usr/src/uts/common/io/devinfo.c

/*
 * 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) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
 */

/*
 * driver for accessing kernel devinfo tree.
 */
#include <sys/types.h>
#include <sys/pathname.h>
#include <sys/debug.h>
#include <sys/autoconf.h>
#include <sys/vmsystm.h>
#include <sys/conf.h>
#include <sys/file.h>
#include <sys/kmem.h>
#include <sys/modctl.h>
#include <sys/stat.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/sunldi_impl.h>
#include <sys/sunndi.h>
#include <sys/esunddi.h>
#include <sys/sunmdi.h>
#include <sys/ddi_impldefs.h>
#include <sys/ndi_impldefs.h>
#include <sys/mdi_impldefs.h>
#include <sys/devinfo_impl.h>
#include <sys/thread.h>
#include <sys/modhash.h>
#include <sys/bitmap.h>
#include <util/qsort.h>
#include <sys/disp.h>
#include <sys/kobj.h>
#include <sys/crc32.h>
#include <sys/ddi_hp.h>
#include <sys/ddi_hp_impl.h>
#include <sys/sysmacros.h>
#include <sys/list.h>


#ifdef DEBUG
static int di_debug;
#define	dcmn_err(args) if (di_debug >= 1) cmn_err args
#define	dcmn_err2(args) if (di_debug >= 2) cmn_err args
#define	dcmn_err3(args) if (di_debug >= 3) cmn_err args
#else
#define	dcmn_err(args) /* nothing */
#define	dcmn_err2(args) /* nothing */
#define	dcmn_err3(args) /* nothing */
#endif

/*
 * We partition the space of devinfo minor nodes equally between the full and
 * unprivileged versions of the driver.  The even-numbered minor nodes are the
 * full version, while the odd-numbered ones are the read-only version.
 */
static int di_max_opens = 32;

static int di_prop_dyn = 1;		/* enable dynamic property support */

#define	DI_FULL_PARENT		0
#define	DI_READONLY_PARENT	1
#define	DI_NODE_SPECIES		2
#define	DI_UNPRIVILEGED_NODE(x)	(((x) % 2) != 0)

#define	IOC_IDLE	0	/* snapshot ioctl states */
#define	IOC_SNAP	1	/* snapshot in progress */
#define	IOC_DONE	2	/* snapshot done, but not copied out */
#define	IOC_COPY	3	/* copyout in progress */

/*
 * Keep max alignment so we can move snapshot to different platforms.
 *
 * NOTE: Most callers should rely on the di_checkmem return value
 * being aligned, and reestablish *off_p with aligned value, instead
 * of trying to align size of their allocations: this approach will
 * minimize memory use.
 */
#define	DI_ALIGN(addr)	((addr + 7l) & ~7l)

/*
 * To avoid wasting memory, make a linked list of memory chunks.
 * Size of each chunk is buf_size.
 */
struct di_mem {
	struct di_mem	*next;		/* link to next chunk */
	char		*buf;		/* contiguous kernel memory */
	size_t		buf_size;	/* size of buf in bytes */
	devmap_cookie_t	cook;		/* cookie from ddi_umem_alloc */
};

/*
 * This is a stack for walking the tree without using recursion.
 * When the devinfo tree height is above some small size, one
 * gets watchdog resets on sun4m.
 */
struct di_stack {
	void		*offset[MAX_TREE_DEPTH];
	struct dev_info *dip[MAX_TREE_DEPTH];
	int		circ[MAX_TREE_DEPTH];
	int		depth;	/* depth of current node to be copied */
};

#define	TOP_OFFSET(stack)	\
	((di_off_t *)(stack)->offset[(stack)->depth - 1])
#define	TOP_NODE(stack)		\
	((stack)->dip[(stack)->depth - 1])
#define	PARENT_OFFSET(stack)	\
	((di_off_t *)(stack)->offset[(stack)->depth - 2])
#define	EMPTY_STACK(stack)	((stack)->depth == 0)
#define	POP_STACK(stack)	{ \
	ndi_devi_exit((dev_info_t *)TOP_NODE(stack), \
		(stack)->circ[(stack)->depth - 1]); \
	((stack)->depth--); \
}
#define	PUSH_STACK(stack, node, off_p)	{ \
	ASSERT(node != NULL); \
	ndi_devi_enter((dev_info_t *)node, &(stack)->circ[(stack)->depth]); \
	(stack)->dip[(stack)->depth] = (node); \
	(stack)->offset[(stack)->depth] = (void *)(off_p); \
	((stack)->depth)++; \
}

#define	DI_ALL_PTR(s)	DI_ALL(di_mem_addr((s), 0))

/*
 * With devfs, the device tree has no global locks. The device tree is
 * dynamic and dips may come and go if they are not locked locally. Under
 * these conditions, pointers are no longer reliable as unique IDs.
 * Specifically, these pointers cannot be used as keys for hash tables
 * as the same devinfo structure may be freed in one part of the tree only
 * to be allocated as the structure for a different device in another
 * part of the tree. This can happen if DR and the snapshot are
 * happening concurrently.
 * The following data structures act as keys for devinfo nodes and
 * pathinfo nodes.
 */

enum di_ktype {
	DI_DKEY = 1,
	DI_PKEY = 2
};

struct di_dkey {
	dev_info_t	*dk_dip;
	major_t		dk_major;
	int		dk_inst;
	pnode_t		dk_nodeid;
};

struct di_pkey {
	mdi_pathinfo_t	*pk_pip;
	char		*pk_path_addr;
	dev_info_t	*pk_client;
	dev_info_t	*pk_phci;
};

struct di_key {
	enum di_ktype	k_type;
	union {
		struct di_dkey dkey;
		struct di_pkey pkey;
	} k_u;
};


struct i_lnode;

typedef struct i_link {
	/*
	 * If a di_link struct representing this i_link struct makes it
	 * into the snapshot, then self will point to the offset of
	 * the di_link struct in the snapshot
	 */
	di_off_t	self;

	int		spec_type;	/* block or char access type */
	struct i_lnode	*src_lnode;	/* src i_lnode */
	struct i_lnode	*tgt_lnode;	/* tgt i_lnode */
	struct i_link	*src_link_next;	/* next src i_link /w same i_lnode */
	struct i_link	*tgt_link_next;	/* next tgt i_link /w same i_lnode */
} i_link_t;

typedef struct i_lnode {
	/*
	 * If a di_lnode struct representing this i_lnode struct makes it
	 * into the snapshot, then self will point to the offset of
	 * the di_lnode struct in the snapshot
	 */
	di_off_t	self;

	/*
	 * used for hashing and comparing i_lnodes
	 */
	int		modid;

	/*
	 * public information describing a link endpoint
	 */
	struct di_node	*di_node;	/* di_node in snapshot */
	dev_t		devt;		/* devt */

	/*
	 * i_link ptr to links coming into this i_lnode node
	 * (this i_lnode is the target of these i_links)
	 */
	i_link_t	*link_in;

	/*
	 * i_link ptr to links going out of this i_lnode node
	 * (this i_lnode is the source of these i_links)
	 */
	i_link_t	*link_out;
} i_lnode_t;

typedef struct i_hp {
	di_off_t	hp_off;		/* Offset of di_hp_t in snapshot */
	dev_info_t	*hp_child;	/* Child devinfo node of the di_hp_t */
	list_node_t	hp_link;	/* List linkage */
} i_hp_t;

/*
 * Soft state associated with each instance of driver open.
 */
static struct di_state {
	di_off_t	mem_size;	/* total # bytes in memlist */
	struct di_mem	*memlist;	/* head of memlist */
	uint_t		command;	/* command from ioctl */
	int		di_iocstate;	/* snapshot ioctl state	*/
	mod_hash_t	*reg_dip_hash;
	mod_hash_t	*reg_pip_hash;
	int		lnode_count;
	int		link_count;

	mod_hash_t	*lnode_hash;
	mod_hash_t	*link_hash;

	list_t		hp_list;
} **di_states;

static kmutex_t di_lock;	/* serialize instance assignment */

typedef enum {
	DI_QUIET = 0,	/* DI_QUIET must always be 0 */
	DI_ERR,
	DI_INFO,
	DI_TRACE,
	DI_TRACE1,
	DI_TRACE2
} di_cache_debug_t;

static uint_t	di_chunk = 32;		/* I/O chunk size in pages */

#define	DI_CACHE_LOCK(c)	(mutex_enter(&(c).cache_lock))
#define	DI_CACHE_UNLOCK(c)	(mutex_exit(&(c).cache_lock))
#define	DI_CACHE_LOCKED(c)	(mutex_owned(&(c).cache_lock))

/*
 * Check that whole device tree is being configured as a pre-condition for
 * cleaning up /etc/devices files.
 */
#define	DEVICES_FILES_CLEANABLE(st)	\
	(((st)->command & DINFOSUBTREE) && ((st)->command & DINFOFORCE) && \
	strcmp(DI_ALL_PTR(st)->root_path, "/") == 0)

#define	CACHE_DEBUG(args)	\
	{ if (di_cache_debug != DI_QUIET) di_cache_print args; }

typedef struct phci_walk_arg {
	di_off_t	off;
	struct di_state	*st;
} phci_walk_arg_t;

static int di_open(dev_t *, int, int, cred_t *);
static int di_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
static int di_close(dev_t, int, int, cred_t *);
static int di_info(dev_info_t *, ddi_info_cmd_t, void *, void **);
static int di_attach(dev_info_t *, ddi_attach_cmd_t);
static int di_detach(dev_info_t *, ddi_detach_cmd_t);

static di_off_t di_copyformat(di_off_t, struct di_state *, intptr_t, int);
static di_off_t di_snapshot_and_clean(struct di_state *);
static di_off_t di_copydevnm(di_off_t *, struct di_state *);
static di_off_t di_copytree(struct dev_info *, di_off_t *, struct di_state *);
static di_off_t di_copynode(struct dev_info *, struct di_stack *,
    struct di_state *);
static di_off_t di_getmdata(struct ddi_minor_data *, di_off_t *, di_off_t,
    struct di_state *);
static di_off_t di_getppdata(struct dev_info *, di_off_t *, struct di_state *);
static di_off_t di_getdpdata(struct dev_info *, di_off_t *, struct di_state *);
static di_off_t di_gethpdata(ddi_hp_cn_handle_t *, di_off_t *,
    struct di_state *);
static di_off_t di_getprop(int, struct ddi_prop **, di_off_t *,
    struct di_state *, struct dev_info *);
static void di_allocmem(struct di_state *, size_t);
static void di_freemem(struct di_state *);
static void di_copymem(struct di_state *st, caddr_t buf, size_t bufsiz);
static di_off_t di_checkmem(struct di_state *, di_off_t, size_t);
static void *di_mem_addr(struct di_state *, di_off_t);
static int di_setstate(struct di_state *, int);
static void di_register_dip(struct di_state *, dev_info_t *, di_off_t);
static void di_register_pip(struct di_state *, mdi_pathinfo_t *, di_off_t);
static di_off_t di_getpath_data(dev_info_t *, di_off_t *, di_off_t,
    struct di_state *, int);
static di_off_t di_getlink_data(di_off_t, struct di_state *);
static int di_dip_find(struct di_state *st, dev_info_t *node, di_off_t *off_p);

static int cache_args_valid(struct di_state *st, int *error);
static int snapshot_is_cacheable(struct di_state *st);
static int di_cache_lookup(struct di_state *st);
static int di_cache_update(struct di_state *st);
static void di_cache_print(di_cache_debug_t msglevel, char *fmt, ...);
static int build_vhci_list(dev_info_t *vh_devinfo, void *arg);
static int build_phci_list(dev_info_t *ph_devinfo, void *arg);
static void di_hotplug_children(struct di_state *st);

extern int modrootloaded;
extern void mdi_walk_vhcis(int (*)(dev_info_t *, void *), void *);
extern void mdi_vhci_walk_phcis(dev_info_t *,
	int (*)(dev_info_t *, void *), void *);


static struct cb_ops di_cb_ops = {
	di_open,		/* open */
	di_close,		/* close */
	nodev,			/* strategy */
	nodev,			/* print */
	nodev,			/* dump */
	nodev,			/* read */
	nodev,			/* write */
	di_ioctl,		/* ioctl */
	nodev,			/* devmap */
	nodev,			/* mmap */
	nodev,			/* segmap */
	nochpoll,		/* poll */
	ddi_prop_op,		/* prop_op */
	NULL,			/* streamtab  */
	D_NEW | D_MP		/* Driver compatibility flag */
};

static struct dev_ops di_ops = {
	DEVO_REV,		/* devo_rev, */
	0,			/* refcnt  */
	di_info,		/* info */
	nulldev,		/* identify */
	nulldev,		/* probe */
	di_attach,		/* attach */
	di_detach,		/* detach */
	nodev,			/* reset */
	&di_cb_ops,		/* driver operations */
	NULL			/* bus operations */
};

/*
 * Module linkage information for the kernel.
 */
static struct modldrv modldrv = {
	&mod_driverops,
	"DEVINFO Driver",
	&di_ops
};

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

int
_init(void)
{
	int	error;

	mutex_init(&di_lock, NULL, MUTEX_DRIVER, NULL);

	error = mod_install(&modlinkage);
	if (error != 0) {
		mutex_destroy(&di_lock);
		return (error);
	}

	return (0);
}

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

int
_fini(void)
{
	int	error;

	error = mod_remove(&modlinkage);
	if (error != 0) {
		return (error);
	}

	mutex_destroy(&di_lock);
	return (0);
}

static dev_info_t *di_dip;

/*ARGSUSED*/
static int
di_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
{
	int	error = DDI_FAILURE;

	switch (infocmd) {
	case DDI_INFO_DEVT2DEVINFO:
		*result = (void *)di_dip;
		error = DDI_SUCCESS;
		break;
	case DDI_INFO_DEVT2INSTANCE:
		/*
		 * All dev_t's map to the same, single instance.
		 */
		*result = (void *)0;
		error = DDI_SUCCESS;
		break;
	default:
		break;
	}

	return (error);
}

static int
di_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
{
	int	error = DDI_FAILURE;

	switch (cmd) {
	case DDI_ATTACH:
		di_states = kmem_zalloc(
		    di_max_opens * sizeof (struct di_state *), KM_SLEEP);

		if (ddi_create_minor_node(dip, "devinfo", S_IFCHR,
		    DI_FULL_PARENT, DDI_PSEUDO, NULL) == DDI_FAILURE ||
		    ddi_create_minor_node(dip, "devinfo,ro", S_IFCHR,
		    DI_READONLY_PARENT, DDI_PSEUDO, NULL) == DDI_FAILURE) {
			kmem_free(di_states,
			    di_max_opens * sizeof (struct di_state *));
			ddi_remove_minor_node(dip, NULL);
			error = DDI_FAILURE;
		} else {
			di_dip = dip;
			ddi_report_dev(dip);

			error = DDI_SUCCESS;
		}
		break;
	default:
		error = DDI_FAILURE;
		break;
	}

	return (error);
}

static int
di_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
{
	int	error = DDI_FAILURE;

	switch (cmd) {
	case DDI_DETACH:
		ddi_remove_minor_node(dip, NULL);
		di_dip = NULL;
		kmem_free(di_states, di_max_opens * sizeof (struct di_state *));

		error = DDI_SUCCESS;
		break;
	default:
		error = DDI_FAILURE;
		break;
	}

	return (error);
}

/*
 * Allow multiple opens by tweaking the dev_t such that it looks like each
 * open is getting a different minor device.  Each minor gets a separate
 * entry in the di_states[] table.  Based on the original minor number, we
 * discriminate opens of the full and read-only nodes.  If all of the instances
 * of the selected minor node are currently open, we return EAGAIN.
 */
/*ARGSUSED*/
static int
di_open(dev_t *devp, int flag, int otyp, cred_t *credp)
{
	int	m;
	minor_t	minor_parent = getminor(*devp);

	if (minor_parent != DI_FULL_PARENT &&
	    minor_parent != DI_READONLY_PARENT)
		return (ENXIO);

	mutex_enter(&di_lock);

	for (m = minor_parent; m < di_max_opens; m += DI_NODE_SPECIES) {
		if (di_states[m] != NULL)
			continue;

		di_states[m] = kmem_zalloc(sizeof (struct di_state), KM_SLEEP);
		break;	/* It's ours. */
	}

	if (m >= di_max_opens) {
		/*
		 * maximum open instance for device reached
		 */
		mutex_exit(&di_lock);
		dcmn_err((CE_WARN, "devinfo: maximum devinfo open reached"));
		return (EAGAIN);
	}
	mutex_exit(&di_lock);

	ASSERT(m < di_max_opens);
	*devp = makedevice(getmajor(*devp), (minor_t)(m + DI_NODE_SPECIES));

	dcmn_err((CE_CONT, "di_open: thread = %p, assigned minor = %d\n",
	    (void *)curthread, m + DI_NODE_SPECIES));

	return (0);
}

/*ARGSUSED*/
static int
di_close(dev_t dev, int flag, int otype, cred_t *cred_p)
{
	struct di_state	*st;
	int		m = (int)getminor(dev) - DI_NODE_SPECIES;

	if (m < 0) {
		cmn_err(CE_WARN, "closing non-existent devinfo minor %d",
		    m + DI_NODE_SPECIES);
		return (ENXIO);
	}

	st = di_states[m];
	ASSERT(m < di_max_opens && st != NULL);

	di_freemem(st);
	kmem_free(st, sizeof (struct di_state));

	/*
	 * empty slot in state table
	 */
	mutex_enter(&di_lock);
	di_states[m] = NULL;
	dcmn_err((CE_CONT, "di_close: thread = %p, assigned minor = %d\n",
	    (void *)curthread, m + DI_NODE_SPECIES));
	mutex_exit(&di_lock);

	return (0);
}


/*ARGSUSED*/
static int
di_ioctl(dev_t dev, int cmd, intptr_t arg, int mode, cred_t *credp, int *rvalp)
{
	int		rv, error;
	di_off_t	off;
	struct di_all	*all;
	struct di_state	*st;
	int		m = (int)getminor(dev) - DI_NODE_SPECIES;
	major_t		i;
	char		*drv_name;
	size_t		map_size, size;
	struct di_mem	*dcp;
	int		ndi_flags;

	if (m < 0 || m >= di_max_opens) {
		return (ENXIO);
	}

	st = di_states[m];
	ASSERT(st != NULL);

	dcmn_err2((CE_CONT, "di_ioctl: mode = %x, cmd = %x\n", mode, cmd));

	switch (cmd) {
	case DINFOIDENT:
		/*
		 * This is called from di_init to verify that the driver
		 * opened is indeed devinfo. The purpose is to guard against
		 * sending ioctl to an unknown driver in case of an
		 * unresolved major number conflict during bfu.
		 */
		*rvalp = DI_MAGIC;
		return (0);

	case DINFOLODRV:
		/*
		 * Hold an installed driver and return the result
		 */
		if (DI_UNPRIVILEGED_NODE(m)) {
			/*
			 * Only the fully enabled instances may issue
			 * DINFOLDDRV.
			 */
			return (EACCES);
		}

		drv_name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
		if (ddi_copyin((void *)arg, drv_name, MAXNAMELEN, mode) != 0) {
			kmem_free(drv_name, MAXNAMELEN);
			return (EFAULT);
		}

		/*
		 * Some 3rd party driver's _init() walks the device tree,
		 * so we load the driver module before configuring driver.
		 */
		i = ddi_name_to_major(drv_name);
		if (ddi_hold_driver(i) == NULL) {
			kmem_free(drv_name, MAXNAMELEN);
			return (ENXIO);
		}

		ndi_flags = NDI_DEVI_PERSIST | NDI_CONFIG | NDI_NO_EVENT;

		/*
		 * i_ddi_load_drvconf() below will trigger a reprobe
		 * via reset_nexus_flags(). NDI_DRV_CONF_REPROBE isn't
		 * needed here.
		 */
		modunload_disable();
		(void) i_ddi_load_drvconf(i);
		(void) ndi_devi_config_driver(ddi_root_node(), ndi_flags, i);
		kmem_free(drv_name, MAXNAMELEN);
		ddi_rele_driver(i);
		rv = i_ddi_devs_attached(i);
		modunload_enable();

		i_ddi_di_cache_invalidate();

		return ((rv == DDI_SUCCESS)? 0 : ENXIO);

	case DINFOUSRLD:
		/*
		 * The case for copying snapshot to userland
		 */
		if (di_setstate(st, IOC_COPY) == -1)
			return (EBUSY);

		map_size = DI_ALL_PTR(st)->map_size;
		if (map_size == 0) {
			(void) di_setstate(st, IOC_DONE);
			return (EFAULT);
		}

		/*
		 * copyout the snapshot
		 */
		map_size = (map_size + PAGEOFFSET) & PAGEMASK;

		/*
		 * Return the map size, so caller may do a sanity
		 * check against the return value of snapshot ioctl()
		 */
		*rvalp = (int)map_size;

		/*
		 * Copy one chunk at a time
		 */
		off = 0;
		dcp = st->memlist;
		while (map_size) {
			size = dcp->buf_size;
			if (map_size <= size) {
				size = map_size;
			}

			if (ddi_copyout(di_mem_addr(st, off),
			    (void *)(arg + off), size, mode) != 0) {
				(void) di_setstate(st, IOC_DONE);
				return (EFAULT);
			}

			map_size -= size;
			off += size;
			dcp = dcp->next;
		}

		di_freemem(st);
		(void) di_setstate(st, IOC_IDLE);
		return (0);

	default:
		if ((cmd & ~DIIOC_MASK) != DIIOC) {
			/*
			 * Invalid ioctl command
			 */
			return (ENOTTY);
		}
		/*
		 * take a snapshot
		 */
		st->command = cmd & DIIOC_MASK;
		/*FALLTHROUGH*/
	}

	/*
	 * Obtain enough memory to hold header + rootpath.  We prevent kernel
	 * memory exhaustion by freeing any previously allocated snapshot and
	 * refusing the operation; otherwise we would be allowing ioctl(),
	 * ioctl(), ioctl(), ..., panic.
	 */
	if (di_setstate(st, IOC_SNAP) == -1)
		return (EBUSY);

	/*
	 * Initial memlist always holds di_all and the root_path - and
	 * is at least a page and size.
	 */
	size = sizeof (struct di_all) +
	    sizeof (((struct dinfo_io *)(NULL))->root_path);
	if (size < PAGESIZE)
		size = PAGESIZE;
	off = di_checkmem(st, 0, size);
	all = DI_ALL_PTR(st);
	off += sizeof (struct di_all);		/* real length of di_all */

	all->devcnt = devcnt;
	all->command = st->command;
	all->version = DI_SNAPSHOT_VERSION;
	all->top_vhci_devinfo = 0;		/* filled by build_vhci_list. */

	/*
	 * Note the endianness in case we need to transport snapshot
	 * over the network.
	 */
#if defined(_LITTLE_ENDIAN)
	all->endianness = DI_LITTLE_ENDIAN;
#else
	all->endianness = DI_BIG_ENDIAN;
#endif

	/* Copyin ioctl args, store in the snapshot. */
	if (copyinstr((void *)arg, all->req_path,
	    sizeof (((struct dinfo_io *)(NULL))->root_path), &size) != 0) {
		di_freemem(st);
		(void) di_setstate(st, IOC_IDLE);
		return (EFAULT);
	}
	(void) strcpy(all->root_path, all->req_path);
	off += size;				/* real length of root_path */

	if ((st->command & DINFOCLEANUP) && !DEVICES_FILES_CLEANABLE(st)) {
		di_freemem(st);
		(void) di_setstate(st, IOC_IDLE);
		return (EINVAL);
	}

	error = 0;
	if ((st->command & DINFOCACHE) && !cache_args_valid(st, &error)) {
		di_freemem(st);
		(void) di_setstate(st, IOC_IDLE);
		return (error);
	}

	/*
	 * Only the fully enabled version may force load drivers or read
	 * the parent private data from a driver.
	 */
	if ((st->command & (DINFOPRIVDATA | DINFOFORCE)) != 0 &&
	    DI_UNPRIVILEGED_NODE(m)) {
		di_freemem(st);
		(void) di_setstate(st, IOC_IDLE);
		return (EACCES);
	}

	/* Do we need private data? */
	if (st->command & DINFOPRIVDATA) {
		arg += sizeof (((struct dinfo_io *)(NULL))->root_path);

#ifdef _MULTI_DATAMODEL
		switch (ddi_model_convert_from(mode & FMODELS)) {
		case DDI_MODEL_ILP32: {
			/*
			 * Cannot copy private data from 64-bit kernel
			 * to 32-bit app
			 */
			di_freemem(st);
			(void) di_setstate(st, IOC_IDLE);
			return (EINVAL);
		}
		case DDI_MODEL_NONE:
			if ((off = di_copyformat(off, st, arg, mode)) == 0) {
				di_freemem(st);
				(void) di_setstate(st, IOC_IDLE);
				return (EFAULT);
			}
			break;
		}
#else /* !_MULTI_DATAMODEL */
		if ((off = di_copyformat(off, st, arg, mode)) == 0) {
			di_freemem(st);
			(void) di_setstate(st, IOC_IDLE);
			return (EFAULT);
		}
#endif /* _MULTI_DATAMODEL */
	}

	all->top_devinfo = DI_ALIGN(off);

	/*
	 * For cache lookups we reallocate memory from scratch,
	 * so the value of "all" is no longer valid.
	 */
	all = NULL;

	if (st->command & DINFOCACHE) {
		*rvalp = di_cache_lookup(st);
	} else if (snapshot_is_cacheable(st)) {
		DI_CACHE_LOCK(di_cache);
		*rvalp = di_cache_update(st);
		DI_CACHE_UNLOCK(di_cache);
	} else
		*rvalp = di_snapshot_and_clean(st);

	if (*rvalp) {
		DI_ALL_PTR(st)->map_size = *rvalp;
		(void) di_setstate(st, IOC_DONE);
	} else {
		di_freemem(st);
		(void) di_setstate(st, IOC_IDLE);
	}

	return (0);
}

/*
 * Get a chunk of memory >= size, for the snapshot
 */
static void
di_allocmem(struct di_state *st, size_t size)
{
	struct di_mem	*mem = kmem_zalloc(sizeof (struct di_mem), KM_SLEEP);

	/*
	 * Round up size to nearest power of 2. If it is less
	 * than st->mem_size, set it to st->mem_size (i.e.,
	 * the mem_size is doubled every time) to reduce the
	 * number of memory allocations.
	 */
	size_t tmp = 1;
	while (tmp < size) {
		tmp <<= 1;
	}
	size = (tmp > st->mem_size) ? tmp : st->mem_size;

	mem->buf = ddi_umem_alloc(size, DDI_UMEM_SLEEP, &mem->cook);
	mem->buf_size = size;

	dcmn_err2((CE_CONT, "di_allocmem: mem_size=%x\n", st->mem_size));

	if (st->mem_size == 0) {	/* first chunk */
		st->memlist = mem;
	} else {
		/*
		 * locate end of linked list and add a chunk at the end
		 */
		struct di_mem *dcp = st->memlist;
		while (dcp->next != NULL) {
			dcp = dcp->next;
		}

		dcp->next = mem;
	}

	st->mem_size += size;
}

/*
 * Copy upto bufsiz bytes of the memlist to buf
 */
static void
di_copymem(struct di_state *st, caddr_t buf, size_t bufsiz)
{
	struct di_mem	*dcp;
	size_t		copysz;

	if (st->mem_size == 0) {
		ASSERT(st->memlist == NULL);
		return;
	}

	copysz = 0;
	for (dcp = st->memlist; dcp; dcp = dcp->next) {

		ASSERT(bufsiz > 0);

		if (bufsiz <= dcp->buf_size)
			copysz = bufsiz;
		else
			copysz = dcp->buf_size;

		bcopy(dcp->buf, buf, copysz);

		buf += copysz;
		bufsiz -= copysz;

		if (bufsiz == 0)
			break;
	}
}

/*
 * Free all memory for the snapshot
 */
static void
di_freemem(struct di_state *st)
{
	struct di_mem	*dcp, *tmp;

	dcmn_err2((CE_CONT, "di_freemem\n"));

	if (st->mem_size) {
		dcp = st->memlist;
		while (dcp) {	/* traverse the linked list */
			tmp = dcp;
			dcp = dcp->next;
			ddi_umem_free(tmp->cook);
			kmem_free(tmp, sizeof (struct di_mem));
		}
		st->mem_size = 0;
		st->memlist = NULL;
	}

	ASSERT(st->mem_size == 0);
	ASSERT(st->memlist == NULL);
}

/*
 * Copies cached data to the di_state structure.
 * Returns:
 *	- size of data copied, on SUCCESS
 *	- 0 on failure
 */
static int
di_cache2mem(struct di_cache *cache, struct di_state *st)
{
	caddr_t	pa;

	ASSERT(st->mem_size == 0);
	ASSERT(st->memlist == NULL);
	ASSERT(!servicing_interrupt());
	ASSERT(DI_CACHE_LOCKED(*cache));

	if (cache->cache_size == 0) {
		ASSERT(cache->cache_data == NULL);
		CACHE_DEBUG((DI_ERR, "Empty cache. Skipping copy"));
		return (0);
	}

	ASSERT(cache->cache_data);

	di_allocmem(st, cache->cache_size);

	pa = di_mem_addr(st, 0);

	ASSERT(pa);

	/*
	 * Verify that di_allocmem() allocates contiguous memory,
	 * so that it is safe to do straight bcopy()
	 */
	ASSERT(st->memlist != NULL);
	ASSERT(st->memlist->next == NULL);
	bcopy(cache->cache_data, pa, cache->cache_size);

	return (cache->cache_size);
}

/*
 * Copies a snapshot from di_state to the cache
 * Returns:
 *	- 0 on failure
 *	- size of copied data on success
 */
static size_t
di_mem2cache(struct di_state *st, struct di_cache *cache)
{
	size_t	map_size;

	ASSERT(cache->cache_size == 0);
	ASSERT(cache->cache_data == NULL);
	ASSERT(!servicing_interrupt());
	ASSERT(DI_CACHE_LOCKED(*cache));

	if (st->mem_size == 0) {
		ASSERT(st->memlist == NULL);
		CACHE_DEBUG((DI_ERR, "Empty memlist. Skipping copy"));
		return (0);
	}

	ASSERT(st->memlist);

	/*
	 * The size of the memory list may be much larger than the
	 * size of valid data (map_size). Cache only the valid data
	 */
	map_size = DI_ALL_PTR(st)->map_size;
	if (map_size == 0 || map_size < sizeof (struct di_all) ||
	    map_size > st->mem_size) {
		CACHE_DEBUG((DI_ERR, "cannot cache: bad size: 0x%x", map_size));
		return (0);
	}

	cache->cache_data = kmem_alloc(map_size, KM_SLEEP);
	cache->cache_size = map_size;
	di_copymem(st, cache->cache_data, cache->cache_size);

	return (map_size);
}

/*
 * Make sure there is at least "size" bytes memory left before
 * going on. Otherwise, start on a new chunk.
 */
static di_off_t
di_checkmem(struct di_state *st, di_off_t off, size_t size)
{
	dcmn_err3((CE_CONT, "di_checkmem: off=%x size=%x\n",
	    off, (int)size));

	/*
	 * di_checkmem() shouldn't be called with a size of zero.
	 * But in case it is, we want to make sure we return a valid
	 * offset within the memlist and not an offset that points us
	 * at the end of the memlist.
	 */
	if (size == 0) {
		dcmn_err((CE_WARN, "di_checkmem: invalid zero size used"));
		size = 1;
	}

	off = DI_ALIGN(off);
	if ((st->mem_size - off) < size) {
		off = st->mem_size;
		di_allocmem(st, size);
	}

	/* verify that return value is aligned */
	ASSERT(off == DI_ALIGN(off));
	return (off);
}

/*
 * Copy the private data format from ioctl arg.
 * On success, the ending offset is returned. On error 0 is returned.
 */
static di_off_t
di_copyformat(di_off_t off, struct di_state *st, intptr_t arg, int mode)
{
	di_off_t		size;
	struct di_priv_data	*priv;
	struct di_all		*all = DI_ALL_PTR(st);

	dcmn_err2((CE_CONT, "di_copyformat: off=%x, arg=%p mode=%x\n",
	    off, (void *)arg, mode));

	/*
	 * Copyin data and check version.
	 * We only handle private data version 0.
	 */
	priv = kmem_alloc(sizeof (struct di_priv_data), KM_SLEEP);
	if ((ddi_copyin((void *)arg, priv, sizeof (struct di_priv_data),
	    mode) != 0) || (priv->version != DI_PRIVDATA_VERSION_0)) {
		kmem_free(priv, sizeof (struct di_priv_data));
		return (0);
	}

	/*
	 * Save di_priv_data copied from userland in snapshot.
	 */
	all->pd_version = priv->version;
	all->n_ppdata = priv->n_parent;
	all->n_dpdata = priv->n_driver;

	/*
	 * copyin private data format, modify offset accordingly
	 */
	if (all->n_ppdata) {	/* parent private data format */
		/*
		 * check memory
		 */
		size = all->n_ppdata * sizeof (struct di_priv_format);
		all->ppdata_format = off = di_checkmem(st, off, size);
		if (ddi_copyin(priv->parent, di_mem_addr(st, off), size,
		    mode) != 0) {
			kmem_free(priv, sizeof (struct di_priv_data));
			return (0);
		}

		off += size;
	}

	if (all->n_dpdata) {	/* driver private data format */
		/*
		 * check memory
		 */
		size = all->n_dpdata * sizeof (struct di_priv_format);
		all->dpdata_format = off = di_checkmem(st, off, size);
		if (ddi_copyin(priv->driver, di_mem_addr(st, off), size,
		    mode) != 0) {
			kmem_free(priv, sizeof (struct di_priv_data));
			return (0);
		}

		off += size;
	}

	kmem_free(priv, sizeof (struct di_priv_data));
	return (off);
}

/*
 * Return the real address based on the offset (off) within snapshot
 */
static void *
di_mem_addr(struct di_state *st, di_off_t off)
{
	struct di_mem	*dcp = st->memlist;

	dcmn_err3((CE_CONT, "di_mem_addr: dcp=%p off=%x\n",
	    (void *)dcp, off));

	ASSERT(off < st->mem_size);

	while (off >= dcp->buf_size) {
		off -= dcp->buf_size;
		dcp = dcp->next;
	}

	dcmn_err3((CE_CONT, "di_mem_addr: new off=%x, return = %p\n",
	    off, (void *)(dcp->buf + off)));

	return (dcp->buf + off);
}

/*
 * Ideally we would use the whole key to derive the hash
 * value. However, the probability that two keys will
 * have the same dip (or pip) is very low, so
 * hashing by dip (or pip) pointer should suffice.
 */
static uint_t
di_hash_byptr(void *arg, mod_hash_key_t key)
{
	struct di_key	*dik = key;
	size_t		rshift;
	void		*ptr;

	ASSERT(arg == NULL);

	switch (dik->k_type) {
	case DI_DKEY:
		ptr = dik->k_u.dkey.dk_dip;
		rshift = highbit(sizeof (struct dev_info));
		break;
	case DI_PKEY:
		ptr = dik->k_u.pkey.pk_pip;
		rshift = highbit(sizeof (struct mdi_pathinfo));
		break;
	default:
		panic("devinfo: unknown key type");
		/*NOTREACHED*/
	}
	return (mod_hash_byptr((void *)rshift, ptr));
}

static void
di_key_dtor(mod_hash_key_t key)
{
	char		*path_addr;
	struct di_key	*dik = key;

	switch (dik->k_type) {
	case DI_DKEY:
		break;
	case DI_PKEY:
		path_addr = dik->k_u.pkey.pk_path_addr;
		if (path_addr)
			kmem_free(path_addr, strlen(path_addr) + 1);
		break;
	default:
		panic("devinfo: unknown key type");
		/*NOTREACHED*/
	}

	kmem_free(dik, sizeof (struct di_key));
}

static int
di_dkey_cmp(struct di_dkey *dk1, struct di_dkey *dk2)
{
	if (dk1->dk_dip !=  dk2->dk_dip)
		return (dk1->dk_dip > dk2->dk_dip ? 1 : -1);

	if (dk1->dk_major != DDI_MAJOR_T_NONE &&
	    dk2->dk_major != DDI_MAJOR_T_NONE) {
		if (dk1->dk_major !=  dk2->dk_major)
			return (dk1->dk_major > dk2->dk_major ? 1 : -1);

		if (dk1->dk_inst !=  dk2->dk_inst)
			return (dk1->dk_inst > dk2->dk_inst ? 1 : -1);
	}

	if (dk1->dk_nodeid != dk2->dk_nodeid)
		return (dk1->dk_nodeid > dk2->dk_nodeid ? 1 : -1);

	return (0);
}

static int
di_pkey_cmp(struct di_pkey *pk1, struct di_pkey *pk2)
{
	char	*p1, *p2;
	int	rv;

	if (pk1->pk_pip !=  pk2->pk_pip)
		return (pk1->pk_pip > pk2->pk_pip ? 1 : -1);

	p1 = pk1->pk_path_addr;
	p2 = pk2->pk_path_addr;

	p1 = p1 ? p1 : "";
	p2 = p2 ? p2 : "";

	rv = strcmp(p1, p2);
	if (rv)
		return (rv > 0  ? 1 : -1);

	if (pk1->pk_client !=  pk2->pk_client)
		return (pk1->pk_client > pk2->pk_client ? 1 : -1);

	if (pk1->pk_phci !=  pk2->pk_phci)
		return (pk1->pk_phci > pk2->pk_phci ? 1 : -1);

	return (0);
}

static int
di_key_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
{
	struct di_key	*dik1, *dik2;

	dik1 = key1;
	dik2 = key2;

	if (dik1->k_type != dik2->k_type) {
		panic("devinfo: mismatched keys");
		/*NOTREACHED*/
	}

	switch (dik1->k_type) {
	case DI_DKEY:
		return (di_dkey_cmp(&(dik1->k_u.dkey), &(dik2->k_u.dkey)));
	case DI_PKEY:
		return (di_pkey_cmp(&(dik1->k_u.pkey), &(dik2->k_u.pkey)));
	default:
		panic("devinfo: unknown key type");
		/*NOTREACHED*/
	}
}

static void
di_copy_aliases(struct di_state *st, alias_pair_t *apair, di_off_t *offp)
{
	di_off_t 		off;
	struct di_all		*all = DI_ALL_PTR(st);
	struct di_alias		*di_alias;
	di_off_t		curroff;
	dev_info_t		*currdip;
	size_t			size;

	currdip = NULL;
	if (resolve_pathname(apair->pair_alias, &currdip, NULL, NULL) != 0) {
		return;
	}

	if (di_dip_find(st, currdip, &curroff) != 0) {
		ndi_rele_devi(currdip);
		return;
	}
	ndi_rele_devi(currdip);

	off = *offp;
	size = sizeof (struct di_alias);
	size += strlen(apair->pair_alias) + 1;
	off = di_checkmem(st, off, size);
	di_alias = DI_ALIAS(di_mem_addr(st, off));

	di_alias->self = off;
	di_alias->next = all->aliases;
	all->aliases = off;
	(void) strcpy(di_alias->alias, apair->pair_alias);
	di_alias->curroff = curroff;

	off += size;

	*offp = off;
}

/*
 * This is the main function that takes a snapshot
 */
static di_off_t
di_snapshot(struct di_state *st)
{
	di_off_t	off;
	struct di_all	*all;
	dev_info_t	*rootnode;
	char		buf[80];
	int		plen;
	char		*path;
	vnode_t		*vp;
	int		i;

	all = DI_ALL_PTR(st);
	dcmn_err((CE_CONT, "Taking a snapshot of devinfo tree...\n"));

	/*
	 * Translate requested root path if an alias and snap-root != "/"
	 */
	if (ddi_aliases_present == B_TRUE && strcmp(all->root_path, "/") != 0) {
		/* If there is no redirected alias, use root_path as is */
		rootnode = ddi_alias_redirect(all->root_path);
		if (rootnode) {
			(void) ddi_pathname(rootnode, all->root_path);
			goto got_root;
		}
	}

	/*
	 * Verify path before entrusting it to e_ddi_hold_devi_by_path because
	 * some platforms have OBP bugs where executing the NDI_PROMNAME code
	 * path against an invalid path results in panic.  The lookupnameat
	 * is done relative to rootdir without a leading '/' on "devices/"
	 * to force the lookup to occur in the global zone.
	 */
	plen = strlen("devices/") + strlen(all->root_path) + 1;
	path = kmem_alloc(plen, KM_SLEEP);
	(void) snprintf(path, plen, "devices/%s", all->root_path);
	if (lookupnameat(path, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp, rootdir)) {
		dcmn_err((CE_CONT, "Devinfo node %s not found\n",
		    all->root_path));
		kmem_free(path, plen);
		return (0);
	}
	kmem_free(path, plen);
	VN_RELE(vp);

	/*
	 * Hold the devinfo node referred by the path.
	 */
	rootnode = e_ddi_hold_devi_by_path(all->root_path, 0);
	if (rootnode == NULL) {
		dcmn_err((CE_CONT, "Devinfo node %s not found\n",
		    all->root_path));
		return (0);
	}

got_root:
	(void) snprintf(buf, sizeof (buf),
	    "devinfo registered dips (statep=%p)", (void *)st);

	st->reg_dip_hash = mod_hash_create_extended(buf, 64,
	    di_key_dtor, mod_hash_null_valdtor, di_hash_byptr,
	    NULL, di_key_cmp, KM_SLEEP);


	(void) snprintf(buf, sizeof (buf),
	    "devinfo registered pips (statep=%p)", (void *)st);

	st->reg_pip_hash = mod_hash_create_extended(buf, 64,
	    di_key_dtor, mod_hash_null_valdtor, di_hash_byptr,
	    NULL, di_key_cmp, KM_SLEEP);

	if (DINFOHP & st->command) {
		list_create(&st->hp_list, sizeof (i_hp_t),
		    offsetof(i_hp_t, hp_link));
	}

	/*
	 * copy the device tree
	 */
	off = di_copytree(DEVI(rootnode), &all->top_devinfo, st);

	if (DINFOPATH & st->command) {
		mdi_walk_vhcis(build_vhci_list, st);
	}

	if (DINFOHP & st->command) {
		di_hotplug_children(st);
	}

	ddi_release_devi(rootnode);

	/*
	 * copy the devnames array
	 */
	all->devnames = off;
	off = di_copydevnm(&all->devnames, st);


	/* initialize the hash tables */
	st->lnode_count = 0;
	st->link_count = 0;

	if (DINFOLYR & st->command) {
		off = di_getlink_data(off, st);
	}

	all->aliases = 0;
	if (ddi_aliases_present == B_FALSE)
		goto done;

	for (i = 0; i < ddi_aliases.dali_num_pairs; i++) {
		di_copy_aliases(st, &(ddi_aliases.dali_alias_pairs[i]), &off);
	}

done:
	/*
	 * Free up hash tables
	 */
	mod_hash_destroy_hash(st->reg_dip_hash);
	mod_hash_destroy_hash(st->reg_pip_hash);

	/*
	 * Record the timestamp now that we are done with snapshot.
	 *
	 * We compute the checksum later and then only if we cache
	 * the snapshot, since checksumming adds some overhead.
	 * The checksum is checked later if we read the cache file.
	 * from disk.
	 *
	 * Set checksum field to 0 as CRC is calculated with that
	 * field set to 0.
	 */
	all->snapshot_time = ddi_get_time();
	all->cache_checksum = 0;

	ASSERT(all->snapshot_time != 0);

	return (off);
}

/*
 * Take a snapshot and clean /etc/devices files if DINFOCLEANUP is set
 */
static di_off_t
di_snapshot_and_clean(struct di_state *st)
{
	di_off_t	off;

	modunload_disable();
	off = di_snapshot(st);
	if (off != 0 && (st->command & DINFOCLEANUP)) {
		ASSERT(DEVICES_FILES_CLEANABLE(st));
		/*
		 * Cleanup /etc/devices files:
		 * In order to accurately account for the system configuration
		 * in /etc/devices files, the appropriate drivers must be
		 * fully configured before the cleanup starts.
		 * So enable modunload only after the cleanup.
		 */
		i_ddi_clean_devices_files();
		/*
		 * Remove backing store nodes for unused devices,
		 * which retain past permissions customizations
		 * and may be undesired for newly configured devices.
		 */
		dev_devices_cleanup();
	}
	modunload_enable();

	return (off);
}

/*
 * construct vhci linkage in the snapshot.
 */
static int
build_vhci_list(dev_info_t *vh_devinfo, void *arg)
{
	struct di_all	*all;
	struct di_node	*me;
	struct di_state	*st;
	di_off_t	off;
	phci_walk_arg_t	pwa;

	dcmn_err3((CE_CONT, "build_vhci list\n"));

	dcmn_err3((CE_CONT, "vhci node %s%d\n",
	    ddi_driver_name(vh_devinfo), ddi_get_instance(vh_devinfo)));

	st = (struct di_state *)arg;
	if (di_dip_find(st, vh_devinfo, &off) != 0) {
		dcmn_err((CE_WARN, "di_dip_find error for the given node\n"));
		return (DDI_WALK_TERMINATE);
	}

	dcmn_err3((CE_CONT, "st->mem_size: %d vh_devinfo off: 0x%x\n",
	    st->mem_size, off));

	all = DI_ALL_PTR(st);
	if (all->top_vhci_devinfo == 0) {
		all->top_vhci_devinfo = off;
	} else {
		me = DI_NODE(di_mem_addr(st, all->top_vhci_devinfo));

		while (me->next_vhci != 0) {
			me = DI_NODE(di_mem_addr(st, me->next_vhci));
		}

		me->next_vhci = off;
	}

	pwa.off = off;
	pwa.st = st;
	mdi_vhci_walk_phcis(vh_devinfo, build_phci_list, &pwa);

	return (DDI_WALK_CONTINUE);
}

/*
 * construct phci linkage for the given vhci in the snapshot.
 */
static int
build_phci_list(dev_info_t *ph_devinfo, void *arg)
{
	struct di_node	*vh_di_node;
	struct di_node	*me;
	phci_walk_arg_t	*pwa;
	di_off_t	off;

	pwa = (phci_walk_arg_t *)arg;

	dcmn_err3((CE_CONT, "build_phci list for vhci at offset: 0x%x\n",
	    pwa->off));

	vh_di_node = DI_NODE(di_mem_addr(pwa->st, pwa->off));
	if (di_dip_find(pwa->st, ph_devinfo, &off) != 0) {
		dcmn_err((CE_WARN, "di_dip_find error for the given node\n"));
		return (DDI_WALK_TERMINATE);
	}

	dcmn_err3((CE_CONT, "phci node %s%d, at offset 0x%x\n",
	    ddi_driver_name(ph_devinfo), ddi_get_instance(ph_devinfo), off));

	if (vh_di_node->top_phci == 0) {
		vh_di_node->top_phci = off;
		return (DDI_WALK_CONTINUE);
	}

	me = DI_NODE(di_mem_addr(pwa->st, vh_di_node->top_phci));

	while (me->next_phci != 0) {
		me = DI_NODE(di_mem_addr(pwa->st, me->next_phci));
	}
	me->next_phci = off;

	return (DDI_WALK_CONTINUE);
}

/*
 * Assumes all devinfo nodes in device tree have been snapshotted
 */
static void
snap_driver_list(struct di_state *st, struct devnames *dnp, di_off_t *off_p)
{
	struct dev_info	*node;
	struct di_node	*me;
	di_off_t	off;

	ASSERT(mutex_owned(&dnp->dn_lock));

	node = DEVI(dnp->dn_head);
	for (; node; node = node->devi_next) {
		if (di_dip_find(st, (dev_info_t *)node, &off) != 0)
			continue;

		ASSERT(off > 0);
		me = DI_NODE(di_mem_addr(st, off));
		ASSERT(me->next == 0 || me->next == -1);
		/*
		 * Only nodes which were BOUND when they were
		 * snapshotted will be added to per-driver list.
		 */
		if (me->next != -1)
			continue;

		*off_p = off;
		off_p = &me->next;
	}

	*off_p = 0;
}

/*
 * Copy the devnames array, so we have a list of drivers in the snapshot.
 * Also makes it possible to locate the per-driver devinfo nodes.
 */
static di_off_t
di_copydevnm(di_off_t *off_p, struct di_state *st)
{
	int		i;
	di_off_t	off;
	size_t		size;
	struct di_devnm	*dnp;

	dcmn_err2((CE_CONT, "di_copydevnm: *off_p = %p\n", (void *)off_p));

	/*
	 * make sure there is some allocated memory
	 */
	size = devcnt * sizeof (struct di_devnm);
	*off_p = off = di_checkmem(st, *off_p, size);
	dnp = DI_DEVNM(di_mem_addr(st, off));
	off += size;

	dcmn_err((CE_CONT, "Start copying devnamesp[%d] at offset 0x%x\n",
	    devcnt, off));

	for (i = 0; i < devcnt; i++) {
		if (devnamesp[i].dn_name == NULL) {
			continue;
		}

		/*
		 * dn_name is not freed during driver unload or removal.
		 *
		 * There is a race condition when make_devname() changes
		 * dn_name during our strcpy. This should be rare since
		 * only add_drv does this. At any rate, we never had a
		 * problem with ddi_name_to_major(), which should have
		 * the same problem.
		 */
		dcmn_err2((CE_CONT, "di_copydevnm: %s%d, off=%x\n",
		    devnamesp[i].dn_name, devnamesp[i].dn_instance, off));

		size = strlen(devnamesp[i].dn_name) + 1;
		dnp[i].name = off = di_checkmem(st, off, size);
		(void) strcpy((char *)di_mem_addr(st, off),
		    devnamesp[i].dn_name);
		off += size;

		mutex_enter(&devnamesp[i].dn_lock);

		/*
		 * Snapshot per-driver node list
		 */
		snap_driver_list(st, &devnamesp[i], &dnp[i].head);

		/*
		 * This is not used by libdevinfo, leave it for now
		 */
		dnp[i].flags = devnamesp[i].dn_flags;
		dnp[i].instance = devnamesp[i].dn_instance;

		/*
		 * get global properties
		 */
		if ((DINFOPROP & st->command) &&
		    devnamesp[i].dn_global_prop_ptr) {
			dnp[i].global_prop = off;
			off = di_getprop(DI_PROP_GLB_LIST,
			    &devnamesp[i].dn_global_prop_ptr->prop_list,
			    &dnp[i].global_prop, st, NULL);
		}

		/*
		 * Bit encode driver ops: & bus_ops, cb_ops, & cb_ops->cb_str
		 */
		if (CB_DRV_INSTALLED(devopsp[i])) {
			if (devopsp[i]->devo_cb_ops) {
				dnp[i].ops |= DI_CB_OPS;
				if (devopsp[i]->devo_cb_ops->cb_str)
					dnp[i].ops |= DI_STREAM_OPS;
			}
			if (NEXUS_DRV(devopsp[i])) {
				dnp[i].ops |= DI_BUS_OPS;
			}
		}

		mutex_exit(&devnamesp[i].dn_lock);
	}

	dcmn_err((CE_CONT, "End copying devnamesp at offset 0x%x\n", off));

	return (off);
}

/*
 * Copy the kernel devinfo tree. The tree and the devnames array forms
 * the entire snapshot (see also di_copydevnm).
 */
static di_off_t
di_copytree(struct dev_info *root, di_off_t *off_p, struct di_state *st)
{
	di_off_t	off;
	struct dev_info	*node;
	struct di_stack	*dsp = kmem_zalloc(sizeof (struct di_stack), KM_SLEEP);

	dcmn_err((CE_CONT, "di_copytree: root = %p, *off_p = %x\n",
	    (void *)root, *off_p));

	/* force attach drivers */
	if (i_ddi_devi_attached((dev_info_t *)root) &&
	    (st->command & DINFOSUBTREE) && (st->command & DINFOFORCE)) {
		(void) ndi_devi_config((dev_info_t *)root,
		    NDI_CONFIG | NDI_DEVI_PERSIST | NDI_NO_EVENT |
		    NDI_DRV_CONF_REPROBE);
	}

	/*
	 * Push top_devinfo onto a stack
	 *
	 * The stack is necessary to avoid recursion, which can overrun
	 * the kernel stack.
	 */
	PUSH_STACK(dsp, root, off_p);

	/*
	 * As long as there is a node on the stack, copy the node.
	 * di_copynode() is responsible for pushing and popping
	 * child and sibling nodes on the stack.
	 */
	while (!EMPTY_STACK(dsp)) {
		node = TOP_NODE(dsp);
		off = di_copynode(node, dsp, st);
	}

	/*
	 * Free the stack structure
	 */
	kmem_free(dsp, sizeof (struct di_stack));

	return (off);
}

/*
 * This is the core function, which copies all data associated with a single
 * node into the snapshot. The amount of information is determined by the
 * ioctl command.
 */
static di_off_t
di_copynode(struct dev_info *node, struct di_stack *dsp, struct di_state *st)
{
	di_off_t	off;
	struct di_node	*me;
	size_t		size;
	struct dev_info *n;

	dcmn_err2((CE_CONT, "di_copynode: depth = %x\n", dsp->depth));
	ASSERT((node != NULL) && (node == TOP_NODE(dsp)));

	/*
	 * check memory usage, and fix offsets accordingly.
	 */
	size = sizeof (struct di_node);
	*(TOP_OFFSET(dsp)) = off = di_checkmem(st, *(TOP_OFFSET(dsp)), size);
	me = DI_NODE(di_mem_addr(st, off));
	me->self = off;
	off += size;

	dcmn_err((CE_CONT, "copy node %s, instance #%d, at offset 0x%x\n",
	    node->devi_node_name, node->devi_instance, off));

	/*
	 * Node parameters:
	 * self		-- offset of current node within snapshot
	 * nodeid	-- pointer to PROM node (tri-valued)
	 * state	-- hot plugging device state
	 * node_state	-- devinfo node state
	 */
	me->instance = node->devi_instance;
	me->nodeid = node->devi_nodeid;
	me->node_class = node->devi_node_class;
	me->attributes = node->devi_node_attributes;
	me->state = node->devi_state;
	me->flags = node->devi_flags;
	me->node_state = node->devi_node_state;
	me->next_vhci = 0;		/* Filled up by build_vhci_list. */
	me->top_phci = 0;		/* Filled up by build_phci_list. */
	me->next_phci = 0;		/* Filled up by build_phci_list. */
	me->multipath_component = MULTIPATH_COMPONENT_NONE; /* set default. */
	me->user_private_data = NULL;

	/*
	 * Get parent's offset in snapshot from the stack
	 * and store it in the current node
	 */
	if (dsp->depth > 1) {
		me->parent = *(PARENT_OFFSET(dsp));
	}

	/*
	 * Save the offset of this di_node in a hash table.
	 * This is used later to resolve references to this
	 * dip from other parts of the tree (per-driver list,
	 * multipathing linkages, layered usage linkages).
	 * The key used for the hash table is derived from
	 * information in the dip.
	 */
	di_register_dip(st, (dev_info_t *)node, me->self);

#ifdef	DEVID_COMPATIBILITY
	/* check for devid as property marker */
	if (node->devi_devid_str) {
		ddi_devid_t	devid;

		/*
		 * The devid is now represented as a property. For
		 * compatibility with di_devid() interface in libdevinfo we
		 * must return it as a binary structure in the snapshot. When
		 * (if) di_devid() is removed from libdevinfo then the code
		 * related to DEVID_COMPATIBILITY can be removed.
		 */
		if (ddi_devid_str_decode(node->devi_devid_str, &devid, NULL) ==
		    DDI_SUCCESS) {
			size = ddi_devid_sizeof(devid);
			off = di_checkmem(st, off, size);
			me->devid = off;
			bcopy(devid, di_mem_addr(st, off), size);
			off += size;
			ddi_devid_free(devid);
		}
	}
#endif	/* DEVID_COMPATIBILITY */

	if (node->devi_node_name) {
		size = strlen(node->devi_node_name) + 1;
		me->node_name = off = di_checkmem(st, off, size);
		(void) strcpy(di_mem_addr(st, off), node->devi_node_name);
		off += size;
	}

	if (node->devi_compat_names && (node->devi_compat_length > 1)) {
		size = node->devi_compat_length;
		me->compat_names = off = di_checkmem(st, off, size);
		me->compat_length = (int)size;
		bcopy(node->devi_compat_names, di_mem_addr(st, off), size);
		off += size;
	}

	if (node->devi_addr) {
		size = strlen(node->devi_addr) + 1;
		me->address = off = di_checkmem(st, off, size);
		(void) strcpy(di_mem_addr(st, off), node->devi_addr);
		off += size;
	}

	if (node->devi_binding_name) {
		size = strlen(node->devi_binding_name) + 1;
		me->bind_name = off = di_checkmem(st, off, size);
		(void) strcpy(di_mem_addr(st, off), node->devi_binding_name);
		off += size;
	}

	me->drv_major = node->devi_major;

	/*
	 * If the dip is BOUND, set the next pointer of the
	 * per-instance list to -1, indicating that it is yet to be resolved.
	 * This will be resolved later in snap_driver_list().
	 */
	if (me->drv_major != -1) {
		me->next = -1;
	} else {
		me->next = 0;
	}

	/*
	 * An optimization to skip mutex_enter when not needed.
	 */
	if (!((DINFOMINOR | DINFOPROP | DINFOPATH | DINFOHP) & st->command)) {
		goto priv_data;
	}

	/*
	 * LOCKING: We already have an active ndi_devi_enter to gather the
	 * minor data, and we will take devi_lock to gather properties as
	 * needed off di_getprop.
	 */
	if (!(DINFOMINOR & st->command)) {
		goto path;
	}

	ASSERT(DEVI_BUSY_OWNED(node));
	if (node->devi_minor) {		/* minor data */
		me->minor_data = off;
		off = di_getmdata(node->devi_minor, &me->minor_data,
		    me->self, st);
	}

path:
	if (!(DINFOPATH & st->command)) {
		goto property;
	}

	if (MDI_VHCI(node)) {
		me->multipath_component = MULTIPATH_COMPONENT_VHCI;
	}

	if (MDI_CLIENT(node)) {
		me->multipath_component = MULTIPATH_COMPONENT_CLIENT;
		me->multipath_client = off;
		off = di_getpath_data((dev_info_t *)node, &me->multipath_client,
		    me->self, st, 1);
		dcmn_err((CE_WARN, "me->multipath_client = %x for node %p "
		    "component type = %d.  off=%d",
		    me->multipath_client,
		    (void *)node, node->devi_mdi_component, off));
	}

	if (MDI_PHCI(node)) {
		me->multipath_component = MULTIPATH_COMPONENT_PHCI;
		me->multipath_phci = off;
		off = di_getpath_data((dev_info_t *)node, &me->multipath_phci,
		    me->self, st, 0);
		dcmn_err((CE_WARN, "me->multipath_phci = %x for node %p "
		    "component type = %d.  off=%d",
		    me->multipath_phci,
		    (void *)node, node->devi_mdi_component, off));
	}

property:
	if (!(DINFOPROP & st->command)) {
		goto hotplug_data;
	}

	if (node->devi_drv_prop_ptr) {	/* driver property list */
		me->drv_prop = off;
		off = di_getprop(DI_PROP_DRV_LIST, &node->devi_drv_prop_ptr,
		    &me->drv_prop, st, node);
	}

	if (node->devi_sys_prop_ptr) {	/* system property list */
		me->sys_prop = off;
		off = di_getprop(DI_PROP_SYS_LIST, &node->devi_sys_prop_ptr,
		    &me->sys_prop, st, node);
	}

	if (node->devi_hw_prop_ptr) {	/* hardware property list */
		me->hw_prop = off;
		off = di_getprop(DI_PROP_HW_LIST, &node->devi_hw_prop_ptr,
		    &me->hw_prop, st, node);
	}

	if (node->devi_global_prop_list == NULL) {
		me->glob_prop = (di_off_t)-1;	/* not global property */
	} else {
		/*
		 * Make copy of global property list if this devinfo refers
		 * global properties different from what's on the devnames
		 * array. It can happen if there has been a forced
		 * driver.conf update. See mod_drv(1M).
		 */
		ASSERT(me->drv_major != -1);
		if (node->devi_global_prop_list !=
		    devnamesp[me->drv_major].dn_global_prop_ptr) {
			me->glob_prop = off;
			off = di_getprop(DI_PROP_GLB_LIST,
			    &node->devi_global_prop_list->prop_list,
			    &me->glob_prop, st, node);
		}
	}

hotplug_data:
	if (!(DINFOHP & st->command)) {
		goto priv_data;
	}

	if (node->devi_hp_hdlp) {	/* hotplug data */
		me->hp_data = off;
		off = di_gethpdata(node->devi_hp_hdlp, &me->hp_data, st);
	}

priv_data:
	if (!(DINFOPRIVDATA & st->command)) {
		goto pm_info;
	}

	if (ddi_get_parent_data((dev_info_t *)node) != NULL) {
		me->parent_data = off;
		off = di_getppdata(node, &me->parent_data, st);
	}

	if (ddi_get_driver_private((dev_info_t *)node) != NULL) {
		me->driver_data = off;
		off = di_getdpdata(node, &me->driver_data, st);
	}

pm_info: /* NOT implemented */

subtree:
	/* keep the stack aligned */
	off = DI_ALIGN(off);

	if (!(DINFOSUBTREE & st->command)) {
		POP_STACK(dsp);
		return (off);
	}

child:
	/*
	 * If there is a visible child--push child onto stack.
	 * Hold the parent (me) busy while doing so.
	 */
	if ((n = node->devi_child) != NULL) {
		/* skip hidden nodes */
		while (n && ndi_dev_is_hidden_node((dev_info_t *)n))
			n = n->devi_sibling;
		if (n) {
			me->child = off;
			PUSH_STACK(dsp, n, &me->child);
			return (me->child);
		}
	}

sibling:
	/*
	 * Done with any child nodes, unroll the stack till a visible
	 * sibling of a parent node is found or root node is reached.
	 */
	POP_STACK(dsp);
	while (!EMPTY_STACK(dsp)) {
		if ((n = node->devi_sibling) != NULL) {
			/* skip hidden nodes */
			while (n && ndi_dev_is_hidden_node((dev_info_t *)n))
				n = n->devi_sibling;
			if (n) {
				me->sibling = DI_ALIGN(off);
				PUSH_STACK(dsp, n, &me->sibling);
				return (me->sibling);
			}
		}
		node = TOP_NODE(dsp);
		me = DI_NODE(di_mem_addr(st, *(TOP_OFFSET(dsp))));
		POP_STACK(dsp);
	}

	/*
	 * DONE with all nodes
	 */
	return (off);
}

static i_lnode_t *
i_lnode_alloc(int modid)
{
	i_lnode_t	*i_lnode;

	i_lnode = kmem_zalloc(sizeof (i_lnode_t), KM_SLEEP);

	ASSERT(modid != -1);
	i_lnode->modid = modid;

	return (i_lnode);
}

static void
i_lnode_free(i_lnode_t *i_lnode)
{
	kmem_free(i_lnode, sizeof (i_lnode_t));
}

static void
i_lnode_check_free(i_lnode_t *i_lnode)
{
	/* This lnode and its dip must have been snapshotted */
	ASSERT(i_lnode->self > 0);
	ASSERT(i_lnode->di_node->self > 0);

	/* at least 1 link (in or out) must exist for this lnode */
	ASSERT(i_lnode->link_in || i_lnode->link_out);

	i_lnode_free(i_lnode);
}

static i_link_t *
i_link_alloc(int spec_type)
{
	i_link_t	*i_link;

	i_link = kmem_zalloc(sizeof (i_link_t), KM_SLEEP);
	i_link->spec_type = spec_type;

	return (i_link);
}

static void
i_link_check_free(i_link_t *i_link)
{
	/* This link must have been snapshotted */
	ASSERT(i_link->self > 0);

	/* Both endpoint lnodes must exist for this link */
	ASSERT(i_link->src_lnode);
	ASSERT(i_link->tgt_lnode);

	kmem_free(i_link, sizeof (i_link_t));
}

/*ARGSUSED*/
static uint_t
i_lnode_hashfunc(void *arg, mod_hash_key_t key)
{
	i_lnode_t	*i_lnode = (i_lnode_t *)key;
	struct di_node	*ptr;
	dev_t		dev;

	dev = i_lnode->devt;
	if (dev != DDI_DEV_T_NONE)
		return (i_lnode->modid + getminor(dev) + getmajor(dev));

	ptr = i_lnode->di_node;
	ASSERT(ptr->self > 0);
	if (ptr) {
		uintptr_t k = (uintptr_t)ptr;
		k >>= (int)highbit(sizeof (struct di_node));
		return ((uint_t)k);
	}

	return (i_lnode->modid);
}

static int
i_lnode_cmp(void *arg1, void *arg2)
{
	i_lnode_t	*i_lnode1 = (i_lnode_t *)arg1;
	i_lnode_t	*i_lnode2 = (i_lnode_t *)arg2;

	if (i_lnode1->modid != i_lnode2->modid) {
		return ((i_lnode1->modid < i_lnode2->modid) ? -1 : 1);
	}

	if (i_lnode1->di_node != i_lnode2->di_node)
		return ((i_lnode1->di_node < i_lnode2->di_node) ? -1 : 1);

	if (i_lnode1->devt != i_lnode2->devt)
		return ((i_lnode1->devt < i_lnode2->devt) ? -1 : 1);

	return (0);
}

/*
 * An lnode represents a {dip, dev_t} tuple. A link represents a
 * {src_lnode, tgt_lnode, spec_type} tuple.
 * The following callback assumes that LDI framework ref-counts the
 * src_dip and tgt_dip while invoking this callback.
 */
static int
di_ldi_callback(const ldi_usage_t *ldi_usage, void *arg)
{
	struct di_state	*st = (struct di_state *)arg;
	i_lnode_t	*src_lnode, *tgt_lnode, *i_lnode;
	i_link_t	**i_link_next, *i_link;
	di_off_t	soff, toff;
	mod_hash_val_t	nodep = NULL;
	int		res;

	/*
	 * if the source or target of this device usage information doesn't
	 * correspond to a device node then we don't report it via
	 * libdevinfo so return.
	 */
	if ((ldi_usage->src_dip == NULL) || (ldi_usage->tgt_dip == NULL))
		return (LDI_USAGE_CONTINUE);

	ASSERT(e_ddi_devi_holdcnt(ldi_usage->src_dip));
	ASSERT(e_ddi_devi_holdcnt(ldi_usage->tgt_dip));

	/*
	 * Skip the ldi_usage if either src or tgt dip is not in the
	 * snapshot. This saves us from pruning bad lnodes/links later.
	 */
	if (di_dip_find(st, ldi_usage->src_dip, &soff) != 0)
		return (LDI_USAGE_CONTINUE);
	if (di_dip_find(st, ldi_usage->tgt_dip, &toff) != 0)
		return (LDI_USAGE_CONTINUE);

	ASSERT(soff > 0);
	ASSERT(toff > 0);

	/*
	 * allocate an i_lnode and add it to the lnode hash
	 * if it is not already present. For this particular
	 * link the lnode is a source, but it may
	 * participate as tgt or src in any number of layered
	 * operations - so it may already be in the hash.
	 */
	i_lnode = i_lnode_alloc(ldi_usage->src_modid);
	i_lnode->di_node = DI_NODE(di_mem_addr(st, soff));
	i_lnode->devt = ldi_usage->src_devt;

	res = mod_hash_find(st->lnode_hash, i_lnode, &nodep);
	if (res == MH_ERR_NOTFOUND) {
		/*
		 * new i_lnode
		 * add it to the hash and increment the lnode count
		 */
		res = mod_hash_insert(st->lnode_hash, i_lnode, i_lnode);
		ASSERT(res == 0);
		st->lnode_count++;
		src_lnode = i_lnode;
	} else {
		/* this i_lnode already exists in the lnode_hash */
		i_lnode_free(i_lnode);
		src_lnode = (i_lnode_t *)nodep;
	}

	/*
	 * allocate a tgt i_lnode and add it to the lnode hash
	 */
	i_lnode = i_lnode_alloc(ldi_usage->tgt_modid);
	i_lnode->di_node = DI_NODE(di_mem_addr(st, toff));
	i_lnode->devt = ldi_usage->tgt_devt;

	res = mod_hash_find(st->lnode_hash, i_lnode, &nodep);
	if (res == MH_ERR_NOTFOUND) {
		/*
		 * new i_lnode
		 * add it to the hash and increment the lnode count
		 */
		res = mod_hash_insert(st->lnode_hash, i_lnode, i_lnode);
		ASSERT(res == 0);
		st->lnode_count++;
		tgt_lnode = i_lnode;
	} else {
		/* this i_lnode already exists in the lnode_hash */
		i_lnode_free(i_lnode);
		tgt_lnode = (i_lnode_t *)nodep;
	}

	/*
	 * allocate a i_link
	 */
	i_link = i_link_alloc(ldi_usage->tgt_spec_type);
	i_link->src_lnode = src_lnode;
	i_link->tgt_lnode = tgt_lnode;

	/*
	 * add this link onto the src i_lnodes outbound i_link list
	 */
	i_link_next = &(src_lnode->link_out);
	while (*i_link_next != NULL) {
		if ((i_lnode_cmp(tgt_lnode, (*i_link_next)->tgt_lnode) == 0) &&
		    (i_link->spec_type == (*i_link_next)->spec_type)) {
			/* this link already exists */
			kmem_free(i_link, sizeof (i_link_t));
			return (LDI_USAGE_CONTINUE);
		}
		i_link_next = &((*i_link_next)->src_link_next);
	}
	*i_link_next = i_link;

	/*
	 * add this link onto the tgt i_lnodes inbound i_link list
	 */
	i_link_next = &(tgt_lnode->link_in);
	while (*i_link_next != NULL) {
		ASSERT(i_lnode_cmp(src_lnode, (*i_link_next)->src_lnode) != 0);
		i_link_next = &((*i_link_next)->tgt_link_next);
	}
	*i_link_next = i_link;

	/*
	 * add this i_link to the link hash
	 */
	res = mod_hash_insert(st->link_hash, i_link, i_link);
	ASSERT(res == 0);
	st->link_count++;

	return (LDI_USAGE_CONTINUE);
}

struct i_layer_data {
	struct di_state	*st;
	int		lnode_count;
	int		link_count;
	di_off_t	lnode_off;
	di_off_t	link_off;
};

/*ARGSUSED*/
static uint_t
i_link_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
{
	i_link_t		*i_link  = (i_link_t *)key;
	struct i_layer_data	*data = arg;
	struct di_link		*me;
	struct di_lnode		*melnode;
	struct di_node		*medinode;

	ASSERT(i_link->self == 0);

	i_link->self = data->link_off +
	    (data->link_count * sizeof (struct di_link));
	data->link_count++;

	ASSERT(data->link_off > 0 && data->link_count > 0);
	ASSERT(data->lnode_count == data->st->lnode_count); /* lnodes done */
	ASSERT(data->link_count <= data->st->link_count);

	/* fill in fields for the di_link snapshot */
	me = DI_LINK(di_mem_addr(data->st, i_link->self));
	me->self = i_link->self;
	me->spec_type = i_link->spec_type;

	/*
	 * The src_lnode and tgt_lnode i_lnode_t for this i_link_t
	 * are created during the LDI table walk. Since we are
	 * walking the link hash, the lnode hash has already been
	 * walked and the lnodes have been snapshotted. Save lnode
	 * offsets.
	 */
	me->src_lnode = i_link->src_lnode->self;
	me->tgt_lnode = i_link->tgt_lnode->self;

	/*
	 * Save this link's offset in the src_lnode snapshot's link_out
	 * field
	 */
	melnode = DI_LNODE(di_mem_addr(data->st, me->src_lnode));
	me->src_link_next = melnode->link_out;
	melnode->link_out = me->self;

	/*
	 * Put this link on the tgt_lnode's link_in field
	 */
	melnode = DI_LNODE(di_mem_addr(data->st, me->tgt_lnode));
	me->tgt_link_next = melnode->link_in;
	melnode->link_in = me->self;

	/*
	 * An i_lnode_t is only created if the corresponding dip exists
	 * in the snapshot. A pointer to the di_node is saved in the
	 * i_lnode_t when it is allocated. For this link, get the di_node
	 * for the source lnode. Then put the link on the di_node's list
	 * of src links
	 */
	medinode = i_link->src_lnode->di_node;
	me->src_node_next = medinode->src_links;
	medinode->src_links = me->self;

	/*
	 * Put this link on the tgt_links list of the target
	 * dip.
	 */
	medinode = i_link->tgt_lnode->di_node;
	me->tgt_node_next = medinode->tgt_links;
	medinode->tgt_links = me->self;

	return (MH_WALK_CONTINUE);
}

/*ARGSUSED*/
static uint_t
i_lnode_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
{
	i_lnode_t		*i_lnode = (i_lnode_t *)key;
	struct i_layer_data	*data = arg;
	struct di_lnode		*me;
	struct di_node		*medinode;

	ASSERT(i_lnode->self == 0);