xref: /spdk/lib/bdev/bdev.c (revision 42dba6047b054d9db99f2a78f1b7100a9b3162a1)

/*-
 *   BSD LICENSE
 *
 *   Copyright (c) Intel Corporation.
 *   All rights reserved.
 *
 *   Redistribution and use in source and binary forms, with or without
 *   modification, are permitted provided that the following conditions
 *   are met:
 *
 *     * Redistributions of source code must retain the above copyright
 *       notice, this list of conditions and the following disclaimer.
 *     * Redistributions in binary form must reproduce the above copyright
 *       notice, this list of conditions and the following disclaimer in
 *       the documentation and/or other materials provided with the
 *       distribution.
 *     * Neither the name of Intel Corporation nor the names of its
 *       contributors may be used to endorse or promote products derived
 *       from this software without specific prior written permission.
 *
 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include "spdk/stdinc.h"

#include "spdk/bdev.h"
#include "spdk/conf.h"

#include "spdk/config.h"
#include "spdk/env.h"
#include "spdk/event.h"
#include "spdk/thread.h"
#include "spdk/likely.h"
#include "spdk/queue.h"
#include "spdk/nvme_spec.h"
#include "spdk/scsi_spec.h"
#include "spdk/util.h"
#include "spdk/trace.h"

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

#ifdef SPDK_CONFIG_VTUNE
#include "ittnotify.h"
#include "ittnotify_types.h"
int __itt_init_ittlib(const char *, __itt_group_id);
#endif

#define SPDK_BDEV_IO_POOL_SIZE			(64 * 1024)
#define SPDK_BDEV_IO_CACHE_SIZE			256
#define BUF_SMALL_POOL_SIZE			8192
#define BUF_LARGE_POOL_SIZE			1024
#define NOMEM_THRESHOLD_COUNT			8
#define ZERO_BUFFER_SIZE			0x100000

#define OWNER_BDEV		0x2

#define OBJECT_BDEV_IO		0x2

#define TRACE_GROUP_BDEV	0x3
#define TRACE_BDEV_IO_START	SPDK_TPOINT_ID(TRACE_GROUP_BDEV, 0x0)
#define TRACE_BDEV_IO_DONE	SPDK_TPOINT_ID(TRACE_GROUP_BDEV, 0x1)

#define SPDK_BDEV_QOS_TIMESLICE_IN_USEC		1000
#define SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE	1
#define SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE	512
#define SPDK_BDEV_QOS_MIN_IOS_PER_SEC		10000
#define SPDK_BDEV_QOS_MIN_BYTES_PER_SEC		(10 * 1024 * 1024)
#define SPDK_BDEV_QOS_LIMIT_NOT_DEFINED		UINT64_MAX

#define SPDK_BDEV_POOL_ALIGNMENT 512

static const char *qos_conf_type[] = {"Limit_IOPS", "Limit_BPS"};
static const char *qos_rpc_type[] = {"rw_ios_per_sec", "rw_mbytes_per_sec"};

TAILQ_HEAD(spdk_bdev_list, spdk_bdev);

struct spdk_bdev_mgr {
	struct spdk_mempool *bdev_io_pool;

	struct spdk_mempool *buf_small_pool;
	struct spdk_mempool *buf_large_pool;

	void *zero_buffer;

	TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;

	struct spdk_bdev_list bdevs;

	bool init_complete;
	bool module_init_complete;

#ifdef SPDK_CONFIG_VTUNE
	__itt_domain	*domain;
#endif
};

static struct spdk_bdev_mgr g_bdev_mgr = {
	.bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
	.bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
	.init_complete = false,
	.module_init_complete = false,
};

static struct spdk_bdev_opts	g_bdev_opts = {
	.bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
	.bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
};

static spdk_bdev_init_cb	g_init_cb_fn = NULL;
static void			*g_init_cb_arg = NULL;

static spdk_bdev_fini_cb	g_fini_cb_fn = NULL;
static void			*g_fini_cb_arg = NULL;
static struct spdk_thread	*g_fini_thread = NULL;

struct spdk_bdev_qos_limit {
	/** IOs or bytes allowed per second (i.e., 1s). */
	uint64_t limit;

	/** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
	 *  For remaining bytes, allowed to run negative if an I/O is submitted when
	 *  some bytes are remaining, but the I/O is bigger than that amount. The
	 *  excess will be deducted from the next timeslice.
	 */
	int64_t remaining_this_timeslice;

	/** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
	uint32_t min_per_timeslice;

	/** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
	uint32_t max_per_timeslice;

	/** Function to check whether to queue the IO. */
	bool (*queue_io)(const struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);

	/** Function to update for the submitted IO. */
	void (*update_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
};

struct spdk_bdev_qos {
	/** Types of structure of rate limits. */
	struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];

	/** The channel that all I/O are funneled through. */
	struct spdk_bdev_channel *ch;

	/** The thread on which the poller is running. */
	struct spdk_thread *thread;

	/** Queue of I/O waiting to be issued. */
	bdev_io_tailq_t queued;

	/** Size of a timeslice in tsc ticks. */
	uint64_t timeslice_size;

	/** Timestamp of start of last timeslice. */
	uint64_t last_timeslice;

	/** Poller that processes queued I/O commands each time slice. */
	struct spdk_poller *poller;
};

struct spdk_bdev_mgmt_channel {
	bdev_io_stailq_t need_buf_small;
	bdev_io_stailq_t need_buf_large;

	/*
	 * Each thread keeps a cache of bdev_io - this allows
	 *  bdev threads which are *not* DPDK threads to still
	 *  benefit from a per-thread bdev_io cache.  Without
	 *  this, non-DPDK threads fetching from the mempool
	 *  incur a cmpxchg on get and put.
	 */
	bdev_io_stailq_t per_thread_cache;
	uint32_t	per_thread_cache_count;
	uint32_t	bdev_io_cache_size;

	TAILQ_HEAD(, spdk_bdev_shared_resource)	shared_resources;
	TAILQ_HEAD(, spdk_bdev_io_wait_entry)	io_wait_queue;
};

/*
 * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
 * will queue here their IO that awaits retry. It makes it possible to retry sending
 * IO to one bdev after IO from other bdev completes.
 */
struct spdk_bdev_shared_resource {
	/* The bdev management channel */
	struct spdk_bdev_mgmt_channel *mgmt_ch;

	/*
	 * Count of I/O submitted to bdev module and waiting for completion.
	 * Incremented before submit_request() is called on an spdk_bdev_io.
	 */
	uint64_t		io_outstanding;

	/*
	 * Queue of IO awaiting retry because of a previous NOMEM status returned
	 *  on this channel.
	 */
	bdev_io_tailq_t		nomem_io;

	/*
	 * Threshold which io_outstanding must drop to before retrying nomem_io.
	 */
	uint64_t		nomem_threshold;

	/* I/O channel allocated by a bdev module */
	struct spdk_io_channel	*shared_ch;

	/* Refcount of bdev channels using this resource */
	uint32_t		ref;

	TAILQ_ENTRY(spdk_bdev_shared_resource) link;
};

#define BDEV_CH_RESET_IN_PROGRESS	(1 << 0)
#define BDEV_CH_QOS_ENABLED		(1 << 1)

struct spdk_bdev_channel {
	struct spdk_bdev	*bdev;

	/* The channel for the underlying device */
	struct spdk_io_channel	*channel;

	/* Per io_device per thread data */
	struct spdk_bdev_shared_resource *shared_resource;

	struct spdk_bdev_io_stat stat;

	/*
	 * Count of I/O submitted through this channel and waiting for completion.
	 * Incremented before submit_request() is called on an spdk_bdev_io.
	 */
	uint64_t		io_outstanding;

	bdev_io_tailq_t		queued_resets;

	uint32_t		flags;

	struct spdk_histogram_data *histogram;

#ifdef SPDK_CONFIG_VTUNE
	uint64_t		start_tsc;
	uint64_t		interval_tsc;
	__itt_string_handle	*handle;
	struct spdk_bdev_io_stat prev_stat;
#endif

};

struct spdk_bdev_desc {
	struct spdk_bdev		*bdev;
	struct spdk_thread		*thread;
	spdk_bdev_remove_cb_t		remove_cb;
	void				*remove_ctx;
	bool				remove_scheduled;
	bool				closed;
	bool				write;
	TAILQ_ENTRY(spdk_bdev_desc)	link;
};

struct spdk_bdev_iostat_ctx {
	struct spdk_bdev_io_stat *stat;
	spdk_bdev_get_device_stat_cb cb;
	void *cb_arg;
};

#define __bdev_to_io_dev(bdev)		(((char *)bdev) + 1)
#define __bdev_from_io_dev(io_dev)	((struct spdk_bdev *)(((char *)io_dev) - 1))

static void _spdk_bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success,
		void *cb_arg);
static void _spdk_bdev_write_zero_buffer_next(void *_bdev_io);

void
spdk_bdev_get_opts(struct spdk_bdev_opts *opts)
{
	*opts = g_bdev_opts;
}

int
spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
{
	uint32_t min_pool_size;

	/*
	 * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
	 *  initialization.  A second mgmt_ch will be created on the same thread when the application starts
	 *  but before the deferred put_io_channel event is executed for the first mgmt_ch.
	 */
	min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
	if (opts->bdev_io_pool_size < min_pool_size) {
		SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
			    " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
			    spdk_thread_get_count());
		SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
		return -1;
	}

	g_bdev_opts = *opts;
	return 0;
}

struct spdk_bdev *
spdk_bdev_first(void)
{
	struct spdk_bdev *bdev;

	bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
	if (bdev) {
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Starting bdev iteration at %s\n", bdev->name);
	}

	return bdev;
}

struct spdk_bdev *
spdk_bdev_next(struct spdk_bdev *prev)
{
	struct spdk_bdev *bdev;

	bdev = TAILQ_NEXT(prev, internal.link);
	if (bdev) {
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Continuing bdev iteration at %s\n", bdev->name);
	}

	return bdev;
}

static struct spdk_bdev *
_bdev_next_leaf(struct spdk_bdev *bdev)
{
	while (bdev != NULL) {
		if (bdev->internal.claim_module == NULL) {
			return bdev;
		} else {
			bdev = TAILQ_NEXT(bdev, internal.link);
		}
	}

	return bdev;
}

struct spdk_bdev *
spdk_bdev_first_leaf(void)
{
	struct spdk_bdev *bdev;

	bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));

	if (bdev) {
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Starting bdev iteration at %s\n", bdev->name);
	}

	return bdev;
}

struct spdk_bdev *
spdk_bdev_next_leaf(struct spdk_bdev *prev)
{
	struct spdk_bdev *bdev;

	bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));

	if (bdev) {
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Continuing bdev iteration at %s\n", bdev->name);
	}

	return bdev;
}

struct spdk_bdev *
spdk_bdev_get_by_name(const char *bdev_name)
{
	struct spdk_bdev_alias *tmp;
	struct spdk_bdev *bdev = spdk_bdev_first();

	while (bdev != NULL) {
		if (strcmp(bdev_name, bdev->name) == 0) {
			return bdev;
		}

		TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
			if (strcmp(bdev_name, tmp->alias) == 0) {
				return bdev;
			}
		}

		bdev = spdk_bdev_next(bdev);
	}

	return NULL;
}

void
spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
{
	struct iovec *iovs;

	iovs = bdev_io->u.bdev.iovs;

	assert(iovs != NULL);
	assert(bdev_io->u.bdev.iovcnt >= 1);

	iovs[0].iov_base = buf;
	iovs[0].iov_len = len;
}

static bool
_is_buf_allocated(struct iovec *iovs)
{
	return iovs[0].iov_base != NULL;
}

static bool
_are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
{
	int i;
	uintptr_t iov_base;

	if (spdk_likely(alignment == 1)) {
		return true;
	}

	for (i = 0; i < iovcnt; i++) {
		iov_base = (uintptr_t)iovs[i].iov_base;
		if ((iov_base & (alignment - 1)) != 0) {
			return false;
		}
	}

	return true;
}

static void
_copy_iovs_to_buf(void *buf, size_t buf_len, struct iovec *iovs, int iovcnt)
{
	int i;
	size_t len;

	for (i = 0; i < iovcnt; i++) {
		len = spdk_min(iovs[i].iov_len, buf_len);
		memcpy(buf, iovs[i].iov_base, len);
		buf += len;
		buf_len -= len;
	}
}

static void
_copy_buf_to_iovs(struct iovec *iovs, int iovcnt, void *buf, size_t buf_len)
{
	int i;
	size_t len;

	for (i = 0; i < iovcnt; i++) {
		len = spdk_min(iovs[i].iov_len, buf_len);
		memcpy(iovs[i].iov_base, buf, len);
		buf += len;
		buf_len -= len;
	}
}

static void
_bdev_io_set_bounce_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
{
	/* save original iovec */
	bdev_io->internal.orig_iovs = bdev_io->u.bdev.iovs;
	bdev_io->internal.orig_iovcnt = bdev_io->u.bdev.iovcnt;
	/* set bounce iov */
	bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_iov;
	bdev_io->u.bdev.iovcnt = 1;
	/* set bounce buffer for this operation */
	bdev_io->u.bdev.iovs[0].iov_base = buf;
	bdev_io->u.bdev.iovs[0].iov_len = len;
	/* if this is write path, copy data from original buffer to bounce buffer */
	if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
		_copy_iovs_to_buf(buf, len, bdev_io->internal.orig_iovs, bdev_io->internal.orig_iovcnt);
	}
}

static void
spdk_bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
{
	struct spdk_mempool *pool;
	struct spdk_bdev_io *tmp;
	void *buf, *aligned_buf;
	bdev_io_stailq_t *stailq;
	struct spdk_bdev_mgmt_channel *ch;
	uint64_t buf_len;
	uint64_t alignment;
	bool buf_allocated;

	buf = bdev_io->internal.buf;
	buf_len = bdev_io->internal.buf_len;
	alignment = spdk_bdev_get_buf_align(bdev_io->bdev);
	ch = bdev_io->internal.ch->shared_resource->mgmt_ch;

	bdev_io->internal.buf = NULL;

	if (buf_len + alignment <= SPDK_BDEV_SMALL_BUF_MAX_SIZE + SPDK_BDEV_POOL_ALIGNMENT) {
		pool = g_bdev_mgr.buf_small_pool;
		stailq = &ch->need_buf_small;
	} else {
		pool = g_bdev_mgr.buf_large_pool;
		stailq = &ch->need_buf_large;
	}

	if (STAILQ_EMPTY(stailq)) {
		spdk_mempool_put(pool, buf);
	} else {
		tmp = STAILQ_FIRST(stailq);

		alignment = spdk_bdev_get_buf_align(tmp->bdev);
		buf_allocated = _is_buf_allocated(tmp->u.bdev.iovs);

		aligned_buf = (void *)(((uintptr_t)buf +
					(alignment - 1)) & ~(alignment - 1));
		if (buf_allocated) {
			_bdev_io_set_bounce_buf(tmp, aligned_buf, tmp->internal.buf_len);
		} else {
			spdk_bdev_io_set_buf(tmp, aligned_buf, tmp->internal.buf_len);
		}

		STAILQ_REMOVE_HEAD(stailq, internal.buf_link);
		tmp->internal.buf = buf;
		tmp->internal.get_buf_cb(tmp->internal.ch->channel, tmp);
	}
}

static void
_bdev_io_unset_bounce_buf(struct spdk_bdev_io *bdev_io)
{
	/* if this is read path, copy data from bounce buffer to original buffer */
	if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ &&
	    bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
		_copy_buf_to_iovs(bdev_io->internal.orig_iovs, bdev_io->internal.orig_iovcnt,
				  bdev_io->internal.bounce_iov.iov_base, bdev_io->internal.bounce_iov.iov_len);
	}
	/* set orignal buffer for this io */
	bdev_io->u.bdev.iovcnt = bdev_io->internal.orig_iovcnt;
	bdev_io->u.bdev.iovs = bdev_io->internal.orig_iovs;
	/* disable bouncing buffer for this io */
	bdev_io->internal.orig_iovcnt = 0;
	bdev_io->internal.orig_iovs = NULL;
	/* return bounce buffer to the pool */
	spdk_bdev_io_put_buf(bdev_io);
}

void
spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
{
	struct spdk_mempool *pool;
	bdev_io_stailq_t *stailq;
	void *buf, *aligned_buf;
	struct spdk_bdev_mgmt_channel *mgmt_ch;
	uint64_t alignment;
	bool buf_allocated;

	assert(cb != NULL);
	assert(bdev_io->u.bdev.iovs != NULL);

	alignment = spdk_bdev_get_buf_align(bdev_io->bdev);
	buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);

	if (buf_allocated &&
	    _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
		/* Buffer already present and aligned */
		cb(bdev_io->internal.ch->channel, bdev_io);
		return;
	}

	assert(len + alignment <= SPDK_BDEV_LARGE_BUF_MAX_SIZE + SPDK_BDEV_POOL_ALIGNMENT);
	mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;

	bdev_io->internal.buf_len = len;
	bdev_io->internal.get_buf_cb = cb;

	if (len + alignment <= SPDK_BDEV_SMALL_BUF_MAX_SIZE + SPDK_BDEV_POOL_ALIGNMENT) {
		pool = g_bdev_mgr.buf_small_pool;
		stailq = &mgmt_ch->need_buf_small;
	} else {
		pool = g_bdev_mgr.buf_large_pool;
		stailq = &mgmt_ch->need_buf_large;
	}

	buf = spdk_mempool_get(pool);

	if (!buf) {
		STAILQ_INSERT_TAIL(stailq, bdev_io, internal.buf_link);
	} else {
		aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));

		if (buf_allocated) {
			_bdev_io_set_bounce_buf(bdev_io, aligned_buf, len);
		} else {
			spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
		}
		bdev_io->internal.buf = buf;
		bdev_io->internal.get_buf_cb(bdev_io->internal.ch->channel, bdev_io);
	}
}

static int
spdk_bdev_module_get_max_ctx_size(void)
{
	struct spdk_bdev_module *bdev_module;
	int max_bdev_module_size = 0;

	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
			max_bdev_module_size = bdev_module->get_ctx_size();
		}
	}

	return max_bdev_module_size;
}

void
spdk_bdev_config_text(FILE *fp)
{
	struct spdk_bdev_module *bdev_module;

	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (bdev_module->config_text) {
			bdev_module->config_text(fp);
		}
	}
}

static void
spdk_bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
{
	int i;
	struct spdk_bdev_qos *qos = bdev->internal.qos;
	uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];

	if (!qos) {
		return;
	}

	spdk_bdev_get_qos_rate_limits(bdev, limits);

	spdk_json_write_object_begin(w);
	spdk_json_write_named_string(w, "method", "set_bdev_qos_limit");
	spdk_json_write_name(w, "params");

	spdk_json_write_object_begin(w);
	spdk_json_write_named_string(w, "name", bdev->name);
	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (limits[i] > 0) {
			spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
		}
	}
	spdk_json_write_object_end(w);

	spdk_json_write_object_end(w);
}

void
spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
{
	struct spdk_bdev_module *bdev_module;
	struct spdk_bdev *bdev;

	assert(w != NULL);

	spdk_json_write_array_begin(w);

	spdk_json_write_object_begin(w);
	spdk_json_write_named_string(w, "method", "set_bdev_options");
	spdk_json_write_name(w, "params");
	spdk_json_write_object_begin(w);
	spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
	spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
	spdk_json_write_object_end(w);
	spdk_json_write_object_end(w);

	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (bdev_module->config_json) {
			bdev_module->config_json(w);
		}
	}

	TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
		spdk_bdev_qos_config_json(bdev, w);

		if (bdev->fn_table->write_config_json) {
			bdev->fn_table->write_config_json(bdev, w);
		}
	}

	spdk_json_write_array_end(w);
}

static int
spdk_bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
{
	struct spdk_bdev_mgmt_channel *ch = ctx_buf;
	struct spdk_bdev_io *bdev_io;
	uint32_t i;

	STAILQ_INIT(&ch->need_buf_small);
	STAILQ_INIT(&ch->need_buf_large);

	STAILQ_INIT(&ch->per_thread_cache);
	ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;

	/* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
	ch->per_thread_cache_count = 0;
	for (i = 0; i < ch->bdev_io_cache_size; i++) {
		bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
		assert(bdev_io != NULL);
		ch->per_thread_cache_count++;
		STAILQ_INSERT_TAIL(&ch->per_thread_cache, bdev_io, internal.buf_link);
	}

	TAILQ_INIT(&ch->shared_resources);
	TAILQ_INIT(&ch->io_wait_queue);

	return 0;
}

static void
spdk_bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
{
	struct spdk_bdev_mgmt_channel *ch = ctx_buf;
	struct spdk_bdev_io *bdev_io;

	if (!STAILQ_EMPTY(&ch->need_buf_small) || !STAILQ_EMPTY(&ch->need_buf_large)) {
		SPDK_ERRLOG("Pending I/O list wasn't empty on mgmt channel free\n");
	}

	if (!TAILQ_EMPTY(&ch->shared_resources)) {
		SPDK_ERRLOG("Module channel list wasn't empty on mgmt channel free\n");
	}

	while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
		bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
		STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
		ch->per_thread_cache_count--;
		spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
	}

	assert(ch->per_thread_cache_count == 0);
}

static void
spdk_bdev_init_complete(int rc)
{
	spdk_bdev_init_cb cb_fn = g_init_cb_fn;
	void *cb_arg = g_init_cb_arg;
	struct spdk_bdev_module *m;

	g_bdev_mgr.init_complete = true;
	g_init_cb_fn = NULL;
	g_init_cb_arg = NULL;

	/*
	 * For modules that need to know when subsystem init is complete,
	 * inform them now.
	 */
	if (rc == 0) {
		TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
			if (m->init_complete) {
				m->init_complete();
			}
		}
	}

	cb_fn(cb_arg, rc);
}

static void
spdk_bdev_module_action_complete(void)
{
	struct spdk_bdev_module *m;

	/*
	 * Don't finish bdev subsystem initialization if
	 * module pre-initialization is still in progress, or
	 * the subsystem been already initialized.
	 */
	if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
		return;
	}

	/*
	 * Check all bdev modules for inits/examinations in progress. If any
	 * exist, return immediately since we cannot finish bdev subsystem
	 * initialization until all are completed.
	 */
	TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (m->internal.action_in_progress > 0) {
			return;
		}
	}

	/*
	 * Modules already finished initialization - now that all
	 * the bdev modules have finished their asynchronous I/O
	 * processing, the entire bdev layer can be marked as complete.
	 */
	spdk_bdev_init_complete(0);
}

static void
spdk_bdev_module_action_done(struct spdk_bdev_module *module)
{
	assert(module->internal.action_in_progress > 0);
	module->internal.action_in_progress--;
	spdk_bdev_module_action_complete();
}

void
spdk_bdev_module_init_done(struct spdk_bdev_module *module)
{
	spdk_bdev_module_action_done(module);
}

void
spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
{
	spdk_bdev_module_action_done(module);
}

/** The last initialized bdev module */
static struct spdk_bdev_module *g_resume_bdev_module = NULL;

static int
spdk_bdev_modules_init(void)
{
	struct spdk_bdev_module *module;
	int rc = 0;

	TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		g_resume_bdev_module = module;
		rc = module->module_init();
		if (rc != 0) {
			return rc;
		}
	}

	g_resume_bdev_module = NULL;
	return 0;
}


static void
spdk_bdev_init_failed_complete(void *cb_arg)
{
	spdk_bdev_init_complete(-1);
}

static void
spdk_bdev_init_failed(void *cb_arg)
{
	spdk_bdev_finish(spdk_bdev_init_failed_complete, NULL);
}

void
spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
{
	struct spdk_conf_section *sp;
	struct spdk_bdev_opts bdev_opts;
	int32_t bdev_io_pool_size, bdev_io_cache_size;
	int cache_size;
	int rc = 0;
	char mempool_name[32];

	assert(cb_fn != NULL);

	sp = spdk_conf_find_section(NULL, "Bdev");
	if (sp != NULL) {
		spdk_bdev_get_opts(&bdev_opts);

		bdev_io_pool_size = spdk_conf_section_get_intval(sp, "BdevIoPoolSize");
		if (bdev_io_pool_size >= 0) {
			bdev_opts.bdev_io_pool_size = bdev_io_pool_size;
		}

		bdev_io_cache_size = spdk_conf_section_get_intval(sp, "BdevIoCacheSize");
		if (bdev_io_cache_size >= 0) {
			bdev_opts.bdev_io_cache_size = bdev_io_cache_size;
		}

		if (spdk_bdev_set_opts(&bdev_opts)) {
			spdk_bdev_init_complete(-1);
			return;
		}

		assert(memcmp(&bdev_opts, &g_bdev_opts, sizeof(bdev_opts)) == 0);
	}

	g_init_cb_fn = cb_fn;
	g_init_cb_arg = cb_arg;

	snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());

	g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
				  g_bdev_opts.bdev_io_pool_size,
				  sizeof(struct spdk_bdev_io) +
				  spdk_bdev_module_get_max_ctx_size(),
				  0,
				  SPDK_ENV_SOCKET_ID_ANY);

	if (g_bdev_mgr.bdev_io_pool == NULL) {
		SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
		spdk_bdev_init_complete(-1);
		return;
	}

	/**
	 * Ensure no more than half of the total buffers end up local caches, by
	 *   using spdk_thread_get_count() to determine how many local caches we need
	 *   to account for.
	 */
	cache_size = BUF_SMALL_POOL_SIZE / (2 * spdk_thread_get_count());
	snprintf(mempool_name, sizeof(mempool_name), "buf_small_pool_%d", getpid());

	g_bdev_mgr.buf_small_pool = spdk_mempool_create(mempool_name,
				    BUF_SMALL_POOL_SIZE,
				    SPDK_BDEV_SMALL_BUF_MAX_SIZE + SPDK_BDEV_POOL_ALIGNMENT,
				    cache_size,
				    SPDK_ENV_SOCKET_ID_ANY);
	if (!g_bdev_mgr.buf_small_pool) {
		SPDK_ERRLOG("create rbuf small pool failed\n");
		spdk_bdev_init_complete(-1);
		return;
	}

	cache_size = BUF_LARGE_POOL_SIZE / (2 * spdk_thread_get_count());
	snprintf(mempool_name, sizeof(mempool_name), "buf_large_pool_%d", getpid());

	g_bdev_mgr.buf_large_pool = spdk_mempool_create(mempool_name,
				    BUF_LARGE_POOL_SIZE,
				    SPDK_BDEV_LARGE_BUF_MAX_SIZE + SPDK_BDEV_POOL_ALIGNMENT,
				    cache_size,
				    SPDK_ENV_SOCKET_ID_ANY);
	if (!g_bdev_mgr.buf_large_pool) {
		SPDK_ERRLOG("create rbuf large pool failed\n");
		spdk_bdev_init_complete(-1);
		return;
	}

	g_bdev_mgr.zero_buffer = spdk_dma_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
				 NULL);
	if (!g_bdev_mgr.zero_buffer) {
		SPDK_ERRLOG("create bdev zero buffer failed\n");
		spdk_bdev_init_complete(-1);
		return;
	}

#ifdef SPDK_CONFIG_VTUNE
	g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
#endif

	spdk_io_device_register(&g_bdev_mgr, spdk_bdev_mgmt_channel_create,
				spdk_bdev_mgmt_channel_destroy,
				sizeof(struct spdk_bdev_mgmt_channel),
				"bdev_mgr");

	rc = spdk_bdev_modules_init();
	g_bdev_mgr.module_init_complete = true;
	if (rc != 0) {
		SPDK_ERRLOG("bdev modules init failed\n");
		spdk_thread_send_msg(spdk_get_thread(), spdk_bdev_init_failed, NULL);
		return;
	}

	spdk_bdev_module_action_complete();
}

static void
spdk_bdev_mgr_unregister_cb(void *io_device)
{
	spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;

	if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
		SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
			    spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
			    g_bdev_opts.bdev_io_pool_size);
	}

	if (spdk_mempool_count(g_bdev_mgr.buf_small_pool) != BUF_SMALL_POOL_SIZE) {
		SPDK_ERRLOG("Small buffer pool count is %zu but should be %u\n",
			    spdk_mempool_count(g_bdev_mgr.buf_small_pool),
			    BUF_SMALL_POOL_SIZE);
		assert(false);
	}

	if (spdk_mempool_count(g_bdev_mgr.buf_large_pool) != BUF_LARGE_POOL_SIZE) {
		SPDK_ERRLOG("Large buffer pool count is %zu but should be %u\n",
			    spdk_mempool_count(g_bdev_mgr.buf_large_pool),
			    BUF_LARGE_POOL_SIZE);
		assert(false);
	}

	spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
	spdk_mempool_free(g_bdev_mgr.buf_small_pool);
	spdk_mempool_free(g_bdev_mgr.buf_large_pool);
	spdk_dma_free(g_bdev_mgr.zero_buffer);

	cb_fn(g_fini_cb_arg);
	g_fini_cb_fn = NULL;
	g_fini_cb_arg = NULL;
	g_bdev_mgr.init_complete = false;
	g_bdev_mgr.module_init_complete = false;
}

static void
spdk_bdev_module_finish_iter(void *arg)
{
	struct spdk_bdev_module *bdev_module;

	/* Start iterating from the last touched module */
	if (!g_resume_bdev_module) {
		bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
	} else {
		bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
					 internal.tailq);
	}

	while (bdev_module) {
		if (bdev_module->async_fini) {
			/* Save our place so we can resume later. We must
			 * save the variable here, before calling module_fini()
			 * below, because in some cases the module may immediately
			 * call spdk_bdev_module_finish_done() and re-enter
			 * this function to continue iterating. */
			g_resume_bdev_module = bdev_module;
		}

		if (bdev_module->module_fini) {
			bdev_module->module_fini();
		}

		if (bdev_module->async_fini) {
			return;
		}

		bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
					 internal.tailq);
	}

	g_resume_bdev_module = NULL;
	spdk_io_device_unregister(&g_bdev_mgr, spdk_bdev_mgr_unregister_cb);
}

void
spdk_bdev_module_finish_done(void)
{
	if (spdk_get_thread() != g_fini_thread) {
		spdk_thread_send_msg(g_fini_thread, spdk_bdev_module_finish_iter, NULL);
	} else {
		spdk_bdev_module_finish_iter(NULL);
	}
}

static void
_spdk_bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
{
	struct spdk_bdev *bdev = cb_arg;

	if (bdeverrno && bdev) {
		SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
			     bdev->name);

		/*
		 * Since the call to spdk_bdev_unregister() failed, we have no way to free this
		 *  bdev; try to continue by manually removing this bdev from the list and continue
		 *  with the next bdev in the list.
		 */
		TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
	}

	if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Done unregistering bdevs\n");
		/*
		 * Bdev module finish need to be deferred as we might be in the middle of some context
		 * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
		 * after returning.
		 */
		spdk_thread_send_msg(spdk_get_thread(), spdk_bdev_module_finish_iter, NULL);
		return;
	}

	/*
	 * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
	 * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
	 * to detect clean shutdown as opposed to run-time hot removal of the underlying
	 * base bdevs.
	 *
	 * Also, walk the list in the reverse order.
	 */
	for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
	     bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
		if (bdev->internal.claim_module != NULL) {
			SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Skipping claimed bdev '%s'(<-'%s').\n",
				      bdev->name, bdev->internal.claim_module->name);
			continue;
		}

		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Unregistering bdev '%s'\n", bdev->name);
		spdk_bdev_unregister(bdev, _spdk_bdev_finish_unregister_bdevs_iter, bdev);
		return;
	}

	/*
	 * If any bdev fails to unclaim underlying bdev properly, we may face the
	 * case of bdev list consisting of claimed bdevs only (if claims are managed
	 * correctly, this would mean there's a loop in the claims graph which is
	 * clearly impossible). Warn and unregister last bdev on the list then.
	 */
	for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
	     bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
		SPDK_ERRLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
		spdk_bdev_unregister(bdev, _spdk_bdev_finish_unregister_bdevs_iter, bdev);
		return;
	}
}

void
spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
{
	struct spdk_bdev_module *m;

	assert(cb_fn != NULL);

	g_fini_thread = spdk_get_thread();

	g_fini_cb_fn = cb_fn;
	g_fini_cb_arg = cb_arg;

	TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (m->fini_start) {
			m->fini_start();
		}
	}

	_spdk_bdev_finish_unregister_bdevs_iter(NULL, 0);
}

static struct spdk_bdev_io *
spdk_bdev_get_io(struct spdk_bdev_channel *channel)
{
	struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
	struct spdk_bdev_io *bdev_io;

	if (ch->per_thread_cache_count > 0) {
		bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
		STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
		ch->per_thread_cache_count--;
	} else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
		/*
		 * Don't try to look for bdev_ios in the global pool if there are
		 * waiters on bdev_ios - we don't want this caller to jump the line.
		 */
		bdev_io = NULL;
	} else {
		bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
	}

	return bdev_io;
}

void
spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
{
	struct spdk_bdev_mgmt_channel *ch = bdev_io->internal.ch->shared_resource->mgmt_ch;

	assert(bdev_io != NULL);
	assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);

	if (bdev_io->internal.buf != NULL) {
		spdk_bdev_io_put_buf(bdev_io);
	}

	if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
		ch->per_thread_cache_count++;
		STAILQ_INSERT_TAIL(&ch->per_thread_cache, bdev_io, internal.buf_link);
		while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
			struct spdk_bdev_io_wait_entry *entry;

			entry = TAILQ_FIRST(&ch->io_wait_queue);
			TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
			entry->cb_fn(entry->cb_arg);
		}
	} else {
		/* We should never have a full cache with entries on the io wait queue. */
		assert(TAILQ_EMPTY(&ch->io_wait_queue));
		spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
	}
}

static bool
_spdk_bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
{
	assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);

	switch (limit) {
	case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
		return true;
	case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
		return false;
	case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
	default:
		return false;
	}
}

static bool
_spdk_bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
{
	switch (bdev_io->type) {
	case SPDK_BDEV_IO_TYPE_NVME_IO:
	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
	case SPDK_BDEV_IO_TYPE_READ:
	case SPDK_BDEV_IO_TYPE_WRITE:
	case SPDK_BDEV_IO_TYPE_UNMAP:
	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
		return true;
	default:
		return false;
	}
}

static uint64_t
_spdk_bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
{
	struct spdk_bdev	*bdev = bdev_io->bdev;

	switch (bdev_io->type) {
	case SPDK_BDEV_IO_TYPE_NVME_IO:
	case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
		return bdev_io->u.nvme_passthru.nbytes;
	case SPDK_BDEV_IO_TYPE_READ:
	case SPDK_BDEV_IO_TYPE_WRITE:
	case SPDK_BDEV_IO_TYPE_UNMAP:
	case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
		return bdev_io->u.bdev.num_blocks * bdev->blocklen;
	default:
		return 0;
	}
}

static bool
_spdk_bdev_qos_rw_queue_io(const struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
{
	if (limit->max_per_timeslice > 0 && limit->remaining_this_timeslice <= 0) {
		return true;
	} else {
		return false;
	}
}

static void
_spdk_bdev_qos_rw_iops_update_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
{
	limit->remaining_this_timeslice--;
}

static void
_spdk_bdev_qos_rw_bps_update_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
{
	limit->remaining_this_timeslice -= _spdk_bdev_get_io_size_in_byte(io);
}

static void
_spdk_bdev_qos_set_ops(struct spdk_bdev_qos *qos)
{
	int i;

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
			qos->rate_limits[i].queue_io = NULL;
			qos->rate_limits[i].update_quota = NULL;
			continue;
		}

		switch (i) {
		case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
			qos->rate_limits[i].queue_io = _spdk_bdev_qos_rw_queue_io;
			qos->rate_limits[i].update_quota = _spdk_bdev_qos_rw_iops_update_quota;
			break;
		case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
			qos->rate_limits[i].queue_io = _spdk_bdev_qos_rw_queue_io;
			qos->rate_limits[i].update_quota = _spdk_bdev_qos_rw_bps_update_quota;
			break;
		default:
			break;
		}
	}
}

static int
_spdk_bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
{
	struct spdk_bdev_io		*bdev_io = NULL, *tmp = NULL;
	struct spdk_bdev		*bdev = ch->bdev;
	struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
	int				i, submitted_ios = 0;

	TAILQ_FOREACH_SAFE(bdev_io, &qos->queued, internal.link, tmp) {
		if (_spdk_bdev_qos_io_to_limit(bdev_io) == true) {
			for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
				if (!qos->rate_limits[i].queue_io) {
					continue;
				}

				if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
								 bdev_io) == true) {
					return submitted_ios;
				}
			}
			for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
				if (!qos->rate_limits[i].update_quota) {
					continue;
				}

				qos->rate_limits[i].update_quota(&qos->rate_limits[i], bdev_io);
			}
		}

		TAILQ_REMOVE(&qos->queued, bdev_io, internal.link);
		ch->io_outstanding++;
		shared_resource->io_outstanding++;
		bdev->fn_table->submit_request(ch->channel, bdev_io);
		submitted_ios++;
	}

	return submitted_ios;
}

static void
_spdk_bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
{
	int rc;

	bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
	bdev_io->internal.waitq_entry.cb_fn = cb_fn;
	bdev_io->internal.waitq_entry.cb_arg = bdev_io;
	rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
				     &bdev_io->internal.waitq_entry);
	if (rc != 0) {
		SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
		bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
	}
}

static bool
_spdk_bdev_io_type_can_split(uint8_t type)
{
	assert(type != SPDK_BDEV_IO_TYPE_INVALID);
	assert(type < SPDK_BDEV_NUM_IO_TYPES);

	/* Only split READ and WRITE I/O.  Theoretically other types of I/O like
	 * UNMAP could be split, but these types of I/O are typically much larger
	 * in size (sometimes the size of the entire block device), and the bdev
	 * module can more efficiently split these types of I/O.  Plus those types
	 * of I/O do not have a payload, which makes the splitting process simpler.
	 */
	if (type == SPDK_BDEV_IO_TYPE_READ || type == SPDK_BDEV_IO_TYPE_WRITE) {
		return true;
	} else {
		return false;
	}
}

static bool
_spdk_bdev_io_should_split(struct spdk_bdev_io *bdev_io)
{
	uint64_t start_stripe, end_stripe;
	uint32_t io_boundary = bdev_io->bdev->optimal_io_boundary;

	if (io_boundary == 0) {
		return false;
	}

	if (!_spdk_bdev_io_type_can_split(bdev_io->type)) {
		return false;
	}

	start_stripe = bdev_io->u.bdev.offset_blocks;
	end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
	/* Avoid expensive div operations if possible.  These spdk_u32 functions are very cheap. */
	if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
		start_stripe >>= spdk_u32log2(io_boundary);
		end_stripe >>= spdk_u32log2(io_boundary);
	} else {
		start_stripe /= io_boundary;
		end_stripe /= io_boundary;
	}
	return (start_stripe != end_stripe);
}

static uint32_t
_to_next_boundary(uint64_t offset, uint32_t boundary)
{
	return (boundary - (offset % boundary));
}

static void
_spdk_bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);

static void
_spdk_bdev_io_split_with_payload(void *_bdev_io)
{
	struct spdk_bdev_io *bdev_io = _bdev_io;
	uint64_t current_offset, remaining;
	uint32_t blocklen, to_next_boundary, to_next_boundary_bytes;
	struct iovec *parent_iov, *iov;
	uint64_t parent_iov_offset, iov_len;
	uint32_t parent_iovpos, parent_iovcnt, child_iovcnt, iovcnt;
	int rc;

	remaining = bdev_io->u.bdev.split_remaining_num_blocks;
	current_offset = bdev_io->u.bdev.split_current_offset_blocks;
	blocklen = bdev_io->bdev->blocklen;
	parent_iov_offset = (current_offset - bdev_io->u.bdev.offset_blocks) * blocklen;
	parent_iovcnt = bdev_io->u.bdev.iovcnt;

	for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
		parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
		if (parent_iov_offset < parent_iov->iov_len) {
			break;
		}
		parent_iov_offset -= parent_iov->iov_len;
	}

	child_iovcnt = 0;
	while (remaining > 0 && parent_iovpos < parent_iovcnt && child_iovcnt < BDEV_IO_NUM_CHILD_IOV) {
		to_next_boundary = _to_next_boundary(current_offset, bdev_io->bdev->optimal_io_boundary);
		to_next_boundary = spdk_min(remaining, to_next_boundary);
		to_next_boundary_bytes = to_next_boundary * blocklen;
		iov = &bdev_io->child_iov[child_iovcnt];
		iovcnt = 0;
		while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
		       child_iovcnt < BDEV_IO_NUM_CHILD_IOV) {
			parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
			iov_len = spdk_min(to_next_boundary_bytes, parent_iov->iov_len - parent_iov_offset);
			to_next_boundary_bytes -= iov_len;

			bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
			bdev_io->child_iov[child_iovcnt].iov_len = iov_len;

			if (iov_len < parent_iov->iov_len - parent_iov_offset) {
				parent_iov_offset += iov_len;
			} else {
				parent_iovpos++;
				parent_iov_offset = 0;
			}
			child_iovcnt++;
			iovcnt++;
		}

		if (to_next_boundary_bytes > 0) {
			/* We had to stop this child I/O early because we ran out of
			 *  child_iov space.  Make sure the iovs collected are valid and
			 *  then adjust to_next_boundary before starting the child I/O.
			 */
			if ((to_next_boundary_bytes % blocklen) != 0) {
				SPDK_ERRLOG("Remaining %" PRIu32 " is not multiple of block size %" PRIu32 "\n",
					    to_next_boundary_bytes, blocklen);
				bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
				if (bdev_io->u.bdev.split_outstanding == 0) {
					bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
				}
				return;
			}
			to_next_boundary -= to_next_boundary_bytes / blocklen;
		}

		bdev_io->u.bdev.split_outstanding++;

		if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
			rc = spdk_bdev_readv_blocks(bdev_io->internal.desc,
						    spdk_io_channel_from_ctx(bdev_io->internal.ch),
						    iov, iovcnt, current_offset, to_next_boundary,
						    _spdk_bdev_io_split_done, bdev_io);
		} else {
			rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
						     spdk_io_channel_from_ctx(bdev_io->internal.ch),
						     iov, iovcnt, current_offset, to_next_boundary,
						     _spdk_bdev_io_split_done, bdev_io);
		}

		if (rc == 0) {
			current_offset += to_next_boundary;
			remaining -= to_next_boundary;
			bdev_io->u.bdev.split_current_offset_blocks = current_offset;
			bdev_io->u.bdev.split_remaining_num_blocks = remaining;
		} else {
			bdev_io->u.bdev.split_outstanding--;
			if (rc == -ENOMEM) {
				if (bdev_io->u.bdev.split_outstanding == 0) {
					/* No I/O is outstanding. Hence we should wait here. */
					_spdk_bdev_queue_io_wait_with_cb(bdev_io,
									 _spdk_bdev_io_split_with_payload);
				}
			} else {
				bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
				if (bdev_io->u.bdev.split_outstanding == 0) {
					bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
				}
			}

			return;
		}
	}
}

static void
_spdk_bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
{
	struct spdk_bdev_io *parent_io = cb_arg;

	spdk_bdev_free_io(bdev_io);

	if (!success) {
		parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
	}
	parent_io->u.bdev.split_outstanding--;
	if (parent_io->u.bdev.split_outstanding != 0) {
		return;
	}

	/*
	 * Parent I/O finishes when all blocks are consumed or there is any failure of
	 * child I/O and no outstanding child I/O.
	 */
	if (parent_io->u.bdev.split_remaining_num_blocks == 0 ||
	    parent_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS) {
		parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
				       parent_io->internal.caller_ctx);
		return;
	}

	/*
	 * Continue with the splitting process.  This function will complete the parent I/O if the
	 * splitting is done.
	 */
	_spdk_bdev_io_split_with_payload(parent_io);
}

static void
_spdk_bdev_io_split(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io)
{
	assert(_spdk_bdev_io_type_can_split(bdev_io->type));

	bdev_io->u.bdev.split_current_offset_blocks = bdev_io->u.bdev.offset_blocks;
	bdev_io->u.bdev.split_remaining_num_blocks = bdev_io->u.bdev.num_blocks;
	bdev_io->u.bdev.split_outstanding = 0;
	bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;

	_spdk_bdev_io_split_with_payload(bdev_io);
}

static void
_spdk_bdev_io_submit(void *ctx)
{
	struct spdk_bdev_io *bdev_io = ctx;
	struct spdk_bdev *bdev = bdev_io->bdev;
	struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
	struct spdk_io_channel *ch = bdev_ch->channel;
	struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
	uint64_t tsc;

	tsc = spdk_get_ticks();
	bdev_io->internal.submit_tsc = tsc;
	spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_START, 0, 0, (uintptr_t)bdev_io, bdev_io->type);
	bdev_ch->io_outstanding++;
	shared_resource->io_outstanding++;
	bdev_io->internal.in_submit_request = true;
	if (spdk_likely(bdev_ch->flags == 0)) {
		if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
			bdev->fn_table->submit_request(ch, bdev_io);
		} else {
			bdev_ch->io_outstanding--;
			shared_resource->io_outstanding--;
			TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
		}
	} else if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
	} else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
		bdev_ch->io_outstanding--;
		shared_resource->io_outstanding--;
		TAILQ_INSERT_TAIL(&bdev->internal.qos->queued, bdev_io, internal.link);
		_spdk_bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
	} else {
		SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
	}
	bdev_io->internal.in_submit_request = false;
}

static void
spdk_bdev_io_submit(struct spdk_bdev_io *bdev_io)
{
	struct spdk_bdev *bdev = bdev_io->bdev;
	struct spdk_thread *thread = spdk_io_channel_get_thread(bdev_io->internal.ch->channel);

	assert(thread != NULL);
	assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);

	if (bdev->split_on_optimal_io_boundary && _spdk_bdev_io_should_split(bdev_io)) {
		if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
			spdk_bdev_io_get_buf(bdev_io, _spdk_bdev_io_split,
					     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
		} else {
			_spdk_bdev_io_split(NULL, bdev_io);
		}
		return;
	}

	if (bdev_io->internal.ch->flags & BDEV_CH_QOS_ENABLED) {
		if ((thread == bdev->internal.qos->thread) || !bdev->internal.qos->thread) {
			_spdk_bdev_io_submit(bdev_io);
		} else {
			bdev_io->internal.io_submit_ch = bdev_io->internal.ch;
			bdev_io->internal.ch = bdev->internal.qos->ch;
			spdk_thread_send_msg(bdev->internal.qos->thread, _spdk_bdev_io_submit, bdev_io);
		}
	} else {
		_spdk_bdev_io_submit(bdev_io);
	}
}

static void
spdk_bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
{
	struct spdk_bdev *bdev = bdev_io->bdev;
	struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
	struct spdk_io_channel *ch = bdev_ch->channel;

	assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);

	bdev_io->internal.in_submit_request = true;
	bdev->fn_table->submit_request(ch, bdev_io);
	bdev_io->internal.in_submit_request = false;
}

static void
spdk_bdev_io_init(struct spdk_bdev_io *bdev_io,
		  struct spdk_bdev *bdev, void *cb_arg,
		  spdk_bdev_io_completion_cb cb)
{
	bdev_io->bdev = bdev;
	bdev_io->internal.caller_ctx = cb_arg;
	bdev_io->internal.cb = cb;
	bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
	bdev_io->internal.in_submit_request = false;
	bdev_io->internal.buf = NULL;
	bdev_io->internal.io_submit_ch = NULL;
	bdev_io->internal.orig_iovs = NULL;
	bdev_io->internal.orig_iovcnt = 0;
}

static bool
_spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
{
	return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
}

bool
spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
{
	bool supported;

	supported = _spdk_bdev_io_type_supported(bdev, io_type);

	if (!supported) {
		switch (io_type) {
		case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
			/* The bdev layer will emulate write zeroes as long as write is supported. */
			supported = _spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
			break;
		default:
			break;
		}
	}

	return supported;
}

int
spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
{
	if (bdev->fn_table->dump_info_json) {
		return bdev->fn_table->dump_info_json(bdev->ctxt, w);
	}

	return 0;
}

static void
spdk_bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
{
	uint32_t max_per_timeslice = 0;
	int i;

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
			qos->rate_limits[i].max_per_timeslice = 0;
			continue;
		}

		max_per_timeslice = qos->rate_limits[i].limit *
				    SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;

		qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
							qos->rate_limits[i].min_per_timeslice);

		qos->rate_limits[i].remaining_this_timeslice = qos->rate_limits[i].max_per_timeslice;
	}

	_spdk_bdev_qos_set_ops(qos);
}

static int
spdk_bdev_channel_poll_qos(void *arg)
{
	struct spdk_bdev_qos *qos = arg;
	uint64_t now = spdk_get_ticks();
	int i;

	if (now < (qos->last_timeslice + qos->timeslice_size)) {
		/* We received our callback earlier than expected - return
		 *  immediately and wait to do accounting until at least one
		 *  timeslice has actually expired.  This should never happen
		 *  with a well-behaved timer implementation.
		 */
		return 0;
	}

	/* Reset for next round of rate limiting */
	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		/* We may have allowed the IOs or bytes to slightly overrun in the last
		 * timeslice. remaining_this_timeslice is signed, so if it's negative
		 * here, we'll account for the overrun so that the next timeslice will
		 * be appropriately reduced.
		 */
		if (qos->rate_limits[i].remaining_this_timeslice > 0) {
			qos->rate_limits[i].remaining_this_timeslice = 0;
		}
	}

	while (now >= (qos->last_timeslice + qos->timeslice_size)) {
		qos->last_timeslice += qos->timeslice_size;
		for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
			qos->rate_limits[i].remaining_this_timeslice +=
				qos->rate_limits[i].max_per_timeslice;
		}
	}

	return _spdk_bdev_qos_io_submit(qos->ch, qos);
}

static void
_spdk_bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
{
	struct spdk_bdev_shared_resource *shared_resource;

	spdk_put_io_channel(ch->channel);

	shared_resource = ch->shared_resource;

	assert(ch->io_outstanding == 0);
	assert(shared_resource->ref > 0);
	shared_resource->ref--;
	if (shared_resource->ref == 0) {
		assert(shared_resource->io_outstanding == 0);
		TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
		spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
		free(shared_resource);
	}
}

/* Caller must hold bdev->internal.mutex. */
static void
_spdk_bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
{
	struct spdk_bdev_qos	*qos = bdev->internal.qos;
	int			i;

	/* Rate limiting on this bdev enabled */
	if (qos) {
		if (qos->ch == NULL) {
			struct spdk_io_channel *io_ch;

			SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
				      bdev->name, spdk_get_thread());

			/* No qos channel has been selected, so set one up */

			/* Take another reference to ch */
			io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
			assert(io_ch != NULL);
			qos->ch = ch;

			qos->thread = spdk_io_channel_get_thread(io_ch);

			TAILQ_INIT(&qos->queued);

			for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
				if (_spdk_bdev_qos_is_iops_rate_limit(i) == true) {
					qos->rate_limits[i].min_per_timeslice =
						SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
				} else {
					qos->rate_limits[i].min_per_timeslice =
						SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
				}

				if (qos->rate_limits[i].limit == 0) {
					qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
				}
			}
			spdk_bdev_qos_update_max_quota_per_timeslice(qos);
			qos->timeslice_size =
				SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
			qos->last_timeslice = spdk_get_ticks();
			qos->poller = spdk_poller_register(spdk_bdev_channel_poll_qos,
							   qos,
							   SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
		}

		ch->flags |= BDEV_CH_QOS_ENABLED;
	}
}

static int
spdk_bdev_channel_create(void *io_device, void *ctx_buf)
{
	struct spdk_bdev		*bdev = __bdev_from_io_dev(io_device);
	struct spdk_bdev_channel	*ch = ctx_buf;
	struct spdk_io_channel		*mgmt_io_ch;
	struct spdk_bdev_mgmt_channel	*mgmt_ch;
	struct spdk_bdev_shared_resource *shared_resource;

	ch->bdev = bdev;
	ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
	if (!ch->channel) {
		return -1;
	}

	assert(ch->histogram == NULL);
	if (bdev->internal.histogram_enabled) {
		ch->histogram = spdk_histogram_data_alloc();
		if (ch->histogram == NULL) {
			SPDK_ERRLOG("Could not allocate histogram\n");
		}
	}

	mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
	if (!mgmt_io_ch) {
		spdk_put_io_channel(ch->channel);
		return -1;
	}

	mgmt_ch = spdk_io_channel_get_ctx(mgmt_io_ch);
	TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
		if (shared_resource->shared_ch == ch->channel) {
			spdk_put_io_channel(mgmt_io_ch);
			shared_resource->ref++;
			break;
		}
	}

	if (shared_resource == NULL) {
		shared_resource = calloc(1, sizeof(*shared_resource));
		if (shared_resource == NULL) {
			spdk_put_io_channel(ch->channel);
			spdk_put_io_channel(mgmt_io_ch);
			return -1;
		}

		shared_resource->mgmt_ch = mgmt_ch;
		shared_resource->io_outstanding = 0;
		TAILQ_INIT(&shared_resource->nomem_io);
		shared_resource->nomem_threshold = 0;
		shared_resource->shared_ch = ch->channel;
		shared_resource->ref = 1;
		TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
	}

	memset(&ch->stat, 0, sizeof(ch->stat));
	ch->stat.ticks_rate = spdk_get_ticks_hz();
	ch->io_outstanding = 0;
	TAILQ_INIT(&ch->queued_resets);
	ch->flags = 0;
	ch->shared_resource = shared_resource;

#ifdef SPDK_CONFIG_VTUNE
	{
		char *name;
		__itt_init_ittlib(NULL, 0);
		name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
		if (!name) {
			_spdk_bdev_channel_destroy_resource(ch);
			return -1;
		}
		ch->handle = __itt_string_handle_create(name);
		free(name);
		ch->start_tsc = spdk_get_ticks();
		ch->interval_tsc = spdk_get_ticks_hz() / 100;
		memset(&ch->prev_stat, 0, sizeof(ch->prev_stat));
	}
#endif

	pthread_mutex_lock(&bdev->internal.mutex);
	_spdk_bdev_enable_qos(bdev, ch);
	pthread_mutex_unlock(&bdev->internal.mutex);

	return 0;
}

/*
 * Abort I/O that are waiting on a data buffer.  These types of I/O are
 *  linked using the spdk_bdev_io internal.buf_link TAILQ_ENTRY.
 */
static void
_spdk_bdev_abort_buf_io(bdev_io_stailq_t *queue, struct spdk_bdev_channel *ch)
{
	bdev_io_stailq_t tmp;
	struct spdk_bdev_io *bdev_io;

	STAILQ_INIT(&tmp);

	while (!STAILQ_EMPTY(queue)) {
		bdev_io = STAILQ_FIRST(queue);
		STAILQ_REMOVE_HEAD(queue, internal.buf_link);
		if (bdev_io->internal.ch == ch) {
			spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
		} else {
			STAILQ_INSERT_TAIL(&tmp, bdev_io, internal.buf_link);
		}
	}

	STAILQ_SWAP(&tmp, queue, spdk_bdev_io);
}

/*
 * Abort I/O that are queued waiting for submission.  These types of I/O are
 *  linked using the spdk_bdev_io link TAILQ_ENTRY.
 */
static void
_spdk_bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
{
	struct spdk_bdev_io *bdev_io, *tmp;

	TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
		if (bdev_io->internal.ch == ch) {
			TAILQ_REMOVE(queue, bdev_io, internal.link);
			/*
			 * spdk_bdev_io_complete() assumes that the completed I/O had
			 *  been submitted to the bdev module.  Since in this case it
			 *  hadn't, bump io_outstanding to account for the decrement
			 *  that spdk_bdev_io_complete() will do.
			 */
			if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
				ch->io_outstanding++;
				ch->shared_resource->io_outstanding++;
			}
			spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
		}
	}
}

static void
spdk_bdev_qos_channel_destroy(void *cb_arg)
{
	struct spdk_bdev_qos *qos = cb_arg;

	spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
	spdk_poller_unregister(&qos->poller);

	SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Free QoS %p.\n", qos);

	free(qos);
}

static int
spdk_bdev_qos_destroy(struct spdk_bdev *bdev)
{
	int i;

	/*
	 * Cleanly shutting down the QoS poller is tricky, because
	 * during the asynchronous operation the user could open
	 * a new descriptor and create a new channel, spawning
	 * a new QoS poller.
	 *
	 * The strategy is to create a new QoS structure here and swap it
	 * in. The shutdown path then continues to refer to the old one
	 * until it completes and then releases it.
	 */
	struct spdk_bdev_qos *new_qos, *old_qos;

	old_qos = bdev->internal.qos;

	new_qos = calloc(1, sizeof(*new_qos));
	if (!new_qos) {
		SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
		return -ENOMEM;
	}

	/* Copy the old QoS data into the newly allocated structure */
	memcpy(new_qos, old_qos, sizeof(*new_qos));

	/* Zero out the key parts of the QoS structure */
	new_qos->ch = NULL;
	new_qos->thread = NULL;
	new_qos->poller = NULL;
	TAILQ_INIT(&new_qos->queued);
	/*
	 * The limit member of spdk_bdev_qos_limit structure is not zeroed.
	 * It will be used later for the new QoS structure.
	 */
	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		new_qos->rate_limits[i].remaining_this_timeslice = 0;
		new_qos->rate_limits[i].min_per_timeslice = 0;
		new_qos->rate_limits[i].max_per_timeslice = 0;
	}

	bdev->internal.qos = new_qos;

	if (old_qos->thread == NULL) {
		free(old_qos);
	} else {
		spdk_thread_send_msg(old_qos->thread, spdk_bdev_qos_channel_destroy,
				     old_qos);
	}

	/* It is safe to continue with destroying the bdev even though the QoS channel hasn't
	 * been destroyed yet. The destruction path will end up waiting for the final
	 * channel to be put before it releases resources. */

	return 0;
}

static void
_spdk_bdev_io_stat_add(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
{
	total->bytes_read += add->bytes_read;
	total->num_read_ops += add->num_read_ops;
	total->bytes_written += add->bytes_written;
	total->num_write_ops += add->num_write_ops;
	total->bytes_unmapped += add->bytes_unmapped;
	total->num_unmap_ops += add->num_unmap_ops;
	total->read_latency_ticks += add->read_latency_ticks;
	total->write_latency_ticks += add->write_latency_ticks;
	total->unmap_latency_ticks += add->unmap_latency_ticks;
}

static void
spdk_bdev_channel_destroy(void *io_device, void *ctx_buf)
{
	struct spdk_bdev_channel	*ch = ctx_buf;
	struct spdk_bdev_mgmt_channel	*mgmt_ch;
	struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;

	SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
		      spdk_get_thread());

	/* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
	pthread_mutex_lock(&ch->bdev->internal.mutex);
	_spdk_bdev_io_stat_add(&ch->bdev->internal.stat, &ch->stat);
	pthread_mutex_unlock(&ch->bdev->internal.mutex);

	mgmt_ch = shared_resource->mgmt_ch;

	_spdk_bdev_abort_queued_io(&ch->queued_resets, ch);
	_spdk_bdev_abort_queued_io(&shared_resource->nomem_io, ch);
	_spdk_bdev_abort_buf_io(&mgmt_ch->need_buf_small, ch);
	_spdk_bdev_abort_buf_io(&mgmt_ch->need_buf_large, ch);

	if (ch->histogram) {
		spdk_histogram_data_free(ch->histogram);
	}

	_spdk_bdev_channel_destroy_resource(ch);
}

int
spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
{
	struct spdk_bdev_alias *tmp;

	if (alias == NULL) {
		SPDK_ERRLOG("Empty alias passed\n");
		return -EINVAL;
	}

	if (spdk_bdev_get_by_name(alias)) {
		SPDK_ERRLOG("Bdev name/alias: %s already exists\n", alias);
		return -EEXIST;
	}

	tmp = calloc(1, sizeof(*tmp));
	if (tmp == NULL) {
		SPDK_ERRLOG("Unable to allocate alias\n");
		return -ENOMEM;
	}

	tmp->alias = strdup(alias);
	if (tmp->alias == NULL) {
		free(tmp);
		SPDK_ERRLOG("Unable to allocate alias\n");
		return -ENOMEM;
	}

	TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);

	return 0;
}

int
spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
{
	struct spdk_bdev_alias *tmp;

	TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
		if (strcmp(alias, tmp->alias) == 0) {
			TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
			free(tmp->alias);
			free(tmp);
			return 0;
		}
	}

	SPDK_INFOLOG(SPDK_LOG_BDEV, "Alias %s does not exists\n", alias);

	return -ENOENT;
}

void
spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
{
	struct spdk_bdev_alias *p, *tmp;

	TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
		TAILQ_REMOVE(&bdev->aliases, p, tailq);
		free(p->alias);
		free(p);
	}
}

struct spdk_io_channel *
spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
{
	return spdk_get_io_channel(__bdev_to_io_dev(desc->bdev));
}

const char *
spdk_bdev_get_name(const struct spdk_bdev *bdev)
{
	return bdev->name;
}

const char *
spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
{
	return bdev->product_name;
}

const struct spdk_bdev_aliases_list *
spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
{
	return &bdev->aliases;
}

uint32_t
spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
{
	return bdev->blocklen;
}

uint64_t
spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
{
	return bdev->blockcnt;
}

const char *
spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
{
	return qos_rpc_type[type];
}

void
spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
{
	int i;

	memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);

	pthread_mutex_lock(&bdev->internal.mutex);
	if (bdev->internal.qos) {
		for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
			if (bdev->internal.qos->rate_limits[i].limit !=
			    SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
				limits[i] = bdev->internal.qos->rate_limits[i].limit;
				if (_spdk_bdev_qos_is_iops_rate_limit(i) == false) {
					/* Change from Byte to Megabyte which is user visible. */
					limits[i] = limits[i] / 1024 / 1024;
				}
			}
		}
	}
	pthread_mutex_unlock(&bdev->internal.mutex);
}

size_t
spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
{
	return 1 << bdev->required_alignment;
}

uint32_t
spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
{
	return bdev->optimal_io_boundary;
}

bool
spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
{
	return bdev->write_cache;
}

const struct spdk_uuid *
spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
{
	return &bdev->uuid;
}

uint64_t
spdk_bdev_get_qd(const struct spdk_bdev *bdev)
{
	return bdev->internal.measured_queue_depth;
}

uint64_t
spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
{
	return bdev->internal.period;
}

uint64_t
spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
{
	return bdev->internal.weighted_io_time;
}

uint64_t
spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
{
	return bdev->internal.io_time;
}

static void
_calculate_measured_qd_cpl(struct spdk_io_channel_iter *i, int status)
{
	struct spdk_bdev *bdev = spdk_io_channel_iter_get_ctx(i);

	bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;

	if (bdev->internal.measured_queue_depth) {
		bdev->internal.io_time += bdev->internal.period;
		bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
	}
}

static void
_calculate_measured_qd(struct spdk_io_channel_iter *i)
{
	struct spdk_bdev *bdev = spdk_io_channel_iter_get_ctx(i);
	struct spdk_io_channel *io_ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *ch = spdk_io_channel_get_ctx(io_ch);

	bdev->internal.temporary_queue_depth += ch->io_outstanding;
	spdk_for_each_channel_continue(i, 0);
}

static int
spdk_bdev_calculate_measured_queue_depth(void *ctx)
{
	struct spdk_bdev *bdev = ctx;
	bdev->internal.temporary_queue_depth = 0;
	spdk_for_each_channel(__bdev_to_io_dev(bdev), _calculate_measured_qd, bdev,
			      _calculate_measured_qd_cpl);
	return 0;
}

void
spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
{
	bdev->internal.period = period;

	if (bdev->internal.qd_poller != NULL) {
		spdk_poller_unregister(&bdev->internal.qd_poller);
		bdev->internal.measured_queue_depth = UINT64_MAX;
	}

	if (period != 0) {
		bdev->internal.qd_poller = spdk_poller_register(spdk_bdev_calculate_measured_queue_depth, bdev,
					   period);
	}
}

int
spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
{
	int ret;

	pthread_mutex_lock(&bdev->internal.mutex);

	/* bdev has open descriptors */
	if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
	    bdev->blockcnt > size) {
		ret = -EBUSY;
	} else {
		bdev->blockcnt = size;
		ret = 0;
	}

	pthread_mutex_unlock(&bdev->internal.mutex);

	return ret;
}

/*
 * Convert I/O offset and length from bytes to blocks.
 *
 * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
 */
static uint64_t
spdk_bdev_bytes_to_blocks(struct spdk_bdev *bdev, uint64_t offset_bytes, uint64_t *offset_blocks,
			  uint64_t num_bytes, uint64_t *num_blocks)
{
	uint32_t block_size = bdev->blocklen;
	uint8_t shift_cnt;

	/* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
	if (spdk_likely(spdk_u32_is_pow2(block_size))) {
		shift_cnt = spdk_u32log2(block_size);
		*offset_blocks = offset_bytes >> shift_cnt;
		*num_blocks = num_bytes >> shift_cnt;
		return (offset_bytes - (*offset_blocks << shift_cnt)) |
		       (num_bytes - (*num_blocks << shift_cnt));
	} else {
		*offset_blocks = offset_bytes / block_size;
		*num_blocks = num_bytes / block_size;
		return (offset_bytes % block_size) | (num_bytes % block_size);
	}
}

static bool
spdk_bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
{
	/* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
	 * has been an overflow and hence the offset has been wrapped around */
	if (offset_blocks + num_blocks < offset_blocks) {
		return false;
	}

	/* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
	if (offset_blocks + num_blocks > bdev->blockcnt) {
		return false;
	}

	return true;
}

int
spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
	       void *buf, uint64_t offset, uint64_t nbytes,
	       spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
}

int
spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		      void *buf, uint64_t offset_blocks, uint64_t num_blocks,
		      spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
	bdev_io->u.bdev.iovs = &bdev_io->iov;
	bdev_io->u.bdev.iovs[0].iov_base = buf;
	bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
	bdev_io->u.bdev.iovcnt = 1;
	bdev_io->u.bdev.num_blocks = num_blocks;
	bdev_io->u.bdev.offset_blocks = offset_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		struct iovec *iov, int iovcnt,
		uint64_t offset, uint64_t nbytes,
		spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
}

int spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
			   struct iovec *iov, int iovcnt,
			   uint64_t offset_blocks, uint64_t num_blocks,
			   spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
	bdev_io->u.bdev.iovs = iov;
	bdev_io->u.bdev.iovcnt = iovcnt;
	bdev_io->u.bdev.num_blocks = num_blocks;
	bdev_io->u.bdev.offset_blocks = offset_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		void *buf, uint64_t offset, uint64_t nbytes,
		spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
}

int
spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		       void *buf, uint64_t offset_blocks, uint64_t num_blocks,
		       spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		return -EBADF;
	}

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
	bdev_io->u.bdev.iovs = &bdev_io->iov;
	bdev_io->u.bdev.iovs[0].iov_base = buf;
	bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
	bdev_io->u.bdev.iovcnt = 1;
	bdev_io->u.bdev.num_blocks = num_blocks;
	bdev_io->u.bdev.offset_blocks = offset_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		 struct iovec *iov, int iovcnt,
		 uint64_t offset, uint64_t len,
		 spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, len, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
}

int
spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
			struct iovec *iov, int iovcnt,
			uint64_t offset_blocks, uint64_t num_blocks,
			spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		return -EBADF;
	}

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
	bdev_io->u.bdev.iovs = iov;
	bdev_io->u.bdev.iovcnt = iovcnt;
	bdev_io->u.bdev.num_blocks = num_blocks;
	bdev_io->u.bdev.offset_blocks = offset_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		       uint64_t offset, uint64_t len,
		       spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, len, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
}

int
spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
			      uint64_t offset_blocks, uint64_t num_blocks,
			      spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		return -EBADF;
	}

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);

	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->u.bdev.offset_blocks = offset_blocks;
	bdev_io->u.bdev.num_blocks = num_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	if (_spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
		spdk_bdev_io_submit(bdev_io);
		return 0;
	} else if (_spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
		assert(spdk_bdev_get_block_size(bdev) <= ZERO_BUFFER_SIZE);
		bdev_io->u.bdev.split_remaining_num_blocks = num_blocks;
		bdev_io->u.bdev.split_current_offset_blocks = offset_blocks;
		_spdk_bdev_write_zero_buffer_next(bdev_io);
		return 0;
	} else {
		spdk_bdev_free_io(bdev_io);
		return -ENOTSUP;
	}
}

int
spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		uint64_t offset, uint64_t nbytes,
		spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
}

int
spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		       uint64_t offset_blocks, uint64_t num_blocks,
		       spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		return -EBADF;
	}

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	if (num_blocks == 0) {
		SPDK_ERRLOG("Can't unmap 0 bytes\n");
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;

	bdev_io->u.bdev.iovs = &bdev_io->iov;
	bdev_io->u.bdev.iovs[0].iov_base = NULL;
	bdev_io->u.bdev.iovs[0].iov_len = 0;
	bdev_io->u.bdev.iovcnt = 1;

	bdev_io->u.bdev.offset_blocks = offset_blocks;
	bdev_io->u.bdev.num_blocks = num_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		uint64_t offset, uint64_t length,
		spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	uint64_t offset_blocks, num_blocks;

	if (spdk_bdev_bytes_to_blocks(desc->bdev, offset, &offset_blocks, length, &num_blocks) != 0) {
		return -EINVAL;
	}

	return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
}

int
spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		       uint64_t offset_blocks, uint64_t num_blocks,
		       spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		return -EBADF;
	}

	if (!spdk_bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
		return -EINVAL;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
	bdev_io->u.bdev.iovs = NULL;
	bdev_io->u.bdev.iovcnt = 0;
	bdev_io->u.bdev.offset_blocks = offset_blocks;
	bdev_io->u.bdev.num_blocks = num_blocks;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

static void
_spdk_bdev_reset_dev(struct spdk_io_channel_iter *i, int status)
{
	struct spdk_bdev_channel *ch = spdk_io_channel_iter_get_ctx(i);
	struct spdk_bdev_io *bdev_io;

	bdev_io = TAILQ_FIRST(&ch->queued_resets);
	TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
	spdk_bdev_io_submit_reset(bdev_io);
}

static void
_spdk_bdev_reset_freeze_channel(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel		*ch;
	struct spdk_bdev_channel	*channel;
	struct spdk_bdev_mgmt_channel	*mgmt_channel;
	struct spdk_bdev_shared_resource *shared_resource;
	bdev_io_tailq_t			tmp_queued;

	TAILQ_INIT(&tmp_queued);

	ch = spdk_io_channel_iter_get_channel(i);
	channel = spdk_io_channel_get_ctx(ch);
	shared_resource = channel->shared_resource;
	mgmt_channel = shared_resource->mgmt_ch;

	channel->flags |= BDEV_CH_RESET_IN_PROGRESS;

	if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
		/* The QoS object is always valid and readable while
		 * the channel flag is set, so the lock here should not
		 * be necessary. We're not in the fast path though, so
		 * just take it anyway. */
		pthread_mutex_lock(&channel->bdev->internal.mutex);
		if (channel->bdev->internal.qos->ch == channel) {
			TAILQ_SWAP(&channel->bdev->internal.qos->queued, &tmp_queued, spdk_bdev_io, internal.link);
		}
		pthread_mutex_unlock(&channel->bdev->internal.mutex);
	}

	_spdk_bdev_abort_queued_io(&shared_resource->nomem_io, channel);
	_spdk_bdev_abort_buf_io(&mgmt_channel->need_buf_small, channel);
	_spdk_bdev_abort_buf_io(&mgmt_channel->need_buf_large, channel);
	_spdk_bdev_abort_queued_io(&tmp_queued, channel);

	spdk_for_each_channel_continue(i, 0);
}

static void
_spdk_bdev_start_reset(void *ctx)
{
	struct spdk_bdev_channel *ch = ctx;

	spdk_for_each_channel(__bdev_to_io_dev(ch->bdev), _spdk_bdev_reset_freeze_channel,
			      ch, _spdk_bdev_reset_dev);
}

static void
_spdk_bdev_channel_start_reset(struct spdk_bdev_channel *ch)
{
	struct spdk_bdev *bdev = ch->bdev;

	assert(!TAILQ_EMPTY(&ch->queued_resets));

	pthread_mutex_lock(&bdev->internal.mutex);
	if (bdev->internal.reset_in_progress == NULL) {
		bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
		/*
		 * Take a channel reference for the target bdev for the life of this
		 *  reset.  This guards against the channel getting destroyed while
		 *  spdk_for_each_channel() calls related to this reset IO are in
		 *  progress.  We will release the reference when this reset is
		 *  completed.
		 */
		bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
		_spdk_bdev_start_reset(ch);
	}
	pthread_mutex_unlock(&bdev->internal.mutex);
}

int
spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
		spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
	bdev_io->u.reset.ch_ref = NULL;
	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	pthread_mutex_lock(&bdev->internal.mutex);
	TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
	pthread_mutex_unlock(&bdev->internal.mutex);

	_spdk_bdev_channel_start_reset(channel);

	return 0;
}

void
spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
		      struct spdk_bdev_io_stat *stat)
{
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	*stat = channel->stat;
}

static void
_spdk_bdev_get_device_stat_done(struct spdk_io_channel_iter *i, int status)
{
	void *io_device = spdk_io_channel_iter_get_io_device(i);
	struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = spdk_io_channel_iter_get_ctx(i);

	bdev_iostat_ctx->cb(__bdev_from_io_dev(io_device), bdev_iostat_ctx->stat,
			    bdev_iostat_ctx->cb_arg, 0);
	free(bdev_iostat_ctx);
}

static void
_spdk_bdev_get_each_channel_stat(struct spdk_io_channel_iter *i)
{
	struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = spdk_io_channel_iter_get_ctx(i);
	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	_spdk_bdev_io_stat_add(bdev_iostat_ctx->stat, &channel->stat);
	spdk_for_each_channel_continue(i, 0);
}

void
spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
			  spdk_bdev_get_device_stat_cb cb, void *cb_arg)
{
	struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;

	assert(bdev != NULL);
	assert(stat != NULL);
	assert(cb != NULL);

	bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
	if (bdev_iostat_ctx == NULL) {
		SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
		cb(bdev, stat, cb_arg, -ENOMEM);
		return;
	}

	bdev_iostat_ctx->stat = stat;
	bdev_iostat_ctx->cb = cb;
	bdev_iostat_ctx->cb_arg = cb_arg;

	/* Start with the statistics from previously deleted channels. */
	pthread_mutex_lock(&bdev->internal.mutex);
	_spdk_bdev_io_stat_add(bdev_iostat_ctx->stat, &bdev->internal.stat);
	pthread_mutex_unlock(&bdev->internal.mutex);

	/* Then iterate and add the statistics from each existing channel. */
	spdk_for_each_channel(__bdev_to_io_dev(bdev),
			      _spdk_bdev_get_each_channel_stat,
			      bdev_iostat_ctx,
			      _spdk_bdev_get_device_stat_done);
}

int
spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
			      const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
			      spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		return -EBADF;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
	bdev_io->u.nvme_passthru.cmd = *cmd;
	bdev_io->u.nvme_passthru.buf = buf;
	bdev_io->u.nvme_passthru.nbytes = nbytes;
	bdev_io->u.nvme_passthru.md_buf = NULL;
	bdev_io->u.nvme_passthru.md_len = 0;

	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
			   const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
			   spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		/*
		 * Do not try to parse the NVMe command - we could maybe use bits in the opcode
		 *  to easily determine if the command is a read or write, but for now just
		 *  do not allow io_passthru with a read-only descriptor.
		 */
		return -EBADF;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
	bdev_io->u.nvme_passthru.cmd = *cmd;
	bdev_io->u.nvme_passthru.buf = buf;
	bdev_io->u.nvme_passthru.nbytes = nbytes;
	bdev_io->u.nvme_passthru.md_buf = NULL;
	bdev_io->u.nvme_passthru.md_len = 0;

	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
			      const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
			      spdk_bdev_io_completion_cb cb, void *cb_arg)
{
	struct spdk_bdev *bdev = desc->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);

	if (!desc->write) {
		/*
		 * Do not try to parse the NVMe command - we could maybe use bits in the opcode
		 *  to easily determine if the command is a read or write, but for now just
		 *  do not allow io_passthru with a read-only descriptor.
		 */
		return -EBADF;
	}

	bdev_io = spdk_bdev_get_io(channel);
	if (!bdev_io) {
		return -ENOMEM;
	}

	bdev_io->internal.ch = channel;
	bdev_io->internal.desc = desc;
	bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
	bdev_io->u.nvme_passthru.cmd = *cmd;
	bdev_io->u.nvme_passthru.buf = buf;
	bdev_io->u.nvme_passthru.nbytes = nbytes;
	bdev_io->u.nvme_passthru.md_buf = md_buf;
	bdev_io->u.nvme_passthru.md_len = md_len;

	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);

	spdk_bdev_io_submit(bdev_io);
	return 0;
}

int
spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
			struct spdk_bdev_io_wait_entry *entry)
{
	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
	struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;

	if (bdev != entry->bdev) {
		SPDK_ERRLOG("bdevs do not match\n");
		return -EINVAL;
	}

	if (mgmt_ch->per_thread_cache_count > 0) {
		SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
		return -EINVAL;
	}

	TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
	return 0;
}

static void
_spdk_bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
{
	struct spdk_bdev *bdev = bdev_ch->bdev;
	struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
	struct spdk_bdev_io *bdev_io;

	if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
		/*
		 * Allow some more I/O to complete before retrying the nomem_io queue.
		 *  Some drivers (such as nvme) cannot immediately take a new I/O in
		 *  the context of a completion, because the resources for the I/O are
		 *  not released until control returns to the bdev poller.  Also, we
		 *  may require several small I/O to complete before a larger I/O
		 *  (that requires splitting) can be submitted.
		 */
		return;
	}

	while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
		bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
		TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
		bdev_io->internal.ch->io_outstanding++;
		shared_resource->io_outstanding++;
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
		bdev->fn_table->submit_request(bdev_io->internal.ch->channel, bdev_io);
		if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
			break;
		}
	}
}

static inline void
_spdk_bdev_io_complete(void *ctx)
{
	struct spdk_bdev_io *bdev_io = ctx;
	uint64_t tsc, tsc_diff;

	if (spdk_unlikely(bdev_io->internal.in_submit_request || bdev_io->internal.io_submit_ch)) {
		/*
		 * Send the completion to the thread that originally submitted the I/O,
		 * which may not be the current thread in the case of QoS.
		 */
		if (bdev_io->internal.io_submit_ch) {
			bdev_io->internal.ch = bdev_io->internal.io_submit_ch;
			bdev_io->internal.io_submit_ch = NULL;
		}

		/*
		 * Defer completion to avoid potential infinite recursion if the
		 * user's completion callback issues a new I/O.
		 */
		spdk_thread_send_msg(spdk_io_channel_get_thread(bdev_io->internal.ch->channel),
				     _spdk_bdev_io_complete, bdev_io);
		return;
	}

	tsc = spdk_get_ticks();
	tsc_diff = tsc - bdev_io->internal.submit_tsc;
	spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, 0, 0, (uintptr_t)bdev_io, 0);

	if (bdev_io->internal.ch->histogram) {
		spdk_histogram_data_tally(bdev_io->internal.ch->histogram, tsc_diff);
	}

	if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
		switch (bdev_io->type) {
		case SPDK_BDEV_IO_TYPE_READ:
			bdev_io->internal.ch->stat.bytes_read += bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
			bdev_io->internal.ch->stat.num_read_ops++;
			bdev_io->internal.ch->stat.read_latency_ticks += tsc_diff;
			break;
		case SPDK_BDEV_IO_TYPE_WRITE:
			bdev_io->internal.ch->stat.bytes_written += bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
			bdev_io->internal.ch->stat.num_write_ops++;
			bdev_io->internal.ch->stat.write_latency_ticks += tsc_diff;
			break;
		case SPDK_BDEV_IO_TYPE_UNMAP:
			bdev_io->internal.ch->stat.bytes_unmapped += bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
			bdev_io->internal.ch->stat.num_unmap_ops++;
			bdev_io->internal.ch->stat.unmap_latency_ticks += tsc_diff;
		default:
			break;
		}
	}

#ifdef SPDK_CONFIG_VTUNE
	uint64_t now_tsc = spdk_get_ticks();
	if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
		uint64_t data[5];

		data[0] = bdev_io->internal.ch->stat.num_read_ops - bdev_io->internal.ch->prev_stat.num_read_ops;
		data[1] = bdev_io->internal.ch->stat.bytes_read - bdev_io->internal.ch->prev_stat.bytes_read;
		data[2] = bdev_io->internal.ch->stat.num_write_ops - bdev_io->internal.ch->prev_stat.num_write_ops;
		data[3] = bdev_io->internal.ch->stat.bytes_written - bdev_io->internal.ch->prev_stat.bytes_written;
		data[4] = bdev_io->bdev->fn_table->get_spin_time ?
			  bdev_io->bdev->fn_table->get_spin_time(bdev_io->internal.ch->channel) : 0;

		__itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
				   __itt_metadata_u64, 5, data);

		bdev_io->internal.ch->prev_stat = bdev_io->internal.ch->stat;
		bdev_io->internal.ch->start_tsc = now_tsc;
	}
#endif

	assert(bdev_io->internal.cb != NULL);
	assert(spdk_get_thread() == spdk_io_channel_get_thread(bdev_io->internal.ch->channel));

	bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
			     bdev_io->internal.caller_ctx);
}

static void
_spdk_bdev_reset_complete(struct spdk_io_channel_iter *i, int status)
{
	struct spdk_bdev_io *bdev_io = spdk_io_channel_iter_get_ctx(i);

	if (bdev_io->u.reset.ch_ref != NULL) {
		spdk_put_io_channel(bdev_io->u.reset.ch_ref);
		bdev_io->u.reset.ch_ref = NULL;
	}

	_spdk_bdev_io_complete(bdev_io);
}

static void
_spdk_bdev_unfreeze_channel(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *ch = spdk_io_channel_get_ctx(_ch);

	ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
	if (!TAILQ_EMPTY(&ch->queued_resets)) {
		_spdk_bdev_channel_start_reset(ch);
	}

	spdk_for_each_channel_continue(i, 0);
}

void
spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
{
	struct spdk_bdev *bdev = bdev_io->bdev;
	struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
	struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;

	bdev_io->internal.status = status;

	if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
		bool unlock_channels = false;

		if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
			SPDK_ERRLOG("NOMEM returned for reset\n");
		}
		pthread_mutex_lock(&bdev->internal.mutex);
		if (bdev_io == bdev->internal.reset_in_progress) {
			bdev->internal.reset_in_progress = NULL;
			unlock_channels = true;
		}
		pthread_mutex_unlock(&bdev->internal.mutex);

		if (unlock_channels) {
			spdk_for_each_channel(__bdev_to_io_dev(bdev), _spdk_bdev_unfreeze_channel,
					      bdev_io, _spdk_bdev_reset_complete);
			return;
		}
	} else {
		if (spdk_unlikely(bdev_io->internal.orig_iovcnt > 0)) {
			_bdev_io_unset_bounce_buf(bdev_io);
		}

		assert(bdev_ch->io_outstanding > 0);
		assert(shared_resource->io_outstanding > 0);
		bdev_ch->io_outstanding--;
		shared_resource->io_outstanding--;

		if (spdk_unlikely(status == SPDK_BDEV_IO_STATUS_NOMEM)) {
			TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
			/*
			 * Wait for some of the outstanding I/O to complete before we
			 *  retry any of the nomem_io.  Normally we will wait for
			 *  NOMEM_THRESHOLD_COUNT I/O to complete but for low queue
			 *  depth channels we will instead wait for half to complete.
			 */
			shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
							   (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
			return;
		}

		if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
			_spdk_bdev_ch_retry_io(bdev_ch);
		}
	}

	_spdk_bdev_io_complete(bdev_io);
}

void
spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
				  enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
{
	if (sc == SPDK_SCSI_STATUS_GOOD) {
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
	} else {
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
		bdev_io->internal.error.scsi.sc = sc;
		bdev_io->internal.error.scsi.sk = sk;
		bdev_io->internal.error.scsi.asc = asc;
		bdev_io->internal.error.scsi.ascq = ascq;
	}

	spdk_bdev_io_complete(bdev_io, bdev_io->internal.status);
}

void
spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
			     int *sc, int *sk, int *asc, int *ascq)
{
	assert(sc != NULL);
	assert(sk != NULL);
	assert(asc != NULL);
	assert(ascq != NULL);

	switch (bdev_io->internal.status) {
	case SPDK_BDEV_IO_STATUS_SUCCESS:
		*sc = SPDK_SCSI_STATUS_GOOD;
		*sk = SPDK_SCSI_SENSE_NO_SENSE;
		*asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
		*ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
		break;
	case SPDK_BDEV_IO_STATUS_NVME_ERROR:
		spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
		break;
	case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
		*sc = bdev_io->internal.error.scsi.sc;
		*sk = bdev_io->internal.error.scsi.sk;
		*asc = bdev_io->internal.error.scsi.asc;
		*ascq = bdev_io->internal.error.scsi.ascq;
		break;
	default:
		*sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
		*sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
		*asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
		*ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
		break;
	}
}

void
spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, int sct, int sc)
{
	if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS) {
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
	} else {
		bdev_io->internal.error.nvme.sct = sct;
		bdev_io->internal.error.nvme.sc = sc;
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
	}

	spdk_bdev_io_complete(bdev_io, bdev_io->internal.status);
}

void
spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, int *sct, int *sc)
{
	assert(sct != NULL);
	assert(sc != NULL);

	if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
		*sct = bdev_io->internal.error.nvme.sct;
		*sc = bdev_io->internal.error.nvme.sc;
	} else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
		*sct = SPDK_NVME_SCT_GENERIC;
		*sc = SPDK_NVME_SC_SUCCESS;
	} else {
		*sct = SPDK_NVME_SCT_GENERIC;
		*sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
	}
}

struct spdk_thread *
spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
{
	return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
}

static void
_spdk_bdev_qos_config_limit(struct spdk_bdev *bdev, uint64_t *limits)
{
	uint64_t	min_qos_set;
	int		i;

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
			break;
		}
	}

	if (i == SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES) {
		SPDK_ERRLOG("Invalid rate limits set.\n");
		return;
	}

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
			continue;
		}

		if (_spdk_bdev_qos_is_iops_rate_limit(i) == true) {
			min_qos_set = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
		} else {
			min_qos_set = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
		}

		if (limits[i] == 0 || limits[i] % min_qos_set) {
			SPDK_ERRLOG("Assigned limit %" PRIu64 " on bdev %s is not multiple of %" PRIu64 "\n",
				    limits[i], bdev->name, min_qos_set);
			SPDK_ERRLOG("Failed to enable QoS on this bdev %s\n", bdev->name);
			return;
		}
	}

	if (!bdev->internal.qos) {
		bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
		if (!bdev->internal.qos) {
			SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
			return;
		}
	}

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		bdev->internal.qos->rate_limits[i].limit = limits[i];
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Bdev:%s QoS type:%d set:%lu\n",
			      bdev->name, i, limits[i]);
	}

	return;
}

static void
_spdk_bdev_qos_config(struct spdk_bdev *bdev)
{
	struct spdk_conf_section	*sp = NULL;
	const char			*val = NULL;
	int				i = 0, j = 0;
	uint64_t			limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES] = {};
	bool				config_qos = false;

	sp = spdk_conf_find_section(NULL, "QoS");
	if (!sp) {
		return;
	}

	while (j < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES) {
		limits[j] = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;

		i = 0;
		while (true) {
			val = spdk_conf_section_get_nmval(sp, qos_conf_type[j], i, 0);
			if (!val) {
				break;
			}

			if (strcmp(bdev->name, val) != 0) {
				i++;
				continue;
			}

			val = spdk_conf_section_get_nmval(sp, qos_conf_type[j], i, 1);
			if (val) {
				if (_spdk_bdev_qos_is_iops_rate_limit(j) == true) {
					limits[j] = strtoull(val, NULL, 10);
				} else {
					limits[j] = strtoull(val, NULL, 10) * 1024 * 1024;
				}
				config_qos = true;
			}

			break;
		}

		j++;
	}

	if (config_qos == true) {
		_spdk_bdev_qos_config_limit(bdev, limits);
	}

	return;
}

static int
spdk_bdev_init(struct spdk_bdev *bdev)
{
	char *bdev_name;

	assert(bdev->module != NULL);

	if (!bdev->name) {
		SPDK_ERRLOG("Bdev name is NULL\n");
		return -EINVAL;
	}

	if (spdk_bdev_get_by_name(bdev->name)) {
		SPDK_ERRLOG("Bdev name:%s already exists\n", bdev->name);
		return -EEXIST;
	}

	/* Users often register their own I/O devices using the bdev name. In
	 * order to avoid conflicts, prepend bdev_. */
	bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
	if (!bdev_name) {
		SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
		return -ENOMEM;
	}

	bdev->internal.status = SPDK_BDEV_STATUS_READY;
	bdev->internal.measured_queue_depth = UINT64_MAX;
	bdev->internal.claim_module = NULL;
	bdev->internal.qd_poller = NULL;
	bdev->internal.qos = NULL;

	if (spdk_bdev_get_buf_align(bdev) > 1) {
		if (bdev->split_on_optimal_io_boundary) {
			bdev->optimal_io_boundary = spdk_min(bdev->optimal_io_boundary,
							     SPDK_BDEV_LARGE_BUF_MAX_SIZE / bdev->blocklen);
		} else {
			bdev->split_on_optimal_io_boundary = true;
			bdev->optimal_io_boundary = SPDK_BDEV_LARGE_BUF_MAX_SIZE / bdev->blocklen;
		}
	}

	TAILQ_INIT(&bdev->internal.open_descs);

	TAILQ_INIT(&bdev->aliases);

	bdev->internal.reset_in_progress = NULL;

	_spdk_bdev_qos_config(bdev);

	spdk_io_device_register(__bdev_to_io_dev(bdev),
				spdk_bdev_channel_create, spdk_bdev_channel_destroy,
				sizeof(struct spdk_bdev_channel),
				bdev_name);

	free(bdev_name);

	pthread_mutex_init(&bdev->internal.mutex, NULL);
	return 0;
}

static void
spdk_bdev_destroy_cb(void *io_device)
{
	int			rc;
	struct spdk_bdev	*bdev;
	spdk_bdev_unregister_cb	cb_fn;
	void			*cb_arg;

	bdev = __bdev_from_io_dev(io_device);
	cb_fn = bdev->internal.unregister_cb;
	cb_arg = bdev->internal.unregister_ctx;

	rc = bdev->fn_table->destruct(bdev->ctxt);
	if (rc < 0) {
		SPDK_ERRLOG("destruct failed\n");
	}
	if (rc <= 0 && cb_fn != NULL) {
		cb_fn(cb_arg, rc);
	}
}


static void
spdk_bdev_fini(struct spdk_bdev *bdev)
{
	pthread_mutex_destroy(&bdev->internal.mutex);

	free(bdev->internal.qos);

	spdk_io_device_unregister(__bdev_to_io_dev(bdev), spdk_bdev_destroy_cb);
}

static void
spdk_bdev_start(struct spdk_bdev *bdev)
{
	struct spdk_bdev_module *module;
	uint32_t action;

	SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Inserting bdev %s into list\n", bdev->name);
	TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);

	/* Examine configuration before initializing I/O */
	TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (module->examine_config) {
			action = module->internal.action_in_progress;
			module->internal.action_in_progress++;
			module->examine_config(bdev);
			if (action != module->internal.action_in_progress) {
				SPDK_ERRLOG("examine_config for module %s did not call spdk_bdev_module_examine_done()\n",
					    module->name);
			}
		}
	}

	if (bdev->internal.claim_module) {
		return;
	}

	TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (module->examine_disk) {
			module->internal.action_in_progress++;
			module->examine_disk(bdev);
		}
	}
}

int
spdk_bdev_register(struct spdk_bdev *bdev)
{
	int rc = spdk_bdev_init(bdev);

	if (rc == 0) {
		spdk_bdev_start(bdev);
	}

	return rc;
}

int
spdk_vbdev_register(struct spdk_bdev *vbdev, struct spdk_bdev **base_bdevs, int base_bdev_count)
{
	int rc;

	rc = spdk_bdev_init(vbdev);
	if (rc) {
		return rc;
	}

	spdk_bdev_start(vbdev);
	return 0;
}

void
spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
{
	if (bdev->internal.unregister_cb != NULL) {
		bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
	}
}

static void
_remove_notify(void *arg)
{
	struct spdk_bdev_desc *desc = arg;

	desc->remove_scheduled = false;

	if (desc->closed) {
		free(desc);
	} else {
		desc->remove_cb(desc->remove_ctx);
	}
}

void
spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
{
	struct spdk_bdev_desc	*desc, *tmp;
	bool			do_destruct = true;
	struct spdk_thread	*thread;

	SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Removing bdev %s from list\n", bdev->name);

	thread = spdk_get_thread();
	if (!thread) {
		/* The user called this from a non-SPDK thread. */
		if (cb_fn != NULL) {
			cb_fn(cb_arg, -ENOTSUP);
		}
		return;
	}

	pthread_mutex_lock(&bdev->internal.mutex);

	bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
	bdev->internal.unregister_cb = cb_fn;
	bdev->internal.unregister_ctx = cb_arg;

	TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
		if (desc->remove_cb) {
			do_destruct = false;
			/*
			 * Defer invocation of the remove_cb to a separate message that will
			 *  run later on its thread.  This ensures this context unwinds and
			 *  we don't recursively unregister this bdev again if the remove_cb
			 *  immediately closes its descriptor.
			 */
			if (!desc->remove_scheduled) {
				/* Avoid scheduling removal of the same descriptor multiple times. */
				desc->remove_scheduled = true;
				spdk_thread_send_msg(desc->thread, _remove_notify, desc);
			}
		}
	}

	if (!do_destruct) {
		pthread_mutex_unlock(&bdev->internal.mutex);
		return;
	}

	TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
	pthread_mutex_unlock(&bdev->internal.mutex);

	spdk_bdev_fini(bdev);
}

int
spdk_bdev_open(struct spdk_bdev *bdev, bool write, spdk_bdev_remove_cb_t remove_cb,
	       void *remove_ctx, struct spdk_bdev_desc **_desc)
{
	struct spdk_bdev_desc *desc;
	struct spdk_thread *thread;

	thread = spdk_get_thread();
	if (!thread) {
		SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
		return -ENOTSUP;
	}

	desc = calloc(1, sizeof(*desc));
	if (desc == NULL) {
		SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
		return -ENOMEM;
	}

	SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
		      spdk_get_thread());

	desc->bdev = bdev;
	desc->thread = thread;
	desc->remove_cb = remove_cb;
	desc->remove_ctx = remove_ctx;
	desc->write = write;
	*_desc = desc;

	pthread_mutex_lock(&bdev->internal.mutex);

	if (write && bdev->internal.claim_module) {
		SPDK_ERRLOG("Could not open %s - %s module already claimed it\n",
			    bdev->name, bdev->internal.claim_module->name);
		pthread_mutex_unlock(&bdev->internal.mutex);
		free(desc);
		*_desc = NULL;
		return -EPERM;
	}

	TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);

	pthread_mutex_unlock(&bdev->internal.mutex);

	return 0;
}

void
spdk_bdev_close(struct spdk_bdev_desc *desc)
{
	struct spdk_bdev *bdev = desc->bdev;
	bool do_unregister = false;

	SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
		      spdk_get_thread());

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

	pthread_mutex_lock(&bdev->internal.mutex);

	TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);

	desc->closed = true;

	if (!desc->remove_scheduled) {
		free(desc);
	}

	/* If no more descriptors, kill QoS channel */
	if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
		SPDK_DEBUGLOG(SPDK_LOG_BDEV, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
			      bdev->name, spdk_get_thread());

		if (spdk_bdev_qos_destroy(bdev)) {
			/* There isn't anything we can do to recover here. Just let the
			 * old QoS poller keep running. The QoS handling won't change
			 * cores when the user allocates a new channel, but it won't break. */
			SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
		}
	}

	spdk_bdev_set_qd_sampling_period(bdev, 0);

	if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
		do_unregister = true;
	}
	pthread_mutex_unlock(&bdev->internal.mutex);

	if (do_unregister == true) {
		spdk_bdev_unregister(bdev, bdev->internal.unregister_cb, bdev->internal.unregister_ctx);
	}
}

int
spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
			    struct spdk_bdev_module *module)
{
	if (bdev->internal.claim_module != NULL) {
		SPDK_ERRLOG("bdev %s already claimed by module %s\n", bdev->name,
			    bdev->internal.claim_module->name);
		return -EPERM;
	}

	if (desc && !desc->write) {
		desc->write = true;
	}

	bdev->internal.claim_module = module;
	return 0;
}

void
spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
{
	assert(bdev->internal.claim_module != NULL);
	bdev->internal.claim_module = NULL;
}

struct spdk_bdev *
spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
{
	return desc->bdev;
}

void
spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
{
	struct iovec *iovs;
	int iovcnt;

	if (bdev_io == NULL) {
		return;
	}

	switch (bdev_io->type) {
	case SPDK_BDEV_IO_TYPE_READ:
		iovs = bdev_io->u.bdev.iovs;
		iovcnt = bdev_io->u.bdev.iovcnt;
		break;
	case SPDK_BDEV_IO_TYPE_WRITE:
		iovs = bdev_io->u.bdev.iovs;
		iovcnt = bdev_io->u.bdev.iovcnt;
		break;
	default:
		iovs = NULL;
		iovcnt = 0;
		break;
	}

	if (iovp) {
		*iovp = iovs;
	}
	if (iovcntp) {
		*iovcntp = iovcnt;
	}
}

void
spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
{

	if (spdk_bdev_module_list_find(bdev_module->name)) {
		SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
		assert(false);
	}

	if (bdev_module->async_init) {
		bdev_module->internal.action_in_progress = 1;
	}

	/*
	 * Modules with examine callbacks must be initialized first, so they are
	 *  ready to handle examine callbacks from later modules that will
	 *  register physical bdevs.
	 */
	if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
		TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
	} else {
		TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
	}
}

struct spdk_bdev_module *
spdk_bdev_module_list_find(const char *name)
{
	struct spdk_bdev_module *bdev_module;

	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
		if (strcmp(name, bdev_module->name) == 0) {
			break;
		}
	}

	return bdev_module;
}

static void
_spdk_bdev_write_zero_buffer_next(void *_bdev_io)
{
	struct spdk_bdev_io *bdev_io = _bdev_io;
	uint64_t num_bytes, num_blocks;
	int rc;

	num_bytes = spdk_min(spdk_bdev_get_block_size(bdev_io->bdev) *
			     bdev_io->u.bdev.split_remaining_num_blocks,
			     ZERO_BUFFER_SIZE);
	num_blocks = num_bytes / spdk_bdev_get_block_size(bdev_io->bdev);

	rc = spdk_bdev_write_blocks(bdev_io->internal.desc,
				    spdk_io_channel_from_ctx(bdev_io->internal.ch),
				    g_bdev_mgr.zero_buffer,
				    bdev_io->u.bdev.split_current_offset_blocks, num_blocks,
				    _spdk_bdev_write_zero_buffer_done, bdev_io);
	if (rc == 0) {
		bdev_io->u.bdev.split_remaining_num_blocks -= num_blocks;
		bdev_io->u.bdev.split_current_offset_blocks += num_blocks;
	} else if (rc == -ENOMEM) {
		_spdk_bdev_queue_io_wait_with_cb(bdev_io, _spdk_bdev_write_zero_buffer_next);
	} else {
		bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
		bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
	}
}

static void
_spdk_bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
{
	struct spdk_bdev_io *parent_io = cb_arg;

	spdk_bdev_free_io(bdev_io);

	if (!success) {
		parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
		parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
		return;
	}

	if (parent_io->u.bdev.split_remaining_num_blocks == 0) {
		parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
		parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
		return;
	}

	_spdk_bdev_write_zero_buffer_next(parent_io);
}

struct set_qos_limit_ctx {
	void (*cb_fn)(void *cb_arg, int status);
	void *cb_arg;
	struct spdk_bdev *bdev;
};

static void
_spdk_bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
{
	pthread_mutex_lock(&ctx->bdev->internal.mutex);
	ctx->bdev->internal.qos_mod_in_progress = false;
	pthread_mutex_unlock(&ctx->bdev->internal.mutex);

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

static void
_spdk_bdev_disable_qos_done(void *cb_arg)
{
	struct set_qos_limit_ctx *ctx = cb_arg;
	struct spdk_bdev *bdev = ctx->bdev;
	struct spdk_bdev_io *bdev_io;
	struct spdk_bdev_qos *qos;

	pthread_mutex_lock(&bdev->internal.mutex);
	qos = bdev->internal.qos;
	bdev->internal.qos = NULL;
	pthread_mutex_unlock(&bdev->internal.mutex);

	while (!TAILQ_EMPTY(&qos->queued)) {
		/* Send queued I/O back to their original thread for resubmission. */
		bdev_io = TAILQ_FIRST(&qos->queued);
		TAILQ_REMOVE(&qos->queued, bdev_io, internal.link);

		if (bdev_io->internal.io_submit_ch) {
			/*
			 * Channel was changed when sending it to the QoS thread - change it back
			 *  before sending it back to the original thread.
			 */
			bdev_io->internal.ch = bdev_io->internal.io_submit_ch;
			bdev_io->internal.io_submit_ch = NULL;
		}

		spdk_thread_send_msg(spdk_io_channel_get_thread(bdev_io->internal.ch->channel),
				     _spdk_bdev_io_submit, bdev_io);
	}

	if (qos->thread != NULL) {
		spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
		spdk_poller_unregister(&qos->poller);
	}

	free(qos);

	_spdk_bdev_set_qos_limit_done(ctx, 0);
}

static void
_spdk_bdev_disable_qos_msg_done(struct spdk_io_channel_iter *i, int status)
{
	void *io_device = spdk_io_channel_iter_get_io_device(i);
	struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
	struct set_qos_limit_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
	struct spdk_thread *thread;

	pthread_mutex_lock(&bdev->internal.mutex);
	thread = bdev->internal.qos->thread;
	pthread_mutex_unlock(&bdev->internal.mutex);

	if (thread != NULL) {
		spdk_thread_send_msg(thread, _spdk_bdev_disable_qos_done, ctx);
	} else {
		_spdk_bdev_disable_qos_done(ctx);
	}
}

static void
_spdk_bdev_disable_qos_msg(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *bdev_ch = spdk_io_channel_get_ctx(ch);

	bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;

	spdk_for_each_channel_continue(i, 0);
}

static void
_spdk_bdev_update_qos_rate_limit_msg(void *cb_arg)
{
	struct set_qos_limit_ctx *ctx = cb_arg;
	struct spdk_bdev *bdev = ctx->bdev;

	pthread_mutex_lock(&bdev->internal.mutex);
	spdk_bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
	pthread_mutex_unlock(&bdev->internal.mutex);

	_spdk_bdev_set_qos_limit_done(ctx, 0);
}

static void
_spdk_bdev_enable_qos_msg(struct spdk_io_channel_iter *i)
{
	void *io_device = spdk_io_channel_iter_get_io_device(i);
	struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
	struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *bdev_ch = spdk_io_channel_get_ctx(ch);

	pthread_mutex_lock(&bdev->internal.mutex);
	_spdk_bdev_enable_qos(bdev, bdev_ch);
	pthread_mutex_unlock(&bdev->internal.mutex);
	spdk_for_each_channel_continue(i, 0);
}

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

	_spdk_bdev_set_qos_limit_done(ctx, status);
}

static void
_spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
{
	int i;

	assert(bdev->internal.qos != NULL);

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
			bdev->internal.qos->rate_limits[i].limit = limits[i];

			if (limits[i] == 0) {
				bdev->internal.qos->rate_limits[i].limit =
					SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
			}
		}
	}
}

void
spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
			      void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
{
	struct set_qos_limit_ctx	*ctx;
	uint32_t			limit_set_complement;
	uint64_t			min_limit_per_sec;
	int				i;
	bool				disable_rate_limit = true;

	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
		if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
			continue;
		}

		if (limits[i] > 0) {
			disable_rate_limit = false;
		}

		if (_spdk_bdev_qos_is_iops_rate_limit(i) == true) {
			min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
		} else {
			/* Change from megabyte to byte rate limit */
			limits[i] = limits[i] * 1024 * 1024;
			min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
		}

		limit_set_complement = limits[i] % min_limit_per_sec;
		if (limit_set_complement) {
			SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
				    limits[i], min_limit_per_sec);
			limits[i] += min_limit_per_sec - limit_set_complement;
			SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
		}
	}

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

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

	pthread_mutex_lock(&bdev->internal.mutex);
	if (bdev->internal.qos_mod_in_progress) {
		pthread_mutex_unlock(&bdev->internal.mutex);
		free(ctx);
		cb_fn(cb_arg, -EAGAIN);
		return;
	}
	bdev->internal.qos_mod_in_progress = true;

	if (disable_rate_limit == true && bdev->internal.qos) {
		for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
			if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
			    (bdev->internal.qos->rate_limits[i].limit > 0 &&
			     bdev->internal.qos->rate_limits[i].limit !=
			     SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
				disable_rate_limit = false;
				break;
			}
		}
	}

	if (disable_rate_limit == false) {
		if (bdev->internal.qos == NULL) {
			bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
			if (!bdev->internal.qos) {
				pthread_mutex_unlock(&bdev->internal.mutex);
				SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
				free(ctx);
				cb_fn(cb_arg, -ENOMEM);
				return;
			}
		}

		if (bdev->internal.qos->thread == NULL) {
			/* Enabling */
			_spdk_bdev_set_qos_rate_limits(bdev, limits);

			spdk_for_each_channel(__bdev_to_io_dev(bdev),
					      _spdk_bdev_enable_qos_msg, ctx,
					      _spdk_bdev_enable_qos_done);
		} else {
			/* Updating */
			_spdk_bdev_set_qos_rate_limits(bdev, limits);

			spdk_thread_send_msg(bdev->internal.qos->thread,
					     _spdk_bdev_update_qos_rate_limit_msg, ctx);
		}
	} else {
		if (bdev->internal.qos != NULL) {
			_spdk_bdev_set_qos_rate_limits(bdev, limits);

			/* Disabling */
			spdk_for_each_channel(__bdev_to_io_dev(bdev),
					      _spdk_bdev_disable_qos_msg, ctx,
					      _spdk_bdev_disable_qos_msg_done);
		} else {
			pthread_mutex_unlock(&bdev->internal.mutex);
			_spdk_bdev_set_qos_limit_done(ctx, 0);
			return;
		}
	}

	pthread_mutex_unlock(&bdev->internal.mutex);
}

struct spdk_bdev_histogram_ctx {
	spdk_bdev_histogram_status_cb cb_fn;
	void *cb_arg;
	struct spdk_bdev *bdev;
	int status;
};

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

	pthread_mutex_lock(&ctx->bdev->internal.mutex);
	ctx->bdev->internal.histogram_in_progress = false;
	pthread_mutex_unlock(&ctx->bdev->internal.mutex);
	ctx->cb_fn(ctx->cb_arg, ctx->status);
	free(ctx);
}

static void
_spdk_bdev_histogram_disable_channel(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *ch = spdk_io_channel_get_ctx(_ch);

	if (ch->histogram != NULL) {
		spdk_histogram_data_free(ch->histogram);
		ch->histogram = NULL;
	}
	spdk_for_each_channel_continue(i, 0);
}

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

	if (status != 0) {
		ctx->status = status;
		ctx->bdev->internal.histogram_enabled = false;
		spdk_for_each_channel(__bdev_to_io_dev(ctx->bdev), _spdk_bdev_histogram_disable_channel, ctx,
				      _spdk_bdev_histogram_disable_channel_cb);
	} else {
		pthread_mutex_lock(&ctx->bdev->internal.mutex);
		ctx->bdev->internal.histogram_in_progress = false;
		pthread_mutex_unlock(&ctx->bdev->internal.mutex);
		ctx->cb_fn(ctx->cb_arg, ctx->status);
		free(ctx);
	}
}

static void
_spdk_bdev_histogram_enable_channel(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *ch = spdk_io_channel_get_ctx(_ch);
	int status = 0;

	if (ch->histogram == NULL) {
		ch->histogram = spdk_histogram_data_alloc();
		if (ch->histogram == NULL) {
			status = -ENOMEM;
		}
	}

	spdk_for_each_channel_continue(i, status);
}

void
spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
			   void *cb_arg, bool enable)
{
	struct spdk_bdev_histogram_ctx *ctx;

	ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
	if (ctx == NULL) {
		cb_fn(cb_arg, -ENOMEM);
		return;
	}

	ctx->bdev = bdev;
	ctx->status = 0;
	ctx->cb_fn = cb_fn;
	ctx->cb_arg = cb_arg;

	pthread_mutex_lock(&bdev->internal.mutex);
	if (bdev->internal.histogram_in_progress) {
		pthread_mutex_unlock(&bdev->internal.mutex);
		free(ctx);
		cb_fn(cb_arg, -EAGAIN);
		return;
	}

	bdev->internal.histogram_in_progress = true;
	pthread_mutex_unlock(&bdev->internal.mutex);

	bdev->internal.histogram_enabled = enable;

	if (enable) {
		/* Allocate histogram for each channel */
		spdk_for_each_channel(__bdev_to_io_dev(bdev), _spdk_bdev_histogram_enable_channel, ctx,
				      _spdk_bdev_histogram_enable_channel_cb);
	} else {
		spdk_for_each_channel(__bdev_to_io_dev(bdev), _spdk_bdev_histogram_disable_channel, ctx,
				      _spdk_bdev_histogram_disable_channel_cb);
	}
}

struct spdk_bdev_histogram_data_ctx {
	spdk_bdev_histogram_data_cb cb_fn;
	void *cb_arg;
	struct spdk_bdev *bdev;
	/** merged histogram data from all channels */
	struct spdk_histogram_data	*histogram;
};

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

	ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
	free(ctx);
}

static void
_spdk_bdev_histogram_get_channel(struct spdk_io_channel_iter *i)
{
	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
	struct spdk_bdev_channel *ch = spdk_io_channel_get_ctx(_ch);
	struct spdk_bdev_histogram_data_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
	int status = 0;

	if (ch->histogram == NULL) {
		status = -EFAULT;
	} else {
		spdk_histogram_data_merge(ctx->histogram, ch->histogram);
	}

	spdk_for_each_channel_continue(i, status);
}

void
spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
			spdk_bdev_histogram_data_cb cb_fn,
			void *cb_arg)
{
	struct spdk_bdev_histogram_data_ctx *ctx;

	ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
	if (ctx == NULL) {
		cb_fn(cb_arg, -ENOMEM, NULL);
		return;
	}

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

	ctx->histogram = histogram;

	spdk_for_each_channel(__bdev_to_io_dev(bdev), _spdk_bdev_histogram_get_channel, ctx,
			      _spdk_bdev_histogram_get_channel_cb);
}

SPDK_LOG_REGISTER_COMPONENT("bdev", SPDK_LOG_BDEV)

SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
{
	spdk_trace_register_owner(OWNER_BDEV, 'b');
	spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
	spdk_trace_register_description("BDEV_IO_START", "", TRACE_BDEV_IO_START, OWNER_BDEV,
					OBJECT_BDEV_IO, 1, 0, "type:   ");
	spdk_trace_register_description("BDEV_IO_DONE", "", TRACE_BDEV_IO_DONE, OWNER_BDEV,
					OBJECT_BDEV_IO, 0, 0, "");
}