}

/* This function marks the current trid as failed by storing the current ticks
 * and then sets the next trid to the active trid within a controller if exists.
 *
 * The purpose of the boolean return value is to request the caller to disconnect
 * the current trid now to try connecting the next trid.
 */
static bool
bdev_nvme_failover_trid(struct nvme_ctrlr *nvme_ctrlr, bool remove, bool start)
{
	struct nvme_path_id *path_id, *next_path;
	int rc __attribute__((unused));

	path_id = TAILQ_FIRST(&nvme_ctrlr->trids);
	assert(path_id);
	assert(path_id == nvme_ctrlr->active_path_id);
	next_path = TAILQ_NEXT(path_id, link);

	/* Update the last failed time. It means the trid is failed if its last
	 * failed time is non-zero.
	 */
	path_id->last_failed_tsc = spdk_get_ticks();

	if (next_path == NULL) {
		/* There is no alternate trid within a controller. */
		return false;
	}

	if (!start && nvme_ctrlr->opts.reconnect_delay_sec == 0) {
		/* Connect is not retried in a controller reset sequence. Connecting
		 * the next trid will be done by the next bdev_nvme_failover_ctrlr() call.
		 */
		return false;
	}

	assert(path_id->trid.trtype != SPDK_NVME_TRANSPORT_PCIE);

	SPDK_NOTICELOG("Start failover from %s:%s to %s:%s\n", path_id->trid.traddr,
		       path_id->trid.trsvcid,	next_path->trid.traddr, next_path->trid.trsvcid);

	spdk_nvme_ctrlr_fail(nvme_ctrlr->ctrlr);
	nvme_ctrlr->active_path_id = next_path;
	rc = spdk_nvme_ctrlr_set_trid(nvme_ctrlr->ctrlr, &next_path->trid);
	assert(rc == 0);
	TAILQ_REMOVE(&nvme_ctrlr->trids, path_id, link);
	if (!remove) {
		/** Shuffle the old trid to the end of the list and use the new one.
		 * Allows for round robin through multiple connections.
		 */
		TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, path_id, link);
	} else {
		free(path_id);
	}

	if (start || next_path->last_failed_tsc == 0) {
		/* bdev_nvme_failover_ctrlr() is just called or the next trid is not failed
		 * or used yet. Try the next trid now.
		 */
		return true;
	}

	if (spdk_get_ticks() > next_path->last_failed_tsc + spdk_get_ticks_hz() *
	    nvme_ctrlr->opts.reconnect_delay_sec) {
		/* Enough backoff passed since the next trid failed. Try the next trid now. */
		return true;
	}

	/* The next trid will be tried after reconnect_delay_sec seconds. */
	return false;
}

static bool
bdev_nvme_check_ctrlr_loss_timeout(struct nvme_ctrlr *nvme_ctrlr)
{
	int32_t elapsed;

	if (nvme_ctrlr->opts.ctrlr_loss_timeout_sec == 0 ||
	    nvme_ctrlr->opts.ctrlr_loss_timeout_sec == -1) {
		return false;
	}

	elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
	if (elapsed >= nvme_ctrlr->opts.ctrlr_loss_timeout_sec) {
		return true;
	} else {
		return false;
	}
}

static bool
bdev_nvme_check_fast_io_fail_timeout(struct nvme_ctrlr *nvme_ctrlr)
{
	uint32_t elapsed;

	if (nvme_ctrlr->opts.fast_io_fail_timeout_sec == 0) {
		return false;
	}

	elapsed = (spdk_get_ticks() - nvme_ctrlr->reset_start_tsc) / spdk_get_ticks_hz();
	if (elapsed >= nvme_ctrlr->opts.fast_io_fail_timeout_sec) {
		return true;
	} else {
		return false;
	}
}

static void bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success);

static void
nvme_ctrlr_disconnect(struct nvme_ctrlr *nvme_ctrlr, nvme_ctrlr_disconnected_cb cb_fn)
{
	int rc;

	rc = spdk_nvme_ctrlr_disconnect(nvme_ctrlr->ctrlr);
	if (rc != 0) {
		/* Disconnect fails if ctrlr is already resetting or removed. In this case,
		 * fail the reset sequence immediately.
		 */
		bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
		return;
	}

	/* spdk_nvme_ctrlr_disconnect() may complete asynchronously later by polling adminq.
	 * Set callback here to execute the specified operation after ctrlr is really disconnected.
	 */
	assert(nvme_ctrlr->disconnected_cb == NULL);
	nvme_ctrlr->disconnected_cb = cb_fn;

	/* During disconnection, reduce the period to poll adminq more often. */
	bdev_nvme_change_adminq_poll_period(nvme_ctrlr, 0);
}

enum bdev_nvme_op_after_reset {
	OP_NONE,
	OP_COMPLETE_PENDING_DESTRUCT,
	OP_DESTRUCT,
	OP_DELAYED_RECONNECT,
	OP_FAILOVER,
};

typedef enum bdev_nvme_op_after_reset _bdev_nvme_op_after_reset;

static _bdev_nvme_op_after_reset
bdev_nvme_check_op_after_reset(struct nvme_ctrlr *nvme_ctrlr, bool success)
{
	if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
		/* Complete pending destruct after reset completes. */
		return OP_COMPLETE_PENDING_DESTRUCT;
	} else if (nvme_ctrlr->pending_failover) {
		nvme_ctrlr->pending_failover = false;
		nvme_ctrlr->reset_start_tsc = 0;
		return OP_FAILOVER;
	} else if (success || nvme_ctrlr->opts.reconnect_delay_sec == 0) {
		nvme_ctrlr->reset_start_tsc = 0;
		return OP_NONE;
	} else if (bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
		return OP_DESTRUCT;
	} else {
		if (bdev_nvme_check_fast_io_fail_timeout(nvme_ctrlr)) {
			nvme_ctrlr->fast_io_fail_timedout = true;
		}
		return OP_DELAYED_RECONNECT;
	}
}

static int bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug);
static void bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr);

static int
bdev_nvme_reconnect_delay_timer_expired(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect_delay, nvme_ctrlr->nbdev_ctrlr->name);
	pthread_mutex_lock(&nvme_ctrlr->mutex);

	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);

	if (!nvme_ctrlr->reconnect_is_delayed) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return SPDK_POLLER_BUSY;
	}

	nvme_ctrlr->reconnect_is_delayed = false;

	if (nvme_ctrlr->destruct) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return SPDK_POLLER_BUSY;
	}

	assert(nvme_ctrlr->resetting == false);
	nvme_ctrlr->resetting = true;

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);

	bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
	return SPDK_POLLER_BUSY;
}

static void
bdev_nvme_start_reconnect_delay_timer(struct nvme_ctrlr *nvme_ctrlr)
{
	spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);

	assert(nvme_ctrlr->reconnect_is_delayed == false);
	nvme_ctrlr->reconnect_is_delayed = true;

	assert(nvme_ctrlr->reconnect_delay_timer == NULL);
	nvme_ctrlr->reconnect_delay_timer = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_delay_timer_expired,
					    nvme_ctrlr,
					    nvme_ctrlr->opts.reconnect_delay_sec * SPDK_SEC_TO_USEC);
}

static void remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr);

static void
_bdev_nvme_reset_ctrlr_complete(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
	bool success = spdk_io_channel_iter_get_ctx(i) == NULL;
	bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
	void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
	enum bdev_nvme_op_after_reset op_after_reset;

	assert(nvme_ctrlr->thread == spdk_get_thread());

	nvme_ctrlr->ctrlr_op_cb_fn = NULL;
	nvme_ctrlr->ctrlr_op_cb_arg = NULL;

	if (!success) {
		SPDK_ERRLOG("Resetting controller failed.\n");
	} else {
		SPDK_NOTICELOG("Resetting controller successful.\n");
	}

	pthread_mutex_lock(&nvme_ctrlr->mutex);
	nvme_ctrlr->resetting = false;
	nvme_ctrlr->dont_retry = false;
	nvme_ctrlr->in_failover = false;

	op_after_reset = bdev_nvme_check_op_after_reset(nvme_ctrlr, success);
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	/* Delay callbacks when the next operation is a failover. */
	if (ctrlr_op_cb_fn && op_after_reset != OP_FAILOVER) {
		ctrlr_op_cb_fn(ctrlr_op_cb_arg, success ? 0 : -1);
	}

	switch (op_after_reset) {
	case OP_COMPLETE_PENDING_DESTRUCT:
		nvme_ctrlr_unregister(nvme_ctrlr);
		break;
	case OP_DESTRUCT:
		bdev_nvme_delete_ctrlr(nvme_ctrlr, false);
		remove_discovery_entry(nvme_ctrlr);
		break;
	case OP_DELAYED_RECONNECT:
		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_start_reconnect_delay_timer);
		break;
	case OP_FAILOVER:
		nvme_ctrlr->ctrlr_op_cb_fn = ctrlr_op_cb_fn;
		nvme_ctrlr->ctrlr_op_cb_arg = ctrlr_op_cb_arg;
		bdev_nvme_failover_ctrlr(nvme_ctrlr);
		break;
	default:
		break;
	}
}

static void
bdev_nvme_reset_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr, bool success)
{
	pthread_mutex_lock(&nvme_ctrlr->mutex);
	if (!success) {
		/* Connecting the active trid failed. Set the next alternate trid to the
		 * active trid if it exists.
		 */
		if (bdev_nvme_failover_trid(nvme_ctrlr, false, false)) {
			/* The next alternate trid exists and is ready to try. Try it now. */
			pthread_mutex_unlock(&nvme_ctrlr->mutex);

			nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
			return;
		}

		/* We came here if there is no alternate trid or if the next trid exists but
		 * is not ready to try. We will try the active trid after reconnect_delay_sec
		 * seconds if it is non-zero or at the next reset call otherwise.
		 */
	} else {
		/* Connecting the active trid succeeded. Clear the last failed time because it
		 * means the trid is failed if its last failed time is non-zero.
		 */
		nvme_ctrlr->active_path_id->last_failed_tsc = 0;
	}
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	/* Make sure we clear any pending resets before returning. */
	spdk_for_each_channel(nvme_ctrlr,
			      bdev_nvme_complete_pending_resets,
			      success ? NULL : (void *)0x1,
			      _bdev_nvme_reset_ctrlr_complete);
}

static void
bdev_nvme_reset_create_qpairs_failed(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);

	bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
}

static void
bdev_nvme_reset_destroy_qpair(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(ch);
	struct nvme_qpair *nvme_qpair;

	nvme_qpair = ctrlr_ch->qpair;
	assert(nvme_qpair != NULL);

	_bdev_nvme_clear_io_path_cache(nvme_qpair);

	if (nvme_qpair->qpair != NULL) {
		if (nvme_qpair->ctrlr->dont_retry) {
			spdk_nvme_qpair_set_abort_dnr(nvme_qpair->qpair, true);
		}
		spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);

		/* The current full reset sequence will move to the next
		 * ctrlr_channel after the qpair is actually disconnected.
		 */
		assert(ctrlr_ch->reset_iter == NULL);
		ctrlr_ch->reset_iter = i;
	} else {
		spdk_for_each_channel_continue(i, 0);
	}
}

static void
bdev_nvme_reset_create_qpairs_done(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);

	if (status == 0) {
		bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, true);
	} else {
		/* Delete the added qpairs and quiesce ctrlr to make the states clean. */
		spdk_for_each_channel(nvme_ctrlr,
				      bdev_nvme_reset_destroy_qpair,
				      NULL,
				      bdev_nvme_reset_create_qpairs_failed);
	}
}

static int
bdev_nvme_reset_check_qpair_connected(void *ctx)
{
	struct nvme_ctrlr_channel *ctrlr_ch = ctx;

	if (ctrlr_ch->reset_iter == NULL) {
		/* qpair was already failed to connect and the reset sequence is being aborted. */
		assert(ctrlr_ch->connect_poller == NULL);
		assert(ctrlr_ch->qpair->qpair == NULL);
		return SPDK_POLLER_BUSY;
	}

	assert(ctrlr_ch->qpair->qpair != NULL);

	if (!spdk_nvme_qpair_is_connected(ctrlr_ch->qpair->qpair)) {
		return SPDK_POLLER_BUSY;
	}

	spdk_poller_unregister(&ctrlr_ch->connect_poller);

	/* qpair was completed to connect. Move to the next ctrlr_channel */
	spdk_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
	ctrlr_ch->reset_iter = NULL;

	if (!g_opts.disable_auto_failback) {
		_bdev_nvme_clear_io_path_cache(ctrlr_ch->qpair);
	}

	return SPDK_POLLER_BUSY;
}

static void
bdev_nvme_reset_create_qpair(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_ctrlr_channel *ctrlr_ch = spdk_io_channel_get_ctx(_ch);
	int rc;

	rc = bdev_nvme_create_qpair(ctrlr_ch->qpair);
	if (rc == 0) {
		ctrlr_ch->connect_poller = SPDK_POLLER_REGISTER(bdev_nvme_reset_check_qpair_connected,
					   ctrlr_ch, 0);

		/* The current full reset sequence will move to the next
		 * ctrlr_channel after the qpair is actually connected.
		 */
		assert(ctrlr_ch->reset_iter == NULL);
		ctrlr_ch->reset_iter = i;
	} else {
		spdk_for_each_channel_continue(i, rc);
	}
}

static void
nvme_ctrlr_check_namespaces(struct nvme_ctrlr *nvme_ctrlr)
{
	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
	struct nvme_ns *nvme_ns;

	for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
	     nvme_ns != NULL;
	     nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
		if (!spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
			SPDK_DEBUGLOG(bdev_nvme, "NSID %u was removed during reset.\n", nvme_ns->id);
			/* NS can be added again. Just nullify nvme_ns->ns. */
			nvme_ns->ns = NULL;
		}
	}
}


static int
bdev_nvme_reconnect_ctrlr_poll(void *arg)
{
	struct nvme_ctrlr *nvme_ctrlr = arg;
	int rc = -ETIMEDOUT;

	if (!bdev_nvme_check_ctrlr_loss_timeout(nvme_ctrlr)) {
		rc = spdk_nvme_ctrlr_reconnect_poll_async(nvme_ctrlr->ctrlr);
		if (rc == -EAGAIN) {
			return SPDK_POLLER_BUSY;
		}
	}

	spdk_poller_unregister(&nvme_ctrlr->reset_detach_poller);
	if (rc == 0) {
		nvme_ctrlr_check_namespaces(nvme_ctrlr);

		/* Recreate all of the I/O queue pairs */
		spdk_for_each_channel(nvme_ctrlr,
				      bdev_nvme_reset_create_qpair,
				      NULL,
				      bdev_nvme_reset_create_qpairs_done);
	} else {
		bdev_nvme_reset_ctrlr_complete(nvme_ctrlr, false);
	}
	return SPDK_POLLER_BUSY;
}

static void
bdev_nvme_reconnect_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
{
	spdk_nvme_ctrlr_reconnect_async(nvme_ctrlr->ctrlr);

	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reconnect, nvme_ctrlr->nbdev_ctrlr->name);
	assert(nvme_ctrlr->reset_detach_poller == NULL);
	nvme_ctrlr->reset_detach_poller = SPDK_POLLER_REGISTER(bdev_nvme_reconnect_ctrlr_poll,
					  nvme_ctrlr, 0);
}

static void
bdev_nvme_reset_destroy_qpair_done(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);

	SPDK_DTRACE_PROBE1(bdev_nvme_ctrlr_reset, nvme_ctrlr->nbdev_ctrlr->name);
	assert(status == 0);

	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
		bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
	} else {
		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reconnect_ctrlr);
	}
}

static void
bdev_nvme_reset_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
{
	spdk_for_each_channel(nvme_ctrlr,
			      bdev_nvme_reset_destroy_qpair,
			      NULL,
			      bdev_nvme_reset_destroy_qpair_done);
}

static void
bdev_nvme_reconnect_ctrlr_now(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	assert(nvme_ctrlr->resetting == true);
	assert(nvme_ctrlr->thread == spdk_get_thread());

	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);

	spdk_poller_resume(nvme_ctrlr->adminq_timer_poller);

	bdev_nvme_reconnect_ctrlr(nvme_ctrlr);
}

static void
_bdev_nvme_reset_ctrlr(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	assert(nvme_ctrlr->resetting == true);
	assert(nvme_ctrlr->thread == spdk_get_thread());

	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_reset_destroy_qpairs);
	} else {
		bdev_nvme_reset_destroy_qpairs(nvme_ctrlr);
	}
}

static int
bdev_nvme_reset_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
{
	spdk_msg_fn msg_fn;

	pthread_mutex_lock(&nvme_ctrlr->mutex);
	if (nvme_ctrlr->destruct) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -ENXIO;
	}

	if (nvme_ctrlr->resetting) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		SPDK_NOTICELOG("Unable to perform reset, already in progress.\n");
		return -EBUSY;
	}

	if (nvme_ctrlr->disabled) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		SPDK_NOTICELOG("Unable to perform reset. Controller is disabled.\n");
		return -EALREADY;
	}

	nvme_ctrlr->resetting = true;
	nvme_ctrlr->dont_retry = true;

	if (nvme_ctrlr->reconnect_is_delayed) {
		SPDK_DEBUGLOG(bdev_nvme, "Reconnect is already scheduled.\n");
		msg_fn = bdev_nvme_reconnect_ctrlr_now;
		nvme_ctrlr->reconnect_is_delayed = false;
	} else {
		msg_fn = _bdev_nvme_reset_ctrlr;
		assert(nvme_ctrlr->reset_start_tsc == 0);
	}

	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
	return 0;
}

static int
bdev_nvme_enable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
{
	pthread_mutex_lock(&nvme_ctrlr->mutex);
	if (nvme_ctrlr->destruct) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -ENXIO;
	}

	if (nvme_ctrlr->resetting) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -EBUSY;
	}

	if (!nvme_ctrlr->disabled) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -EALREADY;
	}

	nvme_ctrlr->disabled = false;
	nvme_ctrlr->resetting = true;

	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_thread_send_msg(nvme_ctrlr->thread, bdev_nvme_reconnect_ctrlr_now, nvme_ctrlr);
	return 0;
}

static void
_bdev_nvme_disable_ctrlr_complete(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);
	bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn = nvme_ctrlr->ctrlr_op_cb_fn;
	void *ctrlr_op_cb_arg = nvme_ctrlr->ctrlr_op_cb_arg;
	enum bdev_nvme_op_after_reset op_after_disable;

	assert(nvme_ctrlr->thread == spdk_get_thread());

	nvme_ctrlr->ctrlr_op_cb_fn = NULL;
	nvme_ctrlr->ctrlr_op_cb_arg = NULL;

	pthread_mutex_lock(&nvme_ctrlr->mutex);

	nvme_ctrlr->resetting = false;
	nvme_ctrlr->dont_retry = false;

	op_after_disable = bdev_nvme_check_op_after_reset(nvme_ctrlr, true);

	nvme_ctrlr->disabled = true;
	spdk_poller_pause(nvme_ctrlr->adminq_timer_poller);

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	if (ctrlr_op_cb_fn) {
		ctrlr_op_cb_fn(ctrlr_op_cb_arg, 0);
	}

	switch (op_after_disable) {
	case OP_COMPLETE_PENDING_DESTRUCT:
		nvme_ctrlr_unregister(nvme_ctrlr);
		break;
	default:
		break;
	}

}

static void
bdev_nvme_disable_ctrlr_complete(struct nvme_ctrlr *nvme_ctrlr)
{
	/* Make sure we clear any pending resets before returning. */
	spdk_for_each_channel(nvme_ctrlr,
			      bdev_nvme_complete_pending_resets,
			      NULL,
			      _bdev_nvme_disable_ctrlr_complete);
}

static void
bdev_nvme_disable_destroy_qpairs_done(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ctrlr *nvme_ctrlr = spdk_io_channel_iter_get_io_device(i);

	assert(status == 0);

	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
		bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
	} else {
		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_ctrlr_complete);
	}
}

static void
bdev_nvme_disable_destroy_qpairs(struct nvme_ctrlr *nvme_ctrlr)
{
	spdk_for_each_channel(nvme_ctrlr,
			      bdev_nvme_reset_destroy_qpair,
			      NULL,
			      bdev_nvme_disable_destroy_qpairs_done);
}

static void
_bdev_nvme_cancel_reconnect_and_disable_ctrlr(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	assert(nvme_ctrlr->resetting == true);
	assert(nvme_ctrlr->thread == spdk_get_thread());

	spdk_poller_unregister(&nvme_ctrlr->reconnect_delay_timer);

	bdev_nvme_disable_ctrlr_complete(nvme_ctrlr);
}

static void
_bdev_nvme_disconnect_and_disable_ctrlr(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	assert(nvme_ctrlr->resetting == true);
	assert(nvme_ctrlr->thread == spdk_get_thread());

	if (!spdk_nvme_ctrlr_is_fabrics(nvme_ctrlr->ctrlr)) {
		nvme_ctrlr_disconnect(nvme_ctrlr, bdev_nvme_disable_destroy_qpairs);
	} else {
		bdev_nvme_disable_destroy_qpairs(nvme_ctrlr);
	}
}

static int
bdev_nvme_disable_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
{
	spdk_msg_fn msg_fn;

	pthread_mutex_lock(&nvme_ctrlr->mutex);
	if (nvme_ctrlr->destruct) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -ENXIO;
	}

	if (nvme_ctrlr->resetting) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -EBUSY;
	}

	if (nvme_ctrlr->disabled) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -EALREADY;
	}

	nvme_ctrlr->resetting = true;
	nvme_ctrlr->dont_retry = true;

	if (nvme_ctrlr->reconnect_is_delayed) {
		msg_fn = _bdev_nvme_cancel_reconnect_and_disable_ctrlr;
		nvme_ctrlr->reconnect_is_delayed = false;
	} else {
		msg_fn = _bdev_nvme_disconnect_and_disable_ctrlr;
	}

	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
	return 0;
}

static int
nvme_ctrlr_op(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
	      bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
{
	int rc;

	switch (op) {
	case NVME_CTRLR_OP_RESET:
		rc = bdev_nvme_reset_ctrlr(nvme_ctrlr);
		break;
	case NVME_CTRLR_OP_ENABLE:
		rc = bdev_nvme_enable_ctrlr(nvme_ctrlr);
		break;
	case NVME_CTRLR_OP_DISABLE:
		rc = bdev_nvme_disable_ctrlr(nvme_ctrlr);
		break;
	default:
		rc = -EINVAL;
		break;
	}

	if (rc == 0) {
		assert(nvme_ctrlr->ctrlr_op_cb_fn == NULL);
		assert(nvme_ctrlr->ctrlr_op_cb_arg == NULL);
		nvme_ctrlr->ctrlr_op_cb_fn = cb_fn;
		nvme_ctrlr->ctrlr_op_cb_arg = cb_arg;
	}
	return rc;
}

struct nvme_ctrlr_op_rpc_ctx {
	struct nvme_ctrlr *nvme_ctrlr;
	struct spdk_thread *orig_thread;
	enum nvme_ctrlr_op op;
	int rc;
	bdev_nvme_ctrlr_op_cb cb_fn;
	void *cb_arg;
};

static void
_nvme_ctrlr_op_rpc_complete(void *_ctx)
{
	struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;

	assert(ctx != NULL);
	assert(ctx->cb_fn != NULL);

	ctx->cb_fn(ctx->cb_arg, ctx->rc);

	free(ctx);
}

static void
nvme_ctrlr_op_rpc_complete(void *cb_arg, int rc)
{
	struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;

	ctx->rc = rc;

	spdk_thread_send_msg(ctx->orig_thread, _nvme_ctrlr_op_rpc_complete, ctx);
}

void
nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op,
		  bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
{
	struct nvme_ctrlr_op_rpc_ctx *ctx;
	int rc;

	assert(cb_fn != NULL);

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
		cb_fn(cb_arg, -ENOMEM);
		return;
	}

	ctx->orig_thread = spdk_get_thread();
	ctx->cb_fn = cb_fn;
	ctx->cb_arg = cb_arg;

	rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_ctrlr_op_rpc_complete, ctx);
	if (rc == 0) {
		return;
	} else if (rc == -EALREADY) {
		rc = 0;
	}

	nvme_ctrlr_op_rpc_complete(ctx, rc);
}

static void nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc);

static void
_nvme_bdev_ctrlr_op_rpc_continue(void *_ctx)
{
	struct nvme_ctrlr_op_rpc_ctx *ctx = _ctx;
	struct nvme_ctrlr *prev_nvme_ctrlr, *next_nvme_ctrlr;
	int rc;

	prev_nvme_ctrlr = ctx->nvme_ctrlr;
	ctx->nvme_ctrlr = NULL;

	if (ctx->rc != 0) {
		goto complete;
	}

	next_nvme_ctrlr = TAILQ_NEXT(prev_nvme_ctrlr, tailq);
	if (next_nvme_ctrlr == NULL) {
		goto complete;
	}

	rc = nvme_ctrlr_op(next_nvme_ctrlr, ctx->op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
	if (rc == 0) {
		ctx->nvme_ctrlr = next_nvme_ctrlr;
		return;
	} else if (rc == -EALREADY) {
		ctx->nvme_ctrlr = next_nvme_ctrlr;
		rc = 0;
	}

	ctx->rc = rc;

complete:
	ctx->cb_fn(ctx->cb_arg, ctx->rc);
	free(ctx);
}

static void
nvme_bdev_ctrlr_op_rpc_continue(void *cb_arg, int rc)
{
	struct nvme_ctrlr_op_rpc_ctx *ctx = cb_arg;

	ctx->rc = rc;

	spdk_thread_send_msg(ctx->orig_thread, _nvme_bdev_ctrlr_op_rpc_continue, ctx);
}

void
nvme_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op,
		       bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg)
{
	struct nvme_ctrlr_op_rpc_ctx *ctx;
	struct nvme_ctrlr *nvme_ctrlr;
	int rc;

	assert(cb_fn != NULL);

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		SPDK_ERRLOG("Failed to allocate nvme_ctrlr_op_rpc_ctx.\n");
		cb_fn(cb_arg, -ENOMEM);
		return;
	}

	ctx->orig_thread = spdk_get_thread();
	ctx->op = op;
	ctx->cb_fn = cb_fn;
	ctx->cb_arg = cb_arg;

	nvme_ctrlr = TAILQ_FIRST(&nbdev_ctrlr->ctrlrs);
	assert(nvme_ctrlr != NULL);

	rc = nvme_ctrlr_op(nvme_ctrlr, op, nvme_bdev_ctrlr_op_rpc_continue, ctx);
	if (rc == 0) {
		ctx->nvme_ctrlr = nvme_ctrlr;
		return;
	} else if (rc == -EALREADY) {
		ctx->nvme_ctrlr = nvme_ctrlr;
		rc = 0;
	}

	nvme_bdev_ctrlr_op_rpc_continue(ctx, rc);
}

static int _bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio);

static void
_bdev_nvme_reset_io_complete(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_bdev_io *bio = spdk_io_channel_iter_get_ctx(i);
	enum spdk_bdev_io_status io_status;

	if (bio->cpl.cdw0 == 0) {
		io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
	} else {
		io_status = SPDK_BDEV_IO_STATUS_FAILED;
	}

	__bdev_nvme_io_complete(spdk_bdev_io_from_ctx(bio), io_status, NULL);
}

static void
bdev_nvme_abort_bdev_channel(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);

	bdev_nvme_abort_retry_ios(nbdev_ch);

	spdk_for_each_channel_continue(i, 0);
}

static void
bdev_nvme_reset_io_complete(struct nvme_bdev_io *bio)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt;

	/* Abort all queued I/Os for retry. */
	spdk_for_each_channel(nbdev,
			      bdev_nvme_abort_bdev_channel,
			      bio,
			      _bdev_nvme_reset_io_complete);
}

static void
_bdev_nvme_reset_io_continue(void *ctx)
{
	struct nvme_bdev_io *bio = ctx;
	struct nvme_io_path *prev_io_path, *next_io_path;
	int rc;

	prev_io_path = bio->io_path;
	bio->io_path = NULL;

	if (bio->cpl.cdw0 != 0) {
		goto complete;
	}

	next_io_path = STAILQ_NEXT(prev_io_path, stailq);
	if (next_io_path == NULL) {
		goto complete;
	}

	rc = _bdev_nvme_reset_io(next_io_path, bio);
	if (rc == 0) {
		return;
	}

	bio->cpl.cdw0 = 1;

complete:
	bdev_nvme_reset_io_complete(bio);
}

static void
bdev_nvme_reset_io_continue(void *cb_arg, int rc)
{
	struct nvme_bdev_io *bio = cb_arg;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);

	bio->cpl.cdw0 = (rc == 0) ? 0 : 1;

	spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), _bdev_nvme_reset_io_continue, bio);
}

static int
_bdev_nvme_reset_io(struct nvme_io_path *io_path, struct nvme_bdev_io *bio)
{
	struct nvme_ctrlr_channel *ctrlr_ch;
	int rc;

	rc = nvme_ctrlr_op(io_path->qpair->ctrlr, NVME_CTRLR_OP_RESET,
			   bdev_nvme_reset_io_continue, bio);
	if (rc == 0) {
		assert(bio->io_path == NULL);
		bio->io_path = io_path;
	} else if (rc == -EBUSY) {
		ctrlr_ch = io_path->qpair->ctrlr_ch;
		assert(ctrlr_ch != NULL);
		/*
		 * Reset call is queued only if it is from the app framework. This is on purpose so that
		 * we don't interfere with the app framework reset strategy. i.e. we are deferring to the
		 * upper level. If they are in the middle of a reset, we won't try to schedule another one.
		 */
		TAILQ_INSERT_TAIL(&ctrlr_ch->pending_resets, bio, retry_link);
		rc = 0;
	}

	return rc;
}

static void
bdev_nvme_reset_io(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio)
{
	struct nvme_io_path *io_path;
	int rc;

	bio->cpl.cdw0 = 0;

	/* Reset all nvme_ctrlrs of a bdev controller sequentially. */
	io_path = STAILQ_FIRST(&nbdev_ch->io_path_list);
	assert(io_path != NULL);

	rc = _bdev_nvme_reset_io(io_path, bio);
	if (rc != 0) {
		/* If the current nvme_ctrlr is disabled, skip it and move to the next nvme_ctrlr. */
		bdev_nvme_reset_io_continue(bio, rc == -EALREADY);
	}
}

static int
bdev_nvme_failover_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool remove)
{
	if (nvme_ctrlr->destruct) {
		/* Don't bother resetting if the controller is in the process of being destructed. */
		return -ENXIO;
	}

	if (nvme_ctrlr->resetting) {
		if (!nvme_ctrlr->in_failover) {
			SPDK_NOTICELOG("Reset is already in progress. Defer failover until reset completes.\n");

			/* Defer failover until reset completes. */
			nvme_ctrlr->pending_failover = true;
			return -EINPROGRESS;
		} else {
			SPDK_NOTICELOG("Unable to perform failover, already in progress.\n");
			return -EBUSY;
		}
	}

	bdev_nvme_failover_trid(nvme_ctrlr, remove, true);

	if (nvme_ctrlr->reconnect_is_delayed) {
		SPDK_NOTICELOG("Reconnect is already scheduled.\n");

		/* We rely on the next reconnect for the failover. */
		return -EALREADY;
	}

	if (nvme_ctrlr->disabled) {
		SPDK_NOTICELOG("Controller is disabled.\n");

		/* We rely on the enablement for the failover. */
		return -EALREADY;
	}

	nvme_ctrlr->resetting = true;
	nvme_ctrlr->in_failover = true;

	assert(nvme_ctrlr->reset_start_tsc == 0);
	nvme_ctrlr->reset_start_tsc = spdk_get_ticks();

	return 0;
}

static int
bdev_nvme_failover_ctrlr(struct nvme_ctrlr *nvme_ctrlr)
{
	int rc;

	pthread_mutex_lock(&nvme_ctrlr->mutex);
	rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, false);
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	if (rc == 0) {
		spdk_thread_send_msg(nvme_ctrlr->thread, _bdev_nvme_reset_ctrlr, nvme_ctrlr);
	} else if (rc == -EALREADY) {
		rc = 0;
	}

	return rc;
}

static int bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks,
			   uint64_t num_blocks);

static int bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks,
				  uint64_t num_blocks);

static int bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks,
			  uint64_t src_offset_blocks,
			  uint64_t num_blocks);

static void
bdev_nvme_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
		     bool success)
{
	struct nvme_bdev_io *bio = (struct nvme_bdev_io *)bdev_io->driver_ctx;
	int ret;

	if (!success) {
		ret = -EINVAL;
		goto exit;
	}

	if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
		ret = -ENXIO;
		goto exit;
	}

	ret = bdev_nvme_readv(bio,
			      bdev_io->u.bdev.iovs,
			      bdev_io->u.bdev.iovcnt,
			      bdev_io->u.bdev.md_buf,
			      bdev_io->u.bdev.num_blocks,
			      bdev_io->u.bdev.offset_blocks,
			      bdev_io->u.bdev.dif_check_flags,
			      bdev_io->u.bdev.memory_domain,
			      bdev_io->u.bdev.memory_domain_ctx,
			      bdev_io->u.bdev.accel_sequence);

exit:
	if (spdk_unlikely(ret != 0)) {
		bdev_nvme_io_complete(bio, ret);
	}
}

static inline void
_bdev_nvme_submit_request(struct nvme_bdev_channel *nbdev_ch, struct spdk_bdev_io *bdev_io)
{
	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;
	struct spdk_bdev *bdev = bdev_io->bdev;
	struct nvme_bdev_io *nbdev_io_to_abort;
	int rc = 0;

	switch (bdev_io->type) {
	case SPDK_BDEV_IO_TYPE_READ:
		if (bdev_io->u.bdev.iovs && bdev_io->u.bdev.iovs[0].iov_base) {

			rc = bdev_nvme_readv(nbdev_io,
					     bdev_io->u.bdev.iovs,
					     bdev_io->u.bdev.iovcnt,
					     bdev_io->u.bdev.md_buf,
					     bdev_io->u.bdev.num_blocks,
					     bdev_io->u.bdev.offset_blocks,
					     bdev_io->u.bdev.dif_check_flags,
					     bdev_io->u.bdev.memory_domain,
					     bdev_io->u.bdev.memory_domain_ctx,
					     bdev_io->u.bdev.accel_sequence);
		} else {
			spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb,
					     bdev_io->u.bdev.num_blocks * bdev->blocklen);
			rc = 0;
		}
		break;
	case SPDK_BDEV_IO_TYPE_WRITE:
		rc = bdev_nvme_writev(nbdev_io,
				      bdev_io->u.bdev.iovs,
				      bdev_io->u.bdev.iovcnt,
				      bdev_io->u.bdev.md_buf,
				      bdev_io->u.bdev.num_blocks,
				      bdev_io->u.bdev.offset_blocks,
				      bdev_io->u.bdev.dif_check_flags,
				      bdev_io->u.bdev.memory_domain,
				      bdev_io->u.bdev.memory_domain_ctx,
				      bdev_io->u.bdev.accel_sequence);
		break;
	case SPDK_BDEV_IO_TYPE_COMPARE:
		rc = bdev_nvme_comparev(nbdev_io,
					bdev_io->u.bdev.iovs,
					bdev_io->u.bdev.iovcnt,
					bdev_io->u.bdev.md_buf,
					bdev_io->u.bdev.num_blocks,
					bdev_io->u.bdev.offset_blocks,
					bdev_io->u.bdev.dif_check_flags);
		break;
	case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
		rc = bdev_nvme_comparev_and_writev(nbdev_io,
						   bdev_io->u.bdev.iovs,
						   bdev_io->u.bdev.iovcnt,
						   bdev_io->u.bdev.fused_iovs,
						   bdev_io->u.bdev.fused_iovcnt,
						   bdev_io->u.bdev.md_buf,
						   bdev_io->u.bdev.num_blocks,
						   bdev_io->u.bdev.offset_blocks,
						   bdev_io->u.bdev.dif_check_flags);
		break;
	case SPDK_BDEV_IO_TYPE_UNMAP:
		rc = bdev_nvme_unmap(nbdev_io,
				     bdev_io->u.bdev.offset_blocks,
				     bdev_io->u.bdev.num_blocks);
		break;
	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
		rc =  bdev_nvme_write_zeroes(nbdev_io,
					     bdev_io->u.bdev.offset_blocks,
					     bdev_io->u.bdev.num_blocks);
		break;
	case SPDK_BDEV_IO_TYPE_RESET:
		nbdev_io->io_path = NULL;
		bdev_nvme_reset_io(nbdev_ch, nbdev_io);
		return;

	case SPDK_BDEV_IO_TYPE_FLUSH:
		bdev_nvme_io_complete(nbdev_io, 0);
		return;

	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
		rc = bdev_nvme_zone_appendv(nbdev_io,
					    bdev_io->u.bdev.iovs,
					    bdev_io->u.bdev.iovcnt,
					    bdev_io->u.bdev.md_buf,
					    bdev_io->u.bdev.num_blocks,
					    bdev_io->u.bdev.offset_blocks,
					    bdev_io->u.bdev.dif_check_flags);
		break;
	case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
		rc = bdev_nvme_get_zone_info(nbdev_io,
					     bdev_io->u.zone_mgmt.zone_id,
					     bdev_io->u.zone_mgmt.num_zones,
					     bdev_io->u.zone_mgmt.buf);
		break;
	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
		rc = bdev_nvme_zone_management(nbdev_io,
					       bdev_io->u.zone_mgmt.zone_id,
					       bdev_io->u.zone_mgmt.zone_action);
		break;
	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
		nbdev_io->io_path = NULL;
		bdev_nvme_admin_passthru(nbdev_ch,
					 nbdev_io,
					 &bdev_io->u.nvme_passthru.cmd,
					 bdev_io->u.nvme_passthru.buf,
					 bdev_io->u.nvme_passthru.nbytes);
		return;

	case SPDK_BDEV_IO_TYPE_NVME_IO:
		rc = bdev_nvme_io_passthru(nbdev_io,
					   &bdev_io->u.nvme_passthru.cmd,
					   bdev_io->u.nvme_passthru.buf,
					   bdev_io->u.nvme_passthru.nbytes);
		break;
	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
		rc = bdev_nvme_io_passthru_md(nbdev_io,
					      &bdev_io->u.nvme_passthru.cmd,
					      bdev_io->u.nvme_passthru.buf,
					      bdev_io->u.nvme_passthru.nbytes,
					      bdev_io->u.nvme_passthru.md_buf,
					      bdev_io->u.nvme_passthru.md_len);
		break;
	case SPDK_BDEV_IO_TYPE_NVME_IOV_MD:
		rc = bdev_nvme_iov_passthru_md(nbdev_io,
					       &bdev_io->u.nvme_passthru.cmd,
					       bdev_io->u.nvme_passthru.iovs,
					       bdev_io->u.nvme_passthru.iovcnt,
					       bdev_io->u.nvme_passthru.nbytes,
					       bdev_io->u.nvme_passthru.md_buf,
					       bdev_io->u.nvme_passthru.md_len);
		break;
	case SPDK_BDEV_IO_TYPE_ABORT:
		nbdev_io->io_path = NULL;
		nbdev_io_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx;
		bdev_nvme_abort(nbdev_ch,
				nbdev_io,
				nbdev_io_to_abort);
		return;

	case SPDK_BDEV_IO_TYPE_COPY:
		rc = bdev_nvme_copy(nbdev_io,
				    bdev_io->u.bdev.offset_blocks,
				    bdev_io->u.bdev.copy.src_offset_blocks,
				    bdev_io->u.bdev.num_blocks);
		break;
	default:
		rc = -EINVAL;
		break;
	}

	if (spdk_unlikely(rc != 0)) {
		bdev_nvme_io_complete(nbdev_io, rc);
	}
}

static void
bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
{
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
	struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx;

	if (spdk_likely(nbdev_io->submit_tsc == 0)) {
		nbdev_io->submit_tsc = spdk_bdev_io_get_submit_tsc(bdev_io);
	} else {
		/* There are cases where submit_tsc != 0, i.e. retry I/O.
		 * We need to update submit_tsc here.
		 */
		nbdev_io->submit_tsc = spdk_get_ticks();
	}

	spdk_trace_record(TRACE_BDEV_NVME_IO_START, 0, 0, (uintptr_t)nbdev_io, (uintptr_t)bdev_io);
	nbdev_io->io_path = bdev_nvme_find_io_path(nbdev_ch);
	if (spdk_unlikely(!nbdev_io->io_path)) {
		if (!bdev_nvme_io_type_is_admin(bdev_io->type)) {
			bdev_nvme_io_complete(nbdev_io, -ENXIO);
			return;
		}

		/* Admin commands do not use the optimal I/O path.
		 * Simply fall through even if it is not found.
		 */
	}

	_bdev_nvme_submit_request(nbdev_ch, bdev_io);
}

static bool
bdev_nvme_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type)
{
	struct nvme_bdev *nbdev = ctx;
	struct nvme_ns *nvme_ns;
	struct spdk_nvme_ns *ns;
	struct spdk_nvme_ctrlr *ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;

	nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
	assert(nvme_ns != NULL);
	ns = nvme_ns->ns;
	if (ns == NULL) {
		return false;
	}

	ctrlr = spdk_nvme_ns_get_ctrlr(ns);

	switch (io_type) {
	case SPDK_BDEV_IO_TYPE_READ:
	case SPDK_BDEV_IO_TYPE_WRITE:
	case SPDK_BDEV_IO_TYPE_RESET:
	case SPDK_BDEV_IO_TYPE_FLUSH:
	case SPDK_BDEV_IO_TYPE_NVME_ADMIN:
	case SPDK_BDEV_IO_TYPE_NVME_IO:
	case SPDK_BDEV_IO_TYPE_ABORT:
		return true;

	case SPDK_BDEV_IO_TYPE_COMPARE:
		return spdk_nvme_ns_supports_compare(ns);

	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
		return spdk_nvme_ns_get_md_size(ns) ? true : false;

	case SPDK_BDEV_IO_TYPE_UNMAP:
		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
		return cdata->oncs.dsm;

	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
		return cdata->oncs.write_zeroes;

	case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE:
		if (spdk_nvme_ctrlr_get_flags(ctrlr) &
		    SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED) {
			return true;
		}
		return false;

	case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO:
	case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT:
		return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS;

	case SPDK_BDEV_IO_TYPE_ZONE_APPEND:
		return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS &&
		       spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ZONE_APPEND_SUPPORTED;

	case SPDK_BDEV_IO_TYPE_COPY:
		cdata = spdk_nvme_ctrlr_get_data(ctrlr);
		return cdata->oncs.copy;

	default:
		return false;
	}
}

static int
nvme_qpair_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ctrlr_channel *ctrlr_ch)
{
	struct nvme_qpair *nvme_qpair;
	struct spdk_io_channel *pg_ch;
	int rc;

	nvme_qpair = calloc(1, sizeof(*nvme_qpair));
	if (!nvme_qpair) {
		SPDK_ERRLOG("Failed to alloc nvme_qpair.\n");
		return -1;
	}

	TAILQ_INIT(&nvme_qpair->io_path_list);

	nvme_qpair->ctrlr = nvme_ctrlr;
	nvme_qpair->ctrlr_ch = ctrlr_ch;

	pg_ch = spdk_get_io_channel(&g_nvme_bdev_ctrlrs);
	if (!pg_ch) {
		free(nvme_qpair);
		return -1;
	}

	nvme_qpair->group = spdk_io_channel_get_ctx(pg_ch);

#ifdef SPDK_CONFIG_VTUNE
	nvme_qpair->group->collect_spin_stat = true;
#else
	nvme_qpair->group->collect_spin_stat = false;
#endif

	if (!nvme_ctrlr->disabled) {
		/* If a nvme_ctrlr is disabled, don't try to create qpair for it. Qpair will
		 * be created when it's enabled.
		 */
		rc = bdev_nvme_create_qpair(nvme_qpair);
		if (rc != 0) {
			/* nvme_ctrlr can't create IO qpair if connection is down.
			 * If reconnect_delay_sec is non-zero, creating IO qpair is retried
			 * after reconnect_delay_sec seconds. If bdev_retry_count is non-zero,
			 * submitted IO will be queued until IO qpair is successfully created.
			 *
			 * Hence, if both are satisfied, ignore the failure.
			 */
			if (nvme_ctrlr->opts.reconnect_delay_sec == 0 || g_opts.bdev_retry_count == 0) {
				spdk_put_io_channel(pg_ch);
				free(nvme_qpair);
				return rc;
			}
		}
	}

	TAILQ_INSERT_TAIL(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);

	ctrlr_ch->qpair = nvme_qpair;

	pthread_mutex_lock(&nvme_qpair->ctrlr->mutex);
	nvme_qpair->ctrlr->ref++;
	pthread_mutex_unlock(&nvme_qpair->ctrlr->mutex);

	return 0;
}

static int
bdev_nvme_create_ctrlr_channel_cb(void *io_device, void *ctx_buf)
{
	struct nvme_ctrlr *nvme_ctrlr = io_device;
	struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;

	TAILQ_INIT(&ctrlr_ch->pending_resets);

	return nvme_qpair_create(nvme_ctrlr, ctrlr_ch);
}

static void
nvme_qpair_delete(struct nvme_qpair *nvme_qpair)
{
	struct nvme_io_path *io_path, *next;

	assert(nvme_qpair->group != NULL);

	TAILQ_FOREACH_SAFE(io_path, &nvme_qpair->io_path_list, tailq, next) {
		TAILQ_REMOVE(&nvme_qpair->io_path_list, io_path, tailq);
		nvme_io_path_free(io_path);
	}

	TAILQ_REMOVE(&nvme_qpair->group->qpair_list, nvme_qpair, tailq);

	spdk_put_io_channel(spdk_io_channel_from_ctx(nvme_qpair->group));

	nvme_ctrlr_release(nvme_qpair->ctrlr);

	free(nvme_qpair);
}

static void
bdev_nvme_destroy_ctrlr_channel_cb(void *io_device, void *ctx_buf)
{
	struct nvme_ctrlr_channel *ctrlr_ch = ctx_buf;
	struct nvme_qpair *nvme_qpair;

	nvme_qpair = ctrlr_ch->qpair;
	assert(nvme_qpair != NULL);

	_bdev_nvme_clear_io_path_cache(nvme_qpair);

	if (nvme_qpair->qpair != NULL) {
		if (ctrlr_ch->reset_iter == NULL) {
			spdk_nvme_ctrlr_disconnect_io_qpair(nvme_qpair->qpair);
		} else {
			/* Skip current ctrlr_channel in a full reset sequence because
			 * it is being deleted now. The qpair is already being disconnected.
			 * We do not have to restart disconnecting it.
			 */
			spdk_for_each_channel_continue(ctrlr_ch->reset_iter, 0);
		}

		/* We cannot release a reference to the poll group now.
		 * The qpair may be disconnected asynchronously later.
		 * We need to poll it until it is actually disconnected.
		 * Just detach the qpair from the deleting ctrlr_channel.
		 */
		nvme_qpair->ctrlr_ch = NULL;
	} else {
		assert(ctrlr_ch->reset_iter == NULL);

		nvme_qpair_delete(nvme_qpair);
	}
}

static inline struct spdk_io_channel *
bdev_nvme_get_accel_channel(struct nvme_poll_group *group)
{
	if (spdk_unlikely(!group->accel_channel)) {
		group->accel_channel = spdk_accel_get_io_channel();
		if (!group->accel_channel) {
			SPDK_ERRLOG("Cannot get the accel_channel for bdev nvme polling group=%p\n",
				    group);
			return NULL;
		}
	}

	return group->accel_channel;
}

static void
bdev_nvme_submit_accel_crc32c(void *ctx, uint32_t *dst, struct iovec *iov,
			      uint32_t iov_cnt, uint32_t seed,
			      spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
{
	struct spdk_io_channel *accel_ch;
	struct nvme_poll_group *group = ctx;
	int rc;

	assert(cb_fn != NULL);

	accel_ch = bdev_nvme_get_accel_channel(group);
	if (spdk_unlikely(accel_ch == NULL)) {
		cb_fn(cb_arg, -ENOMEM);
		return;
	}

	rc = spdk_accel_submit_crc32cv(accel_ch, dst, iov, iov_cnt, seed, cb_fn, cb_arg);
	if (rc) {
		/* For the two cases, spdk_accel_submit_crc32cv does not call the user's cb_fn */
		if (rc == -ENOMEM || rc == -EINVAL) {
			cb_fn(cb_arg, rc);
		}
		SPDK_ERRLOG("Cannot complete the accelerated crc32c operation with iov=%p\n", iov);
	}
}

static void
bdev_nvme_finish_sequence(void *seq, spdk_nvme_accel_completion_cb cb_fn, void *cb_arg)
{
	spdk_accel_sequence_finish(seq, cb_fn, cb_arg);
}

static void
bdev_nvme_abort_sequence(void *seq)
{
	spdk_accel_sequence_abort(seq);
}

static void
bdev_nvme_reverse_sequence(void *seq)
{
	spdk_accel_sequence_reverse(seq);
}

static int
bdev_nvme_append_crc32c(void *ctx, void **seq, uint32_t *dst, struct iovec *iovs, uint32_t iovcnt,
			struct spdk_memory_domain *domain, void *domain_ctx, uint32_t seed,
			spdk_nvme_accel_step_cb cb_fn, void *cb_arg)
{
	struct spdk_io_channel *ch;
	struct nvme_poll_group *group = ctx;

	ch = bdev_nvme_get_accel_channel(group);
	if (spdk_unlikely(ch == NULL)) {
		return -ENOMEM;
	}

	return spdk_accel_append_crc32c((struct spdk_accel_sequence **)seq, ch, dst, iovs, iovcnt,
					domain, domain_ctx, seed, cb_fn, cb_arg);
}

static struct spdk_nvme_accel_fn_table g_bdev_nvme_accel_fn_table = {
	.table_size		= sizeof(struct spdk_nvme_accel_fn_table),
	.submit_accel_crc32c	= bdev_nvme_submit_accel_crc32c,
	.append_crc32c		= bdev_nvme_append_crc32c,
	.finish_sequence	= bdev_nvme_finish_sequence,
	.reverse_sequence	= bdev_nvme_reverse_sequence,
	.abort_sequence		= bdev_nvme_abort_sequence,
};

static int
bdev_nvme_create_poll_group_cb(void *io_device, void *ctx_buf)
{
	struct nvme_poll_group *group = ctx_buf;

	TAILQ_INIT(&group->qpair_list);

	group->group = spdk_nvme_poll_group_create(group, &g_bdev_nvme_accel_fn_table);
	if (group->group == NULL) {
		return -1;
	}

	group->poller = SPDK_POLLER_REGISTER(bdev_nvme_poll, group, g_opts.nvme_ioq_poll_period_us);

	if (group->poller == NULL) {
		spdk_nvme_poll_group_destroy(group->group);
		return -1;
	}

	return 0;
}

static void
bdev_nvme_destroy_poll_group_cb(void *io_device, void *ctx_buf)
{
	struct nvme_poll_group *group = ctx_buf;

	assert(TAILQ_EMPTY(&group->qpair_list));

	if (group->accel_channel) {
		spdk_put_io_channel(group->accel_channel);
	}

	spdk_poller_unregister(&group->poller);
	if (spdk_nvme_poll_group_destroy(group->group)) {
		SPDK_ERRLOG("Unable to destroy a poll group for the NVMe bdev module.\n");
		assert(false);
	}
}

static struct spdk_io_channel *
bdev_nvme_get_io_channel(void *ctx)
{
	struct nvme_bdev *nvme_bdev = ctx;

	return spdk_get_io_channel(nvme_bdev);
}

static void *
bdev_nvme_get_module_ctx(void *ctx)
{
	struct nvme_bdev *nvme_bdev = ctx;
	struct nvme_ns *nvme_ns;

	if (!nvme_bdev || nvme_bdev->disk.module != &nvme_if) {
		return NULL;
	}

	nvme_ns = TAILQ_FIRST(&nvme_bdev->nvme_ns_list);
	if (!nvme_ns) {
		return NULL;
	}

	return nvme_ns->ns;
}

static const char *
_nvme_ana_state_str(enum spdk_nvme_ana_state ana_state)
{
	switch (ana_state) {
	case SPDK_NVME_ANA_OPTIMIZED_STATE:
		return "optimized";
	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
		return "non_optimized";
	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
		return "inaccessible";
	case SPDK_NVME_ANA_PERSISTENT_LOSS_STATE:
		return "persistent_loss";
	case SPDK_NVME_ANA_CHANGE_STATE:
		return "change";
	default:
		return NULL;
	}
}

static int
bdev_nvme_get_memory_domains(void *ctx, struct spdk_memory_domain **domains, int array_size)
{
	struct spdk_memory_domain **_domains = NULL;
	struct nvme_bdev *nbdev = ctx;
	struct nvme_ns *nvme_ns;
	int i = 0, _array_size = array_size;
	int rc = 0;

	TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
		if (domains && array_size >= i) {
			_domains = &domains[i];
		} else {
			_domains = NULL;
		}
		rc = spdk_nvme_ctrlr_get_memory_domains(nvme_ns->ctrlr->ctrlr, _domains, _array_size);
		if (rc > 0) {
			i += rc;
			if (_array_size >= rc) {
				_array_size -= rc;
			} else {
				_array_size = 0;
			}
		} else if (rc < 0) {
			return rc;
		}
	}

	return i;
}

static const char *
nvme_ctrlr_get_state_str(struct nvme_ctrlr *nvme_ctrlr)
{
	if (nvme_ctrlr->destruct) {
		return "deleting";
	} else if (spdk_nvme_ctrlr_is_failed(nvme_ctrlr->ctrlr)) {
		return "failed";
	} else if (nvme_ctrlr->resetting) {
		return "resetting";
	} else if (nvme_ctrlr->reconnect_is_delayed > 0) {
		return "reconnect_is_delayed";
	} else if (nvme_ctrlr->disabled) {
		return "disabled";
	} else {
		return "enabled";
	}
}

void
nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr)
{
	struct spdk_nvme_transport_id *trid;
	const struct spdk_nvme_ctrlr_opts *opts;
	const struct spdk_nvme_ctrlr_data *cdata;
	struct nvme_path_id *path_id;

	spdk_json_write_object_begin(w);

	spdk_json_write_named_string(w, "state", nvme_ctrlr_get_state_str(nvme_ctrlr));

#ifdef SPDK_CONFIG_NVME_CUSE
	size_t cuse_name_size = 128;
	char cuse_name[cuse_name_size];

	int rc = spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr, cuse_name, &cuse_name_size);
	if (rc == 0) {
		spdk_json_write_named_string(w, "cuse_device", cuse_name);
	}
#endif
	trid = &nvme_ctrlr->active_path_id->trid;
	spdk_json_write_named_object_begin(w, "trid");
	nvme_bdev_dump_trid_json(trid, w);
	spdk_json_write_object_end(w);

	path_id = TAILQ_NEXT(nvme_ctrlr->active_path_id, link);
	if (path_id != NULL) {
		spdk_json_write_named_array_begin(w, "alternate_trids");
		do {
			trid = &path_id->trid;
			spdk_json_write_object_begin(w);
			nvme_bdev_dump_trid_json(trid, w);
			spdk_json_write_object_end(w);

			path_id = TAILQ_NEXT(path_id, link);
		} while (path_id != NULL);
		spdk_json_write_array_end(w);
	}

	cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
	spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);

	opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
	spdk_json_write_named_object_begin(w, "host");
	spdk_json_write_named_string(w, "nqn", opts->hostnqn);
	spdk_json_write_named_string(w, "addr", opts->src_addr);
	spdk_json_write_named_string(w, "svcid", opts->src_svcid);
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

static void
nvme_namespace_info_json(struct spdk_json_write_ctx *w,
			 struct nvme_ns *nvme_ns)
{
	struct spdk_nvme_ns *ns;
	struct spdk_nvme_ctrlr *ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;
	const struct spdk_nvme_transport_id *trid;
	union spdk_nvme_vs_register vs;
	const struct spdk_nvme_ns_data *nsdata;
	char buf[128];

	ns = nvme_ns->ns;
	if (ns == NULL) {
		return;
	}

	ctrlr = spdk_nvme_ns_get_ctrlr(ns);

	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
	trid = spdk_nvme_ctrlr_get_transport_id(ctrlr);
	vs = spdk_nvme_ctrlr_get_regs_vs(ctrlr);

	spdk_json_write_object_begin(w);

	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
		spdk_json_write_named_string(w, "pci_address", trid->traddr);
	}

	spdk_json_write_named_object_begin(w, "trid");

	nvme_bdev_dump_trid_json(trid, w);

	spdk_json_write_object_end(w);

#ifdef SPDK_CONFIG_NVME_CUSE
	size_t cuse_name_size = 128;
	char cuse_name[cuse_name_size];

	int rc = spdk_nvme_cuse_get_ns_name(ctrlr, spdk_nvme_ns_get_id(ns),
					    cuse_name, &cuse_name_size);
	if (rc == 0) {
		spdk_json_write_named_string(w, "cuse_device", cuse_name);
	}
#endif

	spdk_json_write_named_object_begin(w, "ctrlr_data");

	spdk_json_write_named_uint16(w, "cntlid", cdata->cntlid);

	spdk_json_write_named_string_fmt(w, "vendor_id", "0x%04x", cdata->vid);

	snprintf(buf, sizeof(cdata->mn) + 1, "%s", cdata->mn);
	spdk_str_trim(buf);
	spdk_json_write_named_string(w, "model_number", buf);

	snprintf(buf, sizeof(cdata->sn) + 1, "%s", cdata->sn);
	spdk_str_trim(buf);
	spdk_json_write_named_string(w, "serial_number", buf);

	snprintf(buf, sizeof(cdata->fr) + 1, "%s", cdata->fr);
	spdk_str_trim(buf);
	spdk_json_write_named_string(w, "firmware_revision", buf);

	if (cdata->subnqn[0] != '\0') {
		spdk_json_write_named_string(w, "subnqn", cdata->subnqn);
	}

	spdk_json_write_named_object_begin(w, "oacs");

	spdk_json_write_named_uint32(w, "security", cdata->oacs.security);
	spdk_json_write_named_uint32(w, "format", cdata->oacs.format);
	spdk_json_write_named_uint32(w, "firmware", cdata->oacs.firmware);
	spdk_json_write_named_uint32(w, "ns_manage", cdata->oacs.ns_manage);

	spdk_json_write_object_end(w);

	spdk_json_write_named_bool(w, "multi_ctrlr", cdata->cmic.multi_ctrlr);
	spdk_json_write_named_bool(w, "ana_reporting", cdata->cmic.ana_reporting);

	spdk_json_write_object_end(w);

	spdk_json_write_named_object_begin(w, "vs");

	spdk_json_write_name(w, "nvme_version");
	if (vs.bits.ter) {
		spdk_json_write_string_fmt(w, "%u.%u.%u", vs.bits.mjr, vs.bits.mnr, vs.bits.ter);
	} else {
		spdk_json_write_string_fmt(w, "%u.%u", vs.bits.mjr, vs.bits.mnr);
	}

	spdk_json_write_object_end(w);

	nsdata = spdk_nvme_ns_get_data(ns);

	spdk_json_write_named_object_begin(w, "ns_data");

	spdk_json_write_named_uint32(w, "id", spdk_nvme_ns_get_id(ns));

	if (cdata->cmic.ana_reporting) {
		spdk_json_write_named_string(w, "ana_state",
					     _nvme_ana_state_str(nvme_ns->ana_state));
	}

	spdk_json_write_named_bool(w, "can_share", nsdata->nmic.can_share);

	spdk_json_write_object_end(w);

	if (cdata->oacs.security) {
		spdk_json_write_named_object_begin(w, "security");

		spdk_json_write_named_bool(w, "opal", nvme_ns->bdev->opal);

		spdk_json_write_object_end(w);
	}

	spdk_json_write_object_end(w);
}

static const char *
nvme_bdev_get_mp_policy_str(struct nvme_bdev *nbdev)
{
	switch (nbdev->mp_policy) {
	case BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE:
		return "active_passive";
	case BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE:
		return "active_active";
	default:
		assert(false);
		return "invalid";
	}
}

static int
bdev_nvme_dump_info_json(void *ctx, struct spdk_json_write_ctx *w)
{
	struct nvme_bdev *nvme_bdev = ctx;
	struct nvme_ns *nvme_ns;

	pthread_mutex_lock(&nvme_bdev->mutex);
	spdk_json_write_named_array_begin(w, "nvme");
	TAILQ_FOREACH(nvme_ns, &nvme_bdev->nvme_ns_list, tailq) {
		nvme_namespace_info_json(w, nvme_ns);
	}
	spdk_json_write_array_end(w);
	spdk_json_write_named_string(w, "mp_policy", nvme_bdev_get_mp_policy_str(nvme_bdev));
	pthread_mutex_unlock(&nvme_bdev->mutex);

	return 0;
}

static void
bdev_nvme_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
{
	/* No config per bdev needed */
}

static uint64_t
bdev_nvme_get_spin_time(struct spdk_io_channel *ch)
{
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(ch);
	struct nvme_io_path *io_path;
	struct nvme_poll_group *group;
	uint64_t spin_time = 0;

	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
		group = io_path->qpair->group;

		if (!group || !group->collect_spin_stat) {
			continue;
		}

		if (group->end_ticks != 0) {
			group->spin_ticks += (group->end_ticks - group->start_ticks);
			group->end_ticks = 0;
		}

		spin_time += group->spin_ticks;
		group->start_ticks = 0;
		group->spin_ticks = 0;
	}

	return (spin_time * 1000000ULL) / spdk_get_ticks_hz();
}

static void
bdev_nvme_reset_device_stat(void *ctx)
{
	struct nvme_bdev *nbdev = ctx;

	if (nbdev->err_stat != NULL) {
		memset(nbdev->err_stat, 0, sizeof(struct nvme_error_stat));
	}
}

/* JSON string should be lowercases and underscore delimited string. */
static void
bdev_nvme_format_nvme_status(char *dst, const char *src)
{
	char tmp[256];

	spdk_strcpy_replace(dst, 256, src, " - ", "_");
	spdk_strcpy_replace(tmp, 256, dst, "-", "_");
	spdk_strcpy_replace(dst, 256, tmp, " ", "_");
	spdk_strlwr(dst);
}

static void
bdev_nvme_dump_device_stat_json(void *ctx, struct spdk_json_write_ctx *w)
{
	struct nvme_bdev *nbdev = ctx;
	struct spdk_nvme_status status = {};
	uint16_t sct, sc;
	char status_json[256];
	const char *status_str;

	if (nbdev->err_stat == NULL) {
		return;
	}

	spdk_json_write_named_object_begin(w, "nvme_error");

	spdk_json_write_named_object_begin(w, "status_type");
	for (sct = 0; sct < 8; sct++) {
		if (nbdev->err_stat->status_type[sct] == 0) {
			continue;
		}
		status.sct = sct;

		status_str = spdk_nvme_cpl_get_status_type_string(&status);
		assert(status_str != NULL);
		bdev_nvme_format_nvme_status(status_json, status_str);

		spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status_type[sct]);
	}
	spdk_json_write_object_end(w);

	spdk_json_write_named_object_begin(w, "status_code");
	for (sct = 0; sct < 4; sct++) {
		status.sct = sct;
		for (sc = 0; sc < 256; sc++) {
			if (nbdev->err_stat->status[sct][sc] == 0) {
				continue;
			}
			status.sc = sc;

			status_str = spdk_nvme_cpl_get_status_string(&status);
			assert(status_str != NULL);
			bdev_nvme_format_nvme_status(status_json, status_str);

			spdk_json_write_named_uint32(w, status_json, nbdev->err_stat->status[sct][sc]);
		}
	}
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

static bool
bdev_nvme_accel_sequence_supported(void *ctx, enum spdk_bdev_io_type type)
{
	struct nvme_bdev *nbdev = ctx;
	struct spdk_nvme_ctrlr *ctrlr;

	if (!g_opts.allow_accel_sequence) {
		return false;
	}

	switch (type) {
	case SPDK_BDEV_IO_TYPE_WRITE:
	case SPDK_BDEV_IO_TYPE_READ:
		break;
	default:
		return false;
	}

	ctrlr = bdev_nvme_get_ctrlr(&nbdev->disk);
	assert(ctrlr != NULL);

	return spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ACCEL_SEQUENCE_SUPPORTED;
}

static const struct spdk_bdev_fn_table nvmelib_fn_table = {
	.destruct			= bdev_nvme_destruct,
	.submit_request			= bdev_nvme_submit_request,
	.io_type_supported		= bdev_nvme_io_type_supported,
	.get_io_channel			= bdev_nvme_get_io_channel,
	.dump_info_json			= bdev_nvme_dump_info_json,
	.write_config_json		= bdev_nvme_write_config_json,
	.get_spin_time			= bdev_nvme_get_spin_time,
	.get_module_ctx			= bdev_nvme_get_module_ctx,
	.get_memory_domains		= bdev_nvme_get_memory_domains,
	.accel_sequence_supported	= bdev_nvme_accel_sequence_supported,
	.reset_device_stat		= bdev_nvme_reset_device_stat,
	.dump_device_stat_json		= bdev_nvme_dump_device_stat_json,
};

typedef int (*bdev_nvme_parse_ana_log_page_cb)(
	const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg);

static int
bdev_nvme_parse_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
			     bdev_nvme_parse_ana_log_page_cb cb_fn, void *cb_arg)
{
	struct spdk_nvme_ana_group_descriptor *copied_desc;
	uint8_t *orig_desc;
	uint32_t i, desc_size, copy_len;
	int rc = 0;

	if (nvme_ctrlr->ana_log_page == NULL) {
		return -EINVAL;
	}

	copied_desc = nvme_ctrlr->copied_ana_desc;

	orig_desc = (uint8_t *)nvme_ctrlr->ana_log_page + sizeof(struct spdk_nvme_ana_page);
	copy_len = nvme_ctrlr->max_ana_log_page_size - sizeof(struct spdk_nvme_ana_page);

	for (i = 0; i < nvme_ctrlr->ana_log_page->num_ana_group_desc; i++) {
		memcpy(copied_desc, orig_desc, copy_len);

		rc = cb_fn(copied_desc, cb_arg);
		if (rc != 0) {
			break;
		}

		desc_size = sizeof(struct spdk_nvme_ana_group_descriptor) +
			    copied_desc->num_of_nsid * sizeof(uint32_t);
		orig_desc += desc_size;
		copy_len -= desc_size;
	}

	return rc;
}

static int
nvme_ns_ana_transition_timedout(void *ctx)
{
	struct nvme_ns *nvme_ns = ctx;

	spdk_poller_unregister(&nvme_ns->anatt_timer);
	nvme_ns->ana_transition_timedout = true;

	return SPDK_POLLER_BUSY;
}

static void
_nvme_ns_set_ana_state(struct nvme_ns *nvme_ns,
		       const struct spdk_nvme_ana_group_descriptor *desc)
{
	const struct spdk_nvme_ctrlr_data *cdata;

	nvme_ns->ana_group_id = desc->ana_group_id;
	nvme_ns->ana_state = desc->ana_state;
	nvme_ns->ana_state_updating = false;

	switch (nvme_ns->ana_state) {
	case SPDK_NVME_ANA_OPTIMIZED_STATE:
	case SPDK_NVME_ANA_NON_OPTIMIZED_STATE:
		nvme_ns->ana_transition_timedout = false;
		spdk_poller_unregister(&nvme_ns->anatt_timer);
		break;

	case SPDK_NVME_ANA_INACCESSIBLE_STATE:
	case SPDK_NVME_ANA_CHANGE_STATE:
		if (nvme_ns->anatt_timer != NULL) {
			break;
		}

		cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);
		nvme_ns->anatt_timer = SPDK_POLLER_REGISTER(nvme_ns_ana_transition_timedout,
				       nvme_ns,
				       cdata->anatt * SPDK_SEC_TO_USEC);
		break;
	default:
		break;
	}
}

static int
nvme_ns_set_ana_state(const struct spdk_nvme_ana_group_descriptor *desc, void *cb_arg)
{
	struct nvme_ns *nvme_ns = cb_arg;
	uint32_t i;

	assert(nvme_ns->ns != NULL);

	for (i = 0; i < desc->num_of_nsid; i++) {
		if (desc->nsid[i] != spdk_nvme_ns_get_id(nvme_ns->ns)) {
			continue;
		}

		_nvme_ns_set_ana_state(nvme_ns, desc);
		return 1;
	}

	return 0;
}

static int
nvme_generate_uuid(const char *sn, uint32_t nsid, struct spdk_uuid *uuid)
{
	int rc = 0;
	struct spdk_uuid new_uuid, namespace_uuid;
	char merged_str[SPDK_NVME_CTRLR_SN_LEN + NSID_STR_LEN + 1] = {'\0'};
	/* This namespace UUID was generated using uuid_generate() method. */
	const char *namespace_str = {"edaed2de-24bc-4b07-b559-f47ecbe730fd"};
	int size;

	assert(strlen(sn) <= SPDK_NVME_CTRLR_SN_LEN);

	spdk_uuid_set_null(&new_uuid);
	spdk_uuid_set_null(&namespace_uuid);

	size = snprintf(merged_str, sizeof(merged_str), "%s%"PRIu32, sn, nsid);
	if (size <= 0 || (unsigned long)size >= sizeof(merged_str)) {
		return -EINVAL;
	}

	spdk_uuid_parse(&namespace_uuid, namespace_str);

	rc = spdk_uuid_generate_sha1(&new_uuid, &namespace_uuid, merged_str, size);
	if (rc == 0) {
		memcpy(uuid, &new_uuid, sizeof(struct spdk_uuid));
	}

	return rc;
}

static int
nvme_disk_create(struct spdk_bdev *disk, const char *base_name,
		 struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns,
		 uint32_t prchk_flags, void *ctx)
{
	const struct spdk_uuid		*uuid;
	const uint8_t *nguid;
	const struct spdk_nvme_ctrlr_data *cdata;
	const struct spdk_nvme_ns_data	*nsdata;
	const struct spdk_nvme_ctrlr_opts *opts;
	enum spdk_nvme_csi		csi;
	uint32_t atomic_bs, phys_bs, bs;
	char sn_tmp[SPDK_NVME_CTRLR_SN_LEN + 1] = {'\0'};
	int rc;

	cdata = spdk_nvme_ctrlr_get_data(ctrlr);
	csi = spdk_nvme_ns_get_csi(ns);
	opts = spdk_nvme_ctrlr_get_opts(ctrlr);

	switch (csi) {
	case SPDK_NVME_CSI_NVM:
		disk->product_name = "NVMe disk";
		break;
	case SPDK_NVME_CSI_ZNS:
		disk->product_name = "NVMe ZNS disk";
		disk->zoned = true;
		disk->zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
		disk->max_zone_append_size = spdk_nvme_zns_ctrlr_get_max_zone_append_size(ctrlr) /
					     spdk_nvme_ns_get_extended_sector_size(ns);
		disk->max_open_zones = spdk_nvme_zns_ns_get_max_open_zones(ns);
		disk->max_active_zones = spdk_nvme_zns_ns_get_max_active_zones(ns);
		break;
	default:
		SPDK_ERRLOG("unsupported CSI: %u\n", csi);
		return -ENOTSUP;
	}

	nguid = spdk_nvme_ns_get_nguid(ns);
	if (!nguid) {
		uuid = spdk_nvme_ns_get_uuid(ns);
		if (uuid) {
			disk->uuid = *uuid;
		} else if (g_opts.generate_uuids) {
			spdk_strcpy_pad(sn_tmp, cdata->sn, SPDK_NVME_CTRLR_SN_LEN, '\0');
			rc = nvme_generate_uuid(sn_tmp, spdk_nvme_ns_get_id(ns), &disk->uuid);
			if (rc != 0) {
				SPDK_ERRLOG("UUID generation failed (%s)\n", strerror(rc));
				return rc;
			}
		}
	} else {
		memcpy(&disk->uuid, nguid, sizeof(disk->uuid));
	}

	disk->name = spdk_sprintf_alloc("%sn%d", base_name, spdk_nvme_ns_get_id(ns));
	if (!disk->name) {
		return -ENOMEM;
	}

	disk->write_cache = 0;
	if (cdata->vwc.present) {
		/* Enable if the Volatile Write Cache exists */
		disk->write_cache = 1;
	}
	if (cdata->oncs.write_zeroes) {
		disk->max_write_zeroes = UINT16_MAX + 1;
	}
	disk->blocklen = spdk_nvme_ns_get_extended_sector_size(ns);
	disk->blockcnt = spdk_nvme_ns_get_num_sectors(ns);
	disk->max_segment_size = spdk_nvme_ctrlr_get_max_xfer_size(ctrlr);
	/* NVMe driver will split one request into multiple requests
	 * based on MDTS and stripe boundary, the bdev layer will use
	 * max_segment_size and max_num_segments to split one big IO
	 * into multiple requests, then small request can't run out
	 * of NVMe internal requests data structure.
	 */
	if (opts && opts->io_queue_requests) {
		disk->max_num_segments = opts->io_queue_requests / 2;
	}
	if (spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_SGL_SUPPORTED) {
		/* The nvme driver will try to split I/O that have too many
		 * SGEs, but it doesn't work if that last SGE doesn't end on
		 * an aggregate total that is block aligned. The bdev layer has
		 * a more robust splitting framework, so use that instead for
		 * this case. (See issue #3269.)
		 */
		uint16_t max_sges = spdk_nvme_ctrlr_get_max_sges(ctrlr);

		if (disk->max_num_segments == 0) {
			disk->max_num_segments = max_sges;
		} else {
			disk->max_num_segments = spdk_min(disk->max_num_segments, max_sges);
		}
	}
	disk->optimal_io_boundary = spdk_nvme_ns_get_optimal_io_boundary(ns);

	nsdata = spdk_nvme_ns_get_data(ns);
	bs = spdk_nvme_ns_get_sector_size(ns);
	atomic_bs = bs;
	phys_bs = bs;
	if (nsdata->nabo == 0) {
		if (nsdata->nsfeat.ns_atomic_write_unit && nsdata->nawupf) {
			atomic_bs = bs * (1 + nsdata->nawupf);
		} else {
			atomic_bs = bs * (1 + cdata->awupf);
		}
	}
	if (nsdata->nsfeat.optperf) {
		phys_bs = bs * (1 + nsdata->npwg);
	}
	disk->phys_blocklen = spdk_min(phys_bs, atomic_bs);

	disk->md_len = spdk_nvme_ns_get_md_size(ns);
	if (disk->md_len != 0) {
		disk->md_interleave = nsdata->flbas.extended;
		disk->dif_type = (enum spdk_dif_type)spdk_nvme_ns_get_pi_type(ns);
		if (disk->dif_type != SPDK_DIF_DISABLE) {
			disk->dif_is_head_of_md = nsdata->dps.md_start;
			disk->dif_check_flags = prchk_flags;
		}
	}

	if (!(spdk_nvme_ctrlr_get_flags(ctrlr) &
	      SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED)) {
		disk->acwu = 0;
	} else if (nsdata->nsfeat.ns_atomic_write_unit) {
		disk->acwu = nsdata->nacwu + 1; /* 0-based */
	} else {
		disk->acwu = cdata->acwu + 1; /* 0-based */
	}

	if (cdata->oncs.copy) {
		/* For now bdev interface allows only single segment copy */
		disk->max_copy = nsdata->mssrl;
	}

	disk->ctxt = ctx;
	disk->fn_table = &nvmelib_fn_table;
	disk->module = &nvme_if;

	return 0;
}

static struct nvme_bdev *
nvme_bdev_alloc(void)
{
	struct nvme_bdev *bdev;
	int rc;

	bdev = calloc(1, sizeof(*bdev));
	if (!bdev) {
		SPDK_ERRLOG("bdev calloc() failed\n");
		return NULL;
	}

	if (g_opts.nvme_error_stat) {
		bdev->err_stat = calloc(1, sizeof(struct nvme_error_stat));
		if (!bdev->err_stat) {
			SPDK_ERRLOG("err_stat calloc() failed\n");
			free(bdev);
			return NULL;
		}
	}

	rc = pthread_mutex_init(&bdev->mutex, NULL);
	if (rc != 0) {
		free(bdev->err_stat);
		free(bdev);
		return NULL;
	}

	bdev->ref = 1;
	bdev->mp_policy = BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE;
	bdev->mp_selector = BDEV_NVME_MP_SELECTOR_ROUND_ROBIN;
	bdev->rr_min_io = UINT32_MAX;
	TAILQ_INIT(&bdev->nvme_ns_list);

	return bdev;
}

static int
nvme_bdev_create(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
{
	struct nvme_bdev *bdev;
	struct nvme_bdev_ctrlr *nbdev_ctrlr = nvme_ctrlr->nbdev_ctrlr;
	int rc;

	bdev = nvme_bdev_alloc();
	if (bdev == NULL) {
		SPDK_ERRLOG("Failed to allocate NVMe bdev\n");
		return -ENOMEM;
	}

	bdev->opal = nvme_ctrlr->opal_dev != NULL;

	rc = nvme_disk_create(&bdev->disk, nbdev_ctrlr->name, nvme_ctrlr->ctrlr,
			      nvme_ns->ns, nvme_ctrlr->opts.prchk_flags, bdev);
	if (rc != 0) {
		SPDK_ERRLOG("Failed to create NVMe disk\n");
		nvme_bdev_free(bdev);
		return rc;
	}

	spdk_io_device_register(bdev,
				bdev_nvme_create_bdev_channel_cb,
				bdev_nvme_destroy_bdev_channel_cb,
				sizeof(struct nvme_bdev_channel),
				bdev->disk.name);

	nvme_ns->bdev = bdev;
	bdev->nsid = nvme_ns->id;
	TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);

	bdev->nbdev_ctrlr = nbdev_ctrlr;
	TAILQ_INSERT_TAIL(&nbdev_ctrlr->bdevs, bdev, tailq);

	rc = spdk_bdev_register(&bdev->disk);
	if (rc != 0) {
		SPDK_ERRLOG("spdk_bdev_register() failed\n");
		spdk_io_device_unregister(bdev, NULL);
		nvme_ns->bdev = NULL;
		TAILQ_REMOVE(&nbdev_ctrlr->bdevs, bdev, tailq);
		nvme_bdev_free(bdev);
		return rc;
	}

	return 0;
}

static bool
bdev_nvme_compare_ns(struct spdk_nvme_ns *ns1, struct spdk_nvme_ns *ns2)
{
	const struct spdk_nvme_ns_data *nsdata1, *nsdata2;
	const struct spdk_uuid *uuid1, *uuid2;

	nsdata1 = spdk_nvme_ns_get_data(ns1);
	nsdata2 = spdk_nvme_ns_get_data(ns2);
	uuid1 = spdk_nvme_ns_get_uuid(ns1);
	uuid2 = spdk_nvme_ns_get_uuid(ns2);

	return memcmp(nsdata1->nguid, nsdata2->nguid, sizeof(nsdata1->nguid)) == 0 &&
	       nsdata1->eui64 == nsdata2->eui64 &&
	       ((uuid1 == NULL && uuid2 == NULL) ||
		(uuid1 != NULL && uuid2 != NULL && spdk_uuid_compare(uuid1, uuid2) == 0)) &&
	       spdk_nvme_ns_get_csi(ns1) == spdk_nvme_ns_get_csi(ns2);
}

static bool
hotplug_probe_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
		 struct spdk_nvme_ctrlr_opts *opts)
{
	struct nvme_probe_skip_entry *entry;

	TAILQ_FOREACH(entry, &g_skipped_nvme_ctrlrs, tailq) {
		if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
			return false;
		}
	}

	opts->arbitration_burst = (uint8_t)g_opts.arbitration_burst;
	opts->low_priority_weight = (uint8_t)g_opts.low_priority_weight;
	opts->medium_priority_weight = (uint8_t)g_opts.medium_priority_weight;
	opts->high_priority_weight = (uint8_t)g_opts.high_priority_weight;
	opts->disable_read_ana_log_page = true;

	SPDK_DEBUGLOG(bdev_nvme, "Attaching to %s\n", trid->traddr);

	return true;
}

static void
nvme_abort_cpl(void *ctx, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	if (spdk_nvme_cpl_is_error(cpl)) {
		SPDK_WARNLOG("Abort failed. Resetting controller. sc is %u, sct is %u.\n", cpl->status.sc,
			     cpl->status.sct);
		bdev_nvme_reset_ctrlr(nvme_ctrlr);
	} else if (cpl->cdw0 & 0x1) {
		SPDK_WARNLOG("Specified command could not be aborted.\n");
		bdev_nvme_reset_ctrlr(nvme_ctrlr);
	}
}

static void
timeout_cb(void *cb_arg, struct spdk_nvme_ctrlr *ctrlr,
	   struct spdk_nvme_qpair *qpair, uint16_t cid)
{
	struct nvme_ctrlr *nvme_ctrlr = cb_arg;
	union spdk_nvme_csts_register csts;
	int rc;

	assert(nvme_ctrlr->ctrlr == ctrlr);

	SPDK_WARNLOG("Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n", ctrlr, qpair, cid);

	/* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O
	 * queue.  (Note: qpair == NULL when there's an admin cmd timeout.)  Otherwise we
	 * would submit another fabrics cmd on the admin queue to read CSTS and check for its
	 * completion recursively.
	 */
	if (nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE || qpair != NULL) {
		csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr);
		if (csts.bits.cfs) {
			SPDK_ERRLOG("Controller Fatal Status, reset required\n");
			bdev_nvme_reset_ctrlr(nvme_ctrlr);
			return;
		}
	}

	switch (g_opts.action_on_timeout) {
	case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT:
		if (qpair) {
			/* Don't send abort to ctrlr when ctrlr is not available. */
			pthread_mutex_lock(&nvme_ctrlr->mutex);
			if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
				pthread_mutex_unlock(&nvme_ctrlr->mutex);
				SPDK_NOTICELOG("Quit abort. Ctrlr is not available.\n");
				return;
			}
			pthread_mutex_unlock(&nvme_ctrlr->mutex);

			rc = spdk_nvme_ctrlr_cmd_abort(ctrlr, qpair, cid,
						       nvme_abort_cpl, nvme_ctrlr);
			if (rc == 0) {
				return;
			}

			SPDK_ERRLOG("Unable to send abort. Resetting, rc is %d.\n", rc);
		}

	/* FALLTHROUGH */
	case SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET:
		bdev_nvme_reset_ctrlr(nvme_ctrlr);
		break;
	case SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE:
		SPDK_DEBUGLOG(bdev_nvme, "No action for nvme controller timeout.\n");
		break;
	default:
		SPDK_ERRLOG("An invalid timeout action value is found.\n");
		break;
	}
}

static struct nvme_ns *
nvme_ns_alloc(void)
{
	struct nvme_ns *nvme_ns;

	nvme_ns = calloc(1, sizeof(struct nvme_ns));
	if (nvme_ns == NULL) {
		return NULL;
	}

	if (g_opts.io_path_stat) {
		nvme_ns->stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
		if (nvme_ns->stat == NULL) {
			free(nvme_ns);
			return NULL;
		}
		spdk_bdev_reset_io_stat(nvme_ns->stat, SPDK_BDEV_RESET_STAT_MAXMIN);
	}

	return nvme_ns;
}

static void
nvme_ns_free(struct nvme_ns *nvme_ns)
{
	free(nvme_ns->stat);
	free(nvme_ns);
}

static void
nvme_ctrlr_populate_namespace_done(struct nvme_ns *nvme_ns, int rc)
{
	struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;
	struct nvme_async_probe_ctx *ctx = nvme_ns->probe_ctx;

	if (rc == 0) {
		nvme_ns->probe_ctx = NULL;
		pthread_mutex_lock(&nvme_ctrlr->mutex);
		nvme_ctrlr->ref++;
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
	} else {
		RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);
		nvme_ns_free(nvme_ns);
	}

	if (ctx) {
		ctx->populates_in_progress--;
		if (ctx->populates_in_progress == 0) {
			nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
		}
	}
}

static void
bdev_nvme_add_io_path(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
	struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
	int rc;

	rc = _bdev_nvme_add_io_path(nbdev_ch, nvme_ns);
	if (rc != 0) {
		SPDK_ERRLOG("Failed to add I/O path to bdev_channel dynamically.\n");
	}

	spdk_for_each_channel_continue(i, rc);
}

static void
bdev_nvme_delete_io_path(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
	struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
	struct nvme_io_path *io_path;

	io_path = _bdev_nvme_get_io_path(nbdev_ch, nvme_ns);
	if (io_path != NULL) {
		_bdev_nvme_delete_io_path(nbdev_ch, io_path);
	}

	spdk_for_each_channel_continue(i, 0);
}

static void
bdev_nvme_add_io_path_failed(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);

	nvme_ctrlr_populate_namespace_done(nvme_ns, -1);
}

static void
bdev_nvme_add_io_path_done(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);
	struct nvme_bdev *bdev = spdk_io_channel_iter_get_io_device(i);

	if (status == 0) {
		nvme_ctrlr_populate_namespace_done(nvme_ns, 0);
	} else {
		/* Delete the added io_paths and fail populating the namespace. */
		spdk_for_each_channel(bdev,
				      bdev_nvme_delete_io_path,
				      nvme_ns,
				      bdev_nvme_add_io_path_failed);
	}
}

static int
nvme_bdev_add_ns(struct nvme_bdev *bdev, struct nvme_ns *nvme_ns)
{
	struct nvme_ns *tmp_ns;
	const struct spdk_nvme_ns_data *nsdata;

	nsdata = spdk_nvme_ns_get_data(nvme_ns->ns);
	if (!nsdata->nmic.can_share) {
		SPDK_ERRLOG("Namespace cannot be shared.\n");
		return -EINVAL;
	}

	pthread_mutex_lock(&bdev->mutex);

	tmp_ns = TAILQ_FIRST(&bdev->nvme_ns_list);
	assert(tmp_ns != NULL);

	if (tmp_ns->ns != NULL && !bdev_nvme_compare_ns(nvme_ns->ns, tmp_ns->ns)) {
		pthread_mutex_unlock(&bdev->mutex);
		SPDK_ERRLOG("Namespaces are not identical.\n");
		return -EINVAL;
	}

	bdev->ref++;
	TAILQ_INSERT_TAIL(&bdev->nvme_ns_list, nvme_ns, tailq);
	nvme_ns->bdev = bdev;

	pthread_mutex_unlock(&bdev->mutex);

	/* Add nvme_io_path to nvme_bdev_channels dynamically. */
	spdk_for_each_channel(bdev,
			      bdev_nvme_add_io_path,
			      nvme_ns,
			      bdev_nvme_add_io_path_done);

	return 0;
}

static void
nvme_ctrlr_populate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
{
	struct spdk_nvme_ns	*ns;
	struct nvme_bdev	*bdev;
	int			rc = 0;

	ns = spdk_nvme_ctrlr_get_ns(nvme_ctrlr->ctrlr, nvme_ns->id);
	if (!ns) {
		SPDK_DEBUGLOG(bdev_nvme, "Invalid NS %d\n", nvme_ns->id);
		rc = -EINVAL;
		goto done;
	}

	nvme_ns->ns = ns;
	nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;

	if (nvme_ctrlr->ana_log_page != NULL) {
		bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ns_set_ana_state, nvme_ns);
	}

	bdev = nvme_bdev_ctrlr_get_bdev(nvme_ctrlr->nbdev_ctrlr, nvme_ns->id);
	if (bdev == NULL) {
		rc = nvme_bdev_create(nvme_ctrlr, nvme_ns);
	} else {
		rc = nvme_bdev_add_ns(bdev, nvme_ns);
		if (rc == 0) {
			return;
		}
	}
done:
	nvme_ctrlr_populate_namespace_done(nvme_ns, rc);
}

static void
nvme_ctrlr_depopulate_namespace_done(struct nvme_ns *nvme_ns)
{
	struct nvme_ctrlr *nvme_ctrlr = nvme_ns->ctrlr;

	assert(nvme_ctrlr != NULL);

	pthread_mutex_lock(&nvme_ctrlr->mutex);

	RB_REMOVE(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);

	if (nvme_ns->bdev != NULL) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return;
	}

	nvme_ns_free(nvme_ns);
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	nvme_ctrlr_release(nvme_ctrlr);
}

static void
bdev_nvme_delete_io_path_done(struct spdk_io_channel_iter *i, int status)
{
	struct nvme_ns *nvme_ns = spdk_io_channel_iter_get_ctx(i);

	nvme_ctrlr_depopulate_namespace_done(nvme_ns);
}

static void
nvme_ctrlr_depopulate_namespace(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *nvme_ns)
{
	struct nvme_bdev *bdev;

	spdk_poller_unregister(&nvme_ns->anatt_timer);

	bdev = nvme_ns->bdev;
	if (bdev != NULL) {
		pthread_mutex_lock(&bdev->mutex);

		assert(bdev->ref > 0);
		bdev->ref--;
		if (bdev->ref == 0) {
			pthread_mutex_unlock(&bdev->mutex);

			spdk_bdev_unregister(&bdev->disk, NULL, NULL);
		} else {
			/* spdk_bdev_unregister() is not called until the last nvme_ns is
			 * depopulated. Hence we need to remove nvme_ns from bdev->nvme_ns_list
			 * and clear nvme_ns->bdev here.
			 */
			TAILQ_REMOVE(&bdev->nvme_ns_list, nvme_ns, tailq);
			nvme_ns->bdev = NULL;

			pthread_mutex_unlock(&bdev->mutex);

			/* Delete nvme_io_paths from nvme_bdev_channels dynamically. After that,
			 * we call depopulate_namespace_done() to avoid use-after-free.
			 */
			spdk_for_each_channel(bdev,
					      bdev_nvme_delete_io_path,
					      nvme_ns,
					      bdev_nvme_delete_io_path_done);
			return;
		}
	}

	nvme_ctrlr_depopulate_namespace_done(nvme_ns);
}

static void
nvme_ctrlr_populate_namespaces(struct nvme_ctrlr *nvme_ctrlr,
			       struct nvme_async_probe_ctx *ctx)
{
	struct spdk_nvme_ctrlr	*ctrlr = nvme_ctrlr->ctrlr;
	struct nvme_ns	*nvme_ns, *next;
	struct spdk_nvme_ns	*ns;
	struct nvme_bdev	*bdev;
	uint32_t		nsid;
	int			rc;
	uint64_t		num_sectors;

	if (ctx) {
		/* Initialize this count to 1 to handle the populate functions
		 * calling nvme_ctrlr_populate_namespace_done() immediately.
		 */
		ctx->populates_in_progress = 1;
	}

	/* First loop over our existing namespaces and see if they have been
	 * removed. */
	nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
	while (nvme_ns != NULL) {
		next = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);

		if (spdk_nvme_ctrlr_is_active_ns(ctrlr, nvme_ns->id)) {
			/* NS is still there or added again. Its attributes may have changed. */
			ns = spdk_nvme_ctrlr_get_ns(ctrlr, nvme_ns->id);
			if (nvme_ns->ns != ns) {
				assert(nvme_ns->ns == NULL);
				nvme_ns->ns = ns;
				SPDK_DEBUGLOG(bdev_nvme, "NSID %u was added\n", nvme_ns->id);
			}

			num_sectors = spdk_nvme_ns_get_num_sectors(ns);
			bdev = nvme_ns->bdev;
			assert(bdev != NULL);
			if (bdev->disk.blockcnt != num_sectors) {
				SPDK_NOTICELOG("NSID %u is resized: bdev name %s, old size %" PRIu64 ", new size %" PRIu64 "\n",
					       nvme_ns->id,
					       bdev->disk.name,
					       bdev->disk.blockcnt,
					       num_sectors);
				rc = spdk_bdev_notify_blockcnt_change(&bdev->disk, num_sectors);
				if (rc != 0) {
					SPDK_ERRLOG("Could not change num blocks for nvme bdev: name %s, errno: %d.\n",
						    bdev->disk.name, rc);
				}
			}
		} else {
			/* Namespace was removed */
			nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
		}

		nvme_ns = next;
	}

	/* Loop through all of the namespaces at the nvme level and see if any of them are new */
	nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
	while (nsid != 0) {
		nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);

		if (nvme_ns == NULL) {
			/* Found a new one */
			nvme_ns = nvme_ns_alloc();
			if (nvme_ns == NULL) {
				SPDK_ERRLOG("Failed to allocate namespace\n");
				/* This just fails to attach the namespace. It may work on a future attempt. */
				continue;
			}

			nvme_ns->id = nsid;
			nvme_ns->ctrlr = nvme_ctrlr;

			nvme_ns->bdev = NULL;

			if (ctx) {
				ctx->populates_in_progress++;
			}
			nvme_ns->probe_ctx = ctx;

			RB_INSERT(nvme_ns_tree, &nvme_ctrlr->namespaces, nvme_ns);

			nvme_ctrlr_populate_namespace(nvme_ctrlr, nvme_ns);
		}

		nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid);
	}

	if (ctx) {
		/* Decrement this count now that the loop is over to account
		 * for the one we started with.  If the count is then 0, we
		 * know any populate_namespace functions completed immediately,
		 * so we'll kick the callback here.
		 */
		ctx->populates_in_progress--;
		if (ctx->populates_in_progress == 0) {
			nvme_ctrlr_populate_namespaces_done(nvme_ctrlr, ctx);
		}
	}

}

static void
nvme_ctrlr_depopulate_namespaces(struct nvme_ctrlr *nvme_ctrlr)
{
	struct nvme_ns *nvme_ns, *tmp;

	RB_FOREACH_SAFE(nvme_ns, nvme_ns_tree, &nvme_ctrlr->namespaces, tmp) {
		nvme_ctrlr_depopulate_namespace(nvme_ctrlr, nvme_ns);
	}
}

static uint32_t
nvme_ctrlr_get_ana_log_page_size(struct nvme_ctrlr *nvme_ctrlr)
{
	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;
	uint32_t nsid, ns_count = 0;

	cdata = spdk_nvme_ctrlr_get_data(ctrlr);

	for (nsid = spdk_nvme_ctrlr_get_first_active_ns(ctrlr);
	     nsid != 0; nsid = spdk_nvme_ctrlr_get_next_active_ns(ctrlr, nsid)) {
		ns_count++;
	}

	return sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
	       sizeof(struct spdk_nvme_ana_group_descriptor) + ns_count *
	       sizeof(uint32_t);
}

static int
nvme_ctrlr_set_ana_states(const struct spdk_nvme_ana_group_descriptor *desc,
			  void *cb_arg)
{
	struct nvme_ctrlr *nvme_ctrlr = cb_arg;
	struct nvme_ns *nvme_ns;
	uint32_t i, nsid;

	for (i = 0; i < desc->num_of_nsid; i++) {
		nsid = desc->nsid[i];
		if (nsid == 0) {
			continue;
		}

		nvme_ns = nvme_ctrlr_get_ns(nvme_ctrlr, nsid);

		assert(nvme_ns != NULL);
		if (nvme_ns == NULL) {
			/* Target told us that an inactive namespace had an ANA change */
			continue;
		}

		_nvme_ns_set_ana_state(nvme_ns, desc);
	}

	return 0;
}

static void
bdev_nvme_disable_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
{
	struct nvme_ns *nvme_ns;

	spdk_free(nvme_ctrlr->ana_log_page);
	nvme_ctrlr->ana_log_page = NULL;

	for (nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
	     nvme_ns != NULL;
	     nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns)) {
		nvme_ns->ana_state_updating = false;
		nvme_ns->ana_state = SPDK_NVME_ANA_OPTIMIZED_STATE;
	}
}

static void
nvme_ctrlr_read_ana_log_page_done(void *ctx, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	if (cpl != NULL && spdk_nvme_cpl_is_success(cpl)) {
		bdev_nvme_parse_ana_log_page(nvme_ctrlr, nvme_ctrlr_set_ana_states,
					     nvme_ctrlr);
	} else {
		bdev_nvme_disable_read_ana_log_page(nvme_ctrlr);
	}

	pthread_mutex_lock(&nvme_ctrlr->mutex);

	assert(nvme_ctrlr->ana_log_page_updating == true);
	nvme_ctrlr->ana_log_page_updating = false;

	if (nvme_ctrlr_can_be_unregistered(nvme_ctrlr)) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);

		nvme_ctrlr_unregister(nvme_ctrlr);
	} else {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);

		bdev_nvme_clear_io_path_caches(nvme_ctrlr);
	}
}

static int
nvme_ctrlr_read_ana_log_page(struct nvme_ctrlr *nvme_ctrlr)
{
	uint32_t ana_log_page_size;
	int rc;

	if (nvme_ctrlr->ana_log_page == NULL) {
		return -EINVAL;
	}

	ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);

	if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
		SPDK_ERRLOG("ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
			    ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
		return -EINVAL;
	}

	pthread_mutex_lock(&nvme_ctrlr->mutex);
	if (!nvme_ctrlr_is_available(nvme_ctrlr) ||
	    nvme_ctrlr->ana_log_page_updating) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return -EBUSY;
	}

	nvme_ctrlr->ana_log_page_updating = true;
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	rc = spdk_nvme_ctrlr_cmd_get_log_page(nvme_ctrlr->ctrlr,
					      SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
					      SPDK_NVME_GLOBAL_NS_TAG,
					      nvme_ctrlr->ana_log_page,
					      ana_log_page_size, 0,
					      nvme_ctrlr_read_ana_log_page_done,
					      nvme_ctrlr);
	if (rc != 0) {
		nvme_ctrlr_read_ana_log_page_done(nvme_ctrlr, NULL);
	}

	return rc;
}

static void
dummy_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
{
}

struct bdev_nvme_set_preferred_path_ctx {
	struct spdk_bdev_desc *desc;
	struct nvme_ns *nvme_ns;
	bdev_nvme_set_preferred_path_cb cb_fn;
	void *cb_arg;
};

static void
bdev_nvme_set_preferred_path_done(struct spdk_io_channel_iter *i, int status)
{
	struct bdev_nvme_set_preferred_path_ctx *ctx = spdk_io_channel_iter_get_ctx(i);

	assert(ctx != NULL);
	assert(ctx->desc != NULL);
	assert(ctx->cb_fn != NULL);

	spdk_bdev_close(ctx->desc);

	ctx->cb_fn(ctx->cb_arg, status);

	free(ctx);
}

static void
_bdev_nvme_set_preferred_path(struct spdk_io_channel_iter *i)
{
	struct bdev_nvme_set_preferred_path_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
	struct nvme_io_path *io_path, *prev;

	prev = NULL;
	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
		if (io_path->nvme_ns == ctx->nvme_ns) {
			break;
		}
		prev = io_path;
	}

	if (io_path != NULL) {
		if (prev != NULL) {
			STAILQ_REMOVE_AFTER(&nbdev_ch->io_path_list, prev, stailq);
			STAILQ_INSERT_HEAD(&nbdev_ch->io_path_list, io_path, stailq);
		}

		/* We can set io_path to nbdev_ch->current_io_path directly here.
		 * However, it needs to be conditional. To simplify the code,
		 * just clear nbdev_ch->current_io_path and let find_io_path()
		 * fill it.
		 *
		 * Automatic failback may be disabled. Hence even if the io_path is
		 * already at the head, clear nbdev_ch->current_io_path.
		 */
		bdev_nvme_clear_current_io_path(nbdev_ch);
	}

	spdk_for_each_channel_continue(i, 0);
}

static struct nvme_ns *
bdev_nvme_set_preferred_ns(struct nvme_bdev *nbdev, uint16_t cntlid)
{
	struct nvme_ns *nvme_ns, *prev;
	const struct spdk_nvme_ctrlr_data *cdata;

	prev = NULL;
	TAILQ_FOREACH(nvme_ns, &nbdev->nvme_ns_list, tailq) {
		cdata = spdk_nvme_ctrlr_get_data(nvme_ns->ctrlr->ctrlr);

		if (cdata->cntlid == cntlid) {
			break;
		}
		prev = nvme_ns;
	}

	if (nvme_ns != NULL && prev != NULL) {
		TAILQ_REMOVE(&nbdev->nvme_ns_list, nvme_ns, tailq);
		TAILQ_INSERT_HEAD(&nbdev->nvme_ns_list, nvme_ns, tailq);
	}

	return nvme_ns;
}

/* This function supports only multipath mode. There is only a single I/O path
 * for each NVMe-oF controller. Hence, just move the matched I/O path to the
 * head of the I/O path list for each NVMe bdev channel.
 *
 * NVMe bdev channel may be acquired after completing this function. move the
 * matched namespace to the head of the namespace list for the NVMe bdev too.
 */
void
bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid,
			     bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg)
{
	struct bdev_nvme_set_preferred_path_ctx *ctx;
	struct spdk_bdev *bdev;
	struct nvme_bdev *nbdev;
	int rc = 0;

	assert(cb_fn != NULL);

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		SPDK_ERRLOG("Failed to alloc context.\n");
		rc = -ENOMEM;
		goto err_alloc;
	}

	ctx->cb_fn = cb_fn;
	ctx->cb_arg = cb_arg;

	rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
	if (rc != 0) {
		SPDK_ERRLOG("Failed to open bdev %s.\n", name);
		goto err_open;
	}

	bdev = spdk_bdev_desc_get_bdev(ctx->desc);

	if (bdev->module != &nvme_if) {
		SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
		rc = -ENODEV;
		goto err_bdev;
	}

	nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);

	pthread_mutex_lock(&nbdev->mutex);

	ctx->nvme_ns = bdev_nvme_set_preferred_ns(nbdev, cntlid);
	if (ctx->nvme_ns == NULL) {
		pthread_mutex_unlock(&nbdev->mutex);

		SPDK_ERRLOG("bdev %s does not have namespace to controller %u.\n", name, cntlid);
		rc = -ENODEV;
		goto err_bdev;
	}

	pthread_mutex_unlock(&nbdev->mutex);

	spdk_for_each_channel(nbdev,
			      _bdev_nvme_set_preferred_path,
			      ctx,
			      bdev_nvme_set_preferred_path_done);
	return;

err_bdev:
	spdk_bdev_close(ctx->desc);
err_open:
	free(ctx);
err_alloc:
	cb_fn(cb_arg, rc);
}

struct bdev_nvme_set_multipath_policy_ctx {
	struct spdk_bdev_desc *desc;
	bdev_nvme_set_multipath_policy_cb cb_fn;
	void *cb_arg;
};

static void
bdev_nvme_set_multipath_policy_done(struct spdk_io_channel_iter *i, int status)
{
	struct bdev_nvme_set_multipath_policy_ctx *ctx = spdk_io_channel_iter_get_ctx(i);

	assert(ctx != NULL);
	assert(ctx->desc != NULL);
	assert(ctx->cb_fn != NULL);

	spdk_bdev_close(ctx->desc);

	ctx->cb_fn(ctx->cb_arg, status);

	free(ctx);
}

static void
_bdev_nvme_set_multipath_policy(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct nvme_bdev_channel *nbdev_ch = spdk_io_channel_get_ctx(_ch);
	struct nvme_bdev *nbdev = spdk_io_channel_get_io_device(_ch);

	nbdev_ch->mp_policy = nbdev->mp_policy;
	nbdev_ch->mp_selector = nbdev->mp_selector;
	nbdev_ch->rr_min_io = nbdev->rr_min_io;
	bdev_nvme_clear_current_io_path(nbdev_ch);

	spdk_for_each_channel_continue(i, 0);
}

void
bdev_nvme_set_multipath_policy(const char *name, enum bdev_nvme_multipath_policy policy,
			       enum bdev_nvme_multipath_selector selector, uint32_t rr_min_io,
			       bdev_nvme_set_multipath_policy_cb cb_fn, void *cb_arg)
{
	struct bdev_nvme_set_multipath_policy_ctx *ctx;
	struct spdk_bdev *bdev;
	struct nvme_bdev *nbdev;
	int rc;

	assert(cb_fn != NULL);

	if (policy == BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE && selector == BDEV_NVME_MP_SELECTOR_ROUND_ROBIN) {
		if (rr_min_io == UINT32_MAX) {
			rr_min_io = 1;
		} else if (rr_min_io == 0) {
			rc = -EINVAL;
			goto exit;
		}
	} else if (rr_min_io != UINT32_MAX) {
		rc = -EINVAL;
		goto exit;
	}

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		SPDK_ERRLOG("Failed to alloc context.\n");
		rc = -ENOMEM;
		goto exit;
	}

	ctx->cb_fn = cb_fn;
	ctx->cb_arg = cb_arg;

	rc = spdk_bdev_open_ext(name, false, dummy_bdev_event_cb, NULL, &ctx->desc);
	if (rc != 0) {
		SPDK_ERRLOG("Failed to open bdev %s.\n", name);
		rc = -ENODEV;
		goto err_open;
	}

	bdev = spdk_bdev_desc_get_bdev(ctx->desc);
	if (bdev->module != &nvme_if) {
		SPDK_ERRLOG("bdev %s is not registered in this module.\n", name);
		rc = -ENODEV;
		goto err_module;
	}
	nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);

	pthread_mutex_lock(&nbdev->mutex);
	nbdev->mp_policy = policy;
	nbdev->mp_selector = selector;
	nbdev->rr_min_io = rr_min_io;
	pthread_mutex_unlock(&nbdev->mutex);

	spdk_for_each_channel(nbdev,
			      _bdev_nvme_set_multipath_policy,
			      ctx,
			      bdev_nvme_set_multipath_policy_done);
	return;

err_module:
	spdk_bdev_close(ctx->desc);
err_open:
	free(ctx);
exit:
	cb_fn(cb_arg, rc);
}

static void
aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_ctrlr *nvme_ctrlr		= arg;
	union spdk_nvme_async_event_completion	event;

	if (spdk_nvme_cpl_is_error(cpl)) {
		SPDK_WARNLOG("AER request execute failed\n");
		return;
	}

	event.raw = cpl->cdw0;
	if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
	    (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) {
		nvme_ctrlr_populate_namespaces(nvme_ctrlr, NULL);
	} else if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) &&
		   (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_ANA_CHANGE)) {
		nvme_ctrlr_read_ana_log_page(nvme_ctrlr);
	}
}

static void
free_nvme_async_probe_ctx(struct nvme_async_probe_ctx *ctx)
{
	spdk_keyring_put_key(ctx->drv_opts.tls_psk);
	spdk_keyring_put_key(ctx->drv_opts.dhchap_key);
	free(ctx);
}

static void
populate_namespaces_cb(struct nvme_async_probe_ctx *ctx, int rc)
{
	if (ctx->cb_fn) {
		ctx->cb_fn(ctx->cb_ctx, ctx->reported_bdevs, rc);
	}

	ctx->namespaces_populated = true;
	if (ctx->probe_done) {
		/* The probe was already completed, so we need to free the context
		 * here.  This can happen for cases like OCSSD, where we need to
		 * send additional commands to the SSD after attach.
		 */
		free_nvme_async_probe_ctx(ctx);
	}
}

static void
nvme_ctrlr_create_done(struct nvme_ctrlr *nvme_ctrlr,
		       struct nvme_async_probe_ctx *ctx)
{
	spdk_io_device_register(nvme_ctrlr,
				bdev_nvme_create_ctrlr_channel_cb,
				bdev_nvme_destroy_ctrlr_channel_cb,
				sizeof(struct nvme_ctrlr_channel),
				nvme_ctrlr->nbdev_ctrlr->name);

	nvme_ctrlr_populate_namespaces(nvme_ctrlr, ctx);
}

static void
nvme_ctrlr_init_ana_log_page_done(void *_ctx, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_ctrlr *nvme_ctrlr = _ctx;
	struct nvme_async_probe_ctx *ctx = nvme_ctrlr->probe_ctx;

	nvme_ctrlr->probe_ctx = NULL;

	if (spdk_nvme_cpl_is_error(cpl)) {
		nvme_ctrlr_delete(nvme_ctrlr);

		if (ctx != NULL) {
			ctx->reported_bdevs = 0;
			populate_namespaces_cb(ctx, -1);
		}
		return;
	}

	nvme_ctrlr_create_done(nvme_ctrlr, ctx);
}

static int
nvme_ctrlr_init_ana_log_page(struct nvme_ctrlr *nvme_ctrlr,
			     struct nvme_async_probe_ctx *ctx)
{
	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;
	uint32_t ana_log_page_size;

	cdata = spdk_nvme_ctrlr_get_data(ctrlr);

	/* Set buffer size enough to include maximum number of allowed namespaces. */
	ana_log_page_size = sizeof(struct spdk_nvme_ana_page) + cdata->nanagrpid *
			    sizeof(struct spdk_nvme_ana_group_descriptor) + cdata->mnan *
			    sizeof(uint32_t);

	nvme_ctrlr->ana_log_page = spdk_zmalloc(ana_log_page_size, 64, NULL,
						SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
	if (nvme_ctrlr->ana_log_page == NULL) {
		SPDK_ERRLOG("could not allocate ANA log page buffer\n");
		return -ENXIO;
	}

	/* Each descriptor in a ANA log page is not ensured to be 8-bytes aligned.
	 * Hence copy each descriptor to a temporary area when parsing it.
	 *
	 * Allocate a buffer whose size is as large as ANA log page buffer because
	 * we do not know the size of a descriptor until actually reading it.
	 */
	nvme_ctrlr->copied_ana_desc = calloc(1, ana_log_page_size);
	if (nvme_ctrlr->copied_ana_desc == NULL) {
		SPDK_ERRLOG("could not allocate a buffer to parse ANA descriptor\n");
		return -ENOMEM;
	}

	nvme_ctrlr->max_ana_log_page_size = ana_log_page_size;

	nvme_ctrlr->probe_ctx = ctx;

	/* Then, set the read size only to include the current active namespaces. */
	ana_log_page_size = nvme_ctrlr_get_ana_log_page_size(nvme_ctrlr);

	if (ana_log_page_size > nvme_ctrlr->max_ana_log_page_size) {
		SPDK_ERRLOG("ANA log page size %" PRIu32 " is larger than allowed %" PRIu32 "\n",
			    ana_log_page_size, nvme_ctrlr->max_ana_log_page_size);
		return -EINVAL;
	}

	return spdk_nvme_ctrlr_cmd_get_log_page(ctrlr,
						SPDK_NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS,
						SPDK_NVME_GLOBAL_NS_TAG,
						nvme_ctrlr->ana_log_page,
						ana_log_page_size, 0,
						nvme_ctrlr_init_ana_log_page_done,
						nvme_ctrlr);
}

/* hostnqn and subnqn were already verified before attaching a controller.
 * Hence check only the multipath capability and cntlid here.
 */
static bool
bdev_nvme_check_multipath(struct nvme_bdev_ctrlr *nbdev_ctrlr, struct spdk_nvme_ctrlr *ctrlr)
{
	struct nvme_ctrlr *tmp;
	const struct spdk_nvme_ctrlr_data *cdata, *tmp_cdata;

	cdata = spdk_nvme_ctrlr_get_data(ctrlr);

	if (!cdata->cmic.multi_ctrlr) {
		SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
		return false;
	}

	TAILQ_FOREACH(tmp, &nbdev_ctrlr->ctrlrs, tailq) {
		tmp_cdata = spdk_nvme_ctrlr_get_data(tmp->ctrlr);

		if (!tmp_cdata->cmic.multi_ctrlr) {
			SPDK_ERRLOG("Ctrlr%u does not support multipath.\n", cdata->cntlid);
			return false;
		}
		if (cdata->cntlid == tmp_cdata->cntlid) {
			SPDK_ERRLOG("cntlid %u are duplicated.\n", tmp_cdata->cntlid);
			return false;
		}
	}

	return true;
}

static int
nvme_bdev_ctrlr_create(const char *name, struct nvme_ctrlr *nvme_ctrlr)
{
	struct nvme_bdev_ctrlr *nbdev_ctrlr;
	struct spdk_nvme_ctrlr *ctrlr = nvme_ctrlr->ctrlr;
	int rc = 0;

	pthread_mutex_lock(&g_bdev_nvme_mutex);

	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
	if (nbdev_ctrlr != NULL) {
		if (!bdev_nvme_check_multipath(nbdev_ctrlr, ctrlr)) {
			rc = -EINVAL;
			goto exit;
		}
	} else {
		nbdev_ctrlr = calloc(1, sizeof(*nbdev_ctrlr));
		if (nbdev_ctrlr == NULL) {
			SPDK_ERRLOG("Failed to allocate nvme_bdev_ctrlr.\n");
			rc = -ENOMEM;
			goto exit;
		}
		nbdev_ctrlr->name = strdup(name);
		if (nbdev_ctrlr->name == NULL) {
			SPDK_ERRLOG("Failed to allocate name of nvme_bdev_ctrlr.\n");
			free(nbdev_ctrlr);
			goto exit;
		}
		TAILQ_INIT(&nbdev_ctrlr->ctrlrs);
		TAILQ_INIT(&nbdev_ctrlr->bdevs);
		TAILQ_INSERT_TAIL(&g_nvme_bdev_ctrlrs, nbdev_ctrlr, tailq);
	}
	nvme_ctrlr->nbdev_ctrlr = nbdev_ctrlr;
	TAILQ_INSERT_TAIL(&nbdev_ctrlr->ctrlrs, nvme_ctrlr, tailq);
exit:
	pthread_mutex_unlock(&g_bdev_nvme_mutex);
	return rc;
}

static int
nvme_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr,
		  const char *name,
		  const struct spdk_nvme_transport_id *trid,
		  struct nvme_async_probe_ctx *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr;
	struct nvme_path_id *path_id;
	const struct spdk_nvme_ctrlr_data *cdata;
	int rc;

	nvme_ctrlr = calloc(1, sizeof(*nvme_ctrlr));
	if (nvme_ctrlr == NULL) {
		SPDK_ERRLOG("Failed to allocate device struct\n");
		return -ENOMEM;
	}

	rc = pthread_mutex_init(&nvme_ctrlr->mutex, NULL);
	if (rc != 0) {
		free(nvme_ctrlr);
		return rc;
	}

	TAILQ_INIT(&nvme_ctrlr->trids);
	RB_INIT(&nvme_ctrlr->namespaces);

	/* Get another reference to the key, so the first one can be released from probe_ctx */
	if (ctx != NULL) {
		if (ctx->drv_opts.tls_psk != NULL) {
			nvme_ctrlr->psk = spdk_keyring_get_key(
						  spdk_key_get_name(ctx->drv_opts.tls_psk));
			if (nvme_ctrlr->psk == NULL) {
				/* Could only happen if the key was removed in the meantime */
				SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
					    spdk_key_get_name(ctx->drv_opts.tls_psk));
				rc = -ENOKEY;
				goto err;
			}
		}

		if (ctx->drv_opts.dhchap_key != NULL) {
			nvme_ctrlr->dhchap_key = spdk_keyring_get_key(
							 spdk_key_get_name(ctx->drv_opts.dhchap_key));
			if (nvme_ctrlr->dhchap_key == NULL) {
				SPDK_ERRLOG("Couldn't get a reference to the key '%s'\n",
					    spdk_key_get_name(ctx->drv_opts.dhchap_key));
				rc = -ENOKEY;
				goto err;
			}
		}
	}

	path_id = calloc(1, sizeof(*path_id));
	if (path_id == NULL) {
		SPDK_ERRLOG("Failed to allocate trid entry pointer\n");
		rc = -ENOMEM;
		goto err;
	}

	path_id->trid = *trid;
	if (ctx != NULL) {
		memcpy(path_id->hostid.hostaddr, ctx->drv_opts.src_addr, sizeof(path_id->hostid.hostaddr));
		memcpy(path_id->hostid.hostsvcid, ctx->drv_opts.src_svcid, sizeof(path_id->hostid.hostsvcid));
	}
	nvme_ctrlr->active_path_id = path_id;
	TAILQ_INSERT_HEAD(&nvme_ctrlr->trids, path_id, link);

	nvme_ctrlr->thread = spdk_get_thread();
	nvme_ctrlr->ctrlr = ctrlr;
	nvme_ctrlr->ref = 1;

	if (spdk_nvme_ctrlr_is_ocssd_supported(ctrlr)) {
		SPDK_ERRLOG("OCSSDs are not supported");
		rc = -ENOTSUP;
		goto err;
	}

	if (ctx != NULL) {
		memcpy(&nvme_ctrlr->opts, &ctx->bdev_opts, sizeof(ctx->bdev_opts));
	} else {
		bdev_nvme_get_default_ctrlr_opts(&nvme_ctrlr->opts);
	}

	nvme_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq, nvme_ctrlr,
					  g_opts.nvme_adminq_poll_period_us);

	if (g_opts.timeout_us > 0) {
		/* Register timeout callback. Timeout values for IO vs. admin reqs can be different. */
		/* If timeout_admin_us is 0 (not specified), admin uses same timeout as IO. */
		uint64_t adm_timeout_us = (g_opts.timeout_admin_us == 0) ?
					  g_opts.timeout_us : g_opts.timeout_admin_us;
		spdk_nvme_ctrlr_register_timeout_callback(ctrlr, g_opts.timeout_us,
				adm_timeout_us, timeout_cb, nvme_ctrlr);
	}

	spdk_nvme_ctrlr_register_aer_callback(ctrlr, aer_cb, nvme_ctrlr);
	spdk_nvme_ctrlr_set_remove_cb(ctrlr, remove_cb, nvme_ctrlr);

	if (spdk_nvme_ctrlr_get_flags(ctrlr) &
	    SPDK_NVME_CTRLR_SECURITY_SEND_RECV_SUPPORTED) {
		nvme_ctrlr->opal_dev = spdk_opal_dev_construct(ctrlr);
	}

	rc = nvme_bdev_ctrlr_create(name, nvme_ctrlr);
	if (rc != 0) {
		goto err;
	}

	cdata = spdk_nvme_ctrlr_get_data(ctrlr);

	if (cdata->cmic.ana_reporting) {
		rc = nvme_ctrlr_init_ana_log_page(nvme_ctrlr, ctx);
		if (rc == 0) {
			return 0;
		}
	} else {
		nvme_ctrlr_create_done(nvme_ctrlr, ctx);
		return 0;
	}

err:
	nvme_ctrlr_delete(nvme_ctrlr);
	return rc;
}

void
bdev_nvme_get_default_ctrlr_opts(struct nvme_ctrlr_opts *opts)
{
	opts->prchk_flags = 0;
	opts->ctrlr_loss_timeout_sec = g_opts.ctrlr_loss_timeout_sec;
	opts->reconnect_delay_sec = g_opts.reconnect_delay_sec;
	opts->fast_io_fail_timeout_sec = g_opts.fast_io_fail_timeout_sec;
}

static void
attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
	  struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *drv_opts)
{
	char *name;

	name = spdk_sprintf_alloc("HotInNvme%d", g_hot_insert_nvme_controller_index++);
	if (!name) {
		SPDK_ERRLOG("Failed to assign name to NVMe device\n");
		return;
	}

	if (nvme_ctrlr_create(ctrlr, name, trid, NULL) == 0) {
		SPDK_DEBUGLOG(bdev_nvme, "Attached to %s (%s)\n", trid->traddr, name);
	} else {
		SPDK_ERRLOG("Failed to attach to %s (%s)\n", trid->traddr, name);
	}

	free(name);
}

static void
_nvme_ctrlr_destruct(void *ctx)
{
	struct nvme_ctrlr *nvme_ctrlr = ctx;

	nvme_ctrlr_depopulate_namespaces(nvme_ctrlr);
	nvme_ctrlr_release(nvme_ctrlr);
}

static int
bdev_nvme_delete_ctrlr_unsafe(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
{
	struct nvme_probe_skip_entry *entry;

	/* The controller's destruction was already started */
	if (nvme_ctrlr->destruct) {
		return -EALREADY;
	}

	if (!hotplug &&
	    nvme_ctrlr->active_path_id->trid.trtype == SPDK_NVME_TRANSPORT_PCIE) {
		entry = calloc(1, sizeof(*entry));
		if (!entry) {
			return -ENOMEM;
		}
		entry->trid = nvme_ctrlr->active_path_id->trid;
		TAILQ_INSERT_TAIL(&g_skipped_nvme_ctrlrs, entry, tailq);
	}

	nvme_ctrlr->destruct = true;
	return 0;
}

static int
bdev_nvme_delete_ctrlr(struct nvme_ctrlr *nvme_ctrlr, bool hotplug)
{
	int rc;

	pthread_mutex_lock(&nvme_ctrlr->mutex);
	rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, hotplug);
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	if (rc == 0) {
		_nvme_ctrlr_destruct(nvme_ctrlr);
	} else if (rc == -EALREADY) {
		rc = 0;
	}

	return rc;
}

static void
remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr)
{
	struct nvme_ctrlr *nvme_ctrlr = cb_ctx;

	bdev_nvme_delete_ctrlr(nvme_ctrlr, true);
}

static int
bdev_nvme_hotplug_probe(void *arg)
{
	if (g_hotplug_probe_ctx == NULL) {
		spdk_poller_unregister(&g_hotplug_probe_poller);
		return SPDK_POLLER_IDLE;
	}

	if (spdk_nvme_probe_poll_async(g_hotplug_probe_ctx) != -EAGAIN) {
		g_hotplug_probe_ctx = NULL;
		spdk_poller_unregister(&g_hotplug_probe_poller);
	}

	return SPDK_POLLER_BUSY;
}

static int
bdev_nvme_hotplug(void *arg)
{
	struct spdk_nvme_transport_id trid_pcie;

	if (g_hotplug_probe_ctx) {
		return SPDK_POLLER_BUSY;
	}

	memset(&trid_pcie, 0, sizeof(trid_pcie));
	spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE);

	g_hotplug_probe_ctx = spdk_nvme_probe_async(&trid_pcie, NULL,
			      hotplug_probe_cb, attach_cb, NULL);

	if (g_hotplug_probe_ctx) {
		assert(g_hotplug_probe_poller == NULL);
		g_hotplug_probe_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug_probe, NULL, 1000);
	}

	return SPDK_POLLER_BUSY;
}

void
bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts)
{
	*opts = g_opts;
}

static bool bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
		uint32_t reconnect_delay_sec,
		uint32_t fast_io_fail_timeout_sec);

static int
bdev_nvme_validate_opts(const struct spdk_bdev_nvme_opts *opts)
{
	if ((opts->timeout_us == 0) && (opts->timeout_admin_us != 0)) {
		/* Can't set timeout_admin_us without also setting timeout_us */
		SPDK_WARNLOG("Invalid options: Can't have (timeout_us == 0) with (timeout_admin_us > 0)\n");
		return -EINVAL;
	}

	if (opts->bdev_retry_count < -1) {
		SPDK_WARNLOG("Invalid option: bdev_retry_count can't be less than -1.\n");
		return -EINVAL;
	}

	if (!bdev_nvme_check_io_error_resiliency_params(opts->ctrlr_loss_timeout_sec,
			opts->reconnect_delay_sec,
			opts->fast_io_fail_timeout_sec)) {
		return -EINVAL;
	}

	return 0;
}

int
bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts)
{
	int ret;

	ret = bdev_nvme_validate_opts(opts);
	if (ret) {
		SPDK_WARNLOG("Failed to set nvme opts.\n");
		return ret;
	}

	if (g_bdev_nvme_init_thread != NULL) {
		if (!TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
			return -EPERM;
		}
	}

	if (opts->rdma_srq_size != 0 ||
	    opts->rdma_max_cq_size != 0 ||
	    opts->rdma_cm_event_timeout_ms != 0) {
		struct spdk_nvme_transport_opts drv_opts;

		spdk_nvme_transport_get_opts(&drv_opts, sizeof(drv_opts));
		if (opts->rdma_srq_size != 0) {
			drv_opts.rdma_srq_size = opts->rdma_srq_size;
		}
		if (opts->rdma_max_cq_size != 0) {
			drv_opts.rdma_max_cq_size = opts->rdma_max_cq_size;
		}
		if (opts->rdma_cm_event_timeout_ms != 0) {
			drv_opts.rdma_cm_event_timeout_ms = opts->rdma_cm_event_timeout_ms;
		}

		ret = spdk_nvme_transport_set_opts(&drv_opts, sizeof(drv_opts));
		if (ret) {
			SPDK_ERRLOG("Failed to set NVMe transport opts.\n");
			return ret;
		}
	}

	g_opts = *opts;

	return 0;
}

struct set_nvme_hotplug_ctx {
	uint64_t period_us;
	bool enabled;
	spdk_msg_fn fn;
	void *fn_ctx;
};

static void
set_nvme_hotplug_period_cb(void *_ctx)
{
	struct set_nvme_hotplug_ctx *ctx = _ctx;

	spdk_poller_unregister(&g_hotplug_poller);
	if (ctx->enabled) {
		g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug, NULL, ctx->period_us);
	}

	g_nvme_hotplug_poll_period_us = ctx->period_us;
	g_nvme_hotplug_enabled = ctx->enabled;
	if (ctx->fn) {
		ctx->fn(ctx->fn_ctx);
	}

	free(ctx);
}

int
bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx)
{
	struct set_nvme_hotplug_ctx *ctx;

	if (enabled == true && !spdk_process_is_primary()) {
		return -EPERM;
	}

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		return -ENOMEM;
	}

	period_us = period_us == 0 ? NVME_HOTPLUG_POLL_PERIOD_DEFAULT : period_us;
	ctx->period_us = spdk_min(period_us, NVME_HOTPLUG_POLL_PERIOD_MAX);
	ctx->enabled = enabled;
	ctx->fn = cb;
	ctx->fn_ctx = cb_ctx;

	spdk_thread_send_msg(g_bdev_nvme_init_thread, set_nvme_hotplug_period_cb, ctx);
	return 0;
}

static void
nvme_ctrlr_populate_namespaces_done(struct nvme_ctrlr *nvme_ctrlr,
				    struct nvme_async_probe_ctx *ctx)
{
	struct nvme_ns	*nvme_ns;
	struct nvme_bdev	*nvme_bdev;
	size_t			j;

	assert(nvme_ctrlr != NULL);

	if (ctx->names == NULL) {
		ctx->reported_bdevs = 0;
		populate_namespaces_cb(ctx, 0);
		return;
	}

	/*
	 * Report the new bdevs that were created in this call.
	 * There can be more than one bdev per NVMe controller.
	 */
	j = 0;
	nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
	while (nvme_ns != NULL) {
		nvme_bdev = nvme_ns->bdev;
		if (j < ctx->max_bdevs) {
			ctx->names[j] = nvme_bdev->disk.name;
			j++;
		} else {
			SPDK_ERRLOG("Maximum number of namespaces supported per NVMe controller is %du. Unable to return all names of created bdevs\n",
				    ctx->max_bdevs);
			ctx->reported_bdevs = 0;
			populate_namespaces_cb(ctx, -ERANGE);
			return;
		}

		nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
	}

	ctx->reported_bdevs = j;
	populate_namespaces_cb(ctx, 0);
}

static int
bdev_nvme_check_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
			       struct spdk_nvme_ctrlr *new_ctrlr,
			       struct spdk_nvme_transport_id *trid)
{
	struct nvme_path_id *tmp_trid;

	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
		SPDK_ERRLOG("PCIe failover is not supported.\n");
		return -ENOTSUP;
	}

	/* Currently we only support failover to the same transport type. */
	if (nvme_ctrlr->active_path_id->trid.trtype != trid->trtype) {
		SPDK_WARNLOG("Failover from trtype: %s to a different trtype: %s is not supported currently\n",
			     spdk_nvme_transport_id_trtype_str(nvme_ctrlr->active_path_id->trid.trtype),
			     spdk_nvme_transport_id_trtype_str(trid->trtype));
		return -EINVAL;
	}


	/* Currently we only support failover to the same NQN. */
	if (strncmp(trid->subnqn, nvme_ctrlr->active_path_id->trid.subnqn, SPDK_NVMF_NQN_MAX_LEN)) {
		SPDK_WARNLOG("Failover from subnqn: %s to a different subnqn: %s is not supported currently\n",
			     nvme_ctrlr->active_path_id->trid.subnqn, trid->subnqn);
		return -EINVAL;
	}

	/* Skip all the other checks if we've already registered this path. */
	TAILQ_FOREACH(tmp_trid, &nvme_ctrlr->trids, link) {
		if (!spdk_nvme_transport_id_compare(&tmp_trid->trid, trid)) {
			SPDK_WARNLOG("This path (traddr: %s subnqn: %s) is already registered\n", trid->traddr,
				     trid->subnqn);
			return -EEXIST;
		}
	}

	return 0;
}

static int
bdev_nvme_check_secondary_namespace(struct nvme_ctrlr *nvme_ctrlr,
				    struct spdk_nvme_ctrlr *new_ctrlr)
{
	struct nvme_ns *nvme_ns;
	struct spdk_nvme_ns *new_ns;

	nvme_ns = nvme_ctrlr_get_first_active_ns(nvme_ctrlr);
	while (nvme_ns != NULL) {
		new_ns = spdk_nvme_ctrlr_get_ns(new_ctrlr, nvme_ns->id);
		assert(new_ns != NULL);

		if (!bdev_nvme_compare_ns(nvme_ns->ns, new_ns)) {
			return -EINVAL;
		}

		nvme_ns = nvme_ctrlr_get_next_active_ns(nvme_ctrlr, nvme_ns);
	}

	return 0;
}

static int
_bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
			      struct spdk_nvme_transport_id *trid)
{
	struct nvme_path_id *active_id, *new_trid, *tmp_trid;

	new_trid = calloc(1, sizeof(*new_trid));
	if (new_trid == NULL) {
		return -ENOMEM;
	}
	new_trid->trid = *trid;

	active_id = nvme_ctrlr->active_path_id;
	assert(active_id != NULL);
	assert(active_id == TAILQ_FIRST(&nvme_ctrlr->trids));

	/* Skip the active trid not to replace it until it is failed. */
	tmp_trid = TAILQ_NEXT(active_id, link);
	if (tmp_trid == NULL) {
		goto add_tail;
	}

	/* It means the trid is faled if its last failed time is non-zero.
	 * Insert the new alternate trid before any failed trid.
	 */
	TAILQ_FOREACH_FROM(tmp_trid, &nvme_ctrlr->trids, link) {
		if (tmp_trid->last_failed_tsc != 0) {
			TAILQ_INSERT_BEFORE(tmp_trid, new_trid, link);
			return 0;
		}
	}

add_tail:
	TAILQ_INSERT_TAIL(&nvme_ctrlr->trids, new_trid, link);
	return 0;
}

/* This is the case that a secondary path is added to an existing
 * nvme_ctrlr for failover. After checking if it can access the same
 * namespaces as the primary path, it is disconnected until failover occurs.
 */
static int
bdev_nvme_add_secondary_trid(struct nvme_ctrlr *nvme_ctrlr,
			     struct spdk_nvme_ctrlr *new_ctrlr,
			     struct spdk_nvme_transport_id *trid)
{
	int rc;

	assert(nvme_ctrlr != NULL);

	pthread_mutex_lock(&nvme_ctrlr->mutex);

	rc = bdev_nvme_check_secondary_trid(nvme_ctrlr, new_ctrlr, trid);
	if (rc != 0) {
		goto exit;
	}

	rc = bdev_nvme_check_secondary_namespace(nvme_ctrlr, new_ctrlr);
	if (rc != 0) {
		goto exit;
	}

	rc = _bdev_nvme_add_secondary_trid(nvme_ctrlr, trid);

exit:
	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	spdk_nvme_detach(new_ctrlr);

	return rc;
}

static void
connect_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
		  struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
{
	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
	struct nvme_async_probe_ctx *ctx;
	int rc;

	ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
	ctx->ctrlr_attached = true;

	rc = nvme_ctrlr_create(ctrlr, ctx->base_name, &ctx->trid, ctx);
	if (rc != 0) {
		ctx->reported_bdevs = 0;
		populate_namespaces_cb(ctx, rc);
	}
}

static void
connect_set_failover_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
			struct spdk_nvme_ctrlr *ctrlr,
			const struct spdk_nvme_ctrlr_opts *opts)
{
	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
	struct nvme_ctrlr *nvme_ctrlr;
	struct nvme_async_probe_ctx *ctx;
	int rc;

	ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, drv_opts);
	ctx->ctrlr_attached = true;

	nvme_ctrlr = nvme_ctrlr_get_by_name(ctx->base_name);
	if (nvme_ctrlr) {
		rc = bdev_nvme_add_secondary_trid(nvme_ctrlr, ctrlr, &ctx->trid);
	} else {
		rc = -ENODEV;
	}

	ctx->reported_bdevs = 0;
	populate_namespaces_cb(ctx, rc);
}

static int
bdev_nvme_async_poll(void *arg)
{
	struct nvme_async_probe_ctx	*ctx = arg;
	int				rc;

	rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
	if (spdk_unlikely(rc != -EAGAIN)) {
		ctx->probe_done = true;
		spdk_poller_unregister(&ctx->poller);
		if (!ctx->ctrlr_attached) {
			/* The probe is done, but no controller was attached.
			 * That means we had a failure, so report -EIO back to
			 * the caller (usually the RPC). populate_namespaces_cb()
			 * will take care of freeing the nvme_async_probe_ctx.
			 */
			ctx->reported_bdevs = 0;
			populate_namespaces_cb(ctx, -EIO);
		} else if (ctx->namespaces_populated) {
			/* The namespaces for the attached controller were all
			 * populated and the response was already sent to the
			 * caller (usually the RPC).  So free the context here.
			 */
			free_nvme_async_probe_ctx(ctx);
		}
	}

	return SPDK_POLLER_BUSY;
}

static bool
bdev_nvme_check_io_error_resiliency_params(int32_t ctrlr_loss_timeout_sec,
		uint32_t reconnect_delay_sec,
		uint32_t fast_io_fail_timeout_sec)
{
	if (ctrlr_loss_timeout_sec < -1) {
		SPDK_ERRLOG("ctrlr_loss_timeout_sec can't be less than -1.\n");
		return false;
	} else if (ctrlr_loss_timeout_sec == -1) {
		if (reconnect_delay_sec == 0) {
			SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
			return false;
		} else if (fast_io_fail_timeout_sec != 0 &&
			   fast_io_fail_timeout_sec < reconnect_delay_sec) {
			SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io-fail_timeout_sec.\n");
			return false;
		}
	} else if (ctrlr_loss_timeout_sec != 0) {
		if (reconnect_delay_sec == 0) {
			SPDK_ERRLOG("reconnect_delay_sec can't be 0 if ctrlr_loss_timeout_sec is not 0.\n");
			return false;
		} else if (reconnect_delay_sec > (uint32_t)ctrlr_loss_timeout_sec) {
			SPDK_ERRLOG("reconnect_delay_sec can't be more than ctrlr_loss_timeout_sec.\n");
			return false;
		} else if (fast_io_fail_timeout_sec != 0) {
			if (fast_io_fail_timeout_sec < reconnect_delay_sec) {
				SPDK_ERRLOG("reconnect_delay_sec can't be more than fast_io_fail_timeout_sec.\n");
				return false;
			} else if (fast_io_fail_timeout_sec > (uint32_t)ctrlr_loss_timeout_sec) {
				SPDK_ERRLOG("fast_io_fail_timeout_sec can't be more than ctrlr_loss_timeout_sec.\n");
				return false;
			}
		}
	} else if (reconnect_delay_sec != 0 || fast_io_fail_timeout_sec != 0) {
		SPDK_ERRLOG("Both reconnect_delay_sec and fast_io_fail_timeout_sec must be 0 if ctrlr_loss_timeout_sec is 0.\n");
		return false;
	}

	return true;
}

static int
bdev_nvme_load_psk(const char *fname, char *buf, size_t bufsz)
{
	FILE *psk_file;
	struct stat statbuf;
	int rc;
#define TCP_PSK_INVALID_PERMISSIONS 0177

	if (stat(fname, &statbuf) != 0) {
		SPDK_ERRLOG("Could not read permissions for PSK file\n");
		return -EACCES;
	}

	if ((statbuf.st_mode & TCP_PSK_INVALID_PERMISSIONS) != 0) {
		SPDK_ERRLOG("Incorrect permissions for PSK file\n");
		return -EPERM;
	}
	if ((size_t)statbuf.st_size >= bufsz) {
		SPDK_ERRLOG("Invalid PSK: too long\n");
		return -EINVAL;
	}
	psk_file = fopen(fname, "r");
	if (psk_file == NULL) {
		SPDK_ERRLOG("Could not open PSK file\n");
		return -EINVAL;
	}

	memset(buf, 0, bufsz);
	rc = fread(buf, 1, statbuf.st_size, psk_file);
	if (rc != statbuf.st_size) {
		SPDK_ERRLOG("Failed to read PSK\n");
		fclose(psk_file);
		return -EINVAL;
	}

	fclose(psk_file);
	return 0;
}

int
bdev_nvme_create(struct spdk_nvme_transport_id *trid,
		 const char *base_name,
		 const char **names,
		 uint32_t count,
		 spdk_bdev_create_nvme_fn cb_fn,
		 void *cb_ctx,
		 struct spdk_nvme_ctrlr_opts *drv_opts,
		 struct nvme_ctrlr_opts *bdev_opts,
		 bool multipath)
{
	struct nvme_probe_skip_entry *entry, *tmp;
	struct nvme_async_probe_ctx *ctx;
	spdk_nvme_attach_cb attach_cb;
	int rc, len;

	/* TODO expand this check to include both the host and target TRIDs.
	 * Only if both are the same should we fail.
	 */
	if (nvme_ctrlr_get(trid) != NULL) {
		SPDK_ERRLOG("A controller with the provided trid (traddr: %s) already exists.\n", trid->traddr);
		return -EEXIST;
	}

	len = strnlen(base_name, SPDK_CONTROLLER_NAME_MAX);

	if (len == 0 || len == SPDK_CONTROLLER_NAME_MAX) {
		SPDK_ERRLOG("controller name must be between 1 and %d characters\n", SPDK_CONTROLLER_NAME_MAX - 1);
		return -EINVAL;
	}

	if (bdev_opts != NULL &&
	    !bdev_nvme_check_io_error_resiliency_params(bdev_opts->ctrlr_loss_timeout_sec,
			    bdev_opts->reconnect_delay_sec,
			    bdev_opts->fast_io_fail_timeout_sec)) {
		return -EINVAL;
	}

	ctx = calloc(1, sizeof(*ctx));
	if (!ctx) {
		return -ENOMEM;
	}
	ctx->base_name = base_name;
	ctx->names = names;
	ctx->max_bdevs = count;
	ctx->cb_fn = cb_fn;
	ctx->cb_ctx = cb_ctx;
	ctx->trid = *trid;

	if (bdev_opts) {
		memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
	} else {
		bdev_nvme_get_default_ctrlr_opts(&ctx->bdev_opts);
	}

	if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) {
		TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, tmp) {
			if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) {
				TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
				free(entry);
				break;
			}
		}
	}

	if (drv_opts) {
		memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
	} else {
		spdk_nvme_ctrlr_get_default_ctrlr_opts(&ctx->drv_opts, sizeof(ctx->drv_opts));
	}

	ctx->drv_opts.transport_retry_count = g_opts.transport_retry_count;
	ctx->drv_opts.transport_ack_timeout = g_opts.transport_ack_timeout;
	ctx->drv_opts.keep_alive_timeout_ms = g_opts.keep_alive_timeout_ms;
	ctx->drv_opts.disable_read_ana_log_page = true;
	ctx->drv_opts.transport_tos = g_opts.transport_tos;

	if (ctx->bdev_opts.psk[0] != '\0') {
		/* Try to use the keyring first */
		ctx->drv_opts.tls_psk = spdk_keyring_get_key(ctx->bdev_opts.psk);
		if (ctx->drv_opts.tls_psk == NULL) {
			rc = bdev_nvme_load_psk(ctx->bdev_opts.psk,
						ctx->drv_opts.psk, sizeof(ctx->drv_opts.psk));
			if (rc != 0) {
				SPDK_ERRLOG("Could not load PSK from %s\n", ctx->bdev_opts.psk);
				free_nvme_async_probe_ctx(ctx);
				return rc;
			}
		}
	}

	if (ctx->bdev_opts.dhchap_key != NULL) {
		ctx->drv_opts.dhchap_key = spdk_keyring_get_key(ctx->bdev_opts.dhchap_key);
		if (ctx->drv_opts.dhchap_key == NULL) {
			SPDK_ERRLOG("Could not load DH-HMAC-CHAP key: %s\n",
				    ctx->bdev_opts.dhchap_key);
			free_nvme_async_probe_ctx(ctx);
			return -ENOKEY;
		}

		ctx->drv_opts.dhchap_digests = g_opts.dhchap_digests;
		ctx->drv_opts.dhchap_dhgroups = g_opts.dhchap_dhgroups;
	}

	if (nvme_bdev_ctrlr_get_by_name(base_name) == NULL || multipath) {
		attach_cb = connect_attach_cb;
	} else {
		attach_cb = connect_set_failover_cb;
	}

	ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, attach_cb);
	if (ctx->probe_ctx == NULL) {
		SPDK_ERRLOG("No controller was found with provided trid (traddr: %s)\n", trid->traddr);
		free_nvme_async_probe_ctx(ctx);
		return -ENODEV;
	}
	ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_async_poll, ctx, 1000);

	return 0;
}

struct bdev_nvme_delete_ctx {
	char                        *name;
	struct nvme_path_id         path_id;
	bdev_nvme_delete_done_fn    delete_done;
	void                        *delete_done_ctx;
	uint64_t                    timeout_ticks;
	struct spdk_poller          *poller;
};

static void
free_bdev_nvme_delete_ctx(struct bdev_nvme_delete_ctx *ctx)
{
	if (ctx != NULL) {
		free(ctx->name);
		free(ctx);
	}
}

static bool
nvme_path_id_compare(struct nvme_path_id *p, const struct nvme_path_id *path_id)
{
	if (path_id->trid.trtype != 0) {
		if (path_id->trid.trtype == SPDK_NVME_TRANSPORT_CUSTOM) {
			if (strcasecmp(path_id->trid.trstring, p->trid.trstring) != 0) {
				return false;
			}
		} else {
			if (path_id->trid.trtype != p->trid.trtype) {
				return false;
			}
		}
	}

	if (!spdk_mem_all_zero(path_id->trid.traddr, sizeof(path_id->trid.traddr))) {
		if (strcasecmp(path_id->trid.traddr, p->trid.traddr) != 0) {
			return false;
		}
	}

	if (path_id->trid.adrfam != 0) {
		if (path_id->trid.adrfam != p->trid.adrfam) {
			return false;
		}
	}

	if (!spdk_mem_all_zero(path_id->trid.trsvcid, sizeof(path_id->trid.trsvcid))) {
		if (strcasecmp(path_id->trid.trsvcid, p->trid.trsvcid) != 0) {
			return false;
		}
	}

	if (!spdk_mem_all_zero(path_id->trid.subnqn, sizeof(path_id->trid.subnqn))) {
		if (strcmp(path_id->trid.subnqn, p->trid.subnqn) != 0) {
			return false;
		}
	}

	if (!spdk_mem_all_zero(path_id->hostid.hostaddr, sizeof(path_id->hostid.hostaddr))) {
		if (strcmp(path_id->hostid.hostaddr, p->hostid.hostaddr) != 0) {
			return false;
		}
	}

	if (!spdk_mem_all_zero(path_id->hostid.hostsvcid, sizeof(path_id->hostid.hostsvcid))) {
		if (strcmp(path_id->hostid.hostsvcid, p->hostid.hostsvcid) != 0) {
			return false;
		}
	}

	return true;
}

static bool
nvme_path_id_exists(const char *name, const struct nvme_path_id *path_id)
{
	struct nvme_bdev_ctrlr  *nbdev_ctrlr;
	struct nvme_ctrlr       *ctrlr;
	struct nvme_path_id     *p;

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
	if (!nbdev_ctrlr) {
		pthread_mutex_unlock(&g_bdev_nvme_mutex);
		return false;
	}

	TAILQ_FOREACH(ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
		pthread_mutex_lock(&ctrlr->mutex);
		TAILQ_FOREACH(p, &ctrlr->trids, link) {
			if (nvme_path_id_compare(p, path_id)) {
				pthread_mutex_unlock(&ctrlr->mutex);
				pthread_mutex_unlock(&g_bdev_nvme_mutex);
				return true;
			}
		}
		pthread_mutex_unlock(&ctrlr->mutex);
	}
	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	return false;
}

static int
bdev_nvme_delete_complete_poll(void *arg)
{
	struct bdev_nvme_delete_ctx     *ctx = arg;
	int                             rc = 0;

	if (nvme_path_id_exists(ctx->name, &ctx->path_id)) {
		if (ctx->timeout_ticks > spdk_get_ticks()) {
			return SPDK_POLLER_BUSY;
		}

		SPDK_ERRLOG("NVMe path '%s' still exists after delete\n", ctx->name);
		rc = -ETIMEDOUT;
	}

	spdk_poller_unregister(&ctx->poller);

	ctx->delete_done(ctx->delete_done_ctx, rc);
	free_bdev_nvme_delete_ctx(ctx);

	return SPDK_POLLER_BUSY;
}

static int
_bdev_nvme_delete(struct nvme_ctrlr *nvme_ctrlr, const struct nvme_path_id *path_id)
{
	struct nvme_path_id	*p, *t;
	spdk_msg_fn		msg_fn;
	int			rc = -ENXIO;

	pthread_mutex_lock(&nvme_ctrlr->mutex);

	TAILQ_FOREACH_REVERSE_SAFE(p, &nvme_ctrlr->trids, nvme_paths, link, t) {
		if (p == TAILQ_FIRST(&nvme_ctrlr->trids)) {
			break;
		}

		if (!nvme_path_id_compare(p, path_id)) {
			continue;
		}

		/* We are not using the specified path. */
		TAILQ_REMOVE(&nvme_ctrlr->trids, p, link);
		free(p);
		rc = 0;
	}

	if (p == NULL || !nvme_path_id_compare(p, path_id)) {
		pthread_mutex_unlock(&nvme_ctrlr->mutex);
		return rc;
	}

	/* If we made it here, then this path is a match! Now we need to remove it. */

	/* This is the active path in use right now. The active path is always the first in the list. */
	assert(p == nvme_ctrlr->active_path_id);

	if (!TAILQ_NEXT(p, link)) {
		/* The current path is the only path. */
		msg_fn = _nvme_ctrlr_destruct;
		rc = bdev_nvme_delete_ctrlr_unsafe(nvme_ctrlr, false);
	} else {
		/* There is an alternative path. */
		msg_fn = _bdev_nvme_reset_ctrlr;
		rc = bdev_nvme_failover_ctrlr_unsafe(nvme_ctrlr, true);
	}

	pthread_mutex_unlock(&nvme_ctrlr->mutex);

	if (rc == 0) {
		spdk_thread_send_msg(nvme_ctrlr->thread, msg_fn, nvme_ctrlr);
	} else if (rc == -EALREADY) {
		rc = 0;
	}

	return rc;
}

int
bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id,
		 bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx)
{
	struct nvme_bdev_ctrlr		*nbdev_ctrlr;
	struct nvme_ctrlr		*nvme_ctrlr, *tmp_nvme_ctrlr;
	struct bdev_nvme_delete_ctx     *ctx = NULL;
	int				rc = -ENXIO, _rc;

	if (name == NULL || path_id == NULL) {
		rc = -EINVAL;
		goto exit;
	}

	pthread_mutex_lock(&g_bdev_nvme_mutex);

	nbdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name);
	if (nbdev_ctrlr == NULL) {
		pthread_mutex_unlock(&g_bdev_nvme_mutex);

		SPDK_ERRLOG("Failed to find NVMe bdev controller\n");
		rc = -ENODEV;
		goto exit;
	}

	TAILQ_FOREACH_SAFE(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq, tmp_nvme_ctrlr) {
		_rc = _bdev_nvme_delete(nvme_ctrlr, path_id);
		if (_rc < 0 && _rc != -ENXIO) {
			pthread_mutex_unlock(&g_bdev_nvme_mutex);
			rc = _rc;
			goto exit;
		} else if (_rc == 0) {
			/* We traverse all remaining nvme_ctrlrs even if one nvme_ctrlr
			 * was deleted successfully. To remember the successful deletion,
			 * overwrite rc only if _rc is zero.
			 */
			rc = 0;
		}
	}

	pthread_mutex_unlock(&g_bdev_nvme_mutex);

	if (rc != 0 || delete_done == NULL) {
		goto exit;
	}

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		SPDK_ERRLOG("Failed to allocate context for bdev_nvme_delete\n");
		rc = -ENOMEM;
		goto exit;
	}

	ctx->name = strdup(name);
	if (ctx->name == NULL) {
		SPDK_ERRLOG("Failed to copy controller name for deletion\n");
		rc = -ENOMEM;
		goto exit;
	}

	ctx->delete_done = delete_done;
	ctx->delete_done_ctx = delete_done_ctx;
	ctx->path_id = *path_id;
	ctx->timeout_ticks = spdk_get_ticks() + 10 * spdk_get_ticks_hz();
	ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_delete_complete_poll, ctx, 1000);
	if (ctx->poller == NULL) {
		SPDK_ERRLOG("Failed to register bdev_nvme_delete poller\n");
		rc = -ENOMEM;
		goto exit;
	}

exit:
	if (rc != 0) {
		free_bdev_nvme_delete_ctx(ctx);
	}

	return rc;
}

#define DISCOVERY_INFOLOG(ctx, format, ...) \
	SPDK_INFOLOG(bdev_nvme, "Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);

#define DISCOVERY_ERRLOG(ctx, format, ...) \
	SPDK_ERRLOG("Discovery[%s:%s] " format, ctx->trid.traddr, ctx->trid.trsvcid, ##__VA_ARGS__);

struct discovery_entry_ctx {
	char						name[128];
	struct spdk_nvme_transport_id			trid;
	struct spdk_nvme_ctrlr_opts			drv_opts;
	struct spdk_nvmf_discovery_log_page_entry	entry;
	TAILQ_ENTRY(discovery_entry_ctx)		tailq;
	struct discovery_ctx				*ctx;
};

struct discovery_ctx {
	char					*name;
	spdk_bdev_nvme_start_discovery_fn	start_cb_fn;
	spdk_bdev_nvme_stop_discovery_fn	stop_cb_fn;
	void					*cb_ctx;
	struct spdk_nvme_probe_ctx		*probe_ctx;
	struct spdk_nvme_detach_ctx		*detach_ctx;
	struct spdk_nvme_ctrlr			*ctrlr;
	struct spdk_nvme_transport_id		trid;
	struct discovery_entry_ctx		*entry_ctx_in_use;
	struct spdk_poller			*poller;
	struct spdk_nvme_ctrlr_opts		drv_opts;
	struct nvme_ctrlr_opts			bdev_opts;
	struct spdk_nvmf_discovery_log_page	*log_page;
	TAILQ_ENTRY(discovery_ctx)		tailq;
	TAILQ_HEAD(, discovery_entry_ctx)	nvm_entry_ctxs;
	TAILQ_HEAD(, discovery_entry_ctx)	discovery_entry_ctxs;
	int					rc;
	bool					wait_for_attach;
	uint64_t				timeout_ticks;
	/* Denotes that the discovery service is being started. We're waiting
	 * for the initial connection to the discovery controller to be
	 * established and attach discovered NVM ctrlrs.
	 */
	bool					initializing;
	/* Denotes if a discovery is currently in progress for this context.
	 * That includes connecting to newly discovered subsystems.  Used to
	 * ensure we do not start a new discovery until an existing one is
	 * complete.
	 */
	bool					in_progress;

	/* Denotes if another discovery is needed after the one in progress
	 * completes.  Set when we receive an AER completion while a discovery
	 * is already in progress.
	 */
	bool					pending;

	/* Signal to the discovery context poller that it should stop the
	 * discovery service, including detaching from the current discovery
	 * controller.
	 */
	bool					stop;

	struct spdk_thread			*calling_thread;
	uint32_t				index;
	uint32_t				attach_in_progress;
	char					*hostnqn;

	/* Denotes if the discovery service was started by the mdns discovery.
	 */
	bool					from_mdns_discovery_service;
};

TAILQ_HEAD(discovery_ctxs, discovery_ctx);
static struct discovery_ctxs g_discovery_ctxs = TAILQ_HEAD_INITIALIZER(g_discovery_ctxs);

static void get_discovery_log_page(struct discovery_ctx *ctx);

static void
free_discovery_ctx(struct discovery_ctx *ctx)
{
	free(ctx->log_page);
	free(ctx->hostnqn);
	free(ctx->name);
	free(ctx);
}

static void
discovery_complete(struct discovery_ctx *ctx)
{
	ctx->initializing = false;
	ctx->in_progress = false;
	if (ctx->pending) {
		ctx->pending = false;
		get_discovery_log_page(ctx);
	}
}

static void
build_trid_from_log_page_entry(struct spdk_nvme_transport_id *trid,
			       struct spdk_nvmf_discovery_log_page_entry *entry)
{
	char *space;

	trid->trtype = entry->trtype;
	trid->adrfam = entry->adrfam;
	memcpy(trid->traddr, entry->traddr, sizeof(entry->traddr));
	memcpy(trid->trsvcid, entry->trsvcid, sizeof(entry->trsvcid));
	/* Because the source buffer (entry->subnqn) is longer than trid->subnqn, and
	 * before call to this function trid->subnqn is zeroed out, we need
	 * to copy sizeof(trid->subnqn) minus one byte to make sure the last character
	 * remains 0. Then we can shorten the string (replace ' ' with 0) if required
	 */
	memcpy(trid->subnqn, entry->subnqn, sizeof(trid->subnqn) - 1);

	/* We want the traddr, trsvcid and subnqn fields to be NULL-terminated.
	 * But the log page entries typically pad them with spaces, not zeroes.
	 * So add a NULL terminator to each of these fields at the appropriate
	 * location.
	 */
	space = strchr(trid->traddr, ' ');
	if (space) {
		*space = 0;
	}
	space = strchr(trid->trsvcid, ' ');
	if (space) {
		*space = 0;
	}
	space = strchr(trid->subnqn, ' ');
	if (space) {
		*space = 0;
	}
}

static void
_stop_discovery(void *_ctx)
{
	struct discovery_ctx *ctx = _ctx;

	if (ctx->attach_in_progress > 0) {
		spdk_thread_send_msg(spdk_get_thread(), _stop_discovery, ctx);
		return;
	}

	ctx->stop = true;

	while (!TAILQ_EMPTY(&ctx->nvm_entry_ctxs)) {
		struct discovery_entry_ctx *entry_ctx;
		struct nvme_path_id path = {};

		entry_ctx = TAILQ_FIRST(&ctx->nvm_entry_ctxs);
		path.trid = entry_ctx->trid;
		bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
		TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
		free(entry_ctx);
	}

	while (!TAILQ_EMPTY(&ctx->discovery_entry_ctxs)) {
		struct discovery_entry_ctx *entry_ctx;

		entry_ctx = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
		TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
		free(entry_ctx);
	}

	free(ctx->entry_ctx_in_use);
	ctx->entry_ctx_in_use = NULL;
}

static void
stop_discovery(struct discovery_ctx *ctx, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
{
	ctx->stop_cb_fn = cb_fn;
	ctx->cb_ctx = cb_ctx;

	if (ctx->attach_in_progress > 0) {
		DISCOVERY_INFOLOG(ctx, "stopping discovery with attach_in_progress: %"PRIu32"\n",
				  ctx->attach_in_progress);
	}

	_stop_discovery(ctx);
}

static void
remove_discovery_entry(struct nvme_ctrlr *nvme_ctrlr)
{
	struct discovery_ctx *d_ctx;
	struct nvme_path_id *path_id;
	struct spdk_nvme_transport_id trid = {};
	struct discovery_entry_ctx *entry_ctx, *tmp;

	path_id = TAILQ_FIRST(&nvme_ctrlr->trids);

	TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
		TAILQ_FOREACH_SAFE(entry_ctx, &d_ctx->nvm_entry_ctxs, tailq, tmp) {
			build_trid_from_log_page_entry(&trid, &entry_ctx->entry);
			if (spdk_nvme_transport_id_compare(&trid, &path_id->trid) != 0) {
				continue;
			}

			TAILQ_REMOVE(&d_ctx->nvm_entry_ctxs, entry_ctx, tailq);
			free(entry_ctx);
			DISCOVERY_INFOLOG(d_ctx, "Remove discovery entry: %s:%s:%s\n",
					  trid.subnqn, trid.traddr, trid.trsvcid);

			/* Fail discovery ctrlr to force reattach attempt */
			spdk_nvme_ctrlr_fail(d_ctx->ctrlr);
		}
	}
}

static void
discovery_remove_controllers(struct discovery_ctx *ctx)
{
	struct spdk_nvmf_discovery_log_page *log_page = ctx->log_page;
	struct discovery_entry_ctx *entry_ctx, *tmp;
	struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
	struct spdk_nvme_transport_id old_trid = {};
	uint64_t numrec, i;
	bool found;

	numrec = from_le64(&log_page->numrec);
	TAILQ_FOREACH_SAFE(entry_ctx, &ctx->nvm_entry_ctxs, tailq, tmp) {
		found = false;
		old_entry = &entry_ctx->entry;
		build_trid_from_log_page_entry(&old_trid, old_entry);
		for (i = 0; i < numrec; i++) {
			new_entry = &log_page->entries[i];
			if (!memcmp(old_entry, new_entry, sizeof(*old_entry))) {
				DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s found again\n",
						  old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);
				found = true;
				break;
			}
		}
		if (!found) {
			struct nvme_path_id path = {};

			DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s not found\n",
					  old_trid.subnqn, old_trid.traddr, old_trid.trsvcid);

			path.trid = entry_ctx->trid;
			bdev_nvme_delete(entry_ctx->name, &path, NULL, NULL);
			TAILQ_REMOVE(&ctx->nvm_entry_ctxs, entry_ctx, tailq);
			free(entry_ctx);
		}
	}
	free(log_page);
	ctx->log_page = NULL;
	discovery_complete(ctx);
}

static void
complete_discovery_start(struct discovery_ctx *ctx, int status)
{
	ctx->timeout_ticks = 0;
	ctx->rc = status;
	if (ctx->start_cb_fn) {
		ctx->start_cb_fn(ctx->cb_ctx, status);
		ctx->start_cb_fn = NULL;
		ctx->cb_ctx = NULL;
	}
}

static void
discovery_attach_controller_done(void *cb_ctx, size_t bdev_count, int rc)
{
	struct discovery_entry_ctx *entry_ctx = cb_ctx;
	struct discovery_ctx *ctx = entry_ctx->ctx;

	DISCOVERY_INFOLOG(ctx, "attach %s done\n", entry_ctx->name);
	ctx->attach_in_progress--;
	if (ctx->attach_in_progress == 0) {
		complete_discovery_start(ctx, ctx->rc);
		if (ctx->initializing && ctx->rc != 0) {
			DISCOVERY_ERRLOG(ctx, "stopping discovery due to errors: %d\n", ctx->rc);
			stop_discovery(ctx, NULL, ctx->cb_ctx);
		} else {
			discovery_remove_controllers(ctx);
		}
	}
}

static struct discovery_entry_ctx *
create_discovery_entry_ctx(struct discovery_ctx *ctx, struct spdk_nvme_transport_id *trid)
{
	struct discovery_entry_ctx *new_ctx;

	new_ctx = calloc(1, sizeof(*new_ctx));
	if (new_ctx == NULL) {
		DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
		return NULL;
	}

	new_ctx->ctx = ctx;
	memcpy(&new_ctx->trid, trid, sizeof(*trid));
	spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
	snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
	return new_ctx;
}

static void
discovery_log_page_cb(void *cb_arg, int rc, const struct spdk_nvme_cpl *cpl,
		      struct spdk_nvmf_discovery_log_page *log_page)
{
	struct discovery_ctx *ctx = cb_arg;
	struct discovery_entry_ctx *entry_ctx, *tmp;
	struct spdk_nvmf_discovery_log_page_entry *new_entry, *old_entry;
	uint64_t numrec, i;
	bool found;

	if (rc || spdk_nvme_cpl_is_error(cpl)) {
		DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
		return;
	}

	ctx->log_page = log_page;
	assert(ctx->attach_in_progress == 0);
	numrec = from_le64(&log_page->numrec);
	TAILQ_FOREACH_SAFE(entry_ctx, &ctx->discovery_entry_ctxs, tailq, tmp) {
		TAILQ_REMOVE(&ctx->discovery_entry_ctxs, entry_ctx, tailq);
		free(entry_ctx);
	}
	for (i = 0; i < numrec; i++) {
		found = false;
		new_entry = &log_page->entries[i];
		if (new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY_CURRENT ||
		    new_entry->subtype == SPDK_NVMF_SUBTYPE_DISCOVERY) {
			struct discovery_entry_ctx *new_ctx;
			struct spdk_nvme_transport_id trid = {};

			build_trid_from_log_page_entry(&trid, new_entry);
			new_ctx = create_discovery_entry_ctx(ctx, &trid);
			if (new_ctx == NULL) {
				DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
				break;
			}

			TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, new_ctx, tailq);
			continue;
		}
		TAILQ_FOREACH(entry_ctx, &ctx->nvm_entry_ctxs, tailq) {
			old_entry = &entry_ctx->entry;
			if (!memcmp(new_entry, old_entry, sizeof(*new_entry))) {
				found = true;
				break;
			}
		}
		if (!found) {
			struct discovery_entry_ctx *subnqn_ctx = NULL, *new_ctx;
			struct discovery_ctx *d_ctx;

			TAILQ_FOREACH(d_ctx, &g_discovery_ctxs, tailq) {
				TAILQ_FOREACH(subnqn_ctx, &d_ctx->nvm_entry_ctxs, tailq) {
					if (!memcmp(subnqn_ctx->entry.subnqn, new_entry->subnqn,
						    sizeof(new_entry->subnqn))) {
						break;
					}
				}
				if (subnqn_ctx) {
					break;
				}
			}

			new_ctx = calloc(1, sizeof(*new_ctx));
			if (new_ctx == NULL) {
				DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
				break;
			}

			new_ctx->ctx = ctx;
			memcpy(&new_ctx->entry, new_entry, sizeof(*new_entry));
			build_trid_from_log_page_entry(&new_ctx->trid, new_entry);
			if (subnqn_ctx) {
				snprintf(new_ctx->name, sizeof(new_ctx->name), "%s", subnqn_ctx->name);
				DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new path for %s\n",
						  new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
						  new_ctx->name);
			} else {
				snprintf(new_ctx->name, sizeof(new_ctx->name), "%s%d", ctx->name, ctx->index++);
				DISCOVERY_INFOLOG(ctx, "NVM %s:%s:%s new subsystem %s\n",
						  new_ctx->trid.subnqn, new_ctx->trid.traddr, new_ctx->trid.trsvcid,
						  new_ctx->name);
			}
			spdk_nvme_ctrlr_get_default_ctrlr_opts(&new_ctx->drv_opts, sizeof(new_ctx->drv_opts));
			snprintf(new_ctx->drv_opts.hostnqn, sizeof(new_ctx->drv_opts.hostnqn), "%s", ctx->hostnqn);
			rc = bdev_nvme_create(&new_ctx->trid, new_ctx->name, NULL, 0,
					      discovery_attach_controller_done, new_ctx,
					      &new_ctx->drv_opts, &ctx->bdev_opts, true);
			if (rc == 0) {
				TAILQ_INSERT_TAIL(&ctx->nvm_entry_ctxs, new_ctx, tailq);
				ctx->attach_in_progress++;
			} else {
				DISCOVERY_ERRLOG(ctx, "bdev_nvme_create failed (%s)\n", spdk_strerror(-rc));
			}
		}
	}

	if (ctx->attach_in_progress == 0) {
		discovery_remove_controllers(ctx);
	}
}

static void
get_discovery_log_page(struct discovery_ctx *ctx)
{
	int rc;

	assert(ctx->in_progress == false);
	ctx->in_progress = true;
	rc = spdk_nvme_ctrlr_get_discovery_log_page(ctx->ctrlr, discovery_log_page_cb, ctx);
	if (rc != 0) {
		DISCOVERY_ERRLOG(ctx, "could not get discovery log page\n");
	}
	DISCOVERY_INFOLOG(ctx, "sent discovery log page command\n");
}

static void
discovery_aer_cb(void *arg, const struct spdk_nvme_cpl *cpl)
{
	struct discovery_ctx *ctx = arg;
	uint32_t log_page_id = (cpl->cdw0 & 0xFF0000) >> 16;

	if (spdk_nvme_cpl_is_error(cpl)) {
		DISCOVERY_ERRLOG(ctx, "aer failed\n");
		return;
	}

	if (log_page_id != SPDK_NVME_LOG_DISCOVERY) {
		DISCOVERY_ERRLOG(ctx, "unexpected log page 0x%x\n", log_page_id);
		return;
	}

	DISCOVERY_INFOLOG(ctx, "got aer\n");
	if (ctx->in_progress) {
		ctx->pending = true;
		return;
	}

	get_discovery_log_page(ctx);
}

static void
discovery_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid,
		    struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts)
{
	struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx;
	struct discovery_ctx *ctx;

	ctx = SPDK_CONTAINEROF(user_opts, struct discovery_ctx, drv_opts);

	DISCOVERY_INFOLOG(ctx, "discovery ctrlr attached\n");
	ctx->probe_ctx = NULL;
	ctx->ctrlr = ctrlr;

	if (ctx->rc != 0) {
		DISCOVERY_ERRLOG(ctx, "encountered error while attaching discovery ctrlr: %d\n",
				 ctx->rc);
		return;
	}

	spdk_nvme_ctrlr_register_aer_callback(ctx->ctrlr, discovery_aer_cb, ctx);
}

static int
discovery_poller(void *arg)
{
	struct discovery_ctx *ctx = arg;
	struct spdk_nvme_transport_id *trid;
	int rc;

	if (ctx->detach_ctx) {
		rc = spdk_nvme_detach_poll_async(ctx->detach_ctx);
		if (rc != -EAGAIN) {
			ctx->detach_ctx = NULL;
			ctx->ctrlr = NULL;
		}
	} else if (ctx->stop) {
		if (ctx->ctrlr != NULL) {
			rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
			if (rc == 0) {
				return SPDK_POLLER_BUSY;
			}
			DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
		}
		spdk_poller_unregister(&ctx->poller);
		TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
		assert(ctx->start_cb_fn == NULL);
		if (ctx->stop_cb_fn != NULL) {
			ctx->stop_cb_fn(ctx->cb_ctx);
		}
		free_discovery_ctx(ctx);
	} else if (ctx->probe_ctx == NULL && ctx->ctrlr == NULL) {
		if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
			DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
			assert(ctx->initializing);
			spdk_poller_unregister(&ctx->poller);
			TAILQ_REMOVE(&g_discovery_ctxs, ctx, tailq);
			complete_discovery_start(ctx, -ETIMEDOUT);
			stop_discovery(ctx, NULL, NULL);
			free_discovery_ctx(ctx);
			return SPDK_POLLER_BUSY;
		}

		assert(ctx->entry_ctx_in_use == NULL);
		ctx->entry_ctx_in_use = TAILQ_FIRST(&ctx->discovery_entry_ctxs);
		TAILQ_REMOVE(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
		trid = &ctx->entry_ctx_in_use->trid;
		ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->drv_opts, discovery_attach_cb);
		if (ctx->probe_ctx) {
			spdk_poller_unregister(&ctx->poller);
			ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000);
		} else {
			DISCOVERY_ERRLOG(ctx, "could not start discovery connect\n");
			TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
			ctx->entry_ctx_in_use = NULL;
		}
	} else if (ctx->probe_ctx) {
		if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
			DISCOVERY_ERRLOG(ctx, "timed out while attaching discovery ctrlr\n");
			complete_discovery_start(ctx, -ETIMEDOUT);
			return SPDK_POLLER_BUSY;
		}

		rc = spdk_nvme_probe_poll_async(ctx->probe_ctx);
		if (rc != -EAGAIN) {
			if (ctx->rc != 0) {
				assert(ctx->initializing);
				stop_discovery(ctx, NULL, ctx->cb_ctx);
			} else {
				assert(rc == 0);
				DISCOVERY_INFOLOG(ctx, "discovery ctrlr connected\n");
				ctx->rc = rc;
				get_discovery_log_page(ctx);
			}
		}
	} else {
		if (ctx->timeout_ticks != 0 && ctx->timeout_ticks < spdk_get_ticks()) {
			DISCOVERY_ERRLOG(ctx, "timed out while attaching NVM ctrlrs\n");
			complete_discovery_start(ctx, -ETIMEDOUT);
			/* We need to wait until all NVM ctrlrs are attached before we stop the
			 * discovery service to make sure we don't detach a ctrlr that is still
			 * being attached.
			 */
			if (ctx->attach_in_progress == 0) {
				stop_discovery(ctx, NULL, ctx->cb_ctx);
				return SPDK_POLLER_BUSY;
			}
		}

		rc = spdk_nvme_ctrlr_process_admin_completions(ctx->ctrlr);
		if (rc < 0) {
			spdk_poller_unregister(&ctx->poller);
			ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
			TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, ctx->entry_ctx_in_use, tailq);
			ctx->entry_ctx_in_use = NULL;

			rc = spdk_nvme_detach_async(ctx->ctrlr, &ctx->detach_ctx);
			if (rc != 0) {
				DISCOVERY_ERRLOG(ctx, "could not detach discovery ctrlr\n");
				ctx->ctrlr = NULL;
			}
		}
	}

	return SPDK_POLLER_BUSY;
}

static void
start_discovery_poller(void *arg)
{
	struct discovery_ctx *ctx = arg;

	TAILQ_INSERT_TAIL(&g_discovery_ctxs, ctx, tailq);
	ctx->poller = SPDK_POLLER_REGISTER(discovery_poller, ctx, 1000 * 1000);
}

int
bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid,
			  const char *base_name,
			  struct spdk_nvme_ctrlr_opts *drv_opts,
			  struct nvme_ctrlr_opts *bdev_opts,
			  uint64_t attach_timeout,
			  bool from_mdns,
			  spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx)
{
	struct discovery_ctx *ctx;
	struct discovery_entry_ctx *discovery_entry_ctx;

	snprintf(trid->subnqn, sizeof(trid->subnqn), "%s", SPDK_NVMF_DISCOVERY_NQN);
	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
		if (strcmp(ctx->name, base_name) == 0) {
			return -EEXIST;
		}

		if (ctx->entry_ctx_in_use != NULL) {
			if (!spdk_nvme_transport_id_compare(trid, &ctx->entry_ctx_in_use->trid)) {
				return -EEXIST;
			}
		}

		TAILQ_FOREACH(discovery_entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
			if (!spdk_nvme_transport_id_compare(trid, &discovery_entry_ctx->trid)) {
				return -EEXIST;
			}
		}
	}

	ctx = calloc(1, sizeof(*ctx));
	if (ctx == NULL) {
		return -ENOMEM;
	}

	ctx->name = strdup(base_name);
	if (ctx->name == NULL) {
		free_discovery_ctx(ctx);
		return -ENOMEM;
	}
	memcpy(&ctx->drv_opts, drv_opts, sizeof(*drv_opts));
	memcpy(&ctx->bdev_opts, bdev_opts, sizeof(*bdev_opts));
	ctx->from_mdns_discovery_service = from_mdns;
	ctx->bdev_opts.from_discovery_service = true;
	ctx->calling_thread = spdk_get_thread();
	ctx->start_cb_fn = cb_fn;
	ctx->cb_ctx = cb_ctx;
	ctx->initializing = true;
	if (ctx->start_cb_fn) {
		/* We can use this when dumping json to denote if this RPC parameter
		 * was specified or not.
		 */
		ctx->wait_for_attach = true;
	}
	if (attach_timeout != 0) {
		ctx->timeout_ticks = spdk_get_ticks() + attach_timeout *
				     spdk_get_ticks_hz() / 1000ull;
	}
	TAILQ_INIT(&ctx->nvm_entry_ctxs);
	TAILQ_INIT(&ctx->discovery_entry_ctxs);
	memcpy(&ctx->trid, trid, sizeof(*trid));
	/* Even if user did not specify hostnqn, we can still strdup("\0"); */
	ctx->hostnqn = strdup(ctx->drv_opts.hostnqn);
	if (ctx->hostnqn == NULL) {
		free_discovery_ctx(ctx);
		return -ENOMEM;
	}
	discovery_entry_ctx = create_discovery_entry_ctx(ctx, trid);
	if (discovery_entry_ctx == NULL) {
		DISCOVERY_ERRLOG(ctx, "could not allocate new entry_ctx\n");
		free_discovery_ctx(ctx);
		return -ENOMEM;
	}

	TAILQ_INSERT_TAIL(&ctx->discovery_entry_ctxs, discovery_entry_ctx, tailq);
	spdk_thread_send_msg(g_bdev_nvme_init_thread, start_discovery_poller, ctx);
	return 0;
}

int
bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn, void *cb_ctx)
{
	struct discovery_ctx *ctx;

	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
		if (strcmp(name, ctx->name) == 0) {
			if (ctx->stop) {
				return -EALREADY;
			}
			/* If we're still starting the discovery service and ->rc is non-zero, we're
			 * going to stop it as soon as we can
			 */
			if (ctx->initializing && ctx->rc != 0) {
				return -EALREADY;
			}
			stop_discovery(ctx, cb_fn, cb_ctx);
			return 0;
		}
	}

	return -ENOENT;
}

static int
bdev_nvme_library_init(void)
{
	g_bdev_nvme_init_thread = spdk_get_thread();

	spdk_io_device_register(&g_nvme_bdev_ctrlrs, bdev_nvme_create_poll_group_cb,
				bdev_nvme_destroy_poll_group_cb,
				sizeof(struct nvme_poll_group),  "nvme_poll_groups");

	return 0;
}

static void
bdev_nvme_fini_destruct_ctrlrs(void)
{
	struct nvme_bdev_ctrlr *nbdev_ctrlr;
	struct nvme_ctrlr *nvme_ctrlr;

	pthread_mutex_lock(&g_bdev_nvme_mutex);
	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
		TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
			pthread_mutex_lock(&nvme_ctrlr->mutex);
			if (nvme_ctrlr->destruct) {
				/* This controller's destruction was already started
				 * before the application started shutting down
				 */
				pthread_mutex_unlock(&nvme_ctrlr->mutex);
				continue;
			}
			nvme_ctrlr->destruct = true;
			pthread_mutex_unlock(&nvme_ctrlr->mutex);

			spdk_thread_send_msg(nvme_ctrlr->thread, _nvme_ctrlr_destruct,
					     nvme_ctrlr);
		}
	}

	g_bdev_nvme_module_finish = true;
	if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) {
		pthread_mutex_unlock(&g_bdev_nvme_mutex);
		spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL);
		spdk_bdev_module_fini_done();
		return;
	}

	pthread_mutex_unlock(&g_bdev_nvme_mutex);
}

static void
check_discovery_fini(void *arg)
{
	if (TAILQ_EMPTY(&g_discovery_ctxs)) {
		bdev_nvme_fini_destruct_ctrlrs();
	}
}

static void
bdev_nvme_library_fini(void)
{
	struct nvme_probe_skip_entry *entry, *entry_tmp;
	struct discovery_ctx *ctx;

	spdk_poller_unregister(&g_hotplug_poller);
	free(g_hotplug_probe_ctx);
	g_hotplug_probe_ctx = NULL;

	TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, entry_tmp) {
		TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq);
		free(entry);
	}

	assert(spdk_get_thread() == g_bdev_nvme_init_thread);
	if (TAILQ_EMPTY(&g_discovery_ctxs)) {
		bdev_nvme_fini_destruct_ctrlrs();
	} else {
		TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
			stop_discovery(ctx, check_discovery_fini, NULL);
		}
	}
}

static void
bdev_nvme_verify_pi_error(struct nvme_bdev_io *bio)
{
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	struct spdk_bdev *bdev = bdev_io->bdev;
	struct spdk_dif_ctx dif_ctx;
	struct spdk_dif_error err_blk = {};
	int rc;
	struct spdk_dif_ctx_init_ext_opts dif_opts;

	dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
	dif_opts.dif_pi_format = SPDK_DIF_PI_FORMAT_16;
	rc = spdk_dif_ctx_init(&dif_ctx,
			       bdev->blocklen, bdev->md_len, bdev->md_interleave,
			       bdev->dif_is_head_of_md, bdev->dif_type,
			       bdev_io->u.bdev.dif_check_flags,
			       bdev_io->u.bdev.offset_blocks, 0, 0, 0, 0, &dif_opts);
	if (rc != 0) {
		SPDK_ERRLOG("Initialization of DIF context failed\n");
		return;
	}

	if (bdev->md_interleave) {
		rc = spdk_dif_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
				     bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
	} else {
		struct iovec md_iov = {
			.iov_base	= bdev_io->u.bdev.md_buf,
			.iov_len	= bdev_io->u.bdev.num_blocks * bdev->md_len,
		};

		rc = spdk_dix_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
				     &md_iov, bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk);
	}

	if (rc != 0) {
		SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
			    err_blk.err_type, err_blk.err_offset);
	} else {
		SPDK_ERRLOG("Hardware reported PI error but SPDK could not find any.\n");
	}
}

static void
bdev_nvme_no_pi_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;

	if (spdk_nvme_cpl_is_success(cpl)) {
		/* Run PI verification for read data buffer. */
		bdev_nvme_verify_pi_error(bio);
	}

	/* Return original completion status */
	bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
}

static void
bdev_nvme_readv_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	int ret;

	if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
		SPDK_ERRLOG("readv completed with PI error (sct=%d, sc=%d)\n",
			    cpl->status.sct, cpl->status.sc);

		/* Save completion status to use after verifying PI error. */
		bio->cpl = *cpl;

		if (spdk_likely(nvme_io_path_is_available(bio->io_path))) {
			/* Read without PI checking to verify PI error. */
			ret = bdev_nvme_no_pi_readv(bio,
						    bdev_io->u.bdev.iovs,
						    bdev_io->u.bdev.iovcnt,
						    bdev_io->u.bdev.md_buf,
						    bdev_io->u.bdev.num_blocks,
						    bdev_io->u.bdev.offset_blocks);
			if (ret == 0) {
				return;
			}
		}
	}

	bdev_nvme_io_complete_nvme_status(bio, cpl);
}

static void
bdev_nvme_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;

	if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) {
		SPDK_ERRLOG("writev completed with PI error (sct=%d, sc=%d)\n",
			    cpl->status.sct, cpl->status.sc);
		/* Run PI verification for write data buffer if PI error is detected. */
		bdev_nvme_verify_pi_error(bio);
	}

	bdev_nvme_io_complete_nvme_status(bio, cpl);
}

static void
bdev_nvme_zone_appendv_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);

	/* spdk_bdev_io_get_append_location() requires that the ALBA is stored in offset_blocks.
	 * Additionally, offset_blocks has to be set before calling bdev_nvme_verify_pi_error().
	 */
	bdev_io->u.bdev.offset_blocks = *(uint64_t *)&cpl->cdw0;

	if (spdk_nvme_cpl_is_pi_error(cpl)) {
		SPDK_ERRLOG("zone append completed with PI error (sct=%d, sc=%d)\n",
			    cpl->status.sct, cpl->status.sc);
		/* Run PI verification for zone append data buffer if PI error is detected. */
		bdev_nvme_verify_pi_error(bio);
	}

	bdev_nvme_io_complete_nvme_status(bio, cpl);
}

static void
bdev_nvme_comparev_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;

	if (spdk_nvme_cpl_is_pi_error(cpl)) {
		SPDK_ERRLOG("comparev completed with PI error (sct=%d, sc=%d)\n",
			    cpl->status.sct, cpl->status.sc);
		/* Run PI verification for compare data buffer if PI error is detected. */
		bdev_nvme_verify_pi_error(bio);
	}

	bdev_nvme_io_complete_nvme_status(bio, cpl);
}

static void
bdev_nvme_comparev_and_writev_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;

	/* Compare operation completion */
	if (!bio->first_fused_completed) {
		/* Save compare result for write callback */
		bio->cpl = *cpl;
		bio->first_fused_completed = true;
		return;
	}

	/* Write operation completion */
	if (spdk_nvme_cpl_is_error(&bio->cpl)) {
		/* If bio->cpl is already an error, it means the compare operation failed.  In that case,
		 * complete the IO with the compare operation's status.
		 */
		if (!spdk_nvme_cpl_is_error(cpl)) {
			SPDK_ERRLOG("Unexpected write success after compare failure.\n");
		}

		bdev_nvme_io_complete_nvme_status(bio, &bio->cpl);
	} else {
		bdev_nvme_io_complete_nvme_status(bio, cpl);
	}
}

static void
bdev_nvme_queued_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;

	bdev_nvme_io_complete_nvme_status(bio, cpl);
}

static int
fill_zone_from_report(struct spdk_bdev_zone_info *info, struct spdk_nvme_zns_zone_desc *desc)
{
	switch (desc->zt) {
	case SPDK_NVME_ZONE_TYPE_SEQWR:
		info->type = SPDK_BDEV_ZONE_TYPE_SEQWR;
		break;
	default:
		SPDK_ERRLOG("Invalid zone type: %#x in zone report\n", desc->zt);
		return -EIO;
	}

	switch (desc->zs) {
	case SPDK_NVME_ZONE_STATE_EMPTY:
		info->state = SPDK_BDEV_ZONE_STATE_EMPTY;
		break;
	case SPDK_NVME_ZONE_STATE_IOPEN:
		info->state = SPDK_BDEV_ZONE_STATE_IMP_OPEN;
		break;
	case SPDK_NVME_ZONE_STATE_EOPEN:
		info->state = SPDK_BDEV_ZONE_STATE_EXP_OPEN;
		break;
	case SPDK_NVME_ZONE_STATE_CLOSED:
		info->state = SPDK_BDEV_ZONE_STATE_CLOSED;
		break;
	case SPDK_NVME_ZONE_STATE_RONLY:
		info->state = SPDK_BDEV_ZONE_STATE_READ_ONLY;
		break;
	case SPDK_NVME_ZONE_STATE_FULL:
		info->state = SPDK_BDEV_ZONE_STATE_FULL;
		break;
	case SPDK_NVME_ZONE_STATE_OFFLINE:
		info->state = SPDK_BDEV_ZONE_STATE_OFFLINE;
		break;
	default:
		SPDK_ERRLOG("Invalid zone state: %#x in zone report\n", desc->zs);
		return -EIO;
	}

	info->zone_id = desc->zslba;
	info->write_pointer = desc->wp;
	info->capacity = desc->zcap;

	return 0;
}

static void
bdev_nvme_get_zone_info_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	uint64_t zone_id = bdev_io->u.zone_mgmt.zone_id;
	uint32_t zones_to_copy = bdev_io->u.zone_mgmt.num_zones;
	struct spdk_bdev_zone_info *info = bdev_io->u.zone_mgmt.buf;
	uint64_t max_zones_per_buf, i;
	uint32_t zone_report_bufsize;
	struct spdk_nvme_ns *ns;
	struct spdk_nvme_qpair *qpair;
	int ret;

	if (spdk_nvme_cpl_is_error(cpl)) {
		goto out_complete_io_nvme_cpl;
	}

	if (spdk_unlikely(!nvme_io_path_is_available(bio->io_path))) {
		ret = -ENXIO;
		goto out_complete_io_ret;
	}

	ns = bio->io_path->nvme_ns->ns;
	qpair = bio->io_path->qpair->qpair;

	zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
	max_zones_per_buf = (zone_report_bufsize - sizeof(*bio->zone_report_buf)) /
			    sizeof(bio->zone_report_buf->descs[0]);

	if (bio->zone_report_buf->nr_zones > max_zones_per_buf) {
		ret = -EINVAL;
		goto out_complete_io_ret;
	}

	if (!bio->zone_report_buf->nr_zones) {
		ret = -EINVAL;
		goto out_complete_io_ret;
	}

	for (i = 0; i < bio->zone_report_buf->nr_zones && bio->handled_zones < zones_to_copy; i++) {
		ret = fill_zone_from_report(&info[bio->handled_zones],
					    &bio->zone_report_buf->descs[i]);
		if (ret) {
			goto out_complete_io_ret;
		}
		bio->handled_zones++;
	}

	if (bio->handled_zones < zones_to_copy) {
		uint64_t zone_size_lba = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
		uint64_t slba = zone_id + (zone_size_lba * bio->handled_zones);

		memset(bio->zone_report_buf, 0, zone_report_bufsize);
		ret = spdk_nvme_zns_report_zones(ns, qpair,
						 bio->zone_report_buf, zone_report_bufsize,
						 slba, SPDK_NVME_ZRA_LIST_ALL, true,
						 bdev_nvme_get_zone_info_done, bio);
		if (!ret) {
			return;
		} else {
			goto out_complete_io_ret;
		}
	}

out_complete_io_nvme_cpl:
	free(bio->zone_report_buf);
	bio->zone_report_buf = NULL;
	bdev_nvme_io_complete_nvme_status(bio, cpl);
	return;

out_complete_io_ret:
	free(bio->zone_report_buf);
	bio->zone_report_buf = NULL;
	bdev_nvme_io_complete(bio, ret);
}

static void
bdev_nvme_zone_management_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;

	bdev_nvme_io_complete_nvme_status(bio, cpl);
}

static void
bdev_nvme_admin_passthru_complete_nvme_status(void *ctx)
{
	struct nvme_bdev_io *bio = ctx;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	const struct spdk_nvme_cpl *cpl = &bio->cpl;

	assert(bdev_nvme_io_type_is_admin(bdev_io->type));

	__bdev_nvme_io_complete(bdev_io, 0, cpl);
}

static void
bdev_nvme_abort_complete(void *ctx)
{
	struct nvme_bdev_io *bio = ctx;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);

	if (spdk_nvme_cpl_is_abort_success(&bio->cpl)) {
		__bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS, NULL);
	} else {
		__bdev_nvme_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED, NULL);
	}
}

static void
bdev_nvme_abort_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);

	bio->cpl = *cpl;
	spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io), bdev_nvme_abort_complete, bio);
}

static void
bdev_nvme_admin_passthru_done(void *ref, const struct spdk_nvme_cpl *cpl)
{
	struct nvme_bdev_io *bio = ref;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);

	bio->cpl = *cpl;
	spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
			     bdev_nvme_admin_passthru_complete_nvme_status, bio);
}

static void
bdev_nvme_queued_reset_sgl(void *ref, uint32_t sgl_offset)
{
	struct nvme_bdev_io *bio = ref;
	struct iovec *iov;

	bio->iov_offset = sgl_offset;
	for (bio->iovpos = 0; bio->iovpos < bio->iovcnt; bio->iovpos++) {
		iov = &bio->iovs[bio->iovpos];
		if (bio->iov_offset < iov->iov_len) {
			break;
		}

		bio->iov_offset -= iov->iov_len;
	}
}

static int
bdev_nvme_queued_next_sge(void *ref, void **address, uint32_t *length)
{
	struct nvme_bdev_io *bio = ref;
	struct iovec *iov;

	assert(bio->iovpos < bio->iovcnt);

	iov = &bio->iovs[bio->iovpos];

	*address = iov->iov_base;
	*length = iov->iov_len;

	if (bio->iov_offset) {
		assert(bio->iov_offset <= iov->iov_len);
		*address += bio->iov_offset;
		*length -= bio->iov_offset;
	}

	bio->iov_offset += *length;
	if (bio->iov_offset == iov->iov_len) {
		bio->iovpos++;
		bio->iov_offset = 0;
	}

	return 0;
}

static void
bdev_nvme_queued_reset_fused_sgl(void *ref, uint32_t sgl_offset)
{
	struct nvme_bdev_io *bio = ref;
	struct iovec *iov;

	bio->fused_iov_offset = sgl_offset;
	for (bio->fused_iovpos = 0; bio->fused_iovpos < bio->fused_iovcnt; bio->fused_iovpos++) {
		iov = &bio->fused_iovs[bio->fused_iovpos];
		if (bio->fused_iov_offset < iov->iov_len) {
			break;
		}

		bio->fused_iov_offset -= iov->iov_len;
	}
}

static int
bdev_nvme_queued_next_fused_sge(void *ref, void **address, uint32_t *length)
{
	struct nvme_bdev_io *bio = ref;
	struct iovec *iov;

	assert(bio->fused_iovpos < bio->fused_iovcnt);

	iov = &bio->fused_iovs[bio->fused_iovpos];

	*address = iov->iov_base;
	*length = iov->iov_len;

	if (bio->fused_iov_offset) {
		assert(bio->fused_iov_offset <= iov->iov_len);
		*address += bio->fused_iov_offset;
		*length -= bio->fused_iov_offset;
	}

	bio->fused_iov_offset += *length;
	if (bio->fused_iov_offset == iov->iov_len) {
		bio->fused_iovpos++;
		bio->fused_iov_offset = 0;
	}

	return 0;
}

static int
bdev_nvme_no_pi_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
		      void *md, uint64_t lba_count, uint64_t lba)
{
	int rc;

	SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 " without PI check\n",
		      lba_count, lba);

	bio->iovs = iov;
	bio->iovcnt = iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;

	rc = spdk_nvme_ns_cmd_readv_with_md(bio->io_path->nvme_ns->ns,
					    bio->io_path->qpair->qpair,
					    lba, lba_count,
					    bdev_nvme_no_pi_readv_done, bio, 0,
					    bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
					    md, 0, 0);

	if (rc != 0 && rc != -ENOMEM) {
		SPDK_ERRLOG("no_pi_readv failed: rc = %d\n", rc);
	}
	return rc;
}

static int
bdev_nvme_readv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
		void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
		struct spdk_memory_domain *domain, void *domain_ctx,
		struct spdk_accel_sequence *seq)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	int rc;

	SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 "\n",
		      lba_count, lba);

	bio->iovs = iov;
	bio->iovcnt = iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;

	if (domain != NULL || seq != NULL) {
		bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
		bio->ext_opts.memory_domain = domain;
		bio->ext_opts.memory_domain_ctx = domain_ctx;
		bio->ext_opts.io_flags = flags;
		bio->ext_opts.metadata = md;
		bio->ext_opts.accel_sequence = seq;

		if (iovcnt == 1) {
			rc = spdk_nvme_ns_cmd_read_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_readv_done,
						       bio, &bio->ext_opts);
		} else {
			rc = spdk_nvme_ns_cmd_readv_ext(ns, qpair, lba, lba_count,
							bdev_nvme_readv_done, bio,
							bdev_nvme_queued_reset_sgl,
							bdev_nvme_queued_next_sge,
							&bio->ext_opts);
		}
	} else if (iovcnt == 1) {
		rc = spdk_nvme_ns_cmd_read_with_md(ns, qpair, iov[0].iov_base,
						   md, lba, lba_count, bdev_nvme_readv_done,
						   bio, flags, 0, 0);
	} else {
		rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count,
						    bdev_nvme_readv_done, bio, flags,
						    bdev_nvme_queued_reset_sgl,
						    bdev_nvme_queued_next_sge, md, 0, 0);
	}

	if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
		SPDK_ERRLOG("readv failed: rc = %d\n", rc);
	}
	return rc;
}

static int
bdev_nvme_writev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
		 void *md, uint64_t lba_count, uint64_t lba, uint32_t flags,
		 struct spdk_memory_domain *domain, void *domain_ctx,
		 struct spdk_accel_sequence *seq)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	int rc;

	SPDK_DEBUGLOG(bdev_nvme, "write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
		      lba_count, lba);

	bio->iovs = iov;
	bio->iovcnt = iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;

	if (domain != NULL || seq != NULL) {
		bio->ext_opts.size = SPDK_SIZEOF(&bio->ext_opts, accel_sequence);
		bio->ext_opts.memory_domain = domain;
		bio->ext_opts.memory_domain_ctx = domain_ctx;
		bio->ext_opts.io_flags = flags;
		bio->ext_opts.metadata = md;
		bio->ext_opts.accel_sequence = seq;

		if (iovcnt == 1) {
			rc = spdk_nvme_ns_cmd_write_ext(ns, qpair, iov[0].iov_base, lba, lba_count, bdev_nvme_writev_done,
							bio, &bio->ext_opts);
		} else {
			rc = spdk_nvme_ns_cmd_writev_ext(ns, qpair, lba, lba_count,
							 bdev_nvme_writev_done, bio,
							 bdev_nvme_queued_reset_sgl,
							 bdev_nvme_queued_next_sge,
							 &bio->ext_opts);
		}
	} else if (iovcnt == 1) {
		rc = spdk_nvme_ns_cmd_write_with_md(ns, qpair, iov[0].iov_base,
						    md, lba, lba_count, bdev_nvme_writev_done,
						    bio, flags, 0, 0);
	} else {
		rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
						     bdev_nvme_writev_done, bio, flags,
						     bdev_nvme_queued_reset_sgl,
						     bdev_nvme_queued_next_sge, md, 0, 0);
	}

	if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
		SPDK_ERRLOG("writev failed: rc = %d\n", rc);
	}
	return rc;
}

static int
bdev_nvme_zone_appendv(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
		       void *md, uint64_t lba_count, uint64_t zslba,
		       uint32_t flags)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	int rc;

	SPDK_DEBUGLOG(bdev_nvme, "zone append %" PRIu64 " blocks to zone start lba %#" PRIx64 "\n",
		      lba_count, zslba);

	bio->iovs = iov;
	bio->iovcnt = iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;

	if (iovcnt == 1) {
		rc = spdk_nvme_zns_zone_append_with_md(ns, qpair, iov[0].iov_base, md, zslba,
						       lba_count,
						       bdev_nvme_zone_appendv_done, bio,
						       flags,
						       0, 0);
	} else {
		rc = spdk_nvme_zns_zone_appendv_with_md(ns, qpair, zslba, lba_count,
							bdev_nvme_zone_appendv_done, bio, flags,
							bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
							md, 0, 0);
	}

	if (rc != 0 && rc != -ENOMEM) {
		SPDK_ERRLOG("zone append failed: rc = %d\n", rc);
	}
	return rc;
}

static int
bdev_nvme_comparev(struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt,
		   void *md, uint64_t lba_count, uint64_t lba,
		   uint32_t flags)
{
	int rc;

	SPDK_DEBUGLOG(bdev_nvme, "compare %" PRIu64 " blocks with offset %#" PRIx64 "\n",
		      lba_count, lba);

	bio->iovs = iov;
	bio->iovcnt = iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;

	rc = spdk_nvme_ns_cmd_comparev_with_md(bio->io_path->nvme_ns->ns,
					       bio->io_path->qpair->qpair,
					       lba, lba_count,
					       bdev_nvme_comparev_done, bio, flags,
					       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge,
					       md, 0, 0);

	if (rc != 0 && rc != -ENOMEM) {
		SPDK_ERRLOG("comparev failed: rc = %d\n", rc);
	}
	return rc;
}

static int
bdev_nvme_comparev_and_writev(struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt,
			      struct iovec *write_iov, int write_iovcnt,
			      void *md, uint64_t lba_count, uint64_t lba, uint32_t flags)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio);
	int rc;

	SPDK_DEBUGLOG(bdev_nvme, "compare and write %" PRIu64 " blocks with offset %#" PRIx64 "\n",
		      lba_count, lba);

	bio->iovs = cmp_iov;
	bio->iovcnt = cmp_iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;
	bio->fused_iovs = write_iov;
	bio->fused_iovcnt = write_iovcnt;
	bio->fused_iovpos = 0;
	bio->fused_iov_offset = 0;

	if (bdev_io->num_retries == 0) {
		bio->first_fused_submitted = false;
		bio->first_fused_completed = false;
	}

	if (!bio->first_fused_submitted) {
		flags |= SPDK_NVME_IO_FLAGS_FUSE_FIRST;
		memset(&bio->cpl, 0, sizeof(bio->cpl));

		rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count,
						       bdev_nvme_comparev_and_writev_done, bio, flags,
						       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, md, 0, 0);
		if (rc == 0) {
			bio->first_fused_submitted = true;
			flags &= ~SPDK_NVME_IO_FLAGS_FUSE_FIRST;
		} else {
			if (rc != -ENOMEM) {
				SPDK_ERRLOG("compare failed: rc = %d\n", rc);
			}
			return rc;
		}
	}

	flags |= SPDK_NVME_IO_FLAGS_FUSE_SECOND;

	rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count,
					     bdev_nvme_comparev_and_writev_done, bio, flags,
					     bdev_nvme_queued_reset_fused_sgl, bdev_nvme_queued_next_fused_sge, md, 0, 0);
	if (rc != 0 && rc != -ENOMEM) {
		SPDK_ERRLOG("write failed: rc = %d\n", rc);
		rc = 0;
	}

	return rc;
}

static int
bdev_nvme_unmap(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
{
	struct spdk_nvme_dsm_range dsm_ranges[SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES];
	struct spdk_nvme_dsm_range *range;
	uint64_t offset, remaining;
	uint64_t num_ranges_u64;
	uint16_t num_ranges;
	int rc;

	num_ranges_u64 = (num_blocks + SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS - 1) /
			 SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
	if (num_ranges_u64 > SPDK_COUNTOF(dsm_ranges)) {
		SPDK_ERRLOG("Unmap request for %" PRIu64 " blocks is too large\n", num_blocks);
		return -EINVAL;
	}
	num_ranges = (uint16_t)num_ranges_u64;

	offset = offset_blocks;
	remaining = num_blocks;
	range = &dsm_ranges[0];

	/* Fill max-size ranges until the remaining blocks fit into one range */
	while (remaining > SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS) {
		range->attributes.raw = 0;
		range->length = SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
		range->starting_lba = offset;

		offset += SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
		remaining -= SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS;
		range++;
	}

	/* Final range describes the remaining blocks */
	range->attributes.raw = 0;
	range->length = remaining;
	range->starting_lba = offset;

	rc = spdk_nvme_ns_cmd_dataset_management(bio->io_path->nvme_ns->ns,
			bio->io_path->qpair->qpair,
			SPDK_NVME_DSM_ATTR_DEALLOCATE,
			dsm_ranges, num_ranges,
			bdev_nvme_queued_done, bio);

	return rc;
}

static int
bdev_nvme_write_zeroes(struct nvme_bdev_io *bio, uint64_t offset_blocks, uint64_t num_blocks)
{
	if (num_blocks > UINT16_MAX + 1) {
		SPDK_ERRLOG("NVMe write zeroes is limited to 16-bit block count\n");
		return -EINVAL;
	}

	return spdk_nvme_ns_cmd_write_zeroes(bio->io_path->nvme_ns->ns,
					     bio->io_path->qpair->qpair,
					     offset_blocks, num_blocks,
					     bdev_nvme_queued_done, bio,
					     0);
}

static int
bdev_nvme_get_zone_info(struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones,
			struct spdk_bdev_zone_info *info)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	uint32_t zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns);
	uint64_t zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns);
	uint64_t total_zones = spdk_nvme_zns_ns_get_num_zones(ns);

	if (zone_id % zone_size != 0) {
		return -EINVAL;
	}

	if (num_zones > total_zones || !num_zones) {
		return -EINVAL;
	}

	assert(!bio->zone_report_buf);
	bio->zone_report_buf = calloc(1, zone_report_bufsize);
	if (!bio->zone_report_buf) {
		return -ENOMEM;
	}

	bio->handled_zones = 0;

	return spdk_nvme_zns_report_zones(ns, qpair, bio->zone_report_buf, zone_report_bufsize,
					  zone_id, SPDK_NVME_ZRA_LIST_ALL, true,
					  bdev_nvme_get_zone_info_done, bio);
}

static int
bdev_nvme_zone_management(struct nvme_bdev_io *bio, uint64_t zone_id,
			  enum spdk_bdev_zone_action action)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;

	switch (action) {
	case SPDK_BDEV_ZONE_CLOSE:
		return spdk_nvme_zns_close_zone(ns, qpair, zone_id, false,
						bdev_nvme_zone_management_done, bio);
	case SPDK_BDEV_ZONE_FINISH:
		return spdk_nvme_zns_finish_zone(ns, qpair, zone_id, false,
						 bdev_nvme_zone_management_done, bio);
	case SPDK_BDEV_ZONE_OPEN:
		return spdk_nvme_zns_open_zone(ns, qpair, zone_id, false,
					       bdev_nvme_zone_management_done, bio);
	case SPDK_BDEV_ZONE_RESET:
		return spdk_nvme_zns_reset_zone(ns, qpair, zone_id, false,
						bdev_nvme_zone_management_done, bio);
	case SPDK_BDEV_ZONE_OFFLINE:
		return spdk_nvme_zns_offline_zone(ns, qpair, zone_id, false,
						  bdev_nvme_zone_management_done, bio);
	default:
		return -EINVAL;
	}
}

static void
bdev_nvme_admin_passthru(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
			 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes)
{
	struct nvme_io_path *io_path;
	struct nvme_ctrlr *nvme_ctrlr;
	uint32_t max_xfer_size;
	int rc = -ENXIO;

	/* Choose the first ctrlr which is not failed. */
	STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
		nvme_ctrlr = io_path->qpair->ctrlr;

		/* We should skip any unavailable nvme_ctrlr rather than checking
		 * if the return value of spdk_nvme_ctrlr_cmd_admin_raw() is -ENXIO.
		 */
		if (!nvme_ctrlr_is_available(nvme_ctrlr)) {
			continue;
		}

		max_xfer_size = spdk_nvme_ctrlr_get_max_xfer_size(nvme_ctrlr->ctrlr);

		if (nbytes > max_xfer_size) {
			SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
			rc = -EINVAL;
			goto err;
		}

		rc = spdk_nvme_ctrlr_cmd_admin_raw(nvme_ctrlr->ctrlr, cmd, buf, (uint32_t)nbytes,
						   bdev_nvme_admin_passthru_done, bio);
		if (rc == 0) {
			return;
		}
	}

err:
	bdev_nvme_admin_complete(bio, rc);
}

static int
bdev_nvme_io_passthru(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
		      void *buf, size_t nbytes)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);

	if (nbytes > max_xfer_size) {
		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
		return -EINVAL;
	}

	/*
	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
	 * so fill it out automatically.
	 */
	cmd->nsid = spdk_nvme_ns_get_id(ns);

	return spdk_nvme_ctrlr_cmd_io_raw(ctrlr, qpair, cmd, buf,
					  (uint32_t)nbytes, bdev_nvme_queued_done, bio);
}

static int
bdev_nvme_io_passthru_md(struct nvme_bdev_io *bio, struct spdk_nvme_cmd *cmd,
			 void *buf, size_t nbytes, void *md_buf, size_t md_len)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);

	if (nbytes > max_xfer_size) {
		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
		return -EINVAL;
	}

	if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
		SPDK_ERRLOG("invalid meta data buffer size\n");
		return -EINVAL;
	}

	/*
	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid,
	 * so fill it out automatically.
	 */
	cmd->nsid = spdk_nvme_ns_get_id(ns);

	return spdk_nvme_ctrlr_cmd_io_raw_with_md(ctrlr, qpair, cmd, buf,
			(uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio);
}

static int
bdev_nvme_iov_passthru_md(struct nvme_bdev_io *bio,
			  struct spdk_nvme_cmd *cmd, struct iovec *iov, int iovcnt,
			  size_t nbytes, void *md_buf, size_t md_len)
{
	struct spdk_nvme_ns *ns = bio->io_path->nvme_ns->ns;
	struct spdk_nvme_qpair *qpair = bio->io_path->qpair->qpair;
	size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns);
	uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns);
	struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns);

	bio->iovs = iov;
	bio->iovcnt = iovcnt;
	bio->iovpos = 0;
	bio->iov_offset = 0;

	if (nbytes > max_xfer_size) {
		SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size);
		return -EINVAL;
	}

	if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) {
		SPDK_ERRLOG("invalid meta data buffer size\n");
		return -EINVAL;
	}

	/*
	 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands
	 * require a nsid, so fill it out automatically.
	 */
	cmd->nsid = spdk_nvme_ns_get_id(ns);

	return spdk_nvme_ctrlr_cmd_iov_raw_with_md(
		       ctrlr, qpair, cmd, (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio,
		       bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge);
}

static void
bdev_nvme_abort(struct nvme_bdev_channel *nbdev_ch, struct nvme_bdev_io *bio,
		struct nvme_bdev_io *bio_to_abort)
{
	struct nvme_io_path *io_path;
	int rc = 0;

	rc = bdev_nvme_abort_retry_io(nbdev_ch, bio_to_abort);
	if (rc == 0) {
		bdev_nvme_admin_complete(bio, 0);
		return;
	}

	io_path = bio_to_abort->io_path;
	if (io_path != NULL) {
		rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
						   io_path->qpair->qpair,
						   bio_to_abort,
						   bdev_nvme_abort_done, bio);
	} else {
		STAILQ_FOREACH(io_path, &nbdev_ch->io_path_list, stailq) {
			rc = spdk_nvme_ctrlr_cmd_abort_ext(io_path->qpair->ctrlr->ctrlr,
							   NULL,
							   bio_to_abort,
							   bdev_nvme_abort_done, bio);

			if (rc != -ENOENT) {
				break;
			}
		}
	}

	if (rc != 0) {
		/* If no command was found or there was any error, complete the abort
		 * request with failure.
		 */
		bdev_nvme_admin_complete(bio, rc);
	}
}

static int
bdev_nvme_copy(struct nvme_bdev_io *bio, uint64_t dst_offset_blocks, uint64_t src_offset_blocks,
	       uint64_t num_blocks)
{
	struct spdk_nvme_scc_source_range range = {
		.slba = src_offset_blocks,
		.nlb = num_blocks - 1
	};

	return spdk_nvme_ns_cmd_copy(bio->io_path->nvme_ns->ns,
				     bio->io_path->qpair->qpair,
				     &range, 1, dst_offset_blocks,
				     bdev_nvme_queued_done, bio);
}

static void
bdev_nvme_opts_config_json(struct spdk_json_write_ctx *w)
{
	const char *action;
	uint32_t i;

	if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET) {
		action = "reset";
	} else if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT) {
		action = "abort";
	} else {
		action = "none";
	}

	spdk_json_write_object_begin(w);

	spdk_json_write_named_string(w, "method", "bdev_nvme_set_options");

	spdk_json_write_named_object_begin(w, "params");
	spdk_json_write_named_string(w, "action_on_timeout", action);
	spdk_json_write_named_uint64(w, "timeout_us", g_opts.timeout_us);
	spdk_json_write_named_uint64(w, "timeout_admin_us", g_opts.timeout_admin_us);
	spdk_json_write_named_uint32(w, "keep_alive_timeout_ms", g_opts.keep_alive_timeout_ms);
	spdk_json_write_named_uint32(w, "arbitration_burst", g_opts.arbitration_burst);
	spdk_json_write_named_uint32(w, "low_priority_weight", g_opts.low_priority_weight);
	spdk_json_write_named_uint32(w, "medium_priority_weight", g_opts.medium_priority_weight);
	spdk_json_write_named_uint32(w, "high_priority_weight", g_opts.high_priority_weight);
	spdk_json_write_named_uint64(w, "nvme_adminq_poll_period_us", g_opts.nvme_adminq_poll_period_us);
	spdk_json_write_named_uint64(w, "nvme_ioq_poll_period_us", g_opts.nvme_ioq_poll_period_us);
	spdk_json_write_named_uint32(w, "io_queue_requests", g_opts.io_queue_requests);
	spdk_json_write_named_bool(w, "delay_cmd_submit", g_opts.delay_cmd_submit);
	spdk_json_write_named_uint32(w, "transport_retry_count", g_opts.transport_retry_count);
	spdk_json_write_named_int32(w, "bdev_retry_count", g_opts.bdev_retry_count);
	spdk_json_write_named_uint8(w, "transport_ack_timeout", g_opts.transport_ack_timeout);
	spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", g_opts.ctrlr_loss_timeout_sec);
	spdk_json_write_named_uint32(w, "reconnect_delay_sec", g_opts.reconnect_delay_sec);
	spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec", g_opts.fast_io_fail_timeout_sec);
	spdk_json_write_named_bool(w, "disable_auto_failback", g_opts.disable_auto_failback);
	spdk_json_write_named_bool(w, "generate_uuids", g_opts.generate_uuids);
	spdk_json_write_named_uint8(w, "transport_tos", g_opts.transport_tos);
	spdk_json_write_named_bool(w, "nvme_error_stat", g_opts.nvme_error_stat);
	spdk_json_write_named_uint32(w, "rdma_srq_size", g_opts.rdma_srq_size);
	spdk_json_write_named_bool(w, "io_path_stat", g_opts.io_path_stat);
	spdk_json_write_named_bool(w, "allow_accel_sequence", g_opts.allow_accel_sequence);
	spdk_json_write_named_uint32(w, "rdma_max_cq_size", g_opts.rdma_max_cq_size);
	spdk_json_write_named_uint16(w, "rdma_cm_event_timeout_ms", g_opts.rdma_cm_event_timeout_ms);
	spdk_json_write_named_array_begin(w, "dhchap_digests");
	for (i = 0; i < 32; ++i) {
		if (g_opts.dhchap_digests & SPDK_BIT(i)) {
			spdk_json_write_string(w, spdk_nvme_dhchap_get_digest_name(i));
		}
	}
	spdk_json_write_array_end(w);
	spdk_json_write_named_array_begin(w, "dhchap_dhgroups");
	for (i = 0; i < 32; ++i) {
		if (g_opts.dhchap_dhgroups & SPDK_BIT(i)) {
			spdk_json_write_string(w, spdk_nvme_dhchap_get_dhgroup_name(i));
		}
	}

	spdk_json_write_array_end(w);
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

static void
bdev_nvme_discovery_config_json(struct spdk_json_write_ctx *w, struct discovery_ctx *ctx)
{
	struct spdk_nvme_transport_id trid;

	spdk_json_write_object_begin(w);

	spdk_json_write_named_string(w, "method", "bdev_nvme_start_discovery");

	spdk_json_write_named_object_begin(w, "params");
	spdk_json_write_named_string(w, "name", ctx->name);
	spdk_json_write_named_string(w, "hostnqn", ctx->hostnqn);

	trid = ctx->trid;
	memset(trid.subnqn, 0, sizeof(trid.subnqn));
	nvme_bdev_dump_trid_json(&trid, w);

	spdk_json_write_named_bool(w, "wait_for_attach", ctx->wait_for_attach);
	spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", ctx->bdev_opts.ctrlr_loss_timeout_sec);
	spdk_json_write_named_uint32(w, "reconnect_delay_sec", ctx->bdev_opts.reconnect_delay_sec);
	spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
				     ctx->bdev_opts.fast_io_fail_timeout_sec);
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

#ifdef SPDK_CONFIG_NVME_CUSE
static void
nvme_ctrlr_cuse_config_json(struct spdk_json_write_ctx *w,
			    struct nvme_ctrlr *nvme_ctrlr)
{
	size_t cuse_name_size = 128;
	char cuse_name[cuse_name_size];

	if (spdk_nvme_cuse_get_ctrlr_name(nvme_ctrlr->ctrlr,
					  cuse_name, &cuse_name_size) != 0) {
		return;
	}

	spdk_json_write_object_begin(w);

	spdk_json_write_named_string(w, "method", "bdev_nvme_cuse_register");

	spdk_json_write_named_object_begin(w, "params");
	spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}
#endif

static void
nvme_ctrlr_config_json(struct spdk_json_write_ctx *w,
		       struct nvme_ctrlr *nvme_ctrlr)
{
	struct spdk_nvme_transport_id	*trid;
	const struct spdk_nvme_ctrlr_opts *opts;

	if (nvme_ctrlr->opts.from_discovery_service) {
		/* Do not emit an RPC for this - it will be implicitly
		 * covered by a separate bdev_nvme_start_discovery or
		 * bdev_nvme_start_mdns_discovery RPC.
		 */
		return;
	}

	trid = &nvme_ctrlr->active_path_id->trid;

	spdk_json_write_object_begin(w);

	spdk_json_write_named_string(w, "method", "bdev_nvme_attach_controller");

	spdk_json_write_named_object_begin(w, "params");
	spdk_json_write_named_string(w, "name", nvme_ctrlr->nbdev_ctrlr->name);
	nvme_bdev_dump_trid_json(trid, w);
	spdk_json_write_named_bool(w, "prchk_reftag",
				   (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_REFTAG) != 0);
	spdk_json_write_named_bool(w, "prchk_guard",
				   (nvme_ctrlr->opts.prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_GUARD) != 0);
	spdk_json_write_named_int32(w, "ctrlr_loss_timeout_sec", nvme_ctrlr->opts.ctrlr_loss_timeout_sec);
	spdk_json_write_named_uint32(w, "reconnect_delay_sec", nvme_ctrlr->opts.reconnect_delay_sec);
	spdk_json_write_named_uint32(w, "fast_io_fail_timeout_sec",
				     nvme_ctrlr->opts.fast_io_fail_timeout_sec);
	if (nvme_ctrlr->psk != NULL) {
		spdk_json_write_named_string(w, "psk", spdk_key_get_name(nvme_ctrlr->psk));
	} else if (nvme_ctrlr->opts.psk[0] != '\0') {
		spdk_json_write_named_string(w, "psk", nvme_ctrlr->opts.psk);
	}

	opts = spdk_nvme_ctrlr_get_opts(nvme_ctrlr->ctrlr);
	spdk_json_write_named_string(w, "hostnqn", opts->hostnqn);
	spdk_json_write_named_bool(w, "hdgst", opts->header_digest);
	spdk_json_write_named_bool(w, "ddgst", opts->data_digest);
	if (opts->src_addr[0] != '\0') {
		spdk_json_write_named_string(w, "hostaddr", opts->src_addr);
	}
	if (opts->src_svcid[0] != '\0') {
		spdk_json_write_named_string(w, "hostsvcid", opts->src_svcid);
	}

	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

static void
bdev_nvme_hotplug_config_json(struct spdk_json_write_ctx *w)
{
	spdk_json_write_object_begin(w);
	spdk_json_write_named_string(w, "method", "bdev_nvme_set_hotplug");

	spdk_json_write_named_object_begin(w, "params");
	spdk_json_write_named_uint64(w, "period_us", g_nvme_hotplug_poll_period_us);
	spdk_json_write_named_bool(w, "enable", g_nvme_hotplug_enabled);
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

static int
bdev_nvme_config_json(struct spdk_json_write_ctx *w)
{
	struct nvme_bdev_ctrlr	*nbdev_ctrlr;
	struct nvme_ctrlr	*nvme_ctrlr;
	struct discovery_ctx	*ctx;

	bdev_nvme_opts_config_json(w);

	pthread_mutex_lock(&g_bdev_nvme_mutex);

	TAILQ_FOREACH(nbdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) {
		TAILQ_FOREACH(nvme_ctrlr, &nbdev_ctrlr->ctrlrs, tailq) {
			nvme_ctrlr_config_json(w, nvme_ctrlr);

#ifdef SPDK_CONFIG_NVME_CUSE
			nvme_ctrlr_cuse_config_json(w, nvme_ctrlr);
#endif
		}
	}

	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
		if (!ctx->from_mdns_discovery_service) {
			bdev_nvme_discovery_config_json(w, ctx);
		}
	}

	bdev_nvme_mdns_discovery_config_json(w);

	/* Dump as last parameter to give all NVMe bdevs chance to be constructed
	 * before enabling hotplug poller.
	 */
	bdev_nvme_hotplug_config_json(w);

	pthread_mutex_unlock(&g_bdev_nvme_mutex);
	return 0;
}

struct spdk_nvme_ctrlr *
bdev_nvme_get_ctrlr(struct spdk_bdev *bdev)
{
	struct nvme_bdev *nbdev;
	struct nvme_ns *nvme_ns;

	if (!bdev || bdev->module != &nvme_if) {
		return NULL;
	}

	nbdev = SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk);
	nvme_ns = TAILQ_FIRST(&nbdev->nvme_ns_list);
	assert(nvme_ns != NULL);

	return nvme_ns->ctrlr->ctrlr;
}

void
nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path)
{
	struct nvme_ns *nvme_ns = io_path->nvme_ns;
	struct nvme_ctrlr *nvme_ctrlr = io_path->qpair->ctrlr;
	const struct spdk_nvme_ctrlr_data *cdata;
	const struct spdk_nvme_transport_id *trid;
	const char *adrfam_str;

	spdk_json_write_object_begin(w);

	spdk_json_write_named_string(w, "bdev_name", nvme_ns->bdev->disk.name);

	cdata = spdk_nvme_ctrlr_get_data(nvme_ctrlr->ctrlr);
	trid = spdk_nvme_ctrlr_get_transport_id(nvme_ctrlr->ctrlr);

	spdk_json_write_named_uint32(w, "cntlid", cdata->cntlid);
	spdk_json_write_named_bool(w, "current", io_path->nbdev_ch != NULL &&
				   io_path == io_path->nbdev_ch->current_io_path);
	spdk_json_write_named_bool(w, "connected", nvme_qpair_is_connected(io_path->qpair));
	spdk_json_write_named_bool(w, "accessible", nvme_ns_is_accessible(nvme_ns));

	spdk_json_write_named_object_begin(w, "transport");
	spdk_json_write_named_string(w, "trtype", trid->trstring);
	spdk_json_write_named_string(w, "traddr", trid->traddr);
	if (trid->trsvcid[0] != '\0') {
		spdk_json_write_named_string(w, "trsvcid", trid->trsvcid);
	}
	adrfam_str = spdk_nvme_transport_id_adrfam_str(trid->adrfam);
	if (adrfam_str) {
		spdk_json_write_named_string(w, "adrfam", adrfam_str);
	}
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

void
bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w)
{
	struct discovery_ctx *ctx;
	struct discovery_entry_ctx *entry_ctx;

	spdk_json_write_array_begin(w);
	TAILQ_FOREACH(ctx, &g_discovery_ctxs, tailq) {
		spdk_json_write_object_begin(w);
		spdk_json_write_named_string(w, "name", ctx->name);

		spdk_json_write_named_object_begin(w, "trid");
		nvme_bdev_dump_trid_json(&ctx->trid, w);
		spdk_json_write_object_end(w);

		spdk_json_write_named_array_begin(w, "referrals");
		TAILQ_FOREACH(entry_ctx, &ctx->discovery_entry_ctxs, tailq) {
			spdk_json_write_object_begin(w);
			spdk_json_write_named_object_begin(w, "trid");
			nvme_bdev_dump_trid_json(&entry_ctx->trid, w);
			spdk_json_write_object_end(w);
			spdk_json_write_object_end(w);
		}
		spdk_json_write_array_end(w);

		spdk_json_write_object_end(w);
	}
	spdk_json_write_array_end(w);
}

SPDK_LOG_REGISTER_COMPONENT(bdev_nvme)

SPDK_TRACE_REGISTER_FN(bdev_nvme_trace, "bdev_nvme", TRACE_GROUP_BDEV_NVME)
{
	struct spdk_trace_tpoint_opts opts[] = {
		{
			"BDEV_NVME_IO_START", TRACE_BDEV_NVME_IO_START,
			OWNER_NONE, OBJECT_BDEV_NVME_IO, 1,
			{{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
		},
		{
			"BDEV_NVME_IO_DONE", TRACE_BDEV_NVME_IO_DONE,
			OWNER_NONE, OBJECT_BDEV_NVME_IO, 0,
			{{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
		}
	};


	spdk_trace_register_object(OBJECT_BDEV_NVME_IO, 'N');
	spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
	spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
	spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_SUBMIT, OBJECT_BDEV_NVME_IO, 0);
	spdk_trace_tpoint_register_relation(TRACE_NVME_PCIE_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
	spdk_trace_tpoint_register_relation(TRACE_NVME_TCP_COMPLETE, OBJECT_BDEV_NVME_IO, 0);
}