/*
 * QEMU NVM Express Controller
 *
 * Copyright (c) 2012, Intel Corporation
 *
 * Written by Keith Busch <keith.busch@intel.com>
 *
 * This code is licensed under the GNU GPL v2 or later.
 */

/**
 * Reference Specs: http://www.nvmexpress.org, 1.4, 1.3, 1.2, 1.1, 1.0e
 *
 *  https://nvmexpress.org/developers/nvme-specification/
 *
 *
 * Notes on coding style
 * ---------------------
 * While QEMU coding style prefers lowercase hexadecimals in constants, the
 * NVMe subsystem use thes format from the NVMe specifications in the comments
 * (i.e. 'h' suffix instead of '0x' prefix).
 *
 * Usage
 * -----
 * See docs/system/nvme.rst for extensive documentation.
 *
 * Add options:
 *      -drive file=<file>,if=none,id=<drive_id>
 *      -device nvme-subsys,id=<subsys_id>,nqn=<nqn_id>
 *      -device nvme,serial=<serial>,id=<bus_name>, \
 *              cmb_size_mb=<cmb_size_mb[optional]>, \
 *              [pmrdev=<mem_backend_file_id>,] \
 *              max_ioqpairs=<N[optional]>, \
 *              aerl=<N[optional]>,aer_max_queued=<N[optional]>, \
 *              mdts=<N[optional]>,vsl=<N[optional]>, \
 *              zoned.zasl=<N[optional]>, \
 *              zoned.auto_transition=<on|off[optional]>, \
 *              sriov_max_vfs=<N[optional]> \
 *              sriov_vq_flexible=<N[optional]> \
 *              sriov_vi_flexible=<N[optional]> \
 *              sriov_max_vi_per_vf=<N[optional]> \
 *              sriov_max_vq_per_vf=<N[optional]> \
 *              subsys=<subsys_id>
 *      -device nvme-ns,drive=<drive_id>,bus=<bus_name>,nsid=<nsid>,\
 *              zoned=<true|false[optional]>, \
 *              subsys=<subsys_id>,detached=<true|false[optional]>
 *
 * Note cmb_size_mb denotes size of CMB in MB. CMB is assumed to be at
 * offset 0 in BAR2 and supports only WDS, RDS and SQS for now. By default, the
 * device will use the "v1.4 CMB scheme" - use the `legacy-cmb` parameter to
 * always enable the CMBLOC and CMBSZ registers (v1.3 behavior).
 *
 * Enabling pmr emulation can be achieved by pointing to memory-backend-file.
 * For example:
 * -object memory-backend-file,id=<mem_id>,share=on,mem-path=<file_path>, \
 *  size=<size> .... -device nvme,...,pmrdev=<mem_id>
 *
 * The PMR will use BAR 4/5 exclusively.
 *
 * To place controller(s) and namespace(s) to a subsystem, then provide
 * nvme-subsys device as above.
 *
 * nvme subsystem device parameters
 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 * - `nqn`
 *   This parameter provides the `<nqn_id>` part of the string
 *   `nqn.2019-08.org.qemu:<nqn_id>` which will be reported in the SUBNQN field
 *   of subsystem controllers. Note that `<nqn_id>` should be unique per
 *   subsystem, but this is not enforced by QEMU. If not specified, it will
 *   default to the value of the `id` parameter (`<subsys_id>`).
 *
 * nvme device parameters
 * ~~~~~~~~~~~~~~~~~~~~~~
 * - `subsys`
 *   Specifying this parameter attaches the controller to the subsystem and
 *   the SUBNQN field in the controller will report the NQN of the subsystem
 *   device. This also enables multi controller capability represented in
 *   Identify Controller data structure in CMIC (Controller Multi-path I/O and
 *   Namespace Sharing Capabilities).
 *
 * - `aerl`
 *   The Asynchronous Event Request Limit (AERL). Indicates the maximum number
 *   of concurrently outstanding Asynchronous Event Request commands support
 *   by the controller. This is a 0's based value.
 *
 * - `aer_max_queued`
 *   This is the maximum number of events that the device will enqueue for
 *   completion when there are no outstanding AERs. When the maximum number of
 *   enqueued events are reached, subsequent events will be dropped.
 *
 * - `mdts`
 *   Indicates the maximum data transfer size for a command that transfers data
 *   between host-accessible memory and the controller. The value is specified
 *   as a power of two (2^n) and is in units of the minimum memory page size
 *   (CAP.MPSMIN). The default value is 7 (i.e. 512 KiB).
 *
 * - `vsl`
 *   Indicates the maximum data size limit for the Verify command. Like `mdts`,
 *   this value is specified as a power of two (2^n) and is in units of the
 *   minimum memory page size (CAP.MPSMIN). The default value is 7 (i.e. 512
 *   KiB).
 *
 * - `zoned.zasl`
 *   Indicates the maximum data transfer size for the Zone Append command. Like
 *   `mdts`, the value is specified as a power of two (2^n) and is in units of
 *   the minimum memory page size (CAP.MPSMIN). The default value is 0 (i.e.
 *   defaulting to the value of `mdts`).
 *
 * - `zoned.auto_transition`
 *   Indicates if zones in zone state implicitly opened can be automatically
 *   transitioned to zone state closed for resource management purposes.
 *   Defaults to 'on'.
 *
 * - `sriov_max_vfs`
 *   Indicates the maximum number of PCIe virtual functions supported
 *   by the controller. The default value is 0. Specifying a non-zero value
 *   enables reporting of both SR-IOV and ARI capabilities by the NVMe device.
 *   Virtual function controllers will not report SR-IOV capability.
 *
 *   NOTE: Single Root I/O Virtualization support is experimental.
 *   All the related parameters may be subject to change.
 *
 * - `sriov_vq_flexible`
 *   Indicates the total number of flexible queue resources assignable to all
 *   the secondary controllers. Implicitly sets the number of primary
 *   controller's private resources to `(max_ioqpairs - sriov_vq_flexible)`.
 *
 * - `sriov_vi_flexible`
 *   Indicates the total number of flexible interrupt resources assignable to
 *   all the secondary controllers. Implicitly sets the number of primary
 *   controller's private resources to `(msix_qsize - sriov_vi_flexible)`.
 *
 * - `sriov_max_vi_per_vf`
 *   Indicates the maximum number of virtual interrupt resources assignable
 *   to a secondary controller. The default 0 resolves to
 *   `(sriov_vi_flexible / sriov_max_vfs)`.
 *
 * - `sriov_max_vq_per_vf`
 *   Indicates the maximum number of virtual queue resources assignable to
 *   a secondary controller. The default 0 resolves to
 *   `(sriov_vq_flexible / sriov_max_vfs)`.
 *
 * nvme namespace device parameters
 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 * - `shared`
 *   When the parent nvme device (as defined explicitly by the 'bus' parameter
 *   or implicitly by the most recently defined NvmeBus) is linked to an
 *   nvme-subsys device, the namespace will be attached to all controllers in
 *   the subsystem. If set to 'off' (the default), the namespace will remain a
 *   private namespace and may only be attached to a single controller at a
 *   time.
 *
 * - `detached`
 *   This parameter is only valid together with the `subsys` parameter. If left
 *   at the default value (`false/off`), the namespace will be attached to all
 *   controllers in the NVMe subsystem at boot-up. If set to `true/on`, the
 *   namespace will be available in the subsystem but not attached to any
 *   controllers.
 *
 * Setting `zoned` to true selects Zoned Command Set at the namespace.
 * In this case, the following namespace properties are available to configure
 * zoned operation:
 *     zoned.zone_size=<zone size in bytes, default: 128MiB>
 *         The number may be followed by K, M, G as in kilo-, mega- or giga-.
 *
 *     zoned.zone_capacity=<zone capacity in bytes, default: zone size>
 *         The value 0 (default) forces zone capacity to be the same as zone
 *         size. The value of this property may not exceed zone size.
 *
 *     zoned.descr_ext_size=<zone descriptor extension size, default 0>
 *         This value needs to be specified in 64B units. If it is zero,
 *         namespace(s) will not support zone descriptor extensions.
 *
 *     zoned.max_active=<Maximum Active Resources (zones), default: 0>
 *         The default value means there is no limit to the number of
 *         concurrently active zones.
 *
 *     zoned.max_open=<Maximum Open Resources (zones), default: 0>
 *         The default value means there is no limit to the number of
 *         concurrently open zones.
 *
 *     zoned.cross_read=<enable RAZB, default: false>
 *         Setting this property to true enables Read Across Zone Boundaries.
 */

#include "qemu/osdep.h"
#include "qemu/cutils.h"
#include "qemu/error-report.h"
#include "qemu/log.h"
#include "qemu/units.h"
#include "qemu/range.h"
#include "qapi/error.h"
#include "qapi/visitor.h"
#include "sysemu/sysemu.h"
#include "sysemu/block-backend.h"
#include "sysemu/hostmem.h"
#include "hw/pci/msix.h"
#include "hw/pci/pcie_sriov.h"
#include "migration/vmstate.h"

#include "nvme.h"
#include "dif.h"
#include "trace.h"

#define NVME_MAX_IOQPAIRS 0xffff
#define NVME_DB_SIZE  4
#define NVME_SPEC_VER 0x00010400
#define NVME_CMB_BIR 2
#define NVME_PMR_BIR 4
#define NVME_TEMPERATURE 0x143
#define NVME_TEMPERATURE_WARNING 0x157
#define NVME_TEMPERATURE_CRITICAL 0x175
#define NVME_NUM_FW_SLOTS 1
#define NVME_DEFAULT_MAX_ZA_SIZE (128 * KiB)
#define NVME_MAX_VFS 127
#define NVME_VF_RES_GRANULARITY 1
#define NVME_VF_OFFSET 0x1
#define NVME_VF_STRIDE 1

#define NVME_GUEST_ERR(trace, fmt, ...) \
    do { \
        (trace_##trace)(__VA_ARGS__); \
        qemu_log_mask(LOG_GUEST_ERROR, #trace \
            " in %s: " fmt "\n", __func__, ## __VA_ARGS__); \
    } while (0)

static const bool nvme_feature_support[NVME_FID_MAX] = {
    [NVME_ARBITRATION]              = true,
    [NVME_POWER_MANAGEMENT]         = true,
    [NVME_TEMPERATURE_THRESHOLD]    = true,
    [NVME_ERROR_RECOVERY]           = true,
    [NVME_VOLATILE_WRITE_CACHE]     = true,
    [NVME_NUMBER_OF_QUEUES]         = true,
    [NVME_INTERRUPT_COALESCING]     = true,
    [NVME_INTERRUPT_VECTOR_CONF]    = true,
    [NVME_WRITE_ATOMICITY]          = true,
    [NVME_ASYNCHRONOUS_EVENT_CONF]  = true,
    [NVME_TIMESTAMP]                = true,
    [NVME_HOST_BEHAVIOR_SUPPORT]    = true,
    [NVME_COMMAND_SET_PROFILE]      = true,
};

static const uint32_t nvme_feature_cap[NVME_FID_MAX] = {
    [NVME_TEMPERATURE_THRESHOLD]    = NVME_FEAT_CAP_CHANGE,
    [NVME_ERROR_RECOVERY]           = NVME_FEAT_CAP_CHANGE | NVME_FEAT_CAP_NS,
    [NVME_VOLATILE_WRITE_CACHE]     = NVME_FEAT_CAP_CHANGE,
    [NVME_NUMBER_OF_QUEUES]         = NVME_FEAT_CAP_CHANGE,
    [NVME_ASYNCHRONOUS_EVENT_CONF]  = NVME_FEAT_CAP_CHANGE,
    [NVME_TIMESTAMP]                = NVME_FEAT_CAP_CHANGE,
    [NVME_HOST_BEHAVIOR_SUPPORT]    = NVME_FEAT_CAP_CHANGE,
    [NVME_COMMAND_SET_PROFILE]      = NVME_FEAT_CAP_CHANGE,
};

static const uint32_t nvme_cse_acs[256] = {
    [NVME_ADM_CMD_DELETE_SQ]        = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_CREATE_SQ]        = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_GET_LOG_PAGE]     = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_DELETE_CQ]        = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_CREATE_CQ]        = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_IDENTIFY]         = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_ABORT]            = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_SET_FEATURES]     = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_GET_FEATURES]     = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_ASYNC_EV_REQ]     = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_NS_ATTACHMENT]    = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_NIC,
    [NVME_ADM_CMD_VIRT_MNGMT]       = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_DBBUF_CONFIG]     = NVME_CMD_EFF_CSUPP,
    [NVME_ADM_CMD_FORMAT_NVM]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
};

static const uint32_t nvme_cse_iocs_none[256];

static const uint32_t nvme_cse_iocs_nvm[256] = {
    [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
    [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
    [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
};

static const uint32_t nvme_cse_iocs_zoned[256] = {
    [NVME_CMD_FLUSH]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_WRITE_ZEROES]         = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_WRITE]                = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_READ]                 = NVME_CMD_EFF_CSUPP,
    [NVME_CMD_DSM]                  = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_VERIFY]               = NVME_CMD_EFF_CSUPP,
    [NVME_CMD_COPY]                 = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_COMPARE]              = NVME_CMD_EFF_CSUPP,
    [NVME_CMD_ZONE_APPEND]          = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_ZONE_MGMT_SEND]       = NVME_CMD_EFF_CSUPP | NVME_CMD_EFF_LBCC,
    [NVME_CMD_ZONE_MGMT_RECV]       = NVME_CMD_EFF_CSUPP,
};

static void nvme_process_sq(void *opaque);
static void nvme_ctrl_reset(NvmeCtrl *n, NvmeResetType rst);

static uint16_t nvme_sqid(NvmeRequest *req)
{
    return le16_to_cpu(req->sq->sqid);
}

static void nvme_assign_zone_state(NvmeNamespace *ns, NvmeZone *zone,
                                   NvmeZoneState state)
{
    if (QTAILQ_IN_USE(zone, entry)) {
        switch (nvme_get_zone_state(zone)) {
        case NVME_ZONE_STATE_EXPLICITLY_OPEN:
            QTAILQ_REMOVE(&ns->exp_open_zones, zone, entry);
            break;
        case NVME_ZONE_STATE_IMPLICITLY_OPEN:
            QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
            break;
        case NVME_ZONE_STATE_CLOSED:
            QTAILQ_REMOVE(&ns->closed_zones, zone, entry);
            break;
        case NVME_ZONE_STATE_FULL:
            QTAILQ_REMOVE(&ns->full_zones, zone, entry);
        default:
            ;
        }
    }

    nvme_set_zone_state(zone, state);

    switch (state) {
    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
        QTAILQ_INSERT_TAIL(&ns->exp_open_zones, zone, entry);
        break;
    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
        QTAILQ_INSERT_TAIL(&ns->imp_open_zones, zone, entry);
        break;
    case NVME_ZONE_STATE_CLOSED:
        QTAILQ_INSERT_TAIL(&ns->closed_zones, zone, entry);
        break;
    case NVME_ZONE_STATE_FULL:
        QTAILQ_INSERT_TAIL(&ns->full_zones, zone, entry);
    case NVME_ZONE_STATE_READ_ONLY:
        break;
    default:
        zone->d.za = 0;
    }
}

static uint16_t nvme_zns_check_resources(NvmeNamespace *ns, uint32_t act,
                                         uint32_t opn, uint32_t zrwa)
{
    if (ns->params.max_active_zones != 0 &&
        ns->nr_active_zones + act > ns->params.max_active_zones) {
        trace_pci_nvme_err_insuff_active_res(ns->params.max_active_zones);
        return NVME_ZONE_TOO_MANY_ACTIVE | NVME_DNR;
    }

    if (ns->params.max_open_zones != 0 &&
        ns->nr_open_zones + opn > ns->params.max_open_zones) {
        trace_pci_nvme_err_insuff_open_res(ns->params.max_open_zones);
        return NVME_ZONE_TOO_MANY_OPEN | NVME_DNR;
    }

    if (zrwa > ns->zns.numzrwa) {
        return NVME_NOZRWA | NVME_DNR;
    }

    return NVME_SUCCESS;
}

/*
 * Check if we can open a zone without exceeding open/active limits.
 * AOR stands for "Active and Open Resources" (see TP 4053 section 2.5).
 */
static uint16_t nvme_aor_check(NvmeNamespace *ns, uint32_t act, uint32_t opn)
{
    return nvme_zns_check_resources(ns, act, opn, 0);
}

static bool nvme_addr_is_cmb(NvmeCtrl *n, hwaddr addr)
{
    hwaddr hi, lo;

    if (!n->cmb.cmse) {
        return false;
    }

    lo = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
    hi = lo + int128_get64(n->cmb.mem.size);

    return addr >= lo && addr < hi;
}

static inline void *nvme_addr_to_cmb(NvmeCtrl *n, hwaddr addr)
{
    hwaddr base = n->params.legacy_cmb ? n->cmb.mem.addr : n->cmb.cba;
    return &n->cmb.buf[addr - base];
}

static bool nvme_addr_is_pmr(NvmeCtrl *n, hwaddr addr)
{
    hwaddr hi;

    if (!n->pmr.cmse) {
        return false;
    }

    hi = n->pmr.cba + int128_get64(n->pmr.dev->mr.size);

    return addr >= n->pmr.cba && addr < hi;
}

static inline void *nvme_addr_to_pmr(NvmeCtrl *n, hwaddr addr)
{
    return memory_region_get_ram_ptr(&n->pmr.dev->mr) + (addr - n->pmr.cba);
}

static inline bool nvme_addr_is_iomem(NvmeCtrl *n, hwaddr addr)
{
    hwaddr hi, lo;

    /*
     * The purpose of this check is to guard against invalid "local" access to
     * the iomem (i.e. controller registers). Thus, we check against the range
     * covered by the 'bar0' MemoryRegion since that is currently composed of
     * two subregions (the NVMe "MBAR" and the MSI-X table/pba). Note, however,
     * that if the device model is ever changed to allow the CMB to be located
     * in BAR0 as well, then this must be changed.
     */
    lo = n->bar0.addr;
    hi = lo + int128_get64(n->bar0.size);

    return addr >= lo && addr < hi;
}

static int nvme_addr_read(NvmeCtrl *n, hwaddr addr, void *buf, int size)
{
    hwaddr hi = addr + size - 1;
    if (hi < addr) {
        return 1;
    }

    if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
        memcpy(buf, nvme_addr_to_cmb(n, addr), size);
        return 0;
    }

    if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
        memcpy(buf, nvme_addr_to_pmr(n, addr), size);
        return 0;
    }

    return pci_dma_read(&n->parent_obj, addr, buf, size);
}

static int nvme_addr_write(NvmeCtrl *n, hwaddr addr, const void *buf, int size)
{
    hwaddr hi = addr + size - 1;
    if (hi < addr) {
        return 1;
    }

    if (n->bar.cmbsz && nvme_addr_is_cmb(n, addr) && nvme_addr_is_cmb(n, hi)) {
        memcpy(nvme_addr_to_cmb(n, addr), buf, size);
        return 0;
    }

    if (nvme_addr_is_pmr(n, addr) && nvme_addr_is_pmr(n, hi)) {
        memcpy(nvme_addr_to_pmr(n, addr), buf, size);
        return 0;
    }

    return pci_dma_write(&n->parent_obj, addr, buf, size);
}

static bool nvme_nsid_valid(NvmeCtrl *n, uint32_t nsid)
{
    return nsid &&
        (nsid == NVME_NSID_BROADCAST || nsid <= NVME_MAX_NAMESPACES);
}

static int nvme_check_sqid(NvmeCtrl *n, uint16_t sqid)
{
    return sqid < n->conf_ioqpairs + 1 && n->sq[sqid] != NULL ? 0 : -1;
}

static int nvme_check_cqid(NvmeCtrl *n, uint16_t cqid)
{
    return cqid < n->conf_ioqpairs + 1 && n->cq[cqid] != NULL ? 0 : -1;
}

static void nvme_inc_cq_tail(NvmeCQueue *cq)
{
    cq->tail++;
    if (cq->tail >= cq->size) {
        cq->tail = 0;
        cq->phase = !cq->phase;
    }
}

static void nvme_inc_sq_head(NvmeSQueue *sq)
{
    sq->head = (sq->head + 1) % sq->size;
}

static uint8_t nvme_cq_full(NvmeCQueue *cq)
{
    return (cq->tail + 1) % cq->size == cq->head;
}

static uint8_t nvme_sq_empty(NvmeSQueue *sq)
{
    return sq->head == sq->tail;
}

static void nvme_irq_check(NvmeCtrl *n)
{
    uint32_t intms = ldl_le_p(&n->bar.intms);

    if (msix_enabled(&(n->parent_obj))) {
        return;
    }
    if (~intms & n->irq_status) {
        pci_irq_assert(&n->parent_obj);
    } else {
        pci_irq_deassert(&n->parent_obj);
    }
}

static void nvme_irq_assert(NvmeCtrl *n, NvmeCQueue *cq)
{
    if (cq->irq_enabled) {
        if (msix_enabled(&(n->parent_obj))) {
            trace_pci_nvme_irq_msix(cq->vector);
            msix_notify(&(n->parent_obj), cq->vector);
        } else {
            trace_pci_nvme_irq_pin();
            assert(cq->vector < 32);
            n->irq_status |= 1 << cq->vector;
            nvme_irq_check(n);
        }
    } else {
        trace_pci_nvme_irq_masked();
    }
}

static void nvme_irq_deassert(NvmeCtrl *n, NvmeCQueue *cq)
{
    if (cq->irq_enabled) {
        if (msix_enabled(&(n->parent_obj))) {
            return;
        } else {
            assert(cq->vector < 32);
            if (!n->cq_pending) {
                n->irq_status &= ~(1 << cq->vector);
            }
            nvme_irq_check(n);
        }
    }
}

static void nvme_req_clear(NvmeRequest *req)
{
    req->ns = NULL;
    req->opaque = NULL;
    req->aiocb = NULL;
    memset(&req->cqe, 0x0, sizeof(req->cqe));
    req->status = NVME_SUCCESS;
}

static inline void nvme_sg_init(NvmeCtrl *n, NvmeSg *sg, bool dma)
{
    if (dma) {
        pci_dma_sglist_init(&sg->qsg, &n->parent_obj, 0);
        sg->flags = NVME_SG_DMA;
    } else {
        qemu_iovec_init(&sg->iov, 0);
    }

    sg->flags |= NVME_SG_ALLOC;
}

static inline void nvme_sg_unmap(NvmeSg *sg)
{
    if (!(sg->flags & NVME_SG_ALLOC)) {
        return;
    }

    if (sg->flags & NVME_SG_DMA) {
        qemu_sglist_destroy(&sg->qsg);
    } else {
        qemu_iovec_destroy(&sg->iov);
    }

    memset(sg, 0x0, sizeof(*sg));
}

/*
 * When metadata is transfered as extended LBAs, the DPTR mapped into `sg`
 * holds both data and metadata. This function splits the data and metadata
 * into two separate QSG/IOVs.
 */
static void nvme_sg_split(NvmeSg *sg, NvmeNamespace *ns, NvmeSg *data,
                          NvmeSg *mdata)
{
    NvmeSg *dst = data;
    uint32_t trans_len, count = ns->lbasz;
    uint64_t offset = 0;
    bool dma = sg->flags & NVME_SG_DMA;
    size_t sge_len;
    size_t sg_len = dma ? sg->qsg.size : sg->iov.size;
    int sg_idx = 0;

    assert(sg->flags & NVME_SG_ALLOC);

    while (sg_len) {
        sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;

        trans_len = MIN(sg_len, count);
        trans_len = MIN(trans_len, sge_len - offset);

        if (dst) {
            if (dma) {
                qemu_sglist_add(&dst->qsg, sg->qsg.sg[sg_idx].base + offset,
                                trans_len);
            } else {
                qemu_iovec_add(&dst->iov,
                               sg->iov.iov[sg_idx].iov_base + offset,
                               trans_len);
            }
        }

        sg_len -= trans_len;
        count -= trans_len;
        offset += trans_len;

        if (count == 0) {
            dst = (dst == data) ? mdata : data;
            count = (dst == data) ? ns->lbasz : ns->lbaf.ms;
        }

        if (sge_len == offset) {
            offset = 0;
            sg_idx++;
        }
    }
}

static uint16_t nvme_map_addr_cmb(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
                                  size_t len)
{
    if (!len) {
        return NVME_SUCCESS;
    }

    trace_pci_nvme_map_addr_cmb(addr, len);

    if (!nvme_addr_is_cmb(n, addr) || !nvme_addr_is_cmb(n, addr + len - 1)) {
        return NVME_DATA_TRAS_ERROR;
    }

    qemu_iovec_add(iov, nvme_addr_to_cmb(n, addr), len);

    return NVME_SUCCESS;
}

static uint16_t nvme_map_addr_pmr(NvmeCtrl *n, QEMUIOVector *iov, hwaddr addr,
                                  size_t len)
{
    if (!len) {
        return NVME_SUCCESS;
    }

    if (!nvme_addr_is_pmr(n, addr) || !nvme_addr_is_pmr(n, addr + len - 1)) {
        return NVME_DATA_TRAS_ERROR;
    }

    qemu_iovec_add(iov, nvme_addr_to_pmr(n, addr), len);

    return NVME_SUCCESS;
}

static uint16_t nvme_map_addr(NvmeCtrl *n, NvmeSg *sg, hwaddr addr, size_t len)
{
    bool cmb = false, pmr = false;

    if (!len) {
        return NVME_SUCCESS;
    }

    trace_pci_nvme_map_addr(addr, len);

    if (nvme_addr_is_iomem(n, addr)) {
        return NVME_DATA_TRAS_ERROR;
    }

    if (nvme_addr_is_cmb(n, addr)) {
        cmb = true;
    } else if (nvme_addr_is_pmr(n, addr)) {
        pmr = true;
    }

    if (cmb || pmr) {
        if (sg->flags & NVME_SG_DMA) {
            return NVME_INVALID_USE_OF_CMB | NVME_DNR;
        }

        if (sg->iov.niov + 1 > IOV_MAX) {
            goto max_mappings_exceeded;
        }

        if (cmb) {
            return nvme_map_addr_cmb(n, &sg->iov, addr, len);
        } else {
            return nvme_map_addr_pmr(n, &sg->iov, addr, len);
        }
    }

    if (!(sg->flags & NVME_SG_DMA)) {
        return NVME_INVALID_USE_OF_CMB | NVME_DNR;
    }

    if (sg->qsg.nsg + 1 > IOV_MAX) {
        goto max_mappings_exceeded;
    }

    qemu_sglist_add(&sg->qsg, addr, len);

    return NVME_SUCCESS;

max_mappings_exceeded:
    NVME_GUEST_ERR(pci_nvme_ub_too_many_mappings,
                   "number of mappings exceed 1024");
    return NVME_INTERNAL_DEV_ERROR | NVME_DNR;
}

static inline bool nvme_addr_is_dma(NvmeCtrl *n, hwaddr addr)
{
    return !(nvme_addr_is_cmb(n, addr) || nvme_addr_is_pmr(n, addr));
}

static uint16_t nvme_map_prp(NvmeCtrl *n, NvmeSg *sg, uint64_t prp1,
                             uint64_t prp2, uint32_t len)
{
    hwaddr trans_len = n->page_size - (prp1 % n->page_size);
    trans_len = MIN(len, trans_len);
    int num_prps = (len >> n->page_bits) + 1;
    uint16_t status;
    int ret;

    trace_pci_nvme_map_prp(trans_len, len, prp1, prp2, num_prps);

    nvme_sg_init(n, sg, nvme_addr_is_dma(n, prp1));

    status = nvme_map_addr(n, sg, prp1, trans_len);
    if (status) {
        goto unmap;
    }

    len -= trans_len;
    if (len) {
        if (len > n->page_size) {
            uint64_t prp_list[n->max_prp_ents];
            uint32_t nents, prp_trans;
            int i = 0;

            /*
             * The first PRP list entry, pointed to by PRP2 may contain offset.
             * Hence, we need to calculate the number of entries in based on
             * that offset.
             */
            nents = (n->page_size - (prp2 & (n->page_size - 1))) >> 3;
            prp_trans = MIN(n->max_prp_ents, nents) * sizeof(uint64_t);
            ret = nvme_addr_read(n, prp2, (void *)prp_list, prp_trans);
            if (ret) {
                trace_pci_nvme_err_addr_read(prp2);
                status = NVME_DATA_TRAS_ERROR;
                goto unmap;
            }
            while (len != 0) {
                uint64_t prp_ent = le64_to_cpu(prp_list[i]);

                if (i == nents - 1 && len > n->page_size) {
                    if (unlikely(prp_ent & (n->page_size - 1))) {
                        trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
                        status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
                        goto unmap;
                    }

                    i = 0;
                    nents = (len + n->page_size - 1) >> n->page_bits;
                    nents = MIN(nents, n->max_prp_ents);
                    prp_trans = nents * sizeof(uint64_t);
                    ret = nvme_addr_read(n, prp_ent, (void *)prp_list,
                                         prp_trans);
                    if (ret) {
                        trace_pci_nvme_err_addr_read(prp_ent);
                        status = NVME_DATA_TRAS_ERROR;
                        goto unmap;
                    }
                    prp_ent = le64_to_cpu(prp_list[i]);
                }

                if (unlikely(prp_ent & (n->page_size - 1))) {
                    trace_pci_nvme_err_invalid_prplist_ent(prp_ent);
                    status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
                    goto unmap;
                }

                trans_len = MIN(len, n->page_size);
                status = nvme_map_addr(n, sg, prp_ent, trans_len);
                if (status) {
                    goto unmap;
                }

                len -= trans_len;
                i++;
            }
        } else {
            if (unlikely(prp2 & (n->page_size - 1))) {
                trace_pci_nvme_err_invalid_prp2_align(prp2);
                status = NVME_INVALID_PRP_OFFSET | NVME_DNR;
                goto unmap;
            }
            status = nvme_map_addr(n, sg, prp2, len);
            if (status) {
                goto unmap;
            }
        }
    }

    return NVME_SUCCESS;

unmap:
    nvme_sg_unmap(sg);
    return status;
}

/*
 * Map 'nsgld' data descriptors from 'segment'. The function will subtract the
 * number of bytes mapped in len.
 */
static uint16_t nvme_map_sgl_data(NvmeCtrl *n, NvmeSg *sg,
                                  NvmeSglDescriptor *segment, uint64_t nsgld,
                                  size_t *len, NvmeCmd *cmd)
{
    dma_addr_t addr, trans_len;
    uint32_t dlen;
    uint16_t status;

    for (int i = 0; i < nsgld; i++) {
        uint8_t type = NVME_SGL_TYPE(segment[i].type);

        switch (type) {
        case NVME_SGL_DESCR_TYPE_DATA_BLOCK:
            break;
        case NVME_SGL_DESCR_TYPE_SEGMENT:
        case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
            return NVME_INVALID_NUM_SGL_DESCRS | NVME_DNR;
        default:
            return NVME_SGL_DESCR_TYPE_INVALID | NVME_DNR;
        }

        dlen = le32_to_cpu(segment[i].len);

        if (!dlen) {
            continue;
        }

        if (*len == 0) {
            /*
             * All data has been mapped, but the SGL contains additional
             * segments and/or descriptors. The controller might accept
             * ignoring the rest of the SGL.
             */
            uint32_t sgls = le32_to_cpu(n->id_ctrl.sgls);
            if (sgls & NVME_CTRL_SGLS_EXCESS_LENGTH) {
                break;
            }

            trace_pci_nvme_err_invalid_sgl_excess_length(dlen);
            return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
        }

        trans_len = MIN(*len, dlen);

        addr = le64_to_cpu(segment[i].addr);

        if (UINT64_MAX - addr < dlen) {
            return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
        }

        status = nvme_map_addr(n, sg, addr, trans_len);
        if (status) {
            return status;
        }

        *len -= trans_len;
    }

    return NVME_SUCCESS;
}

static uint16_t nvme_map_sgl(NvmeCtrl *n, NvmeSg *sg, NvmeSglDescriptor sgl,
                             size_t len, NvmeCmd *cmd)
{
    /*
     * Read the segment in chunks of 256 descriptors (one 4k page) to avoid
     * dynamically allocating a potentially huge SGL. The spec allows the SGL
     * to be larger (as in number of bytes required to describe the SGL
     * descriptors and segment chain) than the command transfer size, so it is
     * not bounded by MDTS.
     */
    const int SEG_CHUNK_SIZE = 256;

    NvmeSglDescriptor segment[SEG_CHUNK_SIZE], *sgld, *last_sgld;
    uint64_t nsgld;
    uint32_t seg_len;
    uint16_t status;
    hwaddr addr;
    int ret;

    sgld = &sgl;
    addr = le64_to_cpu(sgl.addr);

    trace_pci_nvme_map_sgl(NVME_SGL_TYPE(sgl.type), len);

    nvme_sg_init(n, sg, nvme_addr_is_dma(n, addr));

    /*
     * If the entire transfer can be described with a single data block it can
     * be mapped directly.
     */
    if (NVME_SGL_TYPE(sgl.type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
        status = nvme_map_sgl_data(n, sg, sgld, 1, &len, cmd);
        if (status) {
            goto unmap;
        }

        goto out;
    }

    for (;;) {
        switch (NVME_SGL_TYPE(sgld->type)) {
        case NVME_SGL_DESCR_TYPE_SEGMENT:
        case NVME_SGL_DESCR_TYPE_LAST_SEGMENT:
            break;
        default:
            return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
        }

        seg_len = le32_to_cpu(sgld->len);

        /* check the length of the (Last) Segment descriptor */
        if (!seg_len || seg_len & 0xf) {
            return NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
        }

        if (UINT64_MAX - addr < seg_len) {
            return NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
        }

        nsgld = seg_len / sizeof(NvmeSglDescriptor);

        while (nsgld > SEG_CHUNK_SIZE) {
            if (nvme_addr_read(n, addr, segment, sizeof(segment))) {
                trace_pci_nvme_err_addr_read(addr);
                status = NVME_DATA_TRAS_ERROR;
                goto unmap;
            }

            status = nvme_map_sgl_data(n, sg, segment, SEG_CHUNK_SIZE,
                                       &len, cmd);
            if (status) {
                goto unmap;
            }

            nsgld -= SEG_CHUNK_SIZE;
            addr += SEG_CHUNK_SIZE * sizeof(NvmeSglDescriptor);
        }

        ret = nvme_addr_read(n, addr, segment, nsgld *
                             sizeof(NvmeSglDescriptor));
        if (ret) {
            trace_pci_nvme_err_addr_read(addr);
            status = NVME_DATA_TRAS_ERROR;
            goto unmap;
        }

        last_sgld = &segment[nsgld - 1];

        /*
         * If the segment ends with a Data Block, then we are done.
         */
        if (NVME_SGL_TYPE(last_sgld->type) == NVME_SGL_DESCR_TYPE_DATA_BLOCK) {
            status = nvme_map_sgl_data(n, sg, segment, nsgld, &len, cmd);
            if (status) {
                goto unmap;
            }

            goto out;

        }

        /*
         * If the last descriptor was not a Data Block, then the current
         * segment must not be a Last Segment.
         */
        if (NVME_SGL_TYPE(sgld->type) == NVME_SGL_DESCR_TYPE_LAST_SEGMENT) {
            status = NVME_INVALID_SGL_SEG_DESCR | NVME_DNR;
            goto unmap;
        }

        sgld = last_sgld;
        addr = le64_to_cpu(sgld->addr);

        /*
         * Do not map the last descriptor; it will be a Segment or Last Segment
         * descriptor and is handled by the next iteration.
         */
        status = nvme_map_sgl_data(n, sg, segment, nsgld - 1, &len, cmd);
        if (status) {
            goto unmap;
        }
    }

out:
    /* if there is any residual left in len, the SGL was too short */
    if (len) {
        status = NVME_DATA_SGL_LEN_INVALID | NVME_DNR;
        goto unmap;
    }

    return NVME_SUCCESS;

unmap:
    nvme_sg_unmap(sg);
    return status;
}

uint16_t nvme_map_dptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
                       NvmeCmd *cmd)
{
    uint64_t prp1, prp2;

    switch (NVME_CMD_FLAGS_PSDT(cmd->flags)) {
    case NVME_PSDT_PRP:
        prp1 = le64_to_cpu(cmd->dptr.prp1);
        prp2 = le64_to_cpu(cmd->dptr.prp2);

        return nvme_map_prp(n, sg, prp1, prp2, len);
    case NVME_PSDT_SGL_MPTR_CONTIGUOUS:
    case NVME_PSDT_SGL_MPTR_SGL:
        return nvme_map_sgl(n, sg, cmd->dptr.sgl, len, cmd);
    default:
        return NVME_INVALID_FIELD;
    }
}

static uint16_t nvme_map_mptr(NvmeCtrl *n, NvmeSg *sg, size_t len,
                              NvmeCmd *cmd)
{
    int psdt = NVME_CMD_FLAGS_PSDT(cmd->flags);
    hwaddr mptr = le64_to_cpu(cmd->mptr);
    uint16_t status;

    if (psdt == NVME_PSDT_SGL_MPTR_SGL) {
        NvmeSglDescriptor sgl;

        if (nvme_addr_read(n, mptr, &sgl, sizeof(sgl))) {
            return NVME_DATA_TRAS_ERROR;
        }

        status = nvme_map_sgl(n, sg, sgl, len, cmd);
        if (status && (status & 0x7ff) == NVME_DATA_SGL_LEN_INVALID) {
            status = NVME_MD_SGL_LEN_INVALID | NVME_DNR;
        }

        return status;
    }

    nvme_sg_init(n, sg, nvme_addr_is_dma(n, mptr));
    status = nvme_map_addr(n, sg, mptr, len);
    if (status) {
        nvme_sg_unmap(sg);
    }

    return status;
}

static uint16_t nvme_map_data(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
{
    NvmeNamespace *ns = req->ns;
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
    bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);
    size_t len = nvme_l2b(ns, nlb);
    uint16_t status;

    if (nvme_ns_ext(ns) &&
        !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
        NvmeSg sg;

        len += nvme_m2b(ns, nlb);

        status = nvme_map_dptr(n, &sg, len, &req->cmd);
        if (status) {
            return status;
        }

        nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
        nvme_sg_split(&sg, ns, &req->sg, NULL);
        nvme_sg_unmap(&sg);

        return NVME_SUCCESS;
    }

    return nvme_map_dptr(n, &req->sg, len, &req->cmd);
}

static uint16_t nvme_map_mdata(NvmeCtrl *n, uint32_t nlb, NvmeRequest *req)
{
    NvmeNamespace *ns = req->ns;
    size_t len = nvme_m2b(ns, nlb);
    uint16_t status;

    if (nvme_ns_ext(ns)) {
        NvmeSg sg;

        len += nvme_l2b(ns, nlb);

        status = nvme_map_dptr(n, &sg, len, &req->cmd);
        if (status) {
            return status;
        }

        nvme_sg_init(n, &req->sg, sg.flags & NVME_SG_DMA);
        nvme_sg_split(&sg, ns, NULL, &req->sg);
        nvme_sg_unmap(&sg);

        return NVME_SUCCESS;
    }

    return nvme_map_mptr(n, &req->sg, len, &req->cmd);
}

static uint16_t nvme_tx_interleaved(NvmeCtrl *n, NvmeSg *sg, uint8_t *ptr,
                                    uint32_t len, uint32_t bytes,
                                    int32_t skip_bytes, int64_t offset,
                                    NvmeTxDirection dir)
{
    hwaddr addr;
    uint32_t trans_len, count = bytes;
    bool dma = sg->flags & NVME_SG_DMA;
    int64_t sge_len;
    int sg_idx = 0;
    int ret;

    assert(sg->flags & NVME_SG_ALLOC);

    while (len) {
        sge_len = dma ? sg->qsg.sg[sg_idx].len : sg->iov.iov[sg_idx].iov_len;

        if (sge_len - offset < 0) {
            offset -= sge_len;
            sg_idx++;
            continue;
        }

        if (sge_len == offset) {
            offset = 0;
            sg_idx++;
            continue;
        }

        trans_len = MIN(len, count);
        trans_len = MIN(trans_len, sge_len - offset);

        if (dma) {
            addr = sg->qsg.sg[sg_idx].base + offset;
        } else {
            addr = (hwaddr)(uintptr_t)sg->iov.iov[sg_idx].iov_base + offset;
        }

        if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
            ret = nvme_addr_read(n, addr, ptr, trans_len);
        } else {
            ret = nvme_addr_write(n, addr, ptr, trans_len);
        }

        if (ret) {
            return NVME_DATA_TRAS_ERROR;
        }

        ptr += trans_len;
        len -= trans_len;
        count -= trans_len;
        offset += trans_len;

        if (count == 0) {
            count = bytes;
            offset += skip_bytes;
        }
    }

    return NVME_SUCCESS;
}

static uint16_t nvme_tx(NvmeCtrl *n, NvmeSg *sg, void *ptr, uint32_t len,
                        NvmeTxDirection dir)
{
    assert(sg->flags & NVME_SG_ALLOC);

    if (sg->flags & NVME_SG_DMA) {
        const MemTxAttrs attrs = MEMTXATTRS_UNSPECIFIED;
        dma_addr_t residual;

        if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
            dma_buf_write(ptr, len, &residual, &sg->qsg, attrs);
        } else {
            dma_buf_read(ptr, len, &residual, &sg->qsg, attrs);
        }

        if (unlikely(residual)) {
            trace_pci_nvme_err_invalid_dma();
            return NVME_INVALID_FIELD | NVME_DNR;
        }
    } else {
        size_t bytes;

        if (dir == NVME_TX_DIRECTION_TO_DEVICE) {
            bytes = qemu_iovec_to_buf(&sg->iov, 0, ptr, len);
        } else {
            bytes = qemu_iovec_from_buf(&sg->iov, 0, ptr, len);
        }

        if (unlikely(bytes != len)) {
            trace_pci_nvme_err_invalid_dma();
            return NVME_INVALID_FIELD | NVME_DNR;
        }
    }

    return NVME_SUCCESS;
}

static inline uint16_t nvme_c2h(NvmeCtrl *n, void *ptr, uint32_t len,
                                NvmeRequest *req)
{
    uint16_t status;

    status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
    if (status) {
        return status;
    }

    return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_FROM_DEVICE);
}

static inline uint16_t nvme_h2c(NvmeCtrl *n, void *ptr, uint32_t len,
                                NvmeRequest *req)
{
    uint16_t status;

    status = nvme_map_dptr(n, &req->sg, len, &req->cmd);
    if (status) {
        return status;
    }

    return nvme_tx(n, &req->sg, ptr, len, NVME_TX_DIRECTION_TO_DEVICE);
}

uint16_t nvme_bounce_data(NvmeCtrl *n, void *ptr, uint32_t len,
                          NvmeTxDirection dir, NvmeRequest *req)
{
    NvmeNamespace *ns = req->ns;
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    bool pi = !!NVME_ID_NS_DPS_TYPE(ns->id_ns.dps);
    bool pract = !!(le16_to_cpu(rw->control) & NVME_RW_PRINFO_PRACT);

    if (nvme_ns_ext(ns) &&
        !(pi && pract && ns->lbaf.ms == nvme_pi_tuple_size(ns))) {
        return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbasz,
                                   ns->lbaf.ms, 0, dir);
    }

    return nvme_tx(n, &req->sg, ptr, len, dir);
}

uint16_t nvme_bounce_mdata(NvmeCtrl *n, void *ptr, uint32_t len,
                           NvmeTxDirection dir, NvmeRequest *req)
{
    NvmeNamespace *ns = req->ns;
    uint16_t status;

    if (nvme_ns_ext(ns)) {
        return nvme_tx_interleaved(n, &req->sg, ptr, len, ns->lbaf.ms,
                                   ns->lbasz, ns->lbasz, dir);
    }

    nvme_sg_unmap(&req->sg);

    status = nvme_map_mptr(n, &req->sg, len, &req->cmd);
    if (status) {
        return status;
    }

    return nvme_tx(n, &req->sg, ptr, len, dir);
}

static inline void nvme_blk_read(BlockBackend *blk, int64_t offset,
                                 BlockCompletionFunc *cb, NvmeRequest *req)
{
    assert(req->sg.flags & NVME_SG_ALLOC);

    if (req->sg.flags & NVME_SG_DMA) {
        req->aiocb = dma_blk_read(blk, &req->sg.qsg, offset, BDRV_SECTOR_SIZE,
                                  cb, req);
    } else {
        req->aiocb = blk_aio_preadv(blk, offset, &req->sg.iov, 0, cb, req);
    }
}

static inline void nvme_blk_write(BlockBackend *blk, int64_t offset,
                                  BlockCompletionFunc *cb, NvmeRequest *req)
{
    assert(req->sg.flags & NVME_SG_ALLOC);

    if (req->sg.flags & NVME_SG_DMA) {
        req->aiocb = dma_blk_write(blk, &req->sg.qsg, offset, BDRV_SECTOR_SIZE,
                                   cb, req);
    } else {
        req->aiocb = blk_aio_pwritev(blk, offset, &req->sg.iov, 0, cb, req);
    }
}

static void nvme_update_cq_head(NvmeCQueue *cq)
{
    pci_dma_read(&cq->ctrl->parent_obj, cq->db_addr, &cq->head,
            sizeof(cq->head));
    trace_pci_nvme_shadow_doorbell_cq(cq->cqid, cq->head);
}

static void nvme_post_cqes(void *opaque)
{
    NvmeCQueue *cq = opaque;
    NvmeCtrl *n = cq->ctrl;
    NvmeRequest *req, *next;
    bool pending = cq->head != cq->tail;
    int ret;

    QTAILQ_FOREACH_SAFE(req, &cq->req_list, entry, next) {
        NvmeSQueue *sq;
        hwaddr addr;

        if (n->dbbuf_enabled) {
            nvme_update_cq_head(cq);
        }

        if (nvme_cq_full(cq)) {
            break;
        }

        sq = req->sq;
        req->cqe.status = cpu_to_le16((req->status << 1) | cq->phase);
        req->cqe.sq_id = cpu_to_le16(sq->sqid);
        req->cqe.sq_head = cpu_to_le16(sq->head);
        addr = cq->dma_addr + cq->tail * n->cqe_size;
        ret = pci_dma_write(&n->parent_obj, addr, (void *)&req->cqe,
                            sizeof(req->cqe));
        if (ret) {
            trace_pci_nvme_err_addr_write(addr);
            trace_pci_nvme_err_cfs();
            stl_le_p(&n->bar.csts, NVME_CSTS_FAILED);
            break;
        }
        QTAILQ_REMOVE(&cq->req_list, req, entry);
        nvme_inc_cq_tail(cq);
        nvme_sg_unmap(&req->sg);
        QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
    }
    if (cq->tail != cq->head) {
        if (cq->irq_enabled && !pending) {
            n->cq_pending++;
        }

        nvme_irq_assert(n, cq);
    }
}

static void nvme_enqueue_req_completion(NvmeCQueue *cq, NvmeRequest *req)
{
    assert(cq->cqid == req->sq->cqid);
    trace_pci_nvme_enqueue_req_completion(nvme_cid(req), cq->cqid,
                                          le32_to_cpu(req->cqe.result),
                                          le32_to_cpu(req->cqe.dw1),
                                          req->status);

    if (req->status) {
        trace_pci_nvme_err_req_status(nvme_cid(req), nvme_nsid(req->ns),
                                      req->status, req->cmd.opcode);
    }

    QTAILQ_REMOVE(&req->sq->out_req_list, req, entry);
    QTAILQ_INSERT_TAIL(&cq->req_list, req, entry);

    qemu_bh_schedule(cq->bh);
}

static void nvme_process_aers(void *opaque)
{
    NvmeCtrl *n = opaque;
    NvmeAsyncEvent *event, *next;

    trace_pci_nvme_process_aers(n->aer_queued);

    QTAILQ_FOREACH_SAFE(event, &n->aer_queue, entry, next) {
        NvmeRequest *req;
        NvmeAerResult *result;

        /* can't post cqe if there is nothing to complete */
        if (!n->outstanding_aers) {
            trace_pci_nvme_no_outstanding_aers();
            break;
        }

        /* ignore if masked (cqe posted, but event not cleared) */
        if (n->aer_mask & (1 << event->result.event_type)) {
            trace_pci_nvme_aer_masked(event->result.event_type, n->aer_mask);
            continue;
        }

        QTAILQ_REMOVE(&n->aer_queue, event, entry);
        n->aer_queued--;

        n->aer_mask |= 1 << event->result.event_type;
        n->outstanding_aers--;

        req = n->aer_reqs[n->outstanding_aers];

        result = (NvmeAerResult *) &req->cqe.result;
        result->event_type = event->result.event_type;
        result->event_info = event->result.event_info;
        result->log_page = event->result.log_page;
        g_free(event);

        trace_pci_nvme_aer_post_cqe(result->event_type, result->event_info,
                                    result->log_page);

        nvme_enqueue_req_completion(&n->admin_cq, req);
    }
}

static void nvme_enqueue_event(NvmeCtrl *n, uint8_t event_type,
                               uint8_t event_info, uint8_t log_page)
{
    NvmeAsyncEvent *event;

    trace_pci_nvme_enqueue_event(event_type, event_info, log_page);

    if (n->aer_queued == n->params.aer_max_queued) {
        trace_pci_nvme_enqueue_event_noqueue(n->aer_queued);
        return;
    }

    event = g_new(NvmeAsyncEvent, 1);
    event->result = (NvmeAerResult) {
        .event_type = event_type,
        .event_info = event_info,
        .log_page   = log_page,
    };

    QTAILQ_INSERT_TAIL(&n->aer_queue, event, entry);
    n->aer_queued++;

    nvme_process_aers(n);
}

static void nvme_smart_event(NvmeCtrl *n, uint8_t event)
{
    uint8_t aer_info;

    /* Ref SPEC <Asynchronous Event Information 0x2013 SMART / Health Status> */
    if (!(NVME_AEC_SMART(n->features.async_config) & event)) {
        return;
    }

    switch (event) {
    case NVME_SMART_SPARE:
        aer_info = NVME_AER_INFO_SMART_SPARE_THRESH;
        break;
    case NVME_SMART_TEMPERATURE:
        aer_info = NVME_AER_INFO_SMART_TEMP_THRESH;
        break;
    case NVME_SMART_RELIABILITY:
    case NVME_SMART_MEDIA_READ_ONLY:
    case NVME_SMART_FAILED_VOLATILE_MEDIA:
    case NVME_SMART_PMR_UNRELIABLE:
        aer_info = NVME_AER_INFO_SMART_RELIABILITY;
        break;
    default:
        return;
    }

    nvme_enqueue_event(n, NVME_AER_TYPE_SMART, aer_info, NVME_LOG_SMART_INFO);
}

static void nvme_clear_events(NvmeCtrl *n, uint8_t event_type)
{
    n->aer_mask &= ~(1 << event_type);
    if (!QTAILQ_EMPTY(&n->aer_queue)) {
        nvme_process_aers(n);
    }
}

static inline uint16_t nvme_check_mdts(NvmeCtrl *n, size_t len)
{
    uint8_t mdts = n->params.mdts;

    if (mdts && len > n->page_size << mdts) {
        trace_pci_nvme_err_mdts(len);
        return NVME_INVALID_FIELD | NVME_DNR;
    }

    return NVME_SUCCESS;
}

static inline uint16_t nvme_check_bounds(NvmeNamespace *ns, uint64_t slba,
                                         uint32_t nlb)
{
    uint64_t nsze = le64_to_cpu(ns->id_ns.nsze);

    if (unlikely(UINT64_MAX - slba < nlb || slba + nlb > nsze)) {
        trace_pci_nvme_err_invalid_lba_range(slba, nlb, nsze);
        return NVME_LBA_RANGE | NVME_DNR;
    }

    return NVME_SUCCESS;
}

static int nvme_block_status_all(NvmeNamespace *ns, uint64_t slba,
                                 uint32_t nlb, int flags)
{
    BlockDriverState *bs = blk_bs(ns->blkconf.blk);

    int64_t pnum = 0, bytes = nvme_l2b(ns, nlb);
    int64_t offset = nvme_l2b(ns, slba);
    int ret;

    /*
     * `pnum` holds the number of bytes after offset that shares the same
     * allocation status as the byte at offset. If `pnum` is different from
     * `bytes`, we should check the allocation status of the next range and
     * continue this until all bytes have been checked.
     */
    do {
        bytes -= pnum;

        ret = bdrv_block_status(bs, offset, bytes, &pnum, NULL, NULL);
        if (ret < 0) {
            return ret;
        }


        trace_pci_nvme_block_status(offset, bytes, pnum, ret,
                                    !!(ret & BDRV_BLOCK_ZERO));

        if (!(ret & flags)) {
            return 1;
        }

        offset += pnum;
    } while (pnum != bytes);

    return 0;
}

static uint16_t nvme_check_dulbe(NvmeNamespace *ns, uint64_t slba,
                                 uint32_t nlb)
{
    int ret;
    Error *err = NULL;

    ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_DATA);
    if (ret) {
        if (ret < 0) {
            error_setg_errno(&err, -ret, "unable to get block status");
            error_report_err(err);

            return NVME_INTERNAL_DEV_ERROR;
        }

        return NVME_DULB;
    }

    return NVME_SUCCESS;
}

static void nvme_aio_err(NvmeRequest *req, int ret)
{
    uint16_t status = NVME_SUCCESS;
    Error *local_err = NULL;

    switch (req->cmd.opcode) {
    case NVME_CMD_READ:
        status = NVME_UNRECOVERED_READ;
        break;
    case NVME_CMD_FLUSH:
    case NVME_CMD_WRITE:
    case NVME_CMD_WRITE_ZEROES:
    case NVME_CMD_ZONE_APPEND:
        status = NVME_WRITE_FAULT;
        break;
    default:
        status = NVME_INTERNAL_DEV_ERROR;
        break;
    }

    trace_pci_nvme_err_aio(nvme_cid(req), strerror(-ret), status);

    error_setg_errno(&local_err, -ret, "aio failed");
    error_report_err(local_err);

    /*
     * Set the command status code to the first encountered error but allow a
     * subsequent Internal Device Error to trump it.
     */
    if (req->status && status != NVME_INTERNAL_DEV_ERROR) {
        return;
    }

    req->status = status;
}

static inline uint32_t nvme_zone_idx(NvmeNamespace *ns, uint64_t slba)
{
    return ns->zone_size_log2 > 0 ? slba >> ns->zone_size_log2 :
                                    slba / ns->zone_size;
}

static inline NvmeZone *nvme_get_zone_by_slba(NvmeNamespace *ns, uint64_t slba)
{
    uint32_t zone_idx = nvme_zone_idx(ns, slba);

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

    return &ns->zone_array[zone_idx];
}

static uint16_t nvme_check_zone_state_for_write(NvmeZone *zone)
{
    uint64_t zslba = zone->d.zslba;

    switch (nvme_get_zone_state(zone)) {
    case NVME_ZONE_STATE_EMPTY:
    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
    case NVME_ZONE_STATE_CLOSED:
        return NVME_SUCCESS;
    case NVME_ZONE_STATE_FULL:
        trace_pci_nvme_err_zone_is_full(zslba);
        return NVME_ZONE_FULL;
    case NVME_ZONE_STATE_OFFLINE:
        trace_pci_nvme_err_zone_is_offline(zslba);
        return NVME_ZONE_OFFLINE;
    case NVME_ZONE_STATE_READ_ONLY:
        trace_pci_nvme_err_zone_is_read_only(zslba);
        return NVME_ZONE_READ_ONLY;
    default:
        assert(false);
    }

    return NVME_INTERNAL_DEV_ERROR;
}

static uint16_t nvme_check_zone_write(NvmeNamespace *ns, NvmeZone *zone,
                                      uint64_t slba, uint32_t nlb)
{
    uint64_t zcap = nvme_zone_wr_boundary(zone);
    uint16_t status;

    status = nvme_check_zone_state_for_write(zone);
    if (status) {
        return status;
    }

    if (zone->d.za & NVME_ZA_ZRWA_VALID) {
        uint64_t ezrwa = zone->w_ptr + 2 * ns->zns.zrwas;

        if (slba < zone->w_ptr || slba + nlb > ezrwa) {
            trace_pci_nvme_err_zone_invalid_write(slba, zone->w_ptr);
            return NVME_ZONE_INVALID_WRITE;
        }
    } else {
        if (unlikely(slba != zone->w_ptr)) {
            trace_pci_nvme_err_write_not_at_wp(slba, zone->d.zslba,
                                               zone->w_ptr);
            return NVME_ZONE_INVALID_WRITE;
        }
    }

    if (unlikely((slba + nlb) > zcap)) {
        trace_pci_nvme_err_zone_boundary(slba, nlb, zcap);
        return NVME_ZONE_BOUNDARY_ERROR;
    }

    return NVME_SUCCESS;
}

static uint16_t nvme_check_zone_state_for_read(NvmeZone *zone)
{
    switch (nvme_get_zone_state(zone)) {
    case NVME_ZONE_STATE_EMPTY:
    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
    case NVME_ZONE_STATE_FULL:
    case NVME_ZONE_STATE_CLOSED:
    case NVME_ZONE_STATE_READ_ONLY:
        return NVME_SUCCESS;
    case NVME_ZONE_STATE_OFFLINE:
        trace_pci_nvme_err_zone_is_offline(zone->d.zslba);
        return NVME_ZONE_OFFLINE;
    default:
        assert(false);
    }

    return NVME_INTERNAL_DEV_ERROR;
}

static uint16_t nvme_check_zone_read(NvmeNamespace *ns, uint64_t slba,
                                     uint32_t nlb)
{
    NvmeZone *zone;
    uint64_t bndry, end;
    uint16_t status;

    zone = nvme_get_zone_by_slba(ns, slba);
    assert(zone);

    bndry = nvme_zone_rd_boundary(ns, zone);
    end = slba + nlb;

    status = nvme_check_zone_state_for_read(zone);
    if (status) {
        ;
    } else if (unlikely(end > bndry)) {
        if (!ns->params.cross_zone_read) {
            status = NVME_ZONE_BOUNDARY_ERROR;
        } else {
            /*
             * Read across zone boundary - check that all subsequent
             * zones that are being read have an appropriate state.
             */
            do {
                zone++;
                status = nvme_check_zone_state_for_read(zone);
                if (status) {
                    break;
                }
            } while (end > nvme_zone_rd_boundary(ns, zone));
        }
    }

    return status;
}

static uint16_t nvme_zrm_finish(NvmeNamespace *ns, NvmeZone *zone)
{
    switch (nvme_get_zone_state(zone)) {
    case NVME_ZONE_STATE_FULL:
        return NVME_SUCCESS;

    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
        nvme_aor_dec_open(ns);
        /* fallthrough */
    case NVME_ZONE_STATE_CLOSED:
        nvme_aor_dec_active(ns);

        if (zone->d.za & NVME_ZA_ZRWA_VALID) {
            zone->d.za &= ~NVME_ZA_ZRWA_VALID;
            if (ns->params.numzrwa) {
                ns->zns.numzrwa++;
            }
        }

        /* fallthrough */
    case NVME_ZONE_STATE_EMPTY:
        nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_FULL);
        return NVME_SUCCESS;

    default:
        return NVME_ZONE_INVAL_TRANSITION;
    }
}

static uint16_t nvme_zrm_close(NvmeNamespace *ns, NvmeZone *zone)
{
    switch (nvme_get_zone_state(zone)) {
    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
        nvme_aor_dec_open(ns);
        nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_CLOSED);
        /* fall through */
    case NVME_ZONE_STATE_CLOSED:
        return NVME_SUCCESS;

    default:
        return NVME_ZONE_INVAL_TRANSITION;
    }
}

static uint16_t nvme_zrm_reset(NvmeNamespace *ns, NvmeZone *zone)
{
    switch (nvme_get_zone_state(zone)) {
    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
        nvme_aor_dec_open(ns);
        /* fallthrough */
    case NVME_ZONE_STATE_CLOSED:
        nvme_aor_dec_active(ns);

        if (zone->d.za & NVME_ZA_ZRWA_VALID) {
            if (ns->params.numzrwa) {
                ns->zns.numzrwa++;
            }
        }

        /* fallthrough */
    case NVME_ZONE_STATE_FULL:
        zone->w_ptr = zone->d.zslba;
        zone->d.wp = zone->w_ptr;
        nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EMPTY);
        /* fallthrough */
    case NVME_ZONE_STATE_EMPTY:
        return NVME_SUCCESS;

    default:
        return NVME_ZONE_INVAL_TRANSITION;
    }
}

static void nvme_zrm_auto_transition_zone(NvmeNamespace *ns)
{
    NvmeZone *zone;

    if (ns->params.max_open_zones &&
        ns->nr_open_zones == ns->params.max_open_zones) {
        zone = QTAILQ_FIRST(&ns->imp_open_zones);
        if (zone) {
            /*
             * Automatically close this implicitly open zone.
             */
            QTAILQ_REMOVE(&ns->imp_open_zones, zone, entry);
            nvme_zrm_close(ns, zone);
        }
    }
}

enum {
    NVME_ZRM_AUTO = 1 << 0,
    NVME_ZRM_ZRWA = 1 << 1,
};

static uint16_t nvme_zrm_open_flags(NvmeCtrl *n, NvmeNamespace *ns,
                                    NvmeZone *zone, int flags)
{
    int act = 0;
    uint16_t status;

    switch (nvme_get_zone_state(zone)) {
    case NVME_ZONE_STATE_EMPTY:
        act = 1;

        /* fallthrough */

    case NVME_ZONE_STATE_CLOSED:
        if (n->params.auto_transition_zones) {
            nvme_zrm_auto_transition_zone(ns);
        }
        status = nvme_zns_check_resources(ns, act, 1,
                                          (flags & NVME_ZRM_ZRWA) ? 1 : 0);
        if (status) {
            return status;
        }

        if (act) {
            nvme_aor_inc_active(ns);
        }

        nvme_aor_inc_open(ns);

        if (flags & NVME_ZRM_AUTO) {
            nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_IMPLICITLY_OPEN);
            return NVME_SUCCESS;
        }

        /* fallthrough */

    case NVME_ZONE_STATE_IMPLICITLY_OPEN:
        if (flags & NVME_ZRM_AUTO) {
            return NVME_SUCCESS;
        }

        nvme_assign_zone_state(ns, zone, NVME_ZONE_STATE_EXPLICITLY_OPEN);

        /* fallthrough */

    case NVME_ZONE_STATE_EXPLICITLY_OPEN:
        if (flags & NVME_ZRM_ZRWA) {
            ns->zns.numzrwa--;

            zone->d.za |= NVME_ZA_ZRWA_VALID;
        }

        return NVME_SUCCESS;

    default:
        return NVME_ZONE_INVAL_TRANSITION;
    }
}

static inline uint16_t nvme_zrm_auto(NvmeCtrl *n, NvmeNamespace *ns,
                                     NvmeZone *zone)
{
    return nvme_zrm_open_flags(n, ns, zone, NVME_ZRM_AUTO);
}

static void nvme_advance_zone_wp(NvmeNamespace *ns, NvmeZone *zone,
                                 uint32_t nlb)
{
    zone->d.wp += nlb;

    if (zone->d.wp == nvme_zone_wr_boundary(zone)) {
        nvme_zrm_finish(ns, zone);
    }
}

static void nvme_zoned_zrwa_implicit_flush(NvmeNamespace *ns, NvmeZone *zone,
                                           uint32_t nlbc)
{
    uint16_t nzrwafgs = DIV_ROUND_UP(nlbc, ns->zns.zrwafg);

    nlbc = nzrwafgs * ns->zns.zrwafg;

    trace_pci_nvme_zoned_zrwa_implicit_flush(zone->d.zslba, nlbc);

    zone->w_ptr += nlbc;

    nvme_advance_zone_wp(ns, zone, nlbc);
}

static void nvme_finalize_zoned_write(NvmeNamespace *ns, NvmeRequest *req)
{
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    NvmeZone *zone;
    uint64_t slba;
    uint32_t nlb;

    slba = le64_to_cpu(rw->slba);
    nlb = le16_to_cpu(rw->nlb) + 1;
    zone = nvme_get_zone_by_slba(ns, slba);
    assert(zone);

    if (zone->d.za & NVME_ZA_ZRWA_VALID) {
        uint64_t ezrwa = zone->w_ptr + ns->zns.zrwas - 1;
        uint64_t elba = slba + nlb - 1;

        if (elba > ezrwa) {
            nvme_zoned_zrwa_implicit_flush(ns, zone, elba - ezrwa);
        }

        return;
    }

    nvme_advance_zone_wp(ns, zone, nlb);
}

static inline bool nvme_is_write(NvmeRequest *req)
{
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;

    return rw->opcode == NVME_CMD_WRITE ||
           rw->opcode == NVME_CMD_ZONE_APPEND ||
           rw->opcode == NVME_CMD_WRITE_ZEROES;
}

static AioContext *nvme_get_aio_context(BlockAIOCB *acb)
{
    return qemu_get_aio_context();
}

static void nvme_misc_cb(void *opaque, int ret)
{
    NvmeRequest *req = opaque;

    trace_pci_nvme_misc_cb(nvme_cid(req));

    if (ret) {
        nvme_aio_err(req, ret);
    }

    nvme_enqueue_req_completion(nvme_cq(req), req);
}

void nvme_rw_complete_cb(void *opaque, int ret)
{
    NvmeRequest *req = opaque;
    NvmeNamespace *ns = req->ns;
    BlockBackend *blk = ns->blkconf.blk;
    BlockAcctCookie *acct = &req->acct;
    BlockAcctStats *stats = blk_get_stats(blk);

    trace_pci_nvme_rw_complete_cb(nvme_cid(req), blk_name(blk));

    if (ret) {
        block_acct_failed(stats, acct);
        nvme_aio_err(req, ret);
    } else {
        block_acct_done(stats, acct);
    }

    if (ns->params.zoned && nvme_is_write(req)) {
        nvme_finalize_zoned_write(ns, req);
    }

    nvme_enqueue_req_completion(nvme_cq(req), req);
}

static void nvme_rw_cb(void *opaque, int ret)
{
    NvmeRequest *req = opaque;
    NvmeNamespace *ns = req->ns;

    BlockBackend *blk = ns->blkconf.blk;

    trace_pci_nvme_rw_cb(nvme_cid(req), blk_name(blk));

    if (ret) {
        goto out;
    }

    if (ns->lbaf.ms) {
        NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
        uint64_t slba = le64_to_cpu(rw->slba);
        uint32_t nlb = (uint32_t)le16_to_cpu(rw->nlb) + 1;
        uint64_t offset = nvme_moff(ns, slba);

        if (req->cmd.opcode == NVME_CMD_WRITE_ZEROES) {
            size_t mlen = nvme_m2b(ns, nlb);

            req->aiocb = blk_aio_pwrite_zeroes(blk, offset, mlen,
                                               BDRV_REQ_MAY_UNMAP,
                                               nvme_rw_complete_cb, req);
            return;
        }

        if (nvme_ns_ext(ns) || req->cmd.mptr) {
            uint16_t status;

            nvme_sg_unmap(&req->sg);
            status = nvme_map_mdata(nvme_ctrl(req), nlb, req);
            if (status) {
                ret = -EFAULT;
                goto out;
            }

            if (req->cmd.opcode == NVME_CMD_READ) {
                return nvme_blk_read(blk, offset, nvme_rw_complete_cb, req);
            }

            return nvme_blk_write(blk, offset, nvme_rw_complete_cb, req);
        }
    }

out:
    nvme_rw_complete_cb(req, ret);
}

static void nvme_verify_cb(void *opaque, int ret)
{
    NvmeBounceContext *ctx = opaque;
    NvmeRequest *req = ctx->req;
    NvmeNamespace *ns = req->ns;
    BlockBackend *blk = ns->blkconf.blk;
    BlockAcctCookie *acct = &req->acct;
    BlockAcctStats *stats = blk_get_stats(blk);
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    uint64_t slba = le64_to_cpu(rw->slba);
    uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
    uint16_t apptag = le16_to_cpu(rw->apptag);
    uint16_t appmask = le16_to_cpu(rw->appmask);
    uint64_t reftag = le32_to_cpu(rw->reftag);
    uint64_t cdw3 = le32_to_cpu(rw->cdw3);
    uint16_t status;

    reftag |= cdw3 << 32;

    trace_pci_nvme_verify_cb(nvme_cid(req), prinfo, apptag, appmask, reftag);

    if (ret) {
        block_acct_failed(stats, acct);
        nvme_aio_err(req, ret);
        goto out;
    }

    block_acct_done(stats, acct);

    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
        status = nvme_dif_mangle_mdata(ns, ctx->mdata.bounce,
                                       ctx->mdata.iov.size, slba);
        if (status) {
            req->status = status;
            goto out;
        }

        req->status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
                                     ctx->mdata.bounce, ctx->mdata.iov.size,
                                     prinfo, slba, apptag, appmask, &reftag);
    }

out:
    qemu_iovec_destroy(&ctx->data.iov);
    g_free(ctx->data.bounce);

    qemu_iovec_destroy(&ctx->mdata.iov);
    g_free(ctx->mdata.bounce);

    g_free(ctx);

    nvme_enqueue_req_completion(nvme_cq(req), req);
}


static void nvme_verify_mdata_in_cb(void *opaque, int ret)
{
    NvmeBounceContext *ctx = opaque;
    NvmeRequest *req = ctx->req;
    NvmeNamespace *ns = req->ns;
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    uint64_t slba = le64_to_cpu(rw->slba);
    uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
    size_t mlen = nvme_m2b(ns, nlb);
    uint64_t offset = nvme_moff(ns, slba);
    BlockBackend *blk = ns->blkconf.blk;

    trace_pci_nvme_verify_mdata_in_cb(nvme_cid(req), blk_name(blk));

    if (ret) {
        goto out;
    }

    ctx->mdata.bounce = g_malloc(mlen);

    qemu_iovec_reset(&ctx->mdata.iov);
    qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);

    req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
                                nvme_verify_cb, ctx);
    return;

out:
    nvme_verify_cb(ctx, ret);
}

struct nvme_compare_ctx {
    struct {
        QEMUIOVector iov;
        uint8_t *bounce;
    } data;

    struct {
        QEMUIOVector iov;
        uint8_t *bounce;
    } mdata;
};

static void nvme_compare_mdata_cb(void *opaque, int ret)
{
    NvmeRequest *req = opaque;
    NvmeNamespace *ns = req->ns;
    NvmeCtrl *n = nvme_ctrl(req);
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
    uint16_t apptag = le16_to_cpu(rw->apptag);
    uint16_t appmask = le16_to_cpu(rw->appmask);
    uint64_t reftag = le32_to_cpu(rw->reftag);
    uint64_t cdw3 = le32_to_cpu(rw->cdw3);
    struct nvme_compare_ctx *ctx = req->opaque;
    g_autofree uint8_t *buf = NULL;
    BlockBackend *blk = ns->blkconf.blk;
    BlockAcctCookie *acct = &req->acct;
    BlockAcctStats *stats = blk_get_stats(blk);
    uint16_t status = NVME_SUCCESS;

    reftag |= cdw3 << 32;

    trace_pci_nvme_compare_mdata_cb(nvme_cid(req));

    if (ret) {
        block_acct_failed(stats, acct);
        nvme_aio_err(req, ret);
        goto out;
    }

    buf = g_malloc(ctx->mdata.iov.size);

    status = nvme_bounce_mdata(n, buf, ctx->mdata.iov.size,
                               NVME_TX_DIRECTION_TO_DEVICE, req);
    if (status) {
        req->status = status;
        goto out;
    }

    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
        uint64_t slba = le64_to_cpu(rw->slba);
        uint8_t *bufp;
        uint8_t *mbufp = ctx->mdata.bounce;
        uint8_t *end = mbufp + ctx->mdata.iov.size;
        int16_t pil = 0;

        status = nvme_dif_check(ns, ctx->data.bounce, ctx->data.iov.size,
                                ctx->mdata.bounce, ctx->mdata.iov.size, prinfo,
                                slba, apptag, appmask, &reftag);
        if (status) {
            req->status = status;
            goto out;
        }

        /*
         * When formatted with protection information, do not compare the DIF
         * tuple.
         */
        if (!(ns->id_ns.dps & NVME_ID_NS_DPS_FIRST_EIGHT)) {
            pil = ns->lbaf.ms - nvme_pi_tuple_size(ns);
        }

        for (bufp = buf; mbufp < end; bufp += ns->lbaf.ms, mbufp += ns->lbaf.ms) {
            if (memcmp(bufp + pil, mbufp + pil, ns->lbaf.ms - pil)) {
                req->status = NVME_CMP_FAILURE;
                goto out;
            }
        }

        goto out;
    }

    if (memcmp(buf, ctx->mdata.bounce, ctx->mdata.iov.size)) {
        req->status = NVME_CMP_FAILURE;
        goto out;
    }

    block_acct_done(stats, acct);

out:
    qemu_iovec_destroy(&ctx->data.iov);
    g_free(ctx->data.bounce);

    qemu_iovec_destroy(&ctx->mdata.iov);
    g_free(ctx->mdata.bounce);

    g_free(ctx);

    nvme_enqueue_req_completion(nvme_cq(req), req);
}

static void nvme_compare_data_cb(void *opaque, int ret)
{
    NvmeRequest *req = opaque;
    NvmeCtrl *n = nvme_ctrl(req);
    NvmeNamespace *ns = req->ns;
    BlockBackend *blk = ns->blkconf.blk;
    BlockAcctCookie *acct = &req->acct;
    BlockAcctStats *stats = blk_get_stats(blk);

    struct nvme_compare_ctx *ctx = req->opaque;
    g_autofree uint8_t *buf = NULL;
    uint16_t status;

    trace_pci_nvme_compare_data_cb(nvme_cid(req));

    if (ret) {
        block_acct_failed(stats, acct);
        nvme_aio_err(req, ret);
        goto out;
    }

    buf = g_malloc(ctx->data.iov.size);

    status = nvme_bounce_data(n, buf, ctx->data.iov.size,
                              NVME_TX_DIRECTION_TO_DEVICE, req);
    if (status) {
        req->status = status;
        goto out;
    }

    if (memcmp(buf, ctx->data.bounce, ctx->data.iov.size)) {
        req->status = NVME_CMP_FAILURE;
        goto out;
    }

    if (ns->lbaf.ms) {
        NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
        uint64_t slba = le64_to_cpu(rw->slba);
        uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
        size_t mlen = nvme_m2b(ns, nlb);
        uint64_t offset = nvme_moff(ns, slba);

        ctx->mdata.bounce = g_malloc(mlen);

        qemu_iovec_init(&ctx->mdata.iov, 1);
        qemu_iovec_add(&ctx->mdata.iov, ctx->mdata.bounce, mlen);

        req->aiocb = blk_aio_preadv(blk, offset, &ctx->mdata.iov, 0,
                                    nvme_compare_mdata_cb, req);
        return;
    }

    block_acct_done(stats, acct);

out:
    qemu_iovec_destroy(&ctx->data.iov);
    g_free(ctx->data.bounce);
    g_free(ctx);

    nvme_enqueue_req_completion(nvme_cq(req), req);
}

typedef struct NvmeDSMAIOCB {
    BlockAIOCB common;
    BlockAIOCB *aiocb;
    NvmeRequest *req;
    int ret;

    NvmeDsmRange *range;
    unsigned int nr;
    unsigned int idx;
} NvmeDSMAIOCB;

static void nvme_dsm_cancel(BlockAIOCB *aiocb)
{
    NvmeDSMAIOCB *iocb = container_of(aiocb, NvmeDSMAIOCB, common);

    /* break nvme_dsm_cb loop */
    iocb->idx = iocb->nr;
    iocb->ret = -ECANCELED;

    if (iocb->aiocb) {
        blk_aio_cancel_async(iocb->aiocb);
        iocb->aiocb = NULL;
    } else {
        /*
         * We only reach this if nvme_dsm_cancel() has already been called or
         * the command ran to completion.
         */
        assert(iocb->idx == iocb->nr);
    }
}

static const AIOCBInfo nvme_dsm_aiocb_info = {
    .aiocb_size   = sizeof(NvmeDSMAIOCB),
    .cancel_async = nvme_dsm_cancel,
};

static void nvme_dsm_cb(void *opaque, int ret);

static void nvme_dsm_md_cb(void *opaque, int ret)
{
    NvmeDSMAIOCB *iocb = opaque;
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    NvmeDsmRange *range;
    uint64_t slba;
    uint32_t nlb;

    if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
        goto done;
    }

    range = &iocb->range[iocb->idx - 1];
    slba = le64_to_cpu(range->slba);
    nlb = le32_to_cpu(range->nlb);

    /*
     * Check that all block were discarded (zeroed); otherwise we do not zero
     * the metadata.
     */

    ret = nvme_block_status_all(ns, slba, nlb, BDRV_BLOCK_ZERO);
    if (ret) {
        if (ret < 0) {
            goto done;
        }

        nvme_dsm_cb(iocb, 0);
        return;
    }

    iocb->aiocb = blk_aio_pwrite_zeroes(ns->blkconf.blk, nvme_moff(ns, slba),
                                        nvme_m2b(ns, nlb), BDRV_REQ_MAY_UNMAP,
                                        nvme_dsm_cb, iocb);
    return;

done:
    nvme_dsm_cb(iocb, ret);
}

static void nvme_dsm_cb(void *opaque, int ret)
{
    NvmeDSMAIOCB *iocb = opaque;
    NvmeRequest *req = iocb->req;
    NvmeCtrl *n = nvme_ctrl(req);
    NvmeNamespace *ns = req->ns;
    NvmeDsmRange *range;
    uint64_t slba;
    uint32_t nlb;

    if (iocb->ret < 0) {
        goto done;
    } else if (ret < 0) {
        iocb->ret = ret;
        goto done;
    }

next:
    if (iocb->idx == iocb->nr) {
        goto done;
    }

    range = &iocb->range[iocb->idx++];
    slba = le64_to_cpu(range->slba);
    nlb = le32_to_cpu(range->nlb);

    trace_pci_nvme_dsm_deallocate(slba, nlb);

    if (nlb > n->dmrsl) {
        trace_pci_nvme_dsm_single_range_limit_exceeded(nlb, n->dmrsl);
        goto next;
    }

    if (nvme_check_bounds(ns, slba, nlb)) {
        trace_pci_nvme_err_invalid_lba_range(slba, nlb,
                                             ns->id_ns.nsze);
        goto next;
    }

    iocb->aiocb = blk_aio_pdiscard(ns->blkconf.blk, nvme_l2b(ns, slba),
                                   nvme_l2b(ns, nlb),
                                   nvme_dsm_md_cb, iocb);
    return;

done:
    iocb->aiocb = NULL;
    iocb->common.cb(iocb->common.opaque, iocb->ret);
    qemu_aio_unref(iocb);
}

static uint16_t nvme_dsm(NvmeCtrl *n, NvmeRequest *req)
{
    NvmeNamespace *ns = req->ns;
    NvmeDsmCmd *dsm = (NvmeDsmCmd *) &req->cmd;
    uint32_t attr = le32_to_cpu(dsm->attributes);
    uint32_t nr = (le32_to_cpu(dsm->nr) & 0xff) + 1;
    uint16_t status = NVME_SUCCESS;

    trace_pci_nvme_dsm(nr, attr);

    if (attr & NVME_DSMGMT_AD) {
        NvmeDSMAIOCB *iocb = blk_aio_get(&nvme_dsm_aiocb_info, ns->blkconf.blk,
                                         nvme_misc_cb, req);

        iocb->req = req;
        iocb->ret = 0;
        iocb->range = g_new(NvmeDsmRange, nr);
        iocb->nr = nr;
        iocb->idx = 0;

        status = nvme_h2c(n, (uint8_t *)iocb->range, sizeof(NvmeDsmRange) * nr,
                          req);
        if (status) {
            return status;
        }

        req->aiocb = &iocb->common;
        nvme_dsm_cb(iocb, 0);

        return NVME_NO_COMPLETE;
    }

    return status;
}

static uint16_t nvme_verify(NvmeCtrl *n, NvmeRequest *req)
{
    NvmeRwCmd *rw = (NvmeRwCmd *)&req->cmd;
    NvmeNamespace *ns = req->ns;
    BlockBackend *blk = ns->blkconf.blk;
    uint64_t slba = le64_to_cpu(rw->slba);
    uint32_t nlb = le16_to_cpu(rw->nlb) + 1;
    size_t len = nvme_l2b(ns, nlb);
    int64_t offset = nvme_l2b(ns, slba);
    uint8_t prinfo = NVME_RW_PRINFO(le16_to_cpu(rw->control));
    uint32_t reftag = le32_to_cpu(rw->reftag);
    NvmeBounceContext *ctx = NULL;
    uint16_t status;

    trace_pci_nvme_verify(nvme_cid(req), nvme_nsid(ns), slba, nlb);

    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
        status = nvme_check_prinfo(ns, prinfo, slba, reftag);
        if (status) {
            return status;
        }

        if (prinfo & NVME_PRINFO_PRACT) {
            return NVME_INVALID_PROT_INFO | NVME_DNR;
        }
    }

    if (len > n->page_size << n->params.vsl) {
        return NVME_INVALID_FIELD | NVME_DNR;
    }

    status = nvme_check_bounds(ns, slba, nlb);
    if (status) {
        return status;
    }

    if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
        status = nvme_check_dulbe(ns, slba, nlb);
        if (status) {
            return status;
        }
    }

    ctx = g_new0(NvmeBounceContext, 1);
    ctx->req = req;

    ctx->data.bounce = g_malloc(len);

    qemu_iovec_init(&ctx->data.iov, 1);
    qemu_iovec_add(&ctx->data.iov, ctx->data.bounce, len);

    block_acct_start(blk_get_stats(blk), &req->acct, ctx->data.iov.size,
                     BLOCK_ACCT_READ);

    req->aiocb = blk_aio_preadv(ns->blkconf.blk, offset, &ctx->data.iov, 0,
                                nvme_verify_mdata_in_cb, ctx);
    return NVME_NO_COMPLETE;
}

typedef struct NvmeCopyAIOCB {
    BlockAIOCB common;
    BlockAIOCB *aiocb;
    NvmeRequest *req;
    int ret;

    void *ranges;
    unsigned int format;
    int nr;
    int idx;

    uint8_t *bounce;
    QEMUIOVector iov;
    struct {
        BlockAcctCookie read;
        BlockAcctCookie write;
    } acct;

    uint64_t reftag;
    uint64_t slba;

    NvmeZone *zone;
} NvmeCopyAIOCB;

static void nvme_copy_cancel(BlockAIOCB *aiocb)
{
    NvmeCopyAIOCB *iocb = container_of(aiocb, NvmeCopyAIOCB, common);

    iocb->ret = -ECANCELED;

    if (iocb->aiocb) {
        blk_aio_cancel_async(iocb->aiocb);
        iocb->aiocb = NULL;
    }
}

static const AIOCBInfo nvme_copy_aiocb_info = {
    .aiocb_size   = sizeof(NvmeCopyAIOCB),
    .cancel_async = nvme_copy_cancel,
};

static void nvme_copy_done(NvmeCopyAIOCB *iocb)
{
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    BlockAcctStats *stats = blk_get_stats(ns->blkconf.blk);

    if (iocb->idx != iocb->nr) {
        req->cqe.result = cpu_to_le32(iocb->idx);
    }

    qemu_iovec_destroy(&iocb->iov);
    g_free(iocb->bounce);

    if (iocb->ret < 0) {
        block_acct_failed(stats, &iocb->acct.read);
        block_acct_failed(stats, &iocb->acct.write);
    } else {
        block_acct_done(stats, &iocb->acct.read);
        block_acct_done(stats, &iocb->acct.write);
    }

    iocb->common.cb(iocb->common.opaque, iocb->ret);
    qemu_aio_unref(iocb);
}

static void nvme_do_copy(NvmeCopyAIOCB *iocb);

static void nvme_copy_source_range_parse_format0(void *ranges, int idx,
                                                 uint64_t *slba, uint32_t *nlb,
                                                 uint16_t *apptag,
                                                 uint16_t *appmask,
                                                 uint64_t *reftag)
{
    NvmeCopySourceRangeFormat0 *_ranges = ranges;

    if (slba) {
        *slba = le64_to_cpu(_ranges[idx].slba);
    }

    if (nlb) {
        *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
    }

    if (apptag) {
        *apptag = le16_to_cpu(_ranges[idx].apptag);
    }

    if (appmask) {
        *appmask = le16_to_cpu(_ranges[idx].appmask);
    }

    if (reftag) {
        *reftag = le32_to_cpu(_ranges[idx].reftag);
    }
}

static void nvme_copy_source_range_parse_format1(void *ranges, int idx,
                                                 uint64_t *slba, uint32_t *nlb,
                                                 uint16_t *apptag,
                                                 uint16_t *appmask,
                                                 uint64_t *reftag)
{
    NvmeCopySourceRangeFormat1 *_ranges = ranges;

    if (slba) {
        *slba = le64_to_cpu(_ranges[idx].slba);
    }

    if (nlb) {
        *nlb = le16_to_cpu(_ranges[idx].nlb) + 1;
    }

    if (apptag) {
        *apptag = le16_to_cpu(_ranges[idx].apptag);
    }

    if (appmask) {
        *appmask = le16_to_cpu(_ranges[idx].appmask);
    }

    if (reftag) {
        *reftag = 0;

        *reftag |= (uint64_t)_ranges[idx].sr[4] << 40;
        *reftag |= (uint64_t)_ranges[idx].sr[5] << 32;
        *reftag |= (uint64_t)_ranges[idx].sr[6] << 24;
        *reftag |= (uint64_t)_ranges[idx].sr[7] << 16;
        *reftag |= (uint64_t)_ranges[idx].sr[8] << 8;
        *reftag |= (uint64_t)_ranges[idx].sr[9];
    }
}

static void nvme_copy_source_range_parse(void *ranges, int idx, uint8_t format,
                                         uint64_t *slba, uint32_t *nlb,
                                         uint16_t *apptag, uint16_t *appmask,
                                         uint64_t *reftag)
{
    switch (format) {
    case NVME_COPY_FORMAT_0:
        nvme_copy_source_range_parse_format0(ranges, idx, slba, nlb, apptag,
                                             appmask, reftag);
        break;

    case NVME_COPY_FORMAT_1:
        nvme_copy_source_range_parse_format1(ranges, idx, slba, nlb, apptag,
                                             appmask, reftag);
        break;

    default:
        abort();
    }
}

static void nvme_copy_out_completed_cb(void *opaque, int ret)
{
    NvmeCopyAIOCB *iocb = opaque;
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    uint32_t nlb;

    nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
                                 &nlb, NULL, NULL, NULL);

    if (ret < 0) {
        iocb->ret = ret;
        goto out;
    } else if (iocb->ret < 0) {
        goto out;
    }

    if (ns->params.zoned) {
        nvme_advance_zone_wp(ns, iocb->zone, nlb);
    }

    iocb->idx++;
    iocb->slba += nlb;
out:
    nvme_do_copy(iocb);
}

static void nvme_copy_out_cb(void *opaque, int ret)
{
    NvmeCopyAIOCB *iocb = opaque;
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    uint32_t nlb;
    size_t mlen;
    uint8_t *mbounce;

    if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
        goto out;
    }

    nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, NULL,
                                 &nlb, NULL, NULL, NULL);

    mlen = nvme_m2b(ns, nlb);
    mbounce = iocb->bounce + nvme_l2b(ns, nlb);

    qemu_iovec_reset(&iocb->iov);
    qemu_iovec_add(&iocb->iov, mbounce, mlen);

    iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_moff(ns, iocb->slba),
                                  &iocb->iov, 0, nvme_copy_out_completed_cb,
                                  iocb);

    return;

out:
    nvme_copy_out_completed_cb(iocb, ret);
}

static void nvme_copy_in_completed_cb(void *opaque, int ret)
{
    NvmeCopyAIOCB *iocb = opaque;
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    uint32_t nlb;
    uint64_t slba;
    uint16_t apptag, appmask;
    uint64_t reftag;
    size_t len;
    uint16_t status;

    if (ret < 0) {
        iocb->ret = ret;
        goto out;
    } else if (iocb->ret < 0) {
        goto out;
    }

    nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
                                 &nlb, &apptag, &appmask, &reftag);
    len = nvme_l2b(ns, nlb);

    trace_pci_nvme_copy_out(iocb->slba, nlb);

    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps)) {
        NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;

        uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
        uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);

        size_t mlen = nvme_m2b(ns, nlb);
        uint8_t *mbounce = iocb->bounce + nvme_l2b(ns, nlb);

        status = nvme_dif_mangle_mdata(ns, mbounce, mlen, slba);
        if (status) {
            goto invalid;
        }
        status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen, prinfor,
                                slba, apptag, appmask, &reftag);
        if (status) {
            goto invalid;
        }

        apptag = le16_to_cpu(copy->apptag);
        appmask = le16_to_cpu(copy->appmask);

        if (prinfow & NVME_PRINFO_PRACT) {
            status = nvme_check_prinfo(ns, prinfow, iocb->slba, iocb->reftag);
            if (status) {
                goto invalid;
            }

            nvme_dif_pract_generate_dif(ns, iocb->bounce, len, mbounce, mlen,
                                        apptag, &iocb->reftag);
        } else {
            status = nvme_dif_check(ns, iocb->bounce, len, mbounce, mlen,
                                    prinfow, iocb->slba, apptag, appmask,
                                    &iocb->reftag);
            if (status) {
                goto invalid;
            }
        }
    }

    status = nvme_check_bounds(ns, iocb->slba, nlb);
    if (status) {
        goto invalid;
    }

    if (ns->params.zoned) {
        status = nvme_check_zone_write(ns, iocb->zone, iocb->slba, nlb);
        if (status) {
            goto invalid;
        }

        if (!(iocb->zone->d.za & NVME_ZA_ZRWA_VALID)) {
            iocb->zone->w_ptr += nlb;
        }
    }

    qemu_iovec_reset(&iocb->iov);
    qemu_iovec_add(&iocb->iov, iocb->bounce, len);

    iocb->aiocb = blk_aio_pwritev(ns->blkconf.blk, nvme_l2b(ns, iocb->slba),
                                  &iocb->iov, 0, nvme_copy_out_cb, iocb);

    return;

invalid:
    req->status = status;
    iocb->ret = -1;
out:
    nvme_do_copy(iocb);
}

static void nvme_copy_in_cb(void *opaque, int ret)
{
    NvmeCopyAIOCB *iocb = opaque;
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    uint64_t slba;
    uint32_t nlb;

    if (ret < 0 || iocb->ret < 0 || !ns->lbaf.ms) {
        goto out;
    }

    nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
                                 &nlb, NULL, NULL, NULL);

    qemu_iovec_reset(&iocb->iov);
    qemu_iovec_add(&iocb->iov, iocb->bounce + nvme_l2b(ns, nlb),
                   nvme_m2b(ns, nlb));

    iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_moff(ns, slba),
                                 &iocb->iov, 0, nvme_copy_in_completed_cb,
                                 iocb);
    return;

out:
    nvme_copy_in_completed_cb(iocb, ret);
}

static void nvme_do_copy(NvmeCopyAIOCB *iocb)
{
    NvmeRequest *req = iocb->req;
    NvmeNamespace *ns = req->ns;
    uint64_t slba;
    uint32_t nlb;
    size_t len;
    uint16_t status;

    if (iocb->ret < 0) {
        goto done;
    }

    if (iocb->idx == iocb->nr) {
        goto done;
    }

    nvme_copy_source_range_parse(iocb->ranges, iocb->idx, iocb->format, &slba,
                                 &nlb, NULL, NULL, NULL);
    len = nvme_l2b(ns, nlb);

    trace_pci_nvme_copy_source_range(slba, nlb);

    if (nlb > le16_to_cpu(ns->id_ns.mssrl)) {
        status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
        goto invalid;
    }

    status = nvme_check_bounds(ns, slba, nlb);
    if (status) {
        goto invalid;
    }

    if (NVME_ERR_REC_DULBE(ns->features.err_rec)) {
        status = nvme_check_dulbe(ns, slba, nlb);
        if (status) {
            goto invalid;
        }
    }

    if (ns->params.zoned) {
        status = nvme_check_zone_read(ns, slba, nlb);
        if (status) {
            goto invalid;
        }
    }

    qemu_iovec_reset(&iocb->iov);
    qemu_iovec_add(&iocb->iov, iocb->bounce, len);

    iocb->aiocb = blk_aio_preadv(ns->blkconf.blk, nvme_l2b(ns, slba),
                                 &iocb->iov, 0, nvme_copy_in_cb, iocb);
    return;

invalid:
    req->status = status;
    iocb->ret = -1;
done:
    nvme_copy_done(iocb);
}

static uint16_t nvme_copy(NvmeCtrl *n, NvmeRequest *req)
{
    NvmeNamespace *ns = req->ns;
    NvmeCopyCmd *copy = (NvmeCopyCmd *)&req->cmd;
    NvmeCopyAIOCB *iocb = blk_aio_get(&nvme_copy_aiocb_info, ns->blkconf.blk,
                                      nvme_misc_cb, req);
    uint16_t nr = copy->nr + 1;
    uint8_t format = copy->control[0] & 0xf;
    uint16_t prinfor = ((copy->control[0] >> 4) & 0xf);
    uint16_t prinfow = ((copy->control[2] >> 2) & 0xf);
    size_t len = sizeof(NvmeCopySourceRangeFormat0);

    uint16_t status;

    trace_pci_nvme_copy(nvme_cid(req), nvme_nsid(ns), nr, format);

    iocb->ranges = NULL;
    iocb->zone = NULL;

    if (NVME_ID_NS_DPS_TYPE(ns->id_ns.dps) &&
        ((prinfor & NVME_PRINFO_PRACT) != (prinfow & NVME_PRINFO_PRACT))) {
        status = NVME_INVALID_FIELD | NVME_DNR;
        goto invalid;
    }

    if (!(n->id_ctrl.ocfs & (1 << format))) {
        trace_pci_nvme_err_copy_invalid_format(format);
        status = NVME_INVALID_FIELD | NVME_DNR;
        goto invalid;
    }

    if (nr > ns->id_ns.msrc + 1) {
        status = NVME_CMD_SIZE_LIMIT | NVME_DNR;
        goto invalid;
    }

    if ((ns->pif == 0x0 && format != 0x0) ||
        (ns->pif != 0x0 && format != 0x1)) {
        status = NVME_INVALID_FORMAT | NVME_DNR;
        goto invalid;
    }

    if (ns->pif) {
        len = sizeof(NvmeCopySourceRangeFormat1);
    }

    iocb->format = format;
    iocb->ranges = g_malloc_n(nr, len);
    status = nvme_h2c(n, (uint8_t *)iocb->ranges, len * nr, req);
    if (status) {
        goto invalid;
    }