xref: /illumos-gate/usr/src/uts/common/io/blkdev/blkdev.c (revision 744642a2)

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright 2012 Garrett D'Amore <garrett@damore.org>.  All rights reserved.
 * Copyright 2012 Alexey Zaytsev <alexey.zaytsev@gmail.com> All rights reserved.
 * Copyright 2017 The MathWorks, Inc.  All rights reserved.
 * Copyright 2020 Joyent, Inc.
 * Copyright 2022 OmniOS Community Edition (OmniOSce) Association.
 * Copyright 2022 Tintri by DDN, Inc. All rights reserved.
 * Copyright 2023 Oxide Computer Company
 */

#include <sys/types.h>
#include <sys/ksynch.h>
#include <sys/kmem.h>
#include <sys/file.h>
#include <sys/errno.h>
#include <sys/open.h>
#include <sys/buf.h>
#include <sys/uio.h>
#include <sys/aio_req.h>
#include <sys/cred.h>
#include <sys/modctl.h>
#include <sys/cmlb.h>
#include <sys/conf.h>
#include <sys/devops.h>
#include <sys/list.h>
#include <sys/sysmacros.h>
#include <sys/dkio.h>
#include <sys/dkioc_free_util.h>
#include <sys/vtoc.h>
#include <sys/scsi/scsi.h>	/* for DTYPE_DIRECT */
#include <sys/kstat.h>
#include <sys/fs/dv_node.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/note.h>
#include <sys/blkdev.h>
#include <sys/scsi/impl/inquiry.h>
#include <sys/taskq.h>
#include <sys/taskq_impl.h>
#include <sys/disp.h>
#include <sys/sysevent/eventdefs.h>
#include <sys/sysevent/dev.h>

/*
 * blkdev is a driver which provides a lot of the common functionality
 * a block device driver may need and helps by removing code which
 * is frequently duplicated in block device drivers.
 *
 * Within this driver all the struct cb_ops functions required for a
 * block device driver are written with appropriate call back functions
 * to be provided by the parent driver.
 *
 * To use blkdev, a driver needs to:
 *	1. Create a bd_ops_t structure which has the call back operations
 *	   blkdev will use.
 *	2. Create a handle by calling bd_alloc_handle(). One of the
 *	   arguments to this function is the bd_ops_t.
 *	3. Call bd_attach_handle(). This will instantiate a blkdev device
 *	   as a child device node of the calling driver.
 *
 * A parent driver is not restricted to just allocating and attaching a
 * single instance, it may attach as many as it wishes. For each handle
 * attached, appropriate entries in /dev/[r]dsk are created.
 *
 * The bd_ops_t routines that a parent of blkdev need to provide are:
 *
 * o_drive_info: Provide information to blkdev such as how many I/O queues
 *		 to create and the size of those queues. Also some device
 *		 specifics such as EUI, vendor, product, model, serial
 *		 number ....
 *
 * o_media_info: Provide information about the media. Eg size and block size.
 *
 * o_devid_init: Creates and initializes the device id. Typically calls
 *		 ddi_devid_init().
 *
 * o_sync_cache: Issues a device appropriate command to flush any write
 *		 caches.
 *
 * o_read:	 Read data as described by bd_xfer_t argument.
 *
 * o_write:	 Write data as described by bd_xfer_t argument.
 *
 * o_free_space: Free the space described by bd_xfer_t argument (optional).
 *
 * Queues
 * ------
 * Part of the drive_info data is a queue count. blkdev will create
 * "queue count" number of waitq/runq pairs. Each waitq/runq pair
 * operates independently. As an I/O is scheduled up to the parent
 * driver via o_read or o_write its queue number is given. If the
 * parent driver supports multiple hardware queues it can then select
 * where to submit the I/O request.
 *
 * Currently blkdev uses a simplistic round-robin queue selection method.
 * It has the advantage that it is lockless. In the future it will be
 * worthwhile reviewing this strategy for something which prioritizes queues
 * depending on how busy they are.
 *
 * Each waitq/runq pair is protected by its mutex (q_iomutex). Incoming
 * I/O requests are initially added to the waitq. They are taken off the
 * waitq, added to the runq and submitted, providing the runq is less
 * than the qsize as specified in the drive_info. As an I/O request
 * completes, the parent driver is required to call bd_xfer_done(), which
 * will remove the I/O request from the runq and pass I/O completion
 * status up the stack.
 *
 * Locks
 * -----
 * There are 5 instance global locks d_ocmutex, d_ksmutex, d_errmutex,
 * d_statemutex and d_dle_mutex. As well a q_iomutex per waitq/runq pair.
 *
 * Lock Hierarchy
 * --------------
 * The only two locks which may be held simultaneously are q_iomutex and
 * d_ksmutex. In all cases q_iomutex must be acquired before d_ksmutex.
 */

#define	BD_MAXPART	64
#define	BDINST(dev)	(getminor(dev) / BD_MAXPART)
#define	BDPART(dev)	(getminor(dev) % BD_MAXPART)

typedef struct bd bd_t;
typedef struct bd_xfer_impl bd_xfer_impl_t;
typedef struct bd_queue bd_queue_t;

typedef enum {
	BD_DLE_PENDING	= 1 << 0,
	BD_DLE_RUNNING	= 1 << 1
} bd_dle_state_t;

struct bd {
	void		*d_private;
	dev_info_t	*d_dip;
	kmutex_t	d_ocmutex;	/* open/close */
	kmutex_t	d_ksmutex;	/* kstat */
	kmutex_t	d_errmutex;
	kmutex_t	d_statemutex;
	kcondvar_t	d_statecv;
	enum dkio_state	d_state;
	cmlb_handle_t	d_cmlbh;
	unsigned	d_open_lyr[BD_MAXPART];	/* open count */
	uint64_t	d_open_excl;	/* bit mask indexed by partition */
	uint64_t	d_open_reg[OTYPCNT];		/* bit mask */
	uint64_t	d_io_counter;

	uint32_t	d_qcount;
	uint32_t	d_qactive;
	uint32_t	d_maxxfer;
	uint32_t	d_blkshift;
	uint32_t	d_pblkshift;
	uint64_t	d_numblks;
	ddi_devid_t	d_devid;

	uint64_t	d_max_free_seg;
	uint64_t	d_max_free_blks;
	uint64_t	d_max_free_seg_blks;
	uint64_t	d_free_align;

	kmem_cache_t	*d_cache;
	bd_queue_t	*d_queues;
	kstat_t		*d_ksp;
	kstat_io_t	*d_kiop;
	kstat_t		*d_errstats;
	struct bd_errstats *d_kerr;

	boolean_t	d_rdonly;
	boolean_t	d_ssd;
	boolean_t	d_removable;
	boolean_t	d_hotpluggable;
	boolean_t	d_use_dma;

	ddi_dma_attr_t	d_dma;
	bd_ops_t	d_ops;
	bd_handle_t	d_handle;

	kmutex_t	d_dle_mutex;
	taskq_ent_t	d_dle_ent;
	bd_dle_state_t	d_dle_state;
};

struct bd_handle {
	bd_ops_t	h_ops;
	ddi_dma_attr_t	*h_dma;
	dev_info_t	*h_parent;
	dev_info_t	*h_child;
	void		*h_private;
	bd_t		*h_bd;
	char		*h_name;
	char		h_addr[50];	/* enough for w%0.32x,%X */
};

struct bd_xfer_impl {
	bd_xfer_t	i_public;
	list_node_t	i_linkage;
	bd_t		*i_bd;
	buf_t		*i_bp;
	bd_queue_t	*i_bq;
	uint_t		i_num_win;
	uint_t		i_cur_win;
	off_t		i_offset;
	int		(*i_func)(void *, bd_xfer_t *);
	uint32_t	i_blkshift;
	size_t		i_len;
	size_t		i_resid;
};

struct bd_queue {
	kmutex_t	q_iomutex;
	uint32_t	q_qsize;
	uint32_t	q_qactive;
	list_t		q_runq;
	list_t		q_waitq;
};

#define	i_dmah		i_public.x_dmah
#define	i_dmac		i_public.x_dmac
#define	i_ndmac		i_public.x_ndmac
#define	i_kaddr		i_public.x_kaddr
#define	i_nblks		i_public.x_nblks
#define	i_blkno		i_public.x_blkno
#define	i_flags		i_public.x_flags
#define	i_qnum		i_public.x_qnum
#define	i_dfl		i_public.x_dfl

#define	CAN_FREESPACE(bd) \
	(((bd)->d_ops.o_free_space == NULL) ? B_FALSE : B_TRUE)

/*
 * Private prototypes.
 */

static void bd_prop_update_inqstring(dev_info_t *, char *, char *, size_t);
static void bd_create_inquiry_props(dev_info_t *, bd_drive_t *);
static void bd_create_errstats(bd_t *, int, bd_drive_t *);
static void bd_destroy_errstats(bd_t *);
static void bd_errstats_setstr(kstat_named_t *, char *, size_t, char *);
static void bd_init_errstats(bd_t *, bd_drive_t *);
static void bd_fini_errstats(bd_t *);

static int bd_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
static int bd_attach(dev_info_t *, ddi_attach_cmd_t);
static int bd_detach(dev_info_t *, ddi_detach_cmd_t);

static int bd_open(dev_t *, int, int, cred_t *);
static int bd_close(dev_t, int, int, cred_t *);
static int bd_strategy(struct buf *);
static int bd_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
static int bd_dump(dev_t, caddr_t, daddr_t, int);
static int bd_read(dev_t, struct uio *, cred_t *);
static int bd_write(dev_t, struct uio *, cred_t *);
static int bd_aread(dev_t, struct aio_req *, cred_t *);
static int bd_awrite(dev_t, struct aio_req *, cred_t *);
static int bd_prop_op(dev_t, dev_info_t *, ddi_prop_op_t, int, char *,
    caddr_t, int *);

static int bd_tg_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t,
    void *);
static int bd_tg_getinfo(dev_info_t *, int, void *, void *);
static int bd_xfer_ctor(void *, void *, int);
static void bd_xfer_dtor(void *, void *);
static void bd_sched(bd_t *, bd_queue_t *);
static void bd_submit(bd_t *, bd_xfer_impl_t *);
static void bd_runq_exit(bd_xfer_impl_t *, int);
static void bd_update_state(bd_t *);
static int bd_check_state(bd_t *, enum dkio_state *);
static int bd_flush_write_cache(bd_t *, struct dk_callback *);
static int bd_check_uio(dev_t, struct uio *);
static int bd_free_space(dev_t, bd_t *, dkioc_free_list_t *);

struct cmlb_tg_ops bd_tg_ops = {
	TG_DK_OPS_VERSION_1,
	bd_tg_rdwr,
	bd_tg_getinfo,
};

static struct cb_ops bd_cb_ops = {
	bd_open,		/* open */
	bd_close,		/* close */
	bd_strategy,		/* strategy */
	nodev,			/* print */
	bd_dump,		/* dump */
	bd_read,		/* read */
	bd_write,		/* write */
	bd_ioctl,		/* ioctl */
	nodev,			/* devmap */
	nodev,			/* mmap */
	nodev,			/* segmap */
	nochpoll,		/* poll */
	bd_prop_op,		/* cb_prop_op */
	0,			/* streamtab  */
	D_64BIT | D_MP,		/* Driver comaptibility flag */
	CB_REV,			/* cb_rev */
	bd_aread,		/* async read */
	bd_awrite		/* async write */
};

struct dev_ops bd_dev_ops = {
	DEVO_REV,		/* devo_rev, */
	0,			/* refcnt  */
	bd_getinfo,		/* getinfo */
	nulldev,		/* identify */
	nulldev,		/* probe */
	bd_attach,		/* attach */
	bd_detach,		/* detach */
	nodev,			/* reset */
	&bd_cb_ops,		/* driver operations */
	NULL,			/* bus operations */
	NULL,			/* power */
	ddi_quiesce_not_needed,	/* quiesce */
};

static struct modldrv modldrv = {
	&mod_driverops,
	"Generic Block Device",
	&bd_dev_ops,
};

static struct modlinkage modlinkage = {
	MODREV_1, { &modldrv, NULL }
};

static void *bd_state;
static krwlock_t bd_lock;
static taskq_t *bd_taskq;

int
_init(void)
{
	char taskq_name[TASKQ_NAMELEN];
	const char *name;
	int rv;

	rv = ddi_soft_state_init(&bd_state, sizeof (struct bd), 2);
	if (rv != DDI_SUCCESS)
		return (rv);

	name = mod_modname(&modlinkage);
	(void) snprintf(taskq_name, sizeof (taskq_name), "%s_taskq", name);
	bd_taskq = taskq_create(taskq_name, 1, minclsyspri, 0, 0, 0);
	if (bd_taskq == NULL) {
		cmn_err(CE_WARN, "%s: unable to create %s", name, taskq_name);
		ddi_soft_state_fini(&bd_state);
		return (DDI_FAILURE);
	}

	rw_init(&bd_lock, NULL, RW_DRIVER, NULL);

	rv = mod_install(&modlinkage);
	if (rv != DDI_SUCCESS) {
		rw_destroy(&bd_lock);
		taskq_destroy(bd_taskq);
		ddi_soft_state_fini(&bd_state);
	}
	return (rv);
}

int
_fini(void)
{
	int	rv;

	rv = mod_remove(&modlinkage);
	if (rv == DDI_SUCCESS) {
		rw_destroy(&bd_lock);
		taskq_destroy(bd_taskq);
		ddi_soft_state_fini(&bd_state);
	}
	return (rv);
}

int
_info(struct modinfo *modinfop)
{
	return (mod_info(&modlinkage, modinfop));
}

static int
bd_getinfo(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **resultp)
{
	bd_t	*bd;
	minor_t	inst;

	_NOTE(ARGUNUSED(dip));

	inst = BDINST((dev_t)arg);

	switch (cmd) {
	case DDI_INFO_DEVT2DEVINFO:
		bd = ddi_get_soft_state(bd_state, inst);
		if (bd == NULL) {
			return (DDI_FAILURE);
		}
		*resultp = (void *)bd->d_dip;
		break;

	case DDI_INFO_DEVT2INSTANCE:
		*resultp = (void *)(intptr_t)inst;
		break;

	default:
		return (DDI_FAILURE);
	}
	return (DDI_SUCCESS);
}

static void
bd_prop_update_inqstring(dev_info_t *dip, char *name, char *data, size_t len)
{
	int	ilen;
	char	*data_string;

	ilen = scsi_ascii_inquiry_len(data, len);
	ASSERT3U(ilen, <=, len);
	if (ilen <= 0)
		return;
	/* ensure null termination */
	data_string = kmem_zalloc(ilen + 1, KM_SLEEP);
	bcopy(data, data_string, ilen);
	(void) ndi_prop_update_string(DDI_DEV_T_NONE, dip, name, data_string);
	kmem_free(data_string, ilen + 1);
}

static void
bd_create_inquiry_props(dev_info_t *dip, bd_drive_t *drive)
{
	if (drive->d_vendor_len > 0)
		bd_prop_update_inqstring(dip, INQUIRY_VENDOR_ID,
		    drive->d_vendor, drive->d_vendor_len);

	if (drive->d_product_len > 0)
		bd_prop_update_inqstring(dip, INQUIRY_PRODUCT_ID,
		    drive->d_product, drive->d_product_len);

	if (drive->d_serial_len > 0)
		bd_prop_update_inqstring(dip, INQUIRY_SERIAL_NO,
		    drive->d_serial, drive->d_serial_len);

	if (drive->d_revision_len > 0)
		bd_prop_update_inqstring(dip, INQUIRY_REVISION_ID,
		    drive->d_revision, drive->d_revision_len);
}

static void
bd_create_errstats(bd_t *bd, int inst, bd_drive_t *drive)
{
	char	ks_module[KSTAT_STRLEN];
	char	ks_name[KSTAT_STRLEN];
	int	ndata = sizeof (struct bd_errstats) / sizeof (kstat_named_t);

	if (bd->d_errstats != NULL)
		return;

	(void) snprintf(ks_module, sizeof (ks_module), "%serr",
	    ddi_driver_name(bd->d_dip));
	(void) snprintf(ks_name, sizeof (ks_name), "%s%d,err",
	    ddi_driver_name(bd->d_dip), inst);

	bd->d_errstats = kstat_create(ks_module, inst, ks_name, "device_error",
	    KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);

	mutex_init(&bd->d_errmutex, NULL, MUTEX_DRIVER, NULL);
	if (bd->d_errstats == NULL) {
		/*
		 * Even if we cannot create the kstat, we create a
		 * scratch kstat.  The reason for this is to ensure
		 * that we can update the kstat all of the time,
		 * without adding an extra branch instruction.
		 */
		bd->d_kerr = kmem_zalloc(sizeof (struct bd_errstats),
		    KM_SLEEP);
	} else {
		bd->d_errstats->ks_lock = &bd->d_errmutex;
		bd->d_kerr = (struct bd_errstats *)bd->d_errstats->ks_data;
	}

	kstat_named_init(&bd->d_kerr->bd_softerrs,	"Soft Errors",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_harderrs,	"Hard Errors",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_transerrs,	"Transport Errors",
	    KSTAT_DATA_UINT32);

	if (drive->d_model_len > 0) {
		kstat_named_init(&bd->d_kerr->bd_model,	"Model",
		    KSTAT_DATA_STRING);
	} else {
		kstat_named_init(&bd->d_kerr->bd_vid,	"Vendor",
		    KSTAT_DATA_STRING);
		kstat_named_init(&bd->d_kerr->bd_pid,	"Product",
		    KSTAT_DATA_STRING);
	}

	kstat_named_init(&bd->d_kerr->bd_revision,	"Revision",
	    KSTAT_DATA_STRING);
	kstat_named_init(&bd->d_kerr->bd_serial,	"Serial No",
	    KSTAT_DATA_STRING);
	kstat_named_init(&bd->d_kerr->bd_capacity,	"Size",
	    KSTAT_DATA_ULONGLONG);
	kstat_named_init(&bd->d_kerr->bd_rq_media_err,	"Media Error",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_rq_ntrdy_err,	"Device Not Ready",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_rq_nodev_err,	"No Device",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_rq_recov_err,	"Recoverable",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_rq_illrq_err,	"Illegal Request",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&bd->d_kerr->bd_rq_pfa_err,
	    "Predictive Failure Analysis", KSTAT_DATA_UINT32);

	bd->d_errstats->ks_private = bd;

	kstat_install(bd->d_errstats);
	bd_init_errstats(bd, drive);
}

static void
bd_destroy_errstats(bd_t *bd)
{
	if (bd->d_errstats != NULL) {
		bd_fini_errstats(bd);
		kstat_delete(bd->d_errstats);
		bd->d_errstats = NULL;
	} else {
		kmem_free(bd->d_kerr, sizeof (struct bd_errstats));
		bd->d_kerr = NULL;
		mutex_destroy(&bd->d_errmutex);
	}
}

static void
bd_errstats_setstr(kstat_named_t *k, char *str, size_t len, char *alt)
{
	char	*tmp;
	size_t	km_len;

	if (KSTAT_NAMED_STR_PTR(k) == NULL) {
		if (len > 0)
			km_len = strnlen(str, len);
		else if (alt != NULL)
			km_len = strlen(alt);
		else
			return;

		tmp = kmem_alloc(km_len + 1, KM_SLEEP);
		bcopy(len > 0 ? str : alt, tmp, km_len);
		tmp[km_len] = '\0';

		kstat_named_setstr(k, tmp);
	}
}

static void
bd_errstats_clrstr(kstat_named_t *k)
{
	if (KSTAT_NAMED_STR_PTR(k) == NULL)
		return;

	kmem_free(KSTAT_NAMED_STR_PTR(k), KSTAT_NAMED_STR_BUFLEN(k));
	kstat_named_setstr(k, NULL);
}

static void
bd_init_errstats(bd_t *bd, bd_drive_t *drive)
{
	struct bd_errstats	*est = bd->d_kerr;

	mutex_enter(&bd->d_errmutex);

	if (drive->d_model_len > 0 &&
	    KSTAT_NAMED_STR_PTR(&est->bd_model) == NULL) {
		bd_errstats_setstr(&est->bd_model, drive->d_model,
		    drive->d_model_len, NULL);
	} else {
		bd_errstats_setstr(&est->bd_vid, drive->d_vendor,
		    drive->d_vendor_len, "Unknown ");
		bd_errstats_setstr(&est->bd_pid, drive->d_product,
		    drive->d_product_len, "Unknown         ");
	}

	bd_errstats_setstr(&est->bd_revision, drive->d_revision,
	    drive->d_revision_len, "0001");
	bd_errstats_setstr(&est->bd_serial, drive->d_serial,
	    drive->d_serial_len, "0               ");

	mutex_exit(&bd->d_errmutex);
}

static void
bd_fini_errstats(bd_t *bd)
{
	struct bd_errstats	*est = bd->d_kerr;

	mutex_enter(&bd->d_errmutex);

	bd_errstats_clrstr(&est->bd_model);
	bd_errstats_clrstr(&est->bd_vid);
	bd_errstats_clrstr(&est->bd_pid);
	bd_errstats_clrstr(&est->bd_revision);
	bd_errstats_clrstr(&est->bd_serial);

	mutex_exit(&bd->d_errmutex);
}

static void
bd_queues_free(bd_t *bd)
{
	uint32_t i;

	for (i = 0; i < bd->d_qcount; i++) {
		bd_queue_t *bq = &bd->d_queues[i];

		mutex_destroy(&bq->q_iomutex);
		list_destroy(&bq->q_waitq);
		list_destroy(&bq->q_runq);
	}

	kmem_free(bd->d_queues, sizeof (*bd->d_queues) * bd->d_qcount);
}

static int
bd_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
{
	int		inst;
	bd_handle_t	hdl;
	bd_t		*bd;
	bd_drive_t	drive;
	uint32_t	i;
	int		rv;
	char		name[16];
	char		kcache[32];
	char		*node_type;

	switch (cmd) {
	case DDI_ATTACH:
		break;
	case DDI_RESUME:
		/* We don't do anything native for suspend/resume */
		return (DDI_SUCCESS);
	default:
		return (DDI_FAILURE);
	}

	inst = ddi_get_instance(dip);
	hdl = ddi_get_parent_data(dip);

	(void) snprintf(name, sizeof (name), "%s%d",
	    ddi_driver_name(dip), ddi_get_instance(dip));
	(void) snprintf(kcache, sizeof (kcache), "%s_xfer", name);

	if (hdl == NULL) {
		cmn_err(CE_WARN, "%s: missing parent data!", name);
		return (DDI_FAILURE);
	}

	if (ddi_soft_state_zalloc(bd_state, inst) != DDI_SUCCESS) {
		cmn_err(CE_WARN, "%s: unable to zalloc soft state!", name);
		return (DDI_FAILURE);
	}
	bd = ddi_get_soft_state(bd_state, inst);

	if (hdl->h_dma) {
		bd->d_dma = *(hdl->h_dma);
		bd->d_dma.dma_attr_granular =
		    max(DEV_BSIZE, bd->d_dma.dma_attr_granular);
		bd->d_use_dma = B_TRUE;

		if (bd->d_maxxfer &&
		    (bd->d_maxxfer != bd->d_dma.dma_attr_maxxfer)) {
			cmn_err(CE_WARN,
			    "%s: inconsistent maximum transfer size!",
			    name);
			/* We force it */
			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
		} else {
			bd->d_maxxfer = bd->d_dma.dma_attr_maxxfer;
		}
	} else {
		bd->d_use_dma = B_FALSE;
		if (bd->d_maxxfer == 0) {
			bd->d_maxxfer = 1024 * 1024;
		}
	}
	bd->d_ops = hdl->h_ops;
	bd->d_private = hdl->h_private;
	bd->d_blkshift = DEV_BSHIFT;	/* 512 bytes, to start */

	if (bd->d_maxxfer % DEV_BSIZE) {
		cmn_err(CE_WARN, "%s: maximum transfer misaligned!", name);
		bd->d_maxxfer &= ~(DEV_BSIZE - 1);
	}
	if (bd->d_maxxfer < DEV_BSIZE) {
		cmn_err(CE_WARN, "%s: maximum transfer size too small!", name);
		ddi_soft_state_free(bd_state, inst);
		return (DDI_FAILURE);
	}

	bd->d_dip = dip;
	bd->d_handle = hdl;
	ddi_set_driver_private(dip, bd);

	mutex_init(&bd->d_ksmutex, NULL, MUTEX_DRIVER, NULL);
	mutex_init(&bd->d_ocmutex, NULL, MUTEX_DRIVER, NULL);
	mutex_init(&bd->d_statemutex, NULL, MUTEX_DRIVER, NULL);
	cv_init(&bd->d_statecv, NULL, CV_DRIVER, NULL);
	mutex_init(&bd->d_dle_mutex, NULL, MUTEX_DRIVER, NULL);
	bd->d_dle_state = 0;

	bd->d_cache = kmem_cache_create(kcache, sizeof (bd_xfer_impl_t), 8,
	    bd_xfer_ctor, bd_xfer_dtor, NULL, bd, NULL, 0);

	bd->d_ksp = kstat_create(ddi_driver_name(dip), inst, NULL, "disk",
	    KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
	if (bd->d_ksp != NULL) {
		bd->d_ksp->ks_lock = &bd->d_ksmutex;
		kstat_install(bd->d_ksp);
		bd->d_kiop = bd->d_ksp->ks_data;
	} else {
		/*
		 * Even if we cannot create the kstat, we create a
		 * scratch kstat.  The reason for this is to ensure
		 * that we can update the kstat all of the time,
		 * without adding an extra branch instruction.
		 */
		bd->d_kiop = kmem_zalloc(sizeof (kstat_io_t), KM_SLEEP);
	}

	cmlb_alloc_handle(&bd->d_cmlbh);

	bd->d_state = DKIO_NONE;

	bzero(&drive, sizeof (drive));
	/*
	 * Default to one queue, and no restrictions on free space requests
	 * (if driver provides method) parent driver can override.
	 */
	drive.d_qcount = 1;
	drive.d_free_align = 1;
	bd->d_ops.o_drive_info(bd->d_private, &drive);

	/*
	 * Several checks to make sure o_drive_info() didn't return bad
	 * values:
	 *
	 * There must be at least one queue
	 */
	if (drive.d_qcount == 0)
		goto fail_drive_info;

	/* FREE/UNMAP/TRIM alignment needs to be at least 1 block */
	if (drive.d_free_align == 0)
		goto fail_drive_info;

	/*
	 * If d_max_free_blks is not unlimited (not 0), then we cannot allow
	 * an unlimited segment size. It is however permissible to not impose
	 * a limit on the total number of blocks freed while limiting the
	 * amount allowed in an individual segment.
	 */
	if ((drive.d_max_free_blks > 0 && drive.d_max_free_seg_blks == 0))
		goto fail_drive_info;

	/*
	 * If a limit is set on d_max_free_blks (by the above check, we know
	 * if there's a limit on d_max_free_blks, d_max_free_seg_blks cannot
	 * be unlimited), it cannot be smaller than the limit on an individual
	 * segment.
	 */
	if ((drive.d_max_free_blks > 0 &&
	    drive.d_max_free_seg_blks > drive.d_max_free_blks)) {
		goto fail_drive_info;
	}

	bd->d_qcount = drive.d_qcount;
	bd->d_removable = drive.d_removable;
	bd->d_hotpluggable = drive.d_hotpluggable;

	if (drive.d_maxxfer && drive.d_maxxfer < bd->d_maxxfer)
		bd->d_maxxfer = drive.d_maxxfer;

	bd->d_free_align = drive.d_free_align;
	bd->d_max_free_seg = drive.d_max_free_seg;
	bd->d_max_free_blks = drive.d_max_free_blks;
	bd->d_max_free_seg_blks = drive.d_max_free_seg_blks;

	bd_create_inquiry_props(dip, &drive);
	bd_create_errstats(bd, inst, &drive);
	bd_update_state(bd);

	bd->d_queues = kmem_alloc(sizeof (*bd->d_queues) * bd->d_qcount,
	    KM_SLEEP);
	for (i = 0; i < bd->d_qcount; i++) {
		bd_queue_t *bq = &bd->d_queues[i];

		bq->q_qsize = drive.d_qsize;
		bq->q_qactive = 0;
		mutex_init(&bq->q_iomutex, NULL, MUTEX_DRIVER, NULL);

		list_create(&bq->q_waitq, sizeof (bd_xfer_impl_t),
		    offsetof(struct bd_xfer_impl, i_linkage));
		list_create(&bq->q_runq, sizeof (bd_xfer_impl_t),
		    offsetof(struct bd_xfer_impl, i_linkage));
	}

	if (*(uint64_t *)drive.d_eui64 != 0 ||
	    *(uint64_t *)drive.d_guid != 0 ||
	    *((uint64_t *)drive.d_guid + 1) != 0)
		node_type = DDI_NT_BLOCK_BLKDEV;
	else if (drive.d_lun >= 0)
		node_type = DDI_NT_BLOCK_CHAN;
	else
		node_type = DDI_NT_BLOCK;

	rv = cmlb_attach(dip, &bd_tg_ops, DTYPE_DIRECT,
	    bd->d_removable, bd->d_hotpluggable, node_type,
	    CMLB_FAKE_LABEL_ONE_PARTITION, bd->d_cmlbh, 0);
	if (rv != 0) {
		goto fail_cmlb_attach;
	}

	if (bd->d_ops.o_devid_init != NULL) {
		rv = bd->d_ops.o_devid_init(bd->d_private, dip, &bd->d_devid);
		if (rv == DDI_SUCCESS) {
			if (ddi_devid_register(dip, bd->d_devid) !=
			    DDI_SUCCESS) {
				cmn_err(CE_WARN,
				    "%s: unable to register devid", name);
			}
		}
	}

	/*
	 * Add a zero-length attribute to tell the world we support
	 * kernel ioctls (for layered drivers).  Also set up properties
	 * used by HAL to identify removable media.
	 */
	(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
	    DDI_KERNEL_IOCTL, NULL, 0);
	if (bd->d_removable) {
		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
		    "removable-media", NULL, 0);
	}
	if (bd->d_hotpluggable) {
		(void) ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
		    "hotpluggable", NULL, 0);
	}

	/*
	 * Before we proceed, we need to ensure that the geometry and labels on
	 * the cmlb disk are reasonable. When cmlb first attaches, it does not
	 * perform label validation and creates minor nodes based on the
	 * assumption of the size. This may not be correct and the rest of the
	 * system assumes that this will have been done before we allow opens
	 * to proceed. Otherwise, on first open, this'll all end up changing
	 * around on users. We do not care if it succeeds or not. It is totally
	 * acceptable for this device to be unlabeled or not to have anything on
	 * it.
	 */
	(void) cmlb_validate(bd->d_cmlbh, 0, 0);

	hdl->h_bd = bd;
	ddi_report_dev(dip);

	return (DDI_SUCCESS);

fail_cmlb_attach:
	bd_queues_free(bd);
	bd_destroy_errstats(bd);

fail_drive_info:
	cmlb_free_handle(&bd->d_cmlbh);

	if (bd->d_ksp != NULL) {
		kstat_delete(bd->d_ksp);
		bd->d_ksp = NULL;
	} else {
		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
	}

	kmem_cache_destroy(bd->d_cache);
	cv_destroy(&bd->d_statecv);
	mutex_destroy(&bd->d_statemutex);
	mutex_destroy(&bd->d_ocmutex);
	mutex_destroy(&bd->d_ksmutex);
	mutex_destroy(&bd->d_dle_mutex);
	ddi_soft_state_free(bd_state, inst);
	return (DDI_FAILURE);
}

static int
bd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
{
	bd_handle_t	hdl;
	bd_t		*bd;

	bd = ddi_get_driver_private(dip);
	hdl = ddi_get_parent_data(dip);

	switch (cmd) {
	case DDI_DETACH:
		break;
	case DDI_SUSPEND:
		/* We don't suspend, but our parent does */
		return (DDI_SUCCESS);
	default:
		return (DDI_FAILURE);
	}

	hdl->h_bd = NULL;

	if (bd->d_ksp != NULL) {
		kstat_delete(bd->d_ksp);
		bd->d_ksp = NULL;
	} else {
		kmem_free(bd->d_kiop, sizeof (kstat_io_t));
	}

	bd_destroy_errstats(bd);
	cmlb_detach(bd->d_cmlbh, 0);
	cmlb_free_handle(&bd->d_cmlbh);
	if (bd->d_devid)
		ddi_devid_free(bd->d_devid);
	kmem_cache_destroy(bd->d_cache);
	mutex_destroy(&bd->d_ksmutex);
	mutex_destroy(&bd->d_ocmutex);
	mutex_destroy(&bd->d_statemutex);
	cv_destroy(&bd->d_statecv);
	mutex_destroy(&bd->d_dle_mutex);
	bd_queues_free(bd);
	ddi_soft_state_free(bd_state, ddi_get_instance(dip));
	return (DDI_SUCCESS);
}

static int
bd_xfer_ctor(void *buf, void *arg, int kmflag)
{
	bd_xfer_impl_t	*xi;
	bd_t		*bd = arg;
	int		(*dcb)(caddr_t);

	if (kmflag == KM_PUSHPAGE || kmflag == KM_SLEEP) {
		dcb = DDI_DMA_SLEEP;
	} else {
		dcb = DDI_DMA_DONTWAIT;
	}

	xi = buf;
	bzero(xi, sizeof (*xi));
	xi->i_bd = bd;

	if (bd->d_use_dma) {
		if (ddi_dma_alloc_handle(bd->d_dip, &bd->d_dma, dcb, NULL,
		    &xi->i_dmah) != DDI_SUCCESS) {
			return (-1);
		}
	}

	return (0);
}

static void
bd_xfer_dtor(void *buf, void *arg)
{
	bd_xfer_impl_t	*xi = buf;

	_NOTE(ARGUNUSED(arg));

	if (xi->i_dmah)
		ddi_dma_free_handle(&xi->i_dmah);
	xi->i_dmah = NULL;
}

static bd_xfer_impl_t *
bd_xfer_alloc(bd_t *bd, struct buf *bp, int (*func)(void *, bd_xfer_t *),
    int kmflag)
{
	bd_xfer_impl_t		*xi;
	int			rv = 0;
	int			status;
	unsigned		dir;
	int			(*cb)(caddr_t);
	size_t			len;
	uint32_t		shift;

	if (kmflag == KM_SLEEP) {
		cb = DDI_DMA_SLEEP;
	} else {
		cb = DDI_DMA_DONTWAIT;
	}

	xi = kmem_cache_alloc(bd->d_cache, kmflag);
	if (xi == NULL) {
		bioerror(bp, ENOMEM);
		return (NULL);
	}

	ASSERT(bp);

	xi->i_bp = bp;
	xi->i_func = func;
	xi->i_blkno = bp->b_lblkno >> (bd->d_blkshift - DEV_BSHIFT);

	if (bp->b_bcount == 0) {
		xi->i_len = 0;
		xi->i_nblks = 0;
		xi->i_kaddr = NULL;
		xi->i_resid = 0;
		xi->i_num_win = 0;
		goto done;
	}

	if (bp->b_flags & B_READ) {
		dir = DDI_DMA_READ;
		xi->i_func = bd->d_ops.o_read;
	} else {
		dir = DDI_DMA_WRITE;
		xi->i_func = bd->d_ops.o_write;
	}

	shift = bd->d_blkshift;
	xi->i_blkshift = shift;

	if (!bd->d_use_dma) {
		bp_mapin(bp);
		rv = 0;
		xi->i_offset = 0;
		xi->i_num_win =
		    (bp->b_bcount + (bd->d_maxxfer - 1)) / bd->d_maxxfer;
		xi->i_cur_win = 0;
		xi->i_len = min(bp->b_bcount, bd->d_maxxfer);
		xi->i_nblks = xi->i_len >> shift;
		xi->i_kaddr = bp->b_un.b_addr;
		xi->i_resid = bp->b_bcount;
	} else {

		/*
		 * We have to use consistent DMA if the address is misaligned.
		 */
		if (((bp->b_flags & (B_PAGEIO | B_REMAPPED)) != B_PAGEIO) &&
		    ((uintptr_t)bp->b_un.b_addr & 0x7)) {
			dir |= DDI_DMA_CONSISTENT | DDI_DMA_PARTIAL;
		} else {
			dir |= DDI_DMA_STREAMING | DDI_DMA_PARTIAL;
		}

		status = ddi_dma_buf_bind_handle(xi->i_dmah, bp, dir, cb,
		    NULL, &xi->i_dmac, &xi->i_ndmac);
		switch (status) {
		case DDI_DMA_MAPPED:
			xi->i_num_win = 1;
			xi->i_cur_win = 0;
			xi->i_offset = 0;
			xi->i_len = bp->b_bcount;
			xi->i_nblks = xi->i_len >> shift;
			xi->i_resid = bp->b_bcount;
			rv = 0;
			break;
		case DDI_DMA_PARTIAL_MAP:
			xi->i_cur_win = 0;

			if ((ddi_dma_numwin(xi->i_dmah, &xi->i_num_win) !=
			    DDI_SUCCESS) ||
			    (ddi_dma_getwin(xi->i_dmah, 0, &xi->i_offset,
			    &len, &xi->i_dmac, &xi->i_ndmac) !=
			    DDI_SUCCESS) ||
			    (P2PHASE(len, (1U << shift)) != 0)) {
				(void) ddi_dma_unbind_handle(xi->i_dmah);
				rv = EFAULT;
				goto done;
			}
			xi->i_len = len;
			xi->i_nblks = xi->i_len >> shift;
			xi->i_resid = bp->b_bcount;
			rv = 0;
			break;
		case DDI_DMA_NORESOURCES:
			rv = EAGAIN;
			goto done;
		case DDI_DMA_TOOBIG:
			rv = EINVAL;
			goto done;
		case DDI_DMA_NOMAPPING:
		case DDI_DMA_INUSE:
		default:
			rv = EFAULT;
			goto done;
		}
	}

done:
	if (rv != 0) {
		kmem_cache_free(bd->d_cache, xi);
		bioerror(bp, rv);
		return (NULL);
	}

	return (xi);
}

static void
bd_xfer_free(bd_xfer_impl_t *xi)
{
	if (xi->i_dmah) {
		(void) ddi_dma_unbind_handle(xi->i_dmah);
	}
	if (xi->i_dfl != NULL) {
		dfl_free((dkioc_free_list_t *)xi->i_dfl);
		xi->i_dfl = NULL;
	}
	kmem_cache_free(xi->i_bd->d_cache, xi);
}

static int
bd_open(dev_t *devp, int flag, int otyp, cred_t *credp)
{
	dev_t		dev = *devp;
	bd_t		*bd;
	minor_t		part;
	minor_t		inst;
	uint64_t	mask;
	boolean_t	ndelay;
	int		rv;
	diskaddr_t	nblks;
	diskaddr_t	lba;

	_NOTE(ARGUNUSED(credp));

	part = BDPART(dev);
	inst = BDINST(dev);

	if (otyp >= OTYPCNT)
		return (EINVAL);

	ndelay = (flag & (FNDELAY | FNONBLOCK)) ? B_TRUE : B_FALSE;

	/*
	 * Block any DR events from changing the set of registered
	 * devices while we function.
	 */
	rw_enter(&bd_lock, RW_READER);
	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
		rw_exit(&bd_lock);
		return (ENXIO);
	}

	mutex_enter(&bd->d_ocmutex);

	ASSERT(part < 64);
	mask = (1U << part);

	bd_update_state(bd);

	if (cmlb_validate(bd->d_cmlbh, 0, 0) != 0) {

		/* non-blocking opens are allowed to succeed */
		if (!ndelay) {
			rv = ENXIO;
			goto done;
		}
	} else if (cmlb_partinfo(bd->d_cmlbh, part, &nblks, &lba,
	    NULL, NULL, 0) == 0) {

		/*
		 * We read the partinfo, verify valid ranges.  If the
		 * partition is invalid, and we aren't blocking or
		 * doing a raw access, then fail. (Non-blocking and
		 * raw accesses can still succeed to allow a disk with
		 * bad partition data to opened by format and fdisk.)
		 */
		if ((!nblks) && ((!ndelay) || (otyp != OTYP_CHR))) {
			rv = ENXIO;
			goto done;
		}
	} else if (!ndelay) {
		/*
		 * cmlb_partinfo failed -- invalid partition or no
		 * disk label.
		 */
		rv = ENXIO;
		goto done;
	}

	if ((flag & FWRITE) && bd->d_rdonly) {
		rv = EROFS;
		goto done;
	}

	if ((bd->d_open_excl) & (mask)) {
		rv = EBUSY;
		goto done;
	}
	if (flag & FEXCL) {
		if (bd->d_open_lyr[part]) {
			rv = EBUSY;
			goto done;
		}
		for (int i = 0; i < OTYP_LYR; i++) {
			if (bd->d_open_reg[i] & mask) {
				rv = EBUSY;
				goto done;
			}
		}
	}

	if (otyp == OTYP_LYR) {
		bd->d_open_lyr[part]++;
	} else {
		bd->d_open_reg[otyp] |= mask;
	}
	if (flag & FEXCL) {
		bd->d_open_excl |= mask;
	}

	rv = 0;
done:
	mutex_exit(&bd->d_ocmutex);
	rw_exit(&bd_lock);

	return (rv);
}

static int
bd_close(dev_t dev, int flag, int otyp, cred_t *credp)
{
	bd_t		*bd;
	minor_t		inst;
	minor_t		part;
	uint64_t	mask;
	boolean_t	last = B_TRUE;

	_NOTE(ARGUNUSED(flag));
	_NOTE(ARGUNUSED(credp));

	part = BDPART(dev);
	inst = BDINST(dev);

	ASSERT(part < 64);
	mask = (1U << part);

	rw_enter(&bd_lock, RW_READER);

	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
		rw_exit(&bd_lock);
		return (ENXIO);
	}

	mutex_enter(&bd->d_ocmutex);
	if (bd->d_open_excl & mask) {
		bd->d_open_excl &= ~mask;
	}
	if (otyp == OTYP_LYR) {
		bd->d_open_lyr[part]--;
	} else {
		bd->d_open_reg[otyp] &= ~mask;
	}
	for (int i = 0; i < 64; i++) {
		if (bd->d_open_lyr[part]) {
			last = B_FALSE;
		}
	}
	for (int i = 0; last && (i < OTYP_LYR); i++) {
		if (bd->d_open_reg[i]) {
			last = B_FALSE;
		}
	}
	mutex_exit(&bd->d_ocmutex);

	if (last) {
		cmlb_invalidate(bd->d_cmlbh, 0);
	}
	rw_exit(&bd_lock);

	return (0);
}

static int
bd_dump(dev_t dev, caddr_t caddr, daddr_t blkno, int nblk)
{
	minor_t		inst;
	minor_t		part;
	diskaddr_t	pstart;
	diskaddr_t	psize;
	bd_t		*bd;
	bd_xfer_impl_t	*xi;
	buf_t		*bp;
	int		rv;
	uint32_t	shift;
	daddr_t		d_blkno;
	int	d_nblk;

	rw_enter(&bd_lock, RW_READER);

	part = BDPART(dev);
	inst = BDINST(dev);

	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
		rw_exit(&bd_lock);
		return (ENXIO);
	}
	shift = bd->d_blkshift;
	d_blkno = blkno >> (shift - DEV_BSHIFT);
	d_nblk = nblk >> (shift - DEV_BSHIFT);
	/*
	 * do cmlb, but do it synchronously unless we already have the
	 * partition (which we probably should.)
	 */
	if (cmlb_partinfo(bd->d_cmlbh, part, &psize, &pstart, NULL, NULL,
	    (void *)1)) {
		rw_exit(&bd_lock);
		return (ENXIO);
	}

	if ((d_blkno + d_nblk) > psize) {
		rw_exit(&bd_lock);
		return (EINVAL);
	}
	bp = getrbuf(KM_NOSLEEP);
	if (bp == NULL) {
		rw_exit(&bd_lock);
		return (ENOMEM);
	}

	bp->b_bcount = nblk << DEV_BSHIFT;
	bp->b_resid = bp->b_bcount;
	bp->b_lblkno = blkno;
	bp->b_un.b_addr = caddr;

	xi = bd_xfer_alloc(bd, bp,  bd->d_ops.o_write, KM_NOSLEEP);
	if (xi == NULL) {
		rw_exit(&bd_lock);
		freerbuf(bp);
		return (ENOMEM);
	}
	xi->i_blkno = d_blkno + pstart;
	xi->i_flags = BD_XFER_POLL;
	bd_submit(bd, xi);
	rw_exit(&bd_lock);

	/*
	 * Generally, we should have run this entirely synchronously
	 * at this point and the biowait call should be a no-op.  If
	 * it didn't happen this way, it's a bug in the underlying
	 * driver not honoring BD_XFER_POLL.
	 */
	(void) biowait(bp);
	rv = geterror(bp);
	freerbuf(bp);
	return (rv);
}

void
bd_minphys(struct buf *bp)
{
	minor_t inst;
	bd_t	*bd;
	inst = BDINST(bp->b_edev);

	bd = ddi_get_soft_state(bd_state, inst);

	/*
	 * In a non-debug kernel, bd_strategy will catch !bd as
	 * well, and will fail nicely.
	 */
	ASSERT(bd);

	if (bp->b_bcount > bd->d_maxxfer)
		bp->b_bcount = bd->d_maxxfer;
}

static int
bd_check_uio(dev_t dev, struct uio *uio)
{
	bd_t		*bd;
	uint32_t	shift;

	if ((bd = ddi_get_soft_state(bd_state, BDINST(dev))) == NULL) {
		return (ENXIO);
	}

	shift = bd->d_blkshift;
	if ((P2PHASE(uio->uio_loffset, (1U << shift)) != 0) ||
	    (P2PHASE(uio->uio_iov->iov_len, (1U << shift)) != 0)) {
		return (EINVAL);
	}

	return (0);
}

static int
bd_read(dev_t dev, struct uio *uio, cred_t *credp)
{
	_NOTE(ARGUNUSED(credp));
	int	ret = bd_check_uio(dev, uio);
	if (ret != 0) {
		return (ret);
	}
	return (physio(bd_strategy, NULL, dev, B_READ, bd_minphys, uio));
}

static int
bd_write(dev_t dev, struct uio *uio, cred_t *credp)
{
	_NOTE(ARGUNUSED(credp));
	int	ret = bd_check_uio(dev, uio);
	if (ret != 0) {
		return (ret);
	}
	return (physio(bd_strategy, NULL, dev, B_WRITE, bd_minphys, uio));
}

static int
bd_aread(dev_t dev, struct aio_req *aio, cred_t *credp)
{
	_NOTE(ARGUNUSED(credp));
	int	ret = bd_check_uio(dev, aio->aio_uio);
	if (ret != 0) {
		return (ret);
	}
	return (aphysio(bd_strategy, anocancel, dev, B_READ, bd_minphys, aio));
}

static int
bd_awrite(dev_t dev, struct aio_req *aio, cred_t *credp)
{
	_NOTE(ARGUNUSED(credp));
	int	ret = bd_check_uio(dev, aio->aio_uio);
	if (ret != 0) {
		return (ret);
	}
	return (aphysio(bd_strategy, anocancel, dev, B_WRITE, bd_minphys, aio));
}

static int
bd_strategy(struct buf *bp)
{
	minor_t		inst;
	minor_t		part;
	bd_t		*bd;
	diskaddr_t	p_lba;
	diskaddr_t	p_nblks;
	diskaddr_t	b_nblks;
	bd_xfer_impl_t	*xi;
	uint32_t	shift;
	int		(*func)(void *, bd_xfer_t *);
	diskaddr_t	lblkno;

	part = BDPART(bp->b_edev);
	inst = BDINST(bp->b_edev);

	ASSERT(bp);

	bp->b_resid = bp->b_bcount;

	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
		bioerror(bp, ENXIO);
		biodone(bp);
		return (0);
	}

	if (cmlb_partinfo(bd->d_cmlbh, part, &p_nblks, &p_lba,
	    NULL, NULL, 0)) {
		bioerror(bp, ENXIO);
		biodone(bp);
		return (0);
	}

	shift = bd->d_blkshift;
	lblkno = bp->b_lblkno >> (shift - DEV_BSHIFT);
	if ((P2PHASE(bp->b_lblkno, (1U << (shift - DEV_BSHIFT))) != 0) ||
	    (P2PHASE(bp->b_bcount, (1U << shift)) != 0) ||
	    (lblkno > p_nblks)) {
		bioerror(bp, EINVAL);
		biodone(bp);
		return (0);
	}
	b_nblks = bp->b_bcount >> shift;
	if ((lblkno == p_nblks) || (bp->b_bcount == 0)) {
		biodone(bp);
		return (0);
	}

	if ((b_nblks + lblkno) > p_nblks) {
		bp->b_resid = ((lblkno + b_nblks - p_nblks) << shift);
		bp->b_bcount -= bp->b_resid;
	} else {
		bp->b_resid = 0;
	}
	func = (bp->b_flags & B_READ) ? bd->d_ops.o_read : bd->d_ops.o_write;

	xi = bd_xfer_alloc(bd, bp, func, KM_NOSLEEP);
	if (xi == NULL) {
		xi = bd_xfer_alloc(bd, bp, func, KM_PUSHPAGE);
	}
	if (xi == NULL) {
		/* bd_request_alloc will have done bioerror */
		biodone(bp);
		return (0);
	}
	xi->i_blkno = lblkno + p_lba;

	bd_submit(bd, xi);

	return (0);
}

static int
bd_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp, int *rvalp)
{
	minor_t		inst;
	uint16_t	part;
	bd_t		*bd;
	void		*ptr = (void *)arg;
	int		rv;

	part = BDPART(dev);
	inst = BDINST(dev);

	if ((bd = ddi_get_soft_state(bd_state, inst)) == NULL) {
		return (ENXIO);
	}

	rv = cmlb_ioctl(bd->d_cmlbh, dev, cmd, arg, flag, credp, rvalp, 0);
	if (rv != ENOTTY)
		return (rv);

	if (rvalp != NULL) {
		/* the return value of the ioctl is 0 by default */
		*rvalp = 0;
	}

	switch (cmd) {
	case DKIOCGMEDIAINFO: {
		struct dk_minfo minfo;

		/* make sure our state information is current */
		bd_update_state(bd);
		bzero(&minfo, sizeof (minfo));
		minfo.dki_media_type = DK_FIXED_DISK;
		minfo.dki_lbsize = (1U << bd->d_blkshift);
		minfo.dki_capacity = bd->d_numblks;
		if (ddi_copyout(&minfo, ptr, sizeof (minfo), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCGMEDIAINFOEXT: {
		struct dk_minfo_ext miext;
		size_t len;

		/* make sure our state information is current */
		bd_update_state(bd);
		bzero(&miext, sizeof (miext));
		miext.dki_media_type = DK_FIXED_DISK;
		miext.dki_lbsize = (1U << bd->d_blkshift);
		miext.dki_pbsize = (1U << bd->d_pblkshift);
		miext.dki_capacity = bd->d_numblks;

		switch (ddi_model_convert_from(flag & FMODELS)) {
		case DDI_MODEL_ILP32:
			len = sizeof (struct dk_minfo_ext32);
			break;
		default:
			len = sizeof (struct dk_minfo_ext);
			break;
		}

		if (ddi_copyout(&miext, ptr, len, flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCINFO: {
		struct dk_cinfo cinfo;
		bzero(&cinfo, sizeof (cinfo));
		cinfo.dki_ctype = DKC_BLKDEV;
		cinfo.dki_cnum = ddi_get_instance(ddi_get_parent(bd->d_dip));
		(void) snprintf(cinfo.dki_cname, sizeof (cinfo.dki_cname),
		    "%s", ddi_driver_name(ddi_get_parent(bd->d_dip)));
		(void) snprintf(cinfo.dki_dname, sizeof (cinfo.dki_dname),
		    "%s", ddi_driver_name(bd->d_dip));
		cinfo.dki_unit = inst;
		cinfo.dki_flags = DKI_FMTVOL;
		cinfo.dki_partition = part;
		cinfo.dki_maxtransfer = bd->d_maxxfer / DEV_BSIZE;
		cinfo.dki_addr = 0;
		cinfo.dki_slave = 0;
		cinfo.dki_space = 0;
		cinfo.dki_prio = 0;
		cinfo.dki_vec = 0;
		if (ddi_copyout(&cinfo, ptr, sizeof (cinfo), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCREMOVABLE: {
		int i;
		i = bd->d_removable ? 1 : 0;
		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCHOTPLUGGABLE: {
		int i;
		i = bd->d_hotpluggable ? 1 : 0;
		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCREADONLY: {
		int i;
		i = bd->d_rdonly ? 1 : 0;
		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCSOLIDSTATE: {
		int i;
		i = bd->d_ssd ? 1 : 0;
		if (ddi_copyout(&i, ptr, sizeof (i), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCSTATE: {
		enum dkio_state	state;
		if (ddi_copyin(ptr, &state, sizeof (state), flag)) {
			return (EFAULT);
		}
		if ((rv = bd_check_state(bd, &state)) != 0) {
			return (rv);
		}
		if (ddi_copyout(&state, ptr, sizeof (state), flag)) {
			return (EFAULT);
		}
		return (0);
	}
	case DKIOCFLUSHWRITECACHE: {
		struct dk_callback *dkc = NULL;

		if (flag & FKIOCTL)
			dkc = (void *)arg;

		rv = bd_flush_write_cache(bd, dkc);
		return (rv);
	}
	case DKIOCFREE: {
		dkioc_free_list_t *dfl = NULL;

		/*
		 * Check free space support early to avoid copyin/allocation
		 * when unnecessary.
		 */
		if (!CAN_FREESPACE(bd))
			return (ENOTSUP);

		rv = dfl_copyin(ptr, &dfl, flag, KM_SLEEP);
		if (rv != 0)
			return (rv);

		/*
		 * bd_free_space() consumes 'dfl'. bd_free_space() will
		 * call dfl_iter() which will normally try to pass dfl through
		 * to bd_free_space_cb() which attaches dfl to the bd_xfer_t
		 * that is then queued for the underlying driver. Once the
		 * driver processes the request, the bd_xfer_t instance is
		 * disposed of, including any attached dkioc_free_list_t.
		 *
		 * If dfl cannot be processed by the underlying driver due to
		 * size or alignment requirements of the driver, dfl_iter()
		 * will replace dfl with one or more new dkioc_free_list_t
		 * instances with the correct alignment and sizes for the driver
		 * (and free the original dkioc_free_list_t).
		 */
		rv = bd_free_space(dev, bd, dfl);
		return (rv);
	}

	case DKIOC_CANFREE: {
		boolean_t supported = CAN_FREESPACE(bd);

		if (ddi_copyout(&supported, (void *)arg, sizeof (supported),
		    flag) != 0) {
			return (EFAULT);
		}

		return (0);
	}

	default:
		break;

	}
	return (ENOTTY);
}

static int
bd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
    char *name, caddr_t valuep, int *lengthp)
{
	bd_t	*bd;

	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));
	if (bd == NULL)
		return (ddi_prop_op(dev, dip, prop_op, mod_flags,
		    name, valuep, lengthp));

	return (cmlb_prop_op(bd->d_cmlbh, dev, dip, prop_op, mod_flags, name,
	    valuep, lengthp, BDPART(dev), 0));
}


static int
bd_tg_rdwr(dev_info_t *dip, uchar_t cmd, void *bufaddr, diskaddr_t start,
    size_t length, void *tg_cookie)
{
	bd_t		*bd;
	buf_t		*bp;
	bd_xfer_impl_t	*xi;
	int		rv;
	int		(*func)(void *, bd_xfer_t *);
	int		kmflag;

	/*
	 * If we are running in polled mode (such as during dump(9e)
	 * execution), then we cannot sleep for kernel allocations.
	 */
	kmflag = tg_cookie ? KM_NOSLEEP : KM_SLEEP;

	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));

	if (P2PHASE(length, (1U << bd->d_blkshift)) != 0) {
		/* We can only transfer whole blocks at a time! */
		return (EINVAL);
	}

	if ((bp = getrbuf(kmflag)) == NULL) {
		return (ENOMEM);
	}

	switch (cmd) {
	case TG_READ:
		bp->b_flags = B_READ;
		func = bd->d_ops.o_read;
		break;
	case TG_WRITE:
		bp->b_flags = B_WRITE;
		func = bd->d_ops.o_write;
		break;
	default:
		freerbuf(bp);
		return (EINVAL);
	}

	bp->b_un.b_addr = bufaddr;
	bp->b_bcount = length;
	xi = bd_xfer_alloc(bd, bp, func, kmflag);
	if (xi == NULL) {
		rv = geterror(bp);
		freerbuf(bp);
		return (rv);
	}
	xi->i_flags = tg_cookie ? BD_XFER_POLL : 0;
	xi->i_blkno = start;
	bd_submit(bd, xi);
	(void) biowait(bp);
	rv = geterror(bp);
	freerbuf(bp);

	return (rv);
}

static int
bd_tg_getinfo(dev_info_t *dip, int cmd, void *arg, void *tg_cookie)
{
	bd_t		*bd;

	_NOTE(ARGUNUSED(tg_cookie));
	bd = ddi_get_soft_state(bd_state, ddi_get_instance(dip));

	switch (cmd) {
	case TG_GETPHYGEOM:
	case TG_GETVIRTGEOM:
		/*
		 * We don't have any "geometry" as such, let cmlb
		 * fabricate something.
		 */
		return (ENOTTY);

	case TG_GETCAPACITY:
		bd_update_state(bd);
		*(diskaddr_t *)arg = bd->d_numblks;
		return (0);

	case TG_GETBLOCKSIZE:
		*(uint32_t *)arg = (1U << bd->d_blkshift);
		return (0);

	case TG_GETATTR:
		/*
		 * It turns out that cmlb really doesn't do much for
		 * non-writable media, but lets make the information
		 * available for it in case it does more in the
		 * future.  (The value is currently used for
		 * triggering special behavior for CD-ROMs.)
		 */
		bd_update_state(bd);
		((tg_attribute_t *)arg)->media_is_writable =
		    bd->d_rdonly ? B_FALSE : B_TRUE;
		((tg_attribute_t *)arg)->media_is_solid_state = bd->d_ssd;
		((tg_attribute_t *)arg)->media_is_rotational = B_FALSE;
		return (0);

	default:
		return (EINVAL);
	}
}


static void
bd_sched(bd_t *bd, bd_queue_t *bq)
{
	bd_xfer_impl_t	*xi;
	struct buf	*bp;
	int		rv;

	mutex_enter(&bq->q_iomutex);

	while ((bq->q_qactive < bq->q_qsize) &&
	    ((xi = list_remove_head(&bq->q_waitq)) != NULL)) {
		mutex_enter(&bd->d_ksmutex);
		kstat_waitq_to_runq(bd->d_kiop);
		mutex_exit(&bd->d_ksmutex);

		bq->q_qactive++;
		list_insert_tail(&bq->q_runq, xi);

		/*
		 * Submit the job to the driver.  We drop the I/O mutex
		 * so that we can deal with the case where the driver
		 * completion routine calls back into us synchronously.
		 */

		mutex_exit(&bq->q_iomutex);

		rv = xi->i_func(bd->d_private, &xi->i_public);
		if (rv != 0) {
			bp = xi->i_bp;
			bioerror(bp, rv);
			biodone(bp);

			atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32);

			mutex_enter(&bq->q_iomutex);

			mutex_enter(&bd->d_ksmutex);
			kstat_runq_exit(bd->d_kiop);
			mutex_exit(&bd->d_ksmutex);

			bq->q_qactive--;
			list_remove(&bq->q_runq, xi);
			bd_xfer_free(xi);
		} else {
			mutex_enter(&bq->q_iomutex);
		}
	}

	mutex_exit(&bq->q_iomutex);
}

static void
bd_submit(bd_t *bd, bd_xfer_impl_t *xi)
{
	uint64_t	nv = atomic_inc_64_nv(&bd->d_io_counter);
	unsigned	q = nv % bd->d_qcount;
	bd_queue_t	*bq = &bd->d_queues[q];

	xi->i_bq = bq;
	xi->i_qnum = q;

	mutex_enter(&bq->q_iomutex);

	list_insert_tail(&bq->q_waitq, xi);

	mutex_enter(&bd->d_ksmutex);
	kstat_waitq_enter(bd->d_kiop);
	mutex_exit(&bd->d_ksmutex);

	mutex_exit(&bq->q_iomutex);

	bd_sched(bd, bq);
}

static void
bd_runq_exit(bd_xfer_impl_t *xi, int err)
{
	bd_t		*bd = xi->i_bd;
	buf_t		*bp = xi->i_bp;
	bd_queue_t	*bq = xi->i_bq;

	mutex_enter(&bq->q_iomutex);
	bq->q_qactive--;

	mutex_enter(&bd->d_ksmutex);
	kstat_runq_exit(bd->d_kiop);
	mutex_exit(&bd->d_ksmutex);

	list_remove(&bq->q_runq, xi);
	mutex_exit(&bq->q_iomutex);

	if (err == 0) {
		if (bp->b_flags & B_READ) {
			atomic_inc_uint(&bd->d_kiop->reads);
			atomic_add_64((uint64_t *)&bd->d_kiop->nread,
			    bp->b_bcount - xi->i_resid);
		} else {
			atomic_inc_uint(&bd->d_kiop->writes);
			atomic_add_64((uint64_t *)&bd->d_kiop->nwritten,
			    bp->b_bcount - xi->i_resid);
		}
	}
	bd_sched(bd, bq);
}

static void
bd_dle_sysevent_task(void *arg)
{
	nvlist_t *attr = NULL;
	char *path = NULL;
	bd_t *bd = arg;
	dev_info_t *dip = bd->d_dip;
	size_t n;

	mutex_enter(&bd->d_dle_mutex);
	bd->d_dle_state &= ~BD_DLE_PENDING;
	bd->d_dle_state |= BD_DLE_RUNNING;
	mutex_exit(&bd->d_dle_mutex);

	dev_err(dip, CE_NOTE, "!dynamic LUN expansion");

	if (nvlist_alloc(&attr, NV_UNIQUE_NAME_TYPE, KM_SLEEP) != 0) {
		mutex_enter(&bd->d_dle_mutex);
		bd->d_dle_state &= ~(BD_DLE_RUNNING|BD_DLE_PENDING);
		mutex_exit(&bd->d_dle_mutex);
		return;
	}

	path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);

	n = snprintf(path, MAXPATHLEN, "/devices");
	(void) ddi_pathname(dip, path + n);
	n = strlen(path);
	n += snprintf(path + n, MAXPATHLEN - n, ":x");

	for (;;) {
		/*
		 * On receipt of this event, the ZFS sysevent module will scan
		 * active zpools for child vdevs matching this physical path.
		 * In order to catch both whole disk pools and those with an
		 * EFI boot partition, generate separate sysevents for minor
		 * node 'a' and 'b'.
		 */
		for (char c = 'a'; c < 'c'; c++) {
			path[n - 1] = c;

			if (nvlist_add_string(attr, DEV_PHYS_PATH, path) != 0)
				break;

			(void) ddi_log_sysevent(dip, DDI_VENDOR_SUNW,
			    EC_DEV_STATUS, ESC_DEV_DLE, attr, NULL, DDI_SLEEP);
		}

		mutex_enter(&bd->d_dle_mutex);
		if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
			bd->d_dle_state &= ~BD_DLE_RUNNING;
			mutex_exit(&bd->d_dle_mutex);
			break;
		}
		bd->d_dle_state &= ~BD_DLE_PENDING;
		mutex_exit(&bd->d_dle_mutex);
	}

	nvlist_free(attr);
	kmem_free(path, MAXPATHLEN);
}

static void
bd_update_state(bd_t *bd)
{
	enum	dkio_state	state = DKIO_INSERTED;
	boolean_t		docmlb = B_FALSE;
	bd_media_t		media;

	bzero(&media, sizeof (media));

	mutex_enter(&bd->d_statemutex);
	if (bd->d_ops.o_media_info(bd->d_private, &media) != 0) {
		bd->d_numblks = 0;
		state = DKIO_EJECTED;
		goto done;
	}

	if ((media.m_blksize < 512) ||
	    (!ISP2(media.m_blksize)) ||
	    (P2PHASE(bd->d_maxxfer, media.m_blksize))) {
		dev_err(bd->d_dip, CE_WARN, "Invalid media block size (%d)",
		    media.m_blksize);
		/*
		 * We can't use the media, treat it as not present.
		 */
		state = DKIO_EJECTED;
		bd->d_numblks = 0;
		goto done;
	}

	if (((1U << bd->d_blkshift) != media.m_blksize) ||
	    (bd->d_numblks != media.m_nblks)) {
		/* Device size changed */
		docmlb = B_TRUE;
	}

	bd->d_blkshift = ddi_ffs(media.m_blksize) - 1;
	bd->d_pblkshift = bd->d_blkshift;
	bd->d_numblks = media.m_nblks;
	bd->d_rdonly = media.m_readonly;
	bd->d_ssd = media.m_solidstate;

	/*
	 * Only use the supplied physical block size if it is non-zero,
	 * greater or equal to the block size, and a power of 2. Ignore it
	 * if not, it's just informational and we can still use the media.
	 */
	if ((media.m_pblksize != 0) &&
	    (media.m_pblksize >= media.m_blksize) &&
	    (ISP2(media.m_pblksize)))
		bd->d_pblkshift = ddi_ffs(media.m_pblksize) - 1;

done:
	if (state != bd->d_state) {
		bd->d_state = state;
		cv_broadcast(&bd->d_statecv);
		docmlb = B_TRUE;
	}
	mutex_exit(&bd->d_statemutex);

	bd->d_kerr->bd_capacity.value.ui64 = bd->d_numblks << bd->d_blkshift;

	if (docmlb) {
		if (state == DKIO_INSERTED) {
			(void) cmlb_validate(bd->d_cmlbh, 0, 0);

			mutex_enter(&bd->d_dle_mutex);
			/*
			 * If there is already an event pending, there's
			 * nothing to do; we coalesce multiple events.
			 */
			if ((bd->d_dle_state & BD_DLE_PENDING) == 0) {
				if ((bd->d_dle_state & BD_DLE_RUNNING) == 0) {
					taskq_dispatch_ent(bd_taskq,
					    bd_dle_sysevent_task, bd, 0,
					    &bd->d_dle_ent);
				}
				bd->d_dle_state |= BD_DLE_PENDING;
			}
			mutex_exit(&bd->d_dle_mutex);
		} else {
			cmlb_invalidate(bd->d_cmlbh, 0);
		}
	}
}

static int
bd_check_state(bd_t *bd, enum dkio_state *state)
{
	clock_t		when;

	for (;;) {

		bd_update_state(bd);

		mutex_enter(&bd->d_statemutex);

		if (bd->d_state != *state) {
			*state = bd->d_state;
			mutex_exit(&bd->d_statemutex);
			break;
		}

		when = drv_usectohz(1000000);
		if (cv_reltimedwait_sig(&bd->d_statecv, &bd->d_statemutex,
		    when, TR_CLOCK_TICK) == 0) {
			mutex_exit(&bd->d_statemutex);
			return (EINTR);
		}

		mutex_exit(&bd->d_statemutex);
	}

	return (0);
}

static int
bd_flush_write_cache_done(struct buf *bp)
{
	struct dk_callback *dc = (void *)bp->b_private;

	(*dc->dkc_callback)(dc->dkc_cookie, geterror(bp));
	kmem_free(dc, sizeof (*dc));
	freerbuf(bp);
	return (0);
}

static int
bd_flush_write_cache(bd_t *bd, struct dk_callback *dkc)
{
	buf_t			*bp;
	struct dk_callback	*dc;
	bd_xfer_impl_t		*xi;
	int			rv;

	if (bd->d_ops.o_sync_cache == NULL) {
		return (ENOTSUP);
	}
	if ((bp = getrbuf(KM_SLEEP)) == NULL) {
		return (ENOMEM);
	}
	bp->b_resid = 0;
	bp->b_bcount = 0;

	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_sync_cache, KM_SLEEP);
	if (xi == NULL) {
		rv = geterror(bp);
		freerbuf(bp);
		return (rv);
	}

	/* Make an asynchronous flush, but only if there is a callback */
	if (dkc != NULL && dkc->dkc_callback != NULL) {
		/* Make a private copy of the callback structure */
		dc = kmem_alloc(sizeof (*dc), KM_SLEEP);
		*dc = *dkc;
		bp->b_private = dc;
		bp->b_iodone = bd_flush_write_cache_done;

		bd_submit(bd, xi);
		return (0);
	}

	/* In case there is no callback, perform a synchronous flush */
	bd_submit(bd, xi);
	(void) biowait(bp);
	rv = geterror(bp);
	freerbuf(bp);

	return (rv);
}

static int
bd_free_space_done(struct buf *bp)
{
	freerbuf(bp);
	return (0);
}

static int
bd_free_space_cb(dkioc_free_list_t *dfl, void *arg, int kmflag)
{
	bd_t		*bd = arg;
	buf_t		*bp = NULL;
	bd_xfer_impl_t	*xi = NULL;
	boolean_t	sync = DFL_ISSYNC(dfl) ?  B_TRUE : B_FALSE;
	int		rv = 0;

	bp = getrbuf(KM_SLEEP);
	bp->b_resid = 0;
	bp->b_bcount = 0;
	bp->b_lblkno = 0;

	xi = bd_xfer_alloc(bd, bp, bd->d_ops.o_free_space, kmflag);
	xi->i_dfl = dfl;

	if (!sync) {
		bp->b_iodone = bd_free_space_done;
		bd_submit(bd, xi);
		return (0);
	}

	xi->i_flags |= BD_XFER_POLL;
	bd_submit(bd, xi);

	(void) biowait(bp);
	rv = geterror(bp);
	freerbuf(bp);

	return (rv);
}

static int
bd_free_space(dev_t dev, bd_t *bd, dkioc_free_list_t *dfl)
{
	diskaddr_t p_len, p_offset;
	uint64_t offset_bytes, len_bytes;
	minor_t part = BDPART(dev);
	const uint_t bshift = bd->d_blkshift;
	dkioc_free_info_t dfi = {
		.dfi_bshift = bshift,
		.dfi_align = bd->d_free_align << bshift,
		.dfi_max_bytes = bd->d_max_free_blks << bshift,
		.dfi_max_ext = bd->d_max_free_seg,
		.dfi_max_ext_bytes = bd->d_max_free_seg_blks << bshift,
	};

	if (cmlb_partinfo(bd->d_cmlbh, part, &p_len, &p_offset, NULL,
	    NULL, 0) != 0) {
		dfl_free(dfl);
		return (ENXIO);
	}

	/*
	 * bd_ioctl created our own copy of dfl, so we can modify as
	 * necessary
	 */
	offset_bytes = (uint64_t)p_offset << bshift;
	len_bytes = (uint64_t)p_len << bshift;

	dfl->dfl_offset += offset_bytes;
	if (dfl->dfl_offset < offset_bytes) {
		dfl_free(dfl);
		return (EOVERFLOW);
	}

	return (dfl_iter(dfl, &dfi, offset_bytes + len_bytes, bd_free_space_cb,
	    bd, KM_SLEEP));
}

/*
 * Nexus support.
 */
int
bd_bus_ctl(dev_info_t *dip, dev_info_t *rdip, ddi_ctl_enum_t ctlop,
    void *arg, void *result)
{
	bd_handle_t	hdl;

	switch (ctlop) {
	case DDI_CTLOPS_REPORTDEV:
		cmn_err(CE_CONT, "?Block device: %s@%s, %s%d\n",
		    ddi_node_name(rdip), ddi_get_name_addr(rdip),
		    ddi_driver_name(rdip), ddi_get_instance(rdip));
		return (DDI_SUCCESS);

	case DDI_CTLOPS_INITCHILD:
		hdl = ddi_get_parent_data((dev_info_t *)arg);
		if (hdl == NULL) {
			return (DDI_NOT_WELL_FORMED);
		}
		ddi_set_name_addr((dev_info_t *)arg, hdl->h_addr);
		return (DDI_SUCCESS);

	case DDI_CTLOPS_UNINITCHILD:
		ddi_set_name_addr((dev_info_t *)arg, NULL);
		ndi_prop_remove_all((dev_info_t *)arg);
		return (DDI_SUCCESS);

	default:
		return (ddi_ctlops(dip, rdip, ctlop, arg, result));
	}
}

/*
 * Functions for device drivers.
 */
bd_handle_t
bd_alloc_handle(void *private, bd_ops_t *ops, ddi_dma_attr_t *dma, int kmflag)
{
	bd_handle_t	hdl;

	switch (ops->o_version) {
	case BD_OPS_VERSION_0:
	case BD_OPS_VERSION_1:
	case BD_OPS_VERSION_2:
		break;

	default:
		/* Unsupported version */
		return (NULL);
	}

	hdl = kmem_zalloc(sizeof (*hdl), kmflag);
	if (hdl == NULL) {
		return (NULL);
	}

	switch (ops->o_version) {
	case BD_OPS_VERSION_2:
		hdl->h_ops.o_free_space = ops->o_free_space;
		/*FALLTHRU*/
	case BD_OPS_VERSION_1:
	case BD_OPS_VERSION_0:
		hdl->h_ops.o_drive_info = ops->o_drive_info;
		hdl->h_ops.o_media_info = ops->o_media_info;
		hdl->h_ops.o_devid_init = ops->o_devid_init;
		hdl->h_ops.o_sync_cache = ops->o_sync_cache;
		hdl->h_ops.o_read = ops->o_read;
		hdl->h_ops.o_write = ops->o_write;
		break;
	}

	hdl->h_dma = dma;
	hdl->h_private = private;

	return (hdl);
}

void
bd_free_handle(bd_handle_t hdl)
{
	kmem_free(hdl, sizeof (*hdl));
}

int
bd_attach_handle(dev_info_t *dip, bd_handle_t hdl)
{
	bd_drive_t	drive = { 0 };
	dev_info_t	*child;
	size_t		len;

	/*
	 * It's not an error if bd_attach_handle() is called on a handle that
	 * already is attached. We just ignore the request to attach and return.
	 * This way drivers using blkdev don't have to keep track about blkdev
	 * state, they can just call this function to make sure it attached.
	 */
	if (hdl->h_child != NULL) {
		return (DDI_SUCCESS);
	}

	/* if drivers don't override this, make it assume none */
	drive.d_lun = -1;
	hdl->h_ops.o_drive_info(hdl->h_private, &drive);

	hdl->h_parent = dip;
	hdl->h_name = "blkdev";

	/*
	 * Prefer the GUID over the EUI64.
	 */
	if (*(uint64_t *)drive.d_guid != 0 ||
	    *((uint64_t *)drive.d_guid + 1) != 0) {
		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
		    "w%02X%02X%02X%02X%02X%02X%02X%02X"
		    "%02X%02X%02X%02X%02X%02X%02X%02X",
		    drive.d_guid[0], drive.d_guid[1], drive.d_guid[2],
		    drive.d_guid[3], drive.d_guid[4], drive.d_guid[5],
		    drive.d_guid[6], drive.d_guid[7], drive.d_guid[8],
		    drive.d_guid[9], drive.d_guid[10], drive.d_guid[11],
		    drive.d_guid[12], drive.d_guid[13], drive.d_guid[14],
		    drive.d_guid[15]);
	} else if (*(uint64_t *)drive.d_eui64 != 0) {
		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
		    "w%02X%02X%02X%02X%02X%02X%02X%02X",
		    drive.d_eui64[0], drive.d_eui64[1],
		    drive.d_eui64[2], drive.d_eui64[3],
		    drive.d_eui64[4], drive.d_eui64[5],
		    drive.d_eui64[6], drive.d_eui64[7]);
	} else {
		len = snprintf(hdl->h_addr, sizeof (hdl->h_addr),
		    "%X", drive.d_target);
	}

	VERIFY(len <= sizeof (hdl->h_addr));

	if (drive.d_lun >= 0) {
		(void) snprintf(hdl->h_addr + len, sizeof (hdl->h_addr) - len,
		    ",%X", drive.d_lun);
	}

	if (ndi_devi_alloc(dip, hdl->h_name, (pnode_t)DEVI_SID_NODEID,
	    &child) != NDI_SUCCESS) {
		cmn_err(CE_WARN, "%s%d: unable to allocate node %s@%s",
		    ddi_driver_name(dip), ddi_get_instance(dip),
		    "blkdev", hdl->h_addr);
		return (DDI_FAILURE);
	}

	ddi_set_parent_data(child, hdl);
	hdl->h_child = child;

	if (ndi_devi_online(child, 0) != NDI_SUCCESS) {
		cmn_err(CE_WARN, "%s%d: failed bringing node %s@%s online",
		    ddi_driver_name(dip), ddi_get_instance(dip),
		    hdl->h_name, hdl->h_addr);
		(void) ndi_devi_free(child);
		hdl->h_child = NULL;
		return (DDI_FAILURE);
	}

	return (DDI_SUCCESS);
}

int
bd_detach_handle(bd_handle_t hdl)
{
	int	rv;
	char	*devnm;

	/*
	 * It's not an error if bd_detach_handle() is called on a handle that
	 * already is detached. We just ignore the request to detach and return.
	 * This way drivers using blkdev don't have to keep track about blkdev
	 * state, they can just call this function to make sure it detached.
	 */
	if (hdl->h_child == NULL) {
		return (DDI_SUCCESS);
	}
	ndi_devi_enter(hdl->h_parent);
	if (i_ddi_node_state(hdl->h_child) < DS_INITIALIZED) {
		rv = ddi_remove_child(hdl->h_child, 0);
	} else {
		devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
		(void) ddi_deviname(hdl->h_child, devnm);
		(void) devfs_clean(hdl->h_parent, devnm + 1, DV_CLEAN_FORCE);
		rv = ndi_devi_unconfig_one(hdl->h_parent, devnm + 1, NULL,
		    NDI_DEVI_REMOVE | NDI_UNCONFIG);
		kmem_free(devnm, MAXNAMELEN + 1);
	}
	if (rv == 0) {
		hdl->h_child = NULL;
	}

	ndi_devi_exit(hdl->h_parent);
	return (rv == NDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
}

void
bd_xfer_done(bd_xfer_t *xfer, int err)
{
	bd_xfer_impl_t	*xi = (void *)xfer;
	buf_t		*bp = xi->i_bp;
	int		rv = DDI_SUCCESS;
	bd_t		*bd = xi->i_bd;
	size_t		len;

	if (err != 0) {
		bd_runq_exit(xi, err);
		atomic_inc_32(&bd->d_kerr->bd_harderrs.value.ui32);

		bp->b_resid += xi->i_resid;
		bd_xfer_free(xi);
		bioerror(bp, err);
		biodone(bp);
		return;
	}

	xi->i_cur_win++;
	xi->i_resid -= xi->i_len;

	if (xi->i_resid == 0) {
		/* Job completed succcessfully! */
		bd_runq_exit(xi, 0);

		bd_xfer_free(xi);
		biodone(bp);
		return;
	}

	xi->i_blkno += xi->i_nblks;

	if (bd->d_use_dma) {
		/* More transfer still pending... advance to next DMA window. */
		rv = ddi_dma_getwin(xi->i_dmah, xi->i_cur_win,
		    &xi->i_offset, &len, &xi->i_dmac, &xi->i_ndmac);
	} else {
		/* Advance memory window. */
		xi->i_kaddr += xi->i_len;
		xi->i_offset += xi->i_len;
		len = min(bp->b_bcount - xi->i_offset, bd->d_maxxfer);
	}


	if ((rv != DDI_SUCCESS) ||
	    (P2PHASE(len, (1U << xi->i_blkshift)) != 0)) {
		bd_runq_exit(xi, EFAULT);

		bp->b_resid += xi->i_resid;
		bd_xfer_free(xi);
		bioerror(bp, EFAULT);
		biodone(bp);
		return;
	}
	xi->i_len = len;
	xi->i_nblks = len >> xi->i_blkshift;

	/* Submit next window to hardware. */
	rv = xi->i_func(bd->d_private, &xi->i_public);
	if (rv != 0) {
		bd_runq_exit(xi, rv);

		atomic_inc_32(&bd->d_kerr->bd_transerrs.value.ui32);

		bp->b_resid += xi->i_resid;
		bd_xfer_free(xi);
		bioerror(bp, rv);
		biodone(bp);
	}
}

void
bd_error(bd_xfer_t *xfer, int error)
{
	bd_xfer_impl_t	*xi = (void *)xfer;
	bd_t		*bd = xi->i_bd;

	switch (error) {
	case BD_ERR_MEDIA:
		atomic_inc_32(&bd->d_kerr->bd_rq_media_err.value.ui32);
		break;
	case BD_ERR_NTRDY:
		atomic_inc_32(&bd->d_kerr->bd_rq_ntrdy_err.value.ui32);
		break;
	case BD_ERR_NODEV:
		atomic_inc_32(&bd->d_kerr->bd_rq_nodev_err.value.ui32);
		break;
	case BD_ERR_RECOV:
		atomic_inc_32(&bd->d_kerr->bd_rq_recov_err.value.ui32);
		break;
	case BD_ERR_ILLRQ:
		atomic_inc_32(&bd->d_kerr->bd_rq_illrq_err.value.ui32);
		break;
	case BD_ERR_PFA:
		atomic_inc_32(&bd->d_kerr->bd_rq_pfa_err.value.ui32);
		break;
	default:
		cmn_err(CE_PANIC, "bd_error: unknown error type %d", error);
		break;
	}
}

void
bd_state_change(bd_handle_t hdl)
{
	bd_t		*bd;

	if ((bd = hdl->h_bd) != NULL) {
		bd_update_state(bd);
	}
}

const char *
bd_address(bd_handle_t hdl)
{
	return (hdl->h_addr);
}

void
bd_mod_init(struct dev_ops *devops)
{
	static struct bus_ops bd_bus_ops = {
		BUSO_REV,		/* busops_rev */
		nullbusmap,		/* bus_map */
		NULL,			/* bus_get_intrspec (OBSOLETE) */
		NULL,			/* bus_add_intrspec (OBSOLETE) */
		NULL,			/* bus_remove_intrspec (OBSOLETE) */
		i_ddi_map_fault,	/* bus_map_fault */
		NULL,			/* bus_dma_map (OBSOLETE) */
		ddi_dma_allochdl,	/* bus_dma_allochdl */
		ddi_dma_freehdl,	/* bus_dma_freehdl */
		ddi_dma_bindhdl,	/* bus_dma_bindhdl */
		ddi_dma_unbindhdl,	/* bus_dma_unbindhdl */
		ddi_dma_flush,		/* bus_dma_flush */
		ddi_dma_win,		/* bus_dma_win */
		ddi_dma_mctl,		/* bus_dma_ctl */
		bd_bus_ctl,		/* bus_ctl */
		ddi_bus_prop_op,	/* bus_prop_op */
		NULL,			/* bus_get_eventcookie */
		NULL,			/* bus_add_eventcall */
		NULL,			/* bus_remove_eventcall */
		NULL,			/* bus_post_event */
		NULL,			/* bus_intr_ctl (OBSOLETE) */
		NULL,			/* bus_config */
		NULL,			/* bus_unconfig */
		NULL,			/* bus_fm_init */
		NULL,			/* bus_fm_fini */
		NULL,			/* bus_fm_access_enter */
		NULL,			/* bus_fm_access_exit */
		NULL,			/* bus_power */
		NULL,			/* bus_intr_op */
	};

	devops->devo_bus_ops = &bd_bus_ops;

	/*
	 * NB: The device driver is free to supply its own
	 * character entry device support.
	 */
}

void
bd_mod_fini(struct dev_ops *devops)
{
	devops->devo_bus_ops = NULL;
}