usr/src/uts/common/inet/ip/ip_ndp.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) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
 */

/*
 * Copyright (c) 2019, Joyent, Inc.
 */

#include <sys/types.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/strsun.h>
#include <sys/sysmacros.h>
#include <sys/errno.h>
#include <sys/dlpi.h>
#include <sys/socket.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/cmn_err.h>
#include <sys/debug.h>
#include <sys/vtrace.h>
#include <sys/kmem.h>
#include <sys/zone.h>
#include <sys/ethernet.h>
#include <sys/sdt.h>
#include <sys/mac.h>

#include <net/if.h>
#include <net/if_types.h>
#include <net/if_dl.h>
#include <net/route.h>
#include <netinet/in.h>
#include <netinet/ip6.h>
#include <netinet/icmp6.h>

#include <inet/common.h>
#include <inet/mi.h>
#include <inet/mib2.h>
#include <inet/nd.h>
#include <inet/ip.h>
#include <inet/ip_impl.h>
#include <inet/ipclassifier.h>
#include <inet/ip_if.h>
#include <inet/ip_ire.h>
#include <inet/ip_rts.h>
#include <inet/ip6.h>
#include <inet/ip_ndp.h>
#include <inet/sctp_ip.h>
#include <inet/ip_arp.h>
#include <inet/ip2mac_impl.h>

#define	ANNOUNCE_INTERVAL(isv6) \
	(isv6 ? ipst->ips_ip_ndp_unsolicit_interval : \
	ipst->ips_ip_arp_publish_interval)

#define	DEFENSE_INTERVAL(isv6) \
	(isv6 ? ipst->ips_ndp_defend_interval : \
	ipst->ips_arp_defend_interval)

/* Non-tunable probe interval, based on link capabilities */
#define	ILL_PROBE_INTERVAL(ill)	((ill)->ill_note_link ? 150 : 1500)

/*
 * The IPv4 Link Local address space is special; we do extra duplicate checking
 * there, as the entire assignment mechanism rests on random numbers.
 */
#define	IS_IPV4_LL_SPACE(ptr)	(((uchar_t *)ptr)[0] == 169 && \
				((uchar_t *)ptr)[1] == 254)

/*
 * NCE_EXTERNAL_FLAGS_MASK defines the set of ncec_flags that may be passed
 * in to the ncec*add* functions.
 *
 * NCE_F_AUTHORITY means that we ignore any incoming adverts for that
 * mapping (though DAD is performed for the mapping). NCE_F_PUBLISH means
 * that we will respond to requests for the protocol address.
 */
#define	NCE_EXTERNAL_FLAGS_MASK \
	(NCE_F_MYADDR | NCE_F_ISROUTER | NCE_F_NONUD | \
	NCE_F_ANYCAST | NCE_F_UNSOL_ADV | NCE_F_BCAST | NCE_F_MCAST | \
	NCE_F_AUTHORITY | NCE_F_PUBLISH | NCE_F_STATIC)

/*
 * Lock ordering:
 *
 *	ndp_g_lock -> ill_lock -> ncec_lock
 *
 * The ndp_g_lock protects the NCE hash (nce_hash_tbl, NCE_HASH_PTR) and
 * ncec_next.  ncec_lock protects the contents of the NCE (particularly
 * ncec_refcnt).
 */

static	void	nce_cleanup_list(ncec_t *ncec);
static	void	nce_set_ll(ncec_t *ncec, uchar_t *ll_addr);
static	ncec_t	*ncec_lookup_illgrp(ill_t *, const in6_addr_t *,
    ncec_t *);
static	nce_t	*nce_lookup_addr(ill_t *, const in6_addr_t *);
static	int	nce_set_multicast_v6(ill_t *ill, const in6_addr_t *addr,
    uint16_t ncec_flags, nce_t **newnce);
static	int	nce_set_multicast_v4(ill_t *ill, const in_addr_t *dst,
    uint16_t ncec_flags, nce_t **newnce);
static	boolean_t	ndp_xmit(ill_t *ill, uint32_t operation,
    uint8_t *hwaddr, uint_t hwaddr_len, const in6_addr_t *sender,
    const in6_addr_t *target, int flag);
static void	ncec_refhold_locked(ncec_t *);
static boolean_t ill_defend_rate_limit(ill_t *, ncec_t *);
static	void	nce_queue_mp_common(ncec_t *, mblk_t *, boolean_t);
static	int	nce_add_common(ill_t *, uchar_t *, uint_t, const in6_addr_t *,
    uint16_t, uint16_t, nce_t **);
static nce_t *nce_add_impl(ill_t *, ncec_t *, nce_t *, mblk_t *, list_t *);
static nce_t *nce_add(ill_t *, ncec_t *, list_t *);
static void nce_inactive(nce_t *);
extern nce_t	*nce_lookup(ill_t *, const in6_addr_t *);
static nce_t *nce_ill_lookup_then_add(ill_t *, ncec_t *);
static int	nce_add_v6(ill_t *, uchar_t *, uint_t, const in6_addr_t *,
    uint16_t, uint16_t, nce_t **);
static int	nce_add_v4(ill_t *, uchar_t *, uint_t, const in_addr_t *,
    uint16_t, uint16_t, nce_t **);
static int  nce_add_v6_postprocess(nce_t *);
static int  nce_add_v4_postprocess(nce_t *);
static ill_t *nce_resolve_src(ncec_t *, in6_addr_t *);
static clock_t nce_fuzz_interval(clock_t, boolean_t);
static void nce_resolv_ipmp_ok(ncec_t *);
static void nce_walk_common(ill_t *, pfi_t, void *);
static void nce_start_timer(ncec_t *, uint_t);
static nce_t *nce_fastpath_create(ill_t *, ncec_t *);
static void nce_fastpath_trigger(nce_t *);
static nce_t *nce_fastpath(ncec_t *, boolean_t, nce_t *);

#ifdef DEBUG
static void	ncec_trace_cleanup(const ncec_t *);
#endif

#define	NCE_HASH_PTR_V4(ipst, addr)					\
	(&((ipst)->ips_ndp4->nce_hash_tbl[IRE_ADDR_HASH(addr, NCE_TABLE_SIZE)]))

#define	NCE_HASH_PTR_V6(ipst, addr)				 \
	(&((ipst)->ips_ndp6->nce_hash_tbl[NCE_ADDR_HASH_V6(addr, \
		NCE_TABLE_SIZE)]))

extern kmem_cache_t *ncec_cache;
extern kmem_cache_t *nce_cache;

/*
 * Send out a IPv6 (unicast) or IPv4 (broadcast) DAD probe
 * If src_ill is not null, the ncec_addr is bound to src_ill. The
 * src_ill is ignored by nce_dad for IPv4 Neighbor Cache entries where
 * the probe is sent on the ncec_ill (in the non-IPMP case) or the
 * IPMP cast_ill (in the IPMP case).
 *
 * Note that the probe interval is based on the src_ill for IPv6, and
 * the ncec_xmit_interval for IPv4.
 */
static void
nce_dad(ncec_t *ncec, ill_t *src_ill, boolean_t send_probe)
{
	boolean_t dropped;
	uint32_t probe_interval;

	ASSERT(!(ncec->ncec_flags & NCE_F_MCAST));
	ASSERT(!(ncec->ncec_flags & NCE_F_BCAST));
	if (ncec->ncec_ipversion == IPV6_VERSION) {
		dropped = ndp_xmit(src_ill, ND_NEIGHBOR_SOLICIT,
		    ncec->ncec_lladdr, ncec->ncec_lladdr_length,
		    &ipv6_all_zeros, &ncec->ncec_addr, NDP_PROBE);
		probe_interval = ILL_PROBE_INTERVAL(src_ill);
	} else {
		/* IPv4 DAD delay the initial probe. */
		if (send_probe)
			dropped = arp_probe(ncec);
		else
			dropped = B_TRUE;
		probe_interval = nce_fuzz_interval(ncec->ncec_xmit_interval,
		    !send_probe);
	}
	if (!dropped) {
		mutex_enter(&ncec->ncec_lock);
		ncec->ncec_pcnt--;
		mutex_exit(&ncec->ncec_lock);
	}
	nce_restart_timer(ncec, probe_interval);
}

/*
 * Compute default flags to use for an advertisement of this ncec's address.
 */
static int
nce_advert_flags(const ncec_t *ncec)
{
	int flag = 0;

	if (ncec->ncec_flags & NCE_F_ISROUTER)
		flag |= NDP_ISROUTER;
	if (!(ncec->ncec_flags & NCE_F_ANYCAST))
		flag |= NDP_ORIDE;

	return (flag);
}

/*
 * NDP Cache Entry creation routine.
 * This routine must always be called with ndp6->ndp_g_lock held.
 */
int
nce_add_v6(ill_t *ill, uchar_t *hw_addr, uint_t hw_addr_len,
    const in6_addr_t *addr, uint16_t flags, uint16_t state, nce_t **newnce)
{
	int		err;
	nce_t		*nce;

	ASSERT(MUTEX_HELD(&ill->ill_ipst->ips_ndp6->ndp_g_lock));
	ASSERT(ill != NULL && ill->ill_isv6);

	err = nce_add_common(ill, hw_addr, hw_addr_len, addr, flags, state,
	    &nce);
	if (err != 0)
		return (err);
	ASSERT(newnce != NULL);
	*newnce = nce;
	return (err);
}

/*
 * Post-processing routine to be executed after nce_add_v6(). This function
 * triggers fastpath (if appropriate) and DAD on the newly added nce entry
 * and must be called without any locks held.
 */
int
nce_add_v6_postprocess(nce_t *nce)
{
	ncec_t		*ncec = nce->nce_common;
	boolean_t	dropped = B_FALSE;
	uchar_t		*hw_addr = ncec->ncec_lladdr;
	uint_t		hw_addr_len = ncec->ncec_lladdr_length;
	ill_t		*ill = ncec->ncec_ill;
	int		err = 0;
	uint16_t	flags = ncec->ncec_flags;
	ip_stack_t	*ipst = ill->ill_ipst;
	boolean_t	trigger_fastpath = B_TRUE;

	/*
	 * If the hw_addr is NULL, typically for ND_INCOMPLETE nces, then
	 * we call nce_fastpath as soon as the ncec is resolved in nce_process.
	 * We call nce_fastpath from nce_update if the link layer address of
	 * the peer changes from nce_update
	 */
	if (NCE_PUBLISH(ncec) || !NCE_ISREACHABLE(ncec) ||
	    (hw_addr == NULL && ill->ill_net_type != IRE_IF_NORESOLVER))
		trigger_fastpath = B_FALSE;

	if (trigger_fastpath)
		nce_fastpath_trigger(nce);
	if (NCE_PUBLISH(ncec) && ncec->ncec_state == ND_PROBE) {
		ill_t *hwaddr_ill;
		/*
		 * Unicast entry that needs DAD.
		 */
		if (IS_IPMP(ill)) {
			hwaddr_ill = ipmp_illgrp_find_ill(ill->ill_grp,
			    hw_addr, hw_addr_len);
		} else {
			hwaddr_ill = ill;
		}
		nce_dad(ncec, hwaddr_ill, B_TRUE);
		err = EINPROGRESS;
	} else if (flags & NCE_F_UNSOL_ADV) {
		/*
		 * We account for the transmit below by assigning one
		 * less than the ndd variable. Subsequent decrements
		 * are done in nce_timer.
		 */
		mutex_enter(&ncec->ncec_lock);
		ncec->ncec_unsolicit_count =
		    ipst->ips_ip_ndp_unsolicit_count - 1;
		mutex_exit(&ncec->ncec_lock);
		dropped = ndp_xmit(ill,
		    ND_NEIGHBOR_ADVERT,
		    hw_addr,
		    hw_addr_len,
		    &ncec->ncec_addr,	/* Source and target of the adv */
		    &ipv6_all_hosts_mcast, /* Destination of the packet */
		    nce_advert_flags(ncec));
		mutex_enter(&ncec->ncec_lock);
		if (dropped)
			ncec->ncec_unsolicit_count++;
		else
			ncec->ncec_last_time_defended = ddi_get_lbolt();
		if (ncec->ncec_unsolicit_count != 0) {
			nce_start_timer(ncec,
			    ipst->ips_ip_ndp_unsolicit_interval);
		}
		mutex_exit(&ncec->ncec_lock);
	}
	return (err);
}

/*
 * Atomically lookup and add (if needed) Neighbor Cache information for
 * an address.
 *
 * IPMP notes: the ncec for non-local (i.e., !NCE_MYADDR(ncec) addresses
 * are always added pointing at the ipmp_ill. Thus, when the ill passed
 * to nce_add_v6 is an under_ill (i.e., IS_UNDER_IPMP(ill)) two nce_t
 * entries will be created, both pointing at the same ncec_t. The nce_t
 * entries will have their nce_ill set to the ipmp_ill and the under_ill
 * respectively, with the ncec_t having its ncec_ill pointing at the ipmp_ill.
 * Local addresses are always created on the ill passed to nce_add_v6.
 */
int
nce_lookup_then_add_v6(ill_t *ill, uchar_t *hw_addr, uint_t hw_addr_len,
    const in6_addr_t *addr, uint16_t flags, uint16_t state, nce_t **newnce)
{
	int		err = 0;
	ip_stack_t	*ipst = ill->ill_ipst;
	nce_t		*nce, *upper_nce = NULL;
	ill_t		*in_ill = ill;
	boolean_t	need_ill_refrele = B_FALSE;

	if (flags & NCE_F_MCAST) {
		/*
		 * hw_addr will be figured out in nce_set_multicast_v6;
		 * caller has to select the cast_ill
		 */
		ASSERT(hw_addr == NULL);
		ASSERT(!IS_IPMP(ill));
		err = nce_set_multicast_v6(ill, addr, flags, newnce);
		return (err);
	}
	ASSERT(ill->ill_isv6);
	if (IS_UNDER_IPMP(ill) && !(flags & NCE_F_MYADDR)) {
		ill = ipmp_ill_hold_ipmp_ill(ill);
		if (ill == NULL)
			return (ENXIO);
		need_ill_refrele = B_TRUE;
	}

	mutex_enter(&ipst->ips_ndp6->ndp_g_lock);
	nce = nce_lookup_addr(ill, addr);
	if (nce == NULL) {
		err = nce_add_v6(ill, hw_addr, hw_addr_len, addr, flags, state,
		    &nce);
	} else {
		err = EEXIST;
	}
	mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
	if (err == 0)
		err = nce_add_v6_postprocess(nce);
	if (in_ill != ill && nce != NULL) {
		nce_t *under_nce = NULL;

		/*
		 * in_ill was the under_ill. Try to create the under_nce.
		 * Hold the ill_g_lock to prevent changes to group membership
		 * until we are done.
		 */
		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
		if (!IS_IN_SAME_ILLGRP(in_ill, ill)) {
			DTRACE_PROBE2(ill__not__in__group, nce_t *, nce,
			    ill_t *, ill);
			rw_exit(&ipst->ips_ill_g_lock);
			err = ENXIO;
			nce_refrele(nce);
			nce = NULL;
			goto bail;
		}
		under_nce = nce_fastpath_create(in_ill, nce->nce_common);
		if (under_nce == NULL) {
			rw_exit(&ipst->ips_ill_g_lock);
			err = EINVAL;
			nce_refrele(nce);
			nce = NULL;
			goto bail;
		}
		rw_exit(&ipst->ips_ill_g_lock);
		upper_nce = nce;
		nce = under_nce; /* will be returned to caller */
		if (NCE_ISREACHABLE(nce->nce_common))
			nce_fastpath_trigger(under_nce);
	}
	/* nce_refrele is deferred until the lock is dropped  */
	if (nce != NULL) {
		if (newnce != NULL)
			*newnce = nce;
		else
			nce_refrele(nce);
	}
bail:
	if (upper_nce != NULL)
		nce_refrele(upper_nce);
	if (need_ill_refrele)
		ill_refrele(ill);
	return (err);
}

/*
 * Remove all the CONDEMNED nces from the appropriate hash table.
 * We create a private list of NCEs, these may have ires pointing
 * to them, so the list will be passed through to clean up dependent
 * ires and only then we can do ncec_refrele() which can make NCE inactive.
 */
static void
nce_remove(ndp_g_t *ndp, ncec_t *ncec, ncec_t **free_nce_list)
{
	ncec_t *ncec1;
	ncec_t **ptpn;

	ASSERT(MUTEX_HELD(&ndp->ndp_g_lock));
	ASSERT(ndp->ndp_g_walker == 0);
	for (; ncec; ncec = ncec1) {
		ncec1 = ncec->ncec_next;
		mutex_enter(&ncec->ncec_lock);
		if (NCE_ISCONDEMNED(ncec)) {
			ptpn = ncec->ncec_ptpn;
			ncec1 = ncec->ncec_next;
			if (ncec1 != NULL)
				ncec1->ncec_ptpn = ptpn;
			*ptpn = ncec1;
			ncec->ncec_ptpn = NULL;
			ncec->ncec_next = NULL;
			ncec->ncec_next = *free_nce_list;
			*free_nce_list = ncec;
		}
		mutex_exit(&ncec->ncec_lock);
	}
}

/*
 * 1. Mark the entry CONDEMNED. This ensures that no new nce_lookup()
 *    will return this NCE. Also no new timeouts will
 *    be started (See nce_restart_timer).
 * 2. Cancel any currently running timeouts.
 * 3. If there is an ndp walker, return. The walker will do the cleanup.
 *    This ensures that walkers see a consistent list of NCEs while walking.
 * 4. Otherwise remove the NCE from the list of NCEs
 */
void
ncec_delete(ncec_t *ncec)
{
	ncec_t	**ptpn;
	ncec_t	*ncec1;
	int	ipversion = ncec->ncec_ipversion;
	ndp_g_t *ndp;
	ip_stack_t	*ipst = ncec->ncec_ipst;

	if (ipversion == IPV4_VERSION)
		ndp = ipst->ips_ndp4;
	else
		ndp = ipst->ips_ndp6;

	/* Serialize deletes */
	mutex_enter(&ncec->ncec_lock);
	if (NCE_ISCONDEMNED(ncec)) {
		/* Some other thread is doing the delete */
		mutex_exit(&ncec->ncec_lock);
		return;
	}
	/*
	 * Caller has a refhold. Also 1 ref for being in the list. Thus
	 * refcnt has to be >= 2
	 */
	ASSERT(ncec->ncec_refcnt >= 2);
	ncec->ncec_flags |= NCE_F_CONDEMNED;
	mutex_exit(&ncec->ncec_lock);

	/* Count how many condemned ires for kmem_cache callback */
	atomic_inc_32(&ipst->ips_num_nce_condemned);
	nce_fastpath_list_delete(ncec->ncec_ill, ncec, NULL);

	/* Complete any waiting callbacks */
	ncec_cb_dispatch(ncec);

	/*
	 * Cancel any running timer. Timeout can't be restarted
	 * since CONDEMNED is set. Can't hold ncec_lock across untimeout.
	 * Passing invalid timeout id is fine.
	 */
	if (ncec->ncec_timeout_id != 0) {
		(void) untimeout(ncec->ncec_timeout_id);
		ncec->ncec_timeout_id = 0;
	}

	mutex_enter(&ndp->ndp_g_lock);
	if (ncec->ncec_ptpn == NULL) {
		/*
		 * The last ndp walker has already removed this ncec from
		 * the list after we marked the ncec CONDEMNED and before
		 * we grabbed the global lock.
		 */
		mutex_exit(&ndp->ndp_g_lock);
		return;
	}
	if (ndp->ndp_g_walker > 0) {
		/*
		 * Can't unlink. The walker will clean up
		 */
		ndp->ndp_g_walker_cleanup = B_TRUE;
		mutex_exit(&ndp->ndp_g_lock);
		return;
	}

	/*
	 * Now remove the ncec from the list. nce_restart_timer won't restart
	 * the timer since it is marked CONDEMNED.
	 */
	ptpn = ncec->ncec_ptpn;
	ncec1 = ncec->ncec_next;
	if (ncec1 != NULL)
		ncec1->ncec_ptpn = ptpn;
	*ptpn = ncec1;
	ncec->ncec_ptpn = NULL;
	ncec->ncec_next = NULL;
	mutex_exit(&ndp->ndp_g_lock);

	/* Removed from ncec_ptpn/ncec_next list */
	ncec_refrele_notr(ncec);
}

void
ncec_inactive(ncec_t *ncec)
{
	mblk_t		**mpp;
	ill_t		*ill = ncec->ncec_ill;
	ip_stack_t	*ipst = ncec->ncec_ipst;

	ASSERT(ncec->ncec_refcnt == 0);
	ASSERT(MUTEX_HELD(&ncec->ncec_lock));

	/* Count how many condemned nces for kmem_cache callback */
	if (NCE_ISCONDEMNED(ncec))
		atomic_add_32(&ipst->ips_num_nce_condemned, -1);

	/* Free all allocated messages */
	mpp = &ncec->ncec_qd_mp;
	while (*mpp != NULL) {
		mblk_t  *mp;

		mp = *mpp;
		*mpp = mp->b_next;

		inet_freemsg(mp);
	}
	/*
	 * must have been cleaned up in ncec_delete
	 */
	ASSERT(list_is_empty(&ncec->ncec_cb));
	list_destroy(&ncec->ncec_cb);
	/*
	 * free the ncec_lladdr if one was allocated in nce_add_common()
	 */
	if (ncec->ncec_lladdr_length > 0)
		kmem_free(ncec->ncec_lladdr, ncec->ncec_lladdr_length);

#ifdef DEBUG
	ncec_trace_cleanup(ncec);
#endif

	mutex_enter(&ill->ill_lock);
	DTRACE_PROBE3(ill__decr__cnt, (ill_t *), ill,
	    (char *), "ncec", (void *), ncec);
	ill->ill_ncec_cnt--;
	ncec->ncec_ill = NULL;
	/*
	 * If the number of ncec's associated with this ill have dropped
	 * to zero, check whether we need to restart any operation that
	 * is waiting for this to happen.
	 */
	if (ILL_DOWN_OK(ill)) {
		/* ipif_ill_refrele_tail drops the ill_lock */
		ipif_ill_refrele_tail(ill);
	} else {
		mutex_exit(&ill->ill_lock);
	}

	mutex_destroy(&ncec->ncec_lock);
	kmem_cache_free(ncec_cache, ncec);
}

/*
 * ncec_walk routine.  Delete the ncec if it is associated with the ill
 * that is going away.  Always called as a writer.
 */
void
ncec_delete_per_ill(ncec_t *ncec, void *arg)
{
	if ((ncec != NULL) && ncec->ncec_ill == arg) {
		ncec_delete(ncec);
	}
}

/*
 * Neighbor Cache cleanup logic for a list of ncec_t entries.
 */
static void
nce_cleanup_list(ncec_t *ncec)
{
	ncec_t *ncec_next;

	ASSERT(ncec != NULL);
	while (ncec != NULL) {
		ncec_next = ncec->ncec_next;
		ncec->ncec_next = NULL;

		/*
		 * It is possible for the last ndp walker (this thread)
		 * to come here after ncec_delete has marked the ncec CONDEMNED
		 * and before it has removed the ncec from the fastpath list
		 * or called untimeout. So we need to do it here. It is safe
		 * for both ncec_delete and this thread to do it twice or
		 * even simultaneously since each of the threads has a
		 * reference on the ncec.
		 */
		nce_fastpath_list_delete(ncec->ncec_ill, ncec, NULL);
		/*
		 * Cancel any running timer. Timeout can't be restarted
		 * since CONDEMNED is set. The ncec_lock can't be
		 * held across untimeout though passing invalid timeout
		 * id is fine.
		 */
		if (ncec->ncec_timeout_id != 0) {
			(void) untimeout(ncec->ncec_timeout_id);
			ncec->ncec_timeout_id = 0;
		}
		/* Removed from ncec_ptpn/ncec_next list */
		ncec_refrele_notr(ncec);
		ncec = ncec_next;
	}
}

/*
 * Restart DAD on given NCE.  Returns B_TRUE if DAD has been restarted.
 */
boolean_t
nce_restart_dad(ncec_t *ncec)
{
	boolean_t started;
	ill_t *ill, *hwaddr_ill;

	if (ncec == NULL)
		return (B_FALSE);
	ill = ncec->ncec_ill;
	mutex_enter(&ncec->ncec_lock);
	if (ncec->ncec_state == ND_PROBE) {
		mutex_exit(&ncec->ncec_lock);
		started = B_TRUE;
	} else if (ncec->ncec_state == ND_REACHABLE) {
		ASSERT(ncec->ncec_lladdr != NULL);
		ncec->ncec_state = ND_PROBE;
		ncec->ncec_pcnt = ND_MAX_UNICAST_SOLICIT;
		/*
		 * Slight cheat here: we don't use the initial probe delay
		 * for IPv4 in this obscure case.
		 */
		mutex_exit(&ncec->ncec_lock);
		if (IS_IPMP(ill)) {
			hwaddr_ill = ipmp_illgrp_find_ill(ill->ill_grp,
			    ncec->ncec_lladdr, ncec->ncec_lladdr_length);
		} else {
			hwaddr_ill = ill;
		}
		nce_dad(ncec, hwaddr_ill, B_TRUE);
		started = B_TRUE;
	} else {
		mutex_exit(&ncec->ncec_lock);
		started = B_FALSE;
	}
	return (started);
}

/*
 * IPv6 Cache entry lookup.  Try to find an ncec matching the parameters passed.
 * If one is found, the refcnt on the ncec will be incremented.
 */
ncec_t *
ncec_lookup_illgrp_v6(ill_t *ill, const in6_addr_t *addr)
{
	ncec_t		*ncec;
	ip_stack_t	*ipst = ill->ill_ipst;

	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
	mutex_enter(&ipst->ips_ndp6->ndp_g_lock);

	/* Get head of v6 hash table */
	ncec = *((ncec_t **)NCE_HASH_PTR_V6(ipst, *addr));
	ncec = ncec_lookup_illgrp(ill, addr, ncec);
	mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
	rw_exit(&ipst->ips_ill_g_lock);
	return (ncec);
}
/*
 * IPv4 Cache entry lookup.  Try to find an ncec matching the parameters passed.
 * If one is found, the refcnt on the ncec will be incremented.
 */
ncec_t *
ncec_lookup_illgrp_v4(ill_t *ill, const in_addr_t *addr)
{
	ncec_t	*ncec = NULL;
	in6_addr_t addr6;
	ip_stack_t *ipst = ill->ill_ipst;

	rw_enter(&ipst->ips_ill_g_lock, RW_READER);
	mutex_enter(&ipst->ips_ndp4->ndp_g_lock);

	/* Get head of v4 hash table */
	ncec = *((ncec_t **)NCE_HASH_PTR_V4(ipst, *addr));
	IN6_IPADDR_TO_V4MAPPED(*addr, &addr6);
	ncec = ncec_lookup_illgrp(ill, &addr6, ncec);
	mutex_exit(&ipst->ips_ndp4->ndp_g_lock);
	rw_exit(&ipst->ips_ill_g_lock);
	return (ncec);
}

/*
 * Cache entry lookup.  Try to find an ncec matching the parameters passed.
 * If an ncec is found, increment the hold count on that ncec.
 * The caller passes in the start of the appropriate hash table, and must
 * be holding the appropriate global lock (ndp_g_lock). In addition, since
 * this function matches ncec_t entries across the illgrp, the ips_ill_g_lock
 * must be held as reader.
 *
 * This function always matches across the ipmp group.
 */
ncec_t *
ncec_lookup_illgrp(ill_t *ill, const in6_addr_t *addr, ncec_t *ncec)
{
	ndp_g_t		*ndp;
	ip_stack_t	*ipst = ill->ill_ipst;

	if (ill->ill_isv6)
		ndp = ipst->ips_ndp6;
	else
		ndp = ipst->ips_ndp4;

	ASSERT(ill != NULL);
	ASSERT(MUTEX_HELD(&ndp->ndp_g_lock));
	if (IN6_IS_ADDR_UNSPECIFIED(addr))
		return (NULL);
	for (; ncec != NULL; ncec = ncec->ncec_next) {
		if (ncec->ncec_ill == ill ||
		    IS_IN_SAME_ILLGRP(ill, ncec->ncec_ill)) {
			if (IN6_ARE_ADDR_EQUAL(&ncec->ncec_addr, addr)) {
				mutex_enter(&ncec->ncec_lock);
				if (!NCE_ISCONDEMNED(ncec)) {
					ncec_refhold_locked(ncec);
					mutex_exit(&ncec->ncec_lock);
					break;
				}
				mutex_exit(&ncec->ncec_lock);
			}
		}
	}
	return (ncec);
}

/*
 * Find an nce_t on ill with nce_addr == addr. Lookup the nce_t
 * entries for ill only, i.e., when ill is part of an ipmp group,
 * nce_lookup_v4 will never try to match across the group.
 */
nce_t *
nce_lookup_v4(ill_t *ill, const in_addr_t *addr)
{
	nce_t *nce;
	in6_addr_t addr6;
	ip_stack_t *ipst = ill->ill_ipst;

	mutex_enter(&ipst->ips_ndp4->ndp_g_lock);
	IN6_IPADDR_TO_V4MAPPED(*addr, &addr6);
	nce = nce_lookup_addr(ill, &addr6);
	mutex_exit(&ipst->ips_ndp4->ndp_g_lock);
	return (nce);
}

/*
 * Find an nce_t on ill with nce_addr == addr. Lookup the nce_t
 * entries for ill only, i.e., when ill is part of an ipmp group,
 * nce_lookup_v6 will never try to match across the group.
 */
nce_t *
nce_lookup_v6(ill_t *ill, const in6_addr_t *addr6)
{
	nce_t *nce;
	ip_stack_t *ipst = ill->ill_ipst;

	mutex_enter(&ipst->ips_ndp6->ndp_g_lock);
	nce = nce_lookup_addr(ill, addr6);
	mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
	return (nce);
}

static nce_t *
nce_lookup_addr(ill_t *ill, const in6_addr_t *addr)
{
	nce_t *nce;

	ASSERT(ill != NULL);
#ifdef DEBUG
	if (ill->ill_isv6)
		ASSERT(MUTEX_HELD(&ill->ill_ipst->ips_ndp6->ndp_g_lock));
	else
		ASSERT(MUTEX_HELD(&ill->ill_ipst->ips_ndp4->ndp_g_lock));
#endif
	mutex_enter(&ill->ill_lock);
	nce = nce_lookup(ill, addr);
	mutex_exit(&ill->ill_lock);
	return (nce);
}


/*
 * Router turned to host.  We need to make sure that cached copies of the ncec
 * are not used for forwarding packets if they were derived from the default
 * route, and that the default route itself is removed, as  required by
 * section 7.2.5 of RFC 2461.
 *
 * Note that the ncec itself probably has valid link-layer information for the
 * nexthop, so that there is no reason to delete the ncec, as long as the
 * ISROUTER flag is turned off.
 */
static void
ncec_router_to_host(ncec_t *ncec)
{
	ire_t		*ire;
	ip_stack_t	*ipst = ncec->ncec_ipst;

	mutex_enter(&ncec->ncec_lock);
	ncec->ncec_flags &= ~NCE_F_ISROUTER;
	mutex_exit(&ncec->ncec_lock);

	ire = ire_ftable_lookup_v6(&ipv6_all_zeros, &ipv6_all_zeros,
	    &ncec->ncec_addr, IRE_DEFAULT, ncec->ncec_ill, ALL_ZONES, NULL,
	    MATCH_IRE_ILL | MATCH_IRE_TYPE | MATCH_IRE_GW, 0, ipst, NULL);
	if (ire != NULL) {
		ip_rts_rtmsg(RTM_DELETE, ire, 0, ipst);
		ire_delete(ire);
		ire_refrele(ire);
	}
}

/*
 * Process passed in parameters either from an incoming packet or via
 * user ioctl.
 */
void
nce_process(ncec_t *ncec, uchar_t *hw_addr, uint32_t flag, boolean_t is_adv)
{
	ill_t	*ill = ncec->ncec_ill;
	uint32_t hw_addr_len = ill->ill_phys_addr_length;
	boolean_t ll_updated = B_FALSE;
	boolean_t ll_changed;
	nce_t	*nce;

	ASSERT(ncec->ncec_ipversion == IPV6_VERSION);
	/*
	 * No updates of link layer address or the neighbor state is
	 * allowed, when the cache is in NONUD state.  This still
	 * allows for responding to reachability solicitation.
	 */
	mutex_enter(&ncec->ncec_lock);
	if (ncec->ncec_state == ND_INCOMPLETE) {
		if (hw_addr == NULL) {
			mutex_exit(&ncec->ncec_lock);
			return;
		}
		nce_set_ll(ncec, hw_addr);
		/*
		 * Update ncec state and send the queued packets
		 * back to ip this time ire will be added.
		 */
		if (flag & ND_NA_FLAG_SOLICITED) {
			nce_update(ncec, ND_REACHABLE, NULL);
		} else {
			nce_update(ncec, ND_STALE, NULL);
		}
		mutex_exit(&ncec->ncec_lock);
		nce = nce_fastpath(ncec, B_TRUE, NULL);
		nce_resolv_ok(ncec);
		if (nce != NULL)
			nce_refrele(nce);
		return;
	}
	ll_changed = nce_cmp_ll_addr(ncec, hw_addr, hw_addr_len);
	if (!is_adv) {
		/* If this is a SOLICITATION request only */
		if (ll_changed)
			nce_update(ncec, ND_STALE, hw_addr);
		mutex_exit(&ncec->ncec_lock);
		ncec_cb_dispatch(ncec);
		return;
	}
	if (!(flag & ND_NA_FLAG_OVERRIDE) && ll_changed) {
		/* If in any other state than REACHABLE, ignore */
		if (ncec->ncec_state == ND_REACHABLE) {
			nce_update(ncec, ND_STALE, NULL);
		}
		mutex_exit(&ncec->ncec_lock);
		ncec_cb_dispatch(ncec);
		return;
	} else {
		if (ll_changed) {
			nce_update(ncec, ND_UNCHANGED, hw_addr);
			ll_updated = B_TRUE;
		}
		if (flag & ND_NA_FLAG_SOLICITED) {
			nce_update(ncec, ND_REACHABLE, NULL);
		} else {
			if (ll_updated) {
				nce_update(ncec, ND_STALE, NULL);
			}
		}
		mutex_exit(&ncec->ncec_lock);
		if (!(flag & ND_NA_FLAG_ROUTER) && (ncec->ncec_flags &
		    NCE_F_ISROUTER)) {
			ncec_router_to_host(ncec);
		} else {
			ncec_cb_dispatch(ncec);
		}
	}
}

/*
 * Pass arg1 to the cbf supplied, along with each ncec in existence.
 * ncec_walk() places a REFHOLD on the ncec and drops the lock when
 * walking the hash list.
 */
void
ncec_walk_common(ndp_g_t *ndp, ill_t *ill, ncec_walk_cb_t cbf,
    void *arg1, boolean_t trace)
{
	ncec_t	*ncec;
	ncec_t	*ncec1;
	ncec_t	**ncep;
	ncec_t	*free_nce_list = NULL;

	mutex_enter(&ndp->ndp_g_lock);
	/* Prevent ncec_delete from unlink and free of NCE */
	ndp->ndp_g_walker++;
	mutex_exit(&ndp->ndp_g_lock);
	for (ncep = ndp->nce_hash_tbl;
	    ncep < A_END(ndp->nce_hash_tbl); ncep++) {
		for (ncec = *ncep; ncec != NULL; ncec = ncec1) {
			ncec1 = ncec->ncec_next;
			if (ill == NULL || ncec->ncec_ill == ill) {
				if (trace) {
					ncec_refhold(ncec);
					(*cbf)(ncec, arg1);
					ncec_refrele(ncec);
				} else {
					ncec_refhold_notr(ncec);
					(*cbf)(ncec, arg1);
					ncec_refrele_notr(ncec);
				}
			}
		}
	}
	mutex_enter(&ndp->ndp_g_lock);
	ndp->ndp_g_walker--;
	if (ndp->ndp_g_walker_cleanup && ndp->ndp_g_walker == 0) {
		/* Time to delete condemned entries */
		for (ncep = ndp->nce_hash_tbl;
		    ncep < A_END(ndp->nce_hash_tbl); ncep++) {
			ncec = *ncep;
			if (ncec != NULL) {
				nce_remove(ndp, ncec, &free_nce_list);
			}
		}
		ndp->ndp_g_walker_cleanup = B_FALSE;
	}

	mutex_exit(&ndp->ndp_g_lock);

	if (free_nce_list != NULL) {
		nce_cleanup_list(free_nce_list);
	}
}

/*
 * Walk everything.
 * Note that ill can be NULL hence can't derive the ipst from it.
 */
void
ncec_walk(ill_t *ill, ncec_walk_cb_t cbf, void *arg1, ip_stack_t *ipst)
{
	ncec_walk_common(ipst->ips_ndp4, ill, cbf, arg1, B_TRUE);
	ncec_walk_common(ipst->ips_ndp6, ill, cbf, arg1, B_TRUE);
}

/*
 * Cheesy globals (i.e. all netstacks) for both a limit on per-ill multicast
 * NCEs, and the number to reclaim if we hit the limit.  Used by
 * nce_set_multicast_v[46]() to limit the linked-list length of ill_nce. Until
 * we solve the multicast-mappings-shouldn't-be-NCEs problem, use this.
 */

/* Maximum number of multicast NCEs on an ill. */
uint_t ip_max_ill_mcast_nces = 16384;
/*
 * Number of NCEs to delete if we hit the maximum above.  0 means *don't* and
 * return an error.  Non-zero means delete so many, and if the number is >=
 * the max above, that means delete them all.
 */
uint_t ip_ill_mcast_reclaim = 256;

/*
 * Encapsulate multicast ill capping in a function, for easier DTrace
 * detections.  Return a list of refheld NCEs to destroy-via-refrele.  That
 * list can be NULL, but can only be non-NULL if we successfully reclaimed.
 *
 * NOTE:  This function must be called while holding the ill_lock AND
 * JUST PRIOR to making the insertion into the ill_nce list.
 *
 * We can't release the ones we delete ourselves because the ill_lock is held
 * by the caller. They are, instead, passed back in a list_t for deletion
 * outside of the ill_lock hold. nce_graveyard_free() actually frees them.
 *
 * While this covers nce_t, ncec_t gets done even further down the road.  See
 * nce_graveyard_free() for why.
 */
static boolean_t
nce_too_many_mcast(ill_t *ill, list_t *graveyard)
{
	uint_t reclaim_count, max_count, reclaimed = 0;
	boolean_t too_many;
	nce_t *nce, *deadman;

	ASSERT(graveyard != NULL);
	ASSERT(list_is_empty(graveyard));
	ASSERT(MUTEX_HELD(&ill->ill_lock));

	/*
	 * NOTE: Some grinning weirdo may have lowered the global max beyond
	 * what this ill currently has.  The behavior in this case will be
	 * trim-back just by the reclaim amount for any new ones.
	 */
	max_count = ip_max_ill_mcast_nces;
	reclaim_count = min(ip_ill_mcast_reclaim, max_count);

	/* All good? */
	if (ill->ill_mcast_nces < max_count)
		return (B_FALSE);	/* Yes, all good. */

	if (reclaim_count == 0)
		return (B_TRUE);	/* Don't bother - we're stuck. */

	/* We need to reclaim now.  Exploit our held ill_lock. */

	/*
	 * Start at the tail and work backwards, new nces are head-inserted,
	 * so we'll be reaping the oldest entries.
	 */
	nce = list_tail(&ill->ill_nce);
	while (reclaimed < reclaim_count) {
		/* Skip ahead to a multicast NCE. */
		while (nce != NULL &&
		    (nce->nce_common->ncec_flags & NCE_F_MCAST) == 0) {
			nce = list_prev(&ill->ill_nce, nce);
		}
		if (nce == NULL)
			break;

		/*
		 * NOTE: For now, we just delete the first one(s) we find.
		 * This is not optimal, and may require some inspection of nce
		 * & its ncec to be better.
		 */
		deadman = nce;
		nce = list_prev(&ill->ill_nce, nce);

		/* nce_delete() requires caller holds... */
		nce_refhold(deadman);
		nce_delete(deadman);	/* Bumps down ill_mcast_nces. */

		/* Link the dead ones singly, still refheld... */
		list_insert_tail(graveyard, deadman);
		reclaimed++;
	}

	if (reclaimed != reclaim_count) {
		/* We didn't have enough to reach reclaim_count. Why?!? */
		DTRACE_PROBE3(ill__mcast__nce__reclaim__mismatch, ill_t *, ill,
		    uint_t, reclaimed, uint_t, reclaim_count);

		/* In case for some REALLY weird reason we found none! */
		too_many = (reclaimed == 0);
	} else {
		too_many = B_FALSE;
	}

	return (too_many);
}

static void
ncec_mcast_reap_one(ncec_t *ncec, void *arg)
{
	boolean_t reapit;
	ill_t *ill = (ill_t *)arg;

	/* Obvious no-lock-needed checks... */
	if (ncec == NULL || ncec->ncec_ill != ill ||
	    (ncec->ncec_flags & NCE_F_MCAST) == 0)
		return;

	mutex_enter(&ncec->ncec_lock);
	/*
	 * It's refheld by the walk infrastructure. It has one reference for
	 * being in the ndp_g_hash, and if an nce_t exists, that's one more.
	 * We want ones without an nce_t, so 2 is the magic number.  If it's
	 * LESS than 2, we have much bigger problems anyway.
	 */
	ASSERT(ncec->ncec_refcnt >= 2);
	reapit = (ncec->ncec_refcnt == 2);
	mutex_exit(&ncec->ncec_lock);

	if (reapit) {
		IP_STAT(ill->ill_ipst, ip_nce_mcast_reclaim_deleted);
		ncec_delete(ncec);
	}
}

/*
 * Attempt to reap stray multicast ncec_t structures left in the wake of
 * nce_graveyard_free(). This is a taskq servicing routine, as it's well
 * outside any netstack-global locks being held - ndp_g_lock in this case.  We
 * have a reference hold on the ill, which will prevent any unplumbing races.
 */
static void
ncec_mcast_reap(void *arg)
{
	ill_t *ill = (ill_t *)arg;

	IP_STAT(ill->ill_ipst, ip_nce_mcast_reclaim_calls);
	ncec_walk(ill, ncec_mcast_reap_one, ill, ill->ill_ipst);
	mutex_enter(&ill->ill_lock);
	ill->ill_mcast_ncec_cleanup = B_FALSE;
	/*
	 * Inline a _notr() version of ill_refrele. See nce_graveyard_free()
	 * below for why.
	 */
	ill->ill_refcnt--;
	if (ill->ill_refcnt == 0)
		ipif_ill_refrele_tail(ill);	/* Drops ill_lock. */
	else
		mutex_exit(&ill->ill_lock);
}

/*
 * Free a list (including handling an empty list or NULL list) of
 * reference-held NCEs that were reaped from a nce_too_many_mcast()
 * call. Separate because the caller must have dropped ndp_g_lock first.
 *
 * This also schedules a taskq task to unlink underlying NCECs from the
 * ndp_g_hash, which are protected by ndp_g_lock.
 */
static void
nce_graveyard_free(list_t *graveyard)
{
	nce_t *deadman, *current;
	ill_t *ill;
	boolean_t doit;

	if (graveyard == NULL)
		return;

	current = list_head(graveyard);
	if (current == NULL) {
		list_destroy(graveyard);
		return;
	}

	ill = current->nce_ill;
	/*
	 * Normally one should ill_refhold(ill) here.  There's no _notr()
	 * variant like there is for ire_t, dce_t, or even ncec_t, but this is
	 * the ONLY case that'll break the mh_trace that IP debugging uses for
	 * reference counts (i.e. they assume same thread releases as
	 * holds). Instead, we inline ill_refhold() here.  We must do the same
	 * in the release done by the ncec_mcast_reap() above.
	 */
	mutex_enter(&ill->ill_lock);
	ill->ill_refcnt++;
	mutex_exit(&ill->ill_lock);

	while (current != NULL) {
		ASSERT3P(ill, ==, current->nce_ill);
		deadman = current;
		current = list_next(graveyard, deadman);
		list_remove(graveyard, deadman);
		ASSERT3U((deadman->nce_common->ncec_flags & NCE_F_MCAST), !=,
		    0);
		nce_refrele(deadman);
	}
	list_destroy(graveyard);

	mutex_enter(&ill->ill_lock);
	if (ill->ill_mcast_ncec_cleanup)
		doit = B_FALSE;
	else {
		ill->ill_mcast_ncec_cleanup = B_TRUE;
		doit = B_TRUE;
	}
	mutex_exit(&ill->ill_lock);
	if (!doit || taskq_dispatch(system_taskq, ncec_mcast_reap,
	    ill, TQ_NOSLEEP) == TASKQID_INVALID) {
		mutex_enter(&ill->ill_lock);
		if (doit) {
			IP_STAT(ill->ill_ipst, ip_nce_mcast_reclaim_tqfail);
			ill->ill_mcast_ncec_cleanup = B_FALSE;
		}
		/* There's no _notr() for ill_refrele(), so inline it here. */
		ill->ill_refcnt--;
		if (ill->ill_refcnt == 0)
			ipif_ill_refrele_tail(ill);	/* Drops ill_lock */
		else
			mutex_exit(&ill->ill_lock);
	}
}

/*
 * For each interface an entry is added for the unspecified multicast group.
 * Here that mapping is used to form the multicast cache entry for a particular
 * multicast destination.
 */
static int
nce_set_multicast_v6(ill_t *ill, const in6_addr_t *dst,
    uint16_t flags, nce_t **newnce)
{
	uchar_t		*hw_addr;
	int		err = 0;
	ip_stack_t	*ipst = ill->ill_ipst;
	nce_t		*nce;

	ASSERT(ill != NULL);
	ASSERT(ill->ill_isv6);
	ASSERT(!(IN6_IS_ADDR_UNSPECIFIED(dst)));

	mutex_enter(&ipst->ips_ndp6->ndp_g_lock);
	nce = nce_lookup_addr(ill, dst);
	if (nce != NULL) {
		mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
		goto done;
	}
	if (ill->ill_net_type == IRE_IF_RESOLVER) {
		/*
		 * For IRE_IF_RESOLVER a hardware mapping can be
		 * generated.
		 */
		hw_addr = kmem_alloc(ill->ill_nd_lla_len, KM_NOSLEEP);
		if (hw_addr == NULL) {
			mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
			return (ENOMEM);
		}
		ip_mcast_mapping(ill, (uchar_t *)dst, hw_addr);
	} else {
		/* No hw_addr is needed for IRE_IF_NORESOLVER. */
		hw_addr = NULL;
	}
	ASSERT((flags & NCE_F_MCAST) != 0);
	ASSERT((flags & NCE_F_NONUD) != 0);
	/* nce_state will be computed by nce_add_common() */
	err = nce_add_v6(ill, hw_addr, ill->ill_phys_addr_length, dst, flags,
	    ND_UNCHANGED, &nce);
	mutex_exit(&ipst->ips_ndp6->ndp_g_lock);
	if (err == 0)
		err = (nce != NULL) ? nce_add_v6_postprocess(nce) : ENOMEM;
	if (hw_addr != NULL)
		kmem_free(hw_addr, ill->ill_nd_lla_len);
	if (err != 0) {
		ip1dbg(("nce_set_multicast_v6: create failed" "%d\n", err));
		return (err);
	}
done:
	ASSERT(nce->nce_common->ncec_state == ND_REACHABLE);
	if (newnce != NULL)
		*newnce = nce;
	else
		nce_refrele(nce);
	return (0);
}

/*
 * Return the link layer address, and any flags of a ncec.
 */
int
ndp_query(ill_t *ill, struct lif_nd_req *lnr)
{
	ncec_t		*ncec;
	in6_addr_t	*addr;
	sin6_t		*sin6;

	ASSERT(ill != NULL && ill->ill_isv6);
	sin6 = (sin6_t *)&lnr->lnr_addr;
	addr =  &sin6->sin6_addr;

	/*
	 * NOTE: if the ill is an IPMP interface, then match against the whole
	 * illgrp.  This e.g. allows in.ndpd to retrieve the link layer
	 * addresses for the data addresses on an IPMP interface even though
	 * ipif_ndp_up() created them with an ncec_ill of ipif_bound_ill.
	 */
	ncec = ncec_lookup_illgrp_v6(ill, addr);
	if (ncec == NULL)
		return (ESRCH);
	/* If no link layer address is available yet, return ESRCH */
	if (!NCE_ISREACHABLE(ncec)) {
		ncec_refrele(ncec);
		return (ESRCH);
	}
	lnr->lnr_hdw_len = ill->ill_phys_addr_length;
	bcopy(ncec->ncec_lladdr, (uchar_t *)&lnr->lnr_hdw_addr,
	    lnr->lnr_hdw_len);
	if (ncec->ncec_flags & NCE_F_ISROUTER)
		lnr->lnr_flags = NDF_ISROUTER_ON;
	if (ncec->ncec_flags & NCE_F_ANYCAST)
		lnr->lnr_flags |= NDF_ANYCAST_ON;
	if (ncec->ncec_flags & NCE_F_STATIC)
		lnr->lnr_flags |= NDF_STATIC;
	ncec_refrele(ncec);
	return (0);
}

/*
 * Finish setting up the Enable/Disable multicast for the driver.
 */
mblk_t *
ndp_mcastreq(ill_t *ill, const in6_addr_t *v6group, uint32_t hw_addr_len,
    uint32_t hw_addr_offset, mblk_t *mp)
{
	uchar_t		*hw_addr;
	ipaddr_t	v4group;
	uchar_t		*addr;

	ASSERT(ill->ill_net_type == IRE_IF_RESOLVER);
	if (IN6_IS_ADDR_V4MAPPED(v6group)) {
		IN6_V4MAPPED_TO_IPADDR(v6group, v4group);

		ASSERT(CLASSD(v4group));
		ASSERT(!(ill->ill_isv6));

		addr = (uchar_t *)&v4group;
	} else {
		ASSERT(IN6_IS_ADDR_MULTICAST(v6group));
		ASSERT(ill->ill_isv6);

		addr = (uchar_t *)v6group;
	}
	hw_addr = mi_offset_paramc(mp, hw_addr_offset, hw_addr_len);
	if (hw_addr == NULL) {
		ip0dbg(("ndp_mcastreq NULL hw_addr\n"));
		freemsg(mp);
		return (NULL);
	}

	ip_mcast_mapping(ill, addr, hw_addr);
	return (mp);
}

void
ip_ndp_resolve(ncec_t *ncec)
{
	in_addr_t	sender4 = INADDR_ANY;
	in6_addr_t	sender6 = ipv6_all_zeros;
	ill_t		*src_ill;
	uint32_t	ms;

	src_ill = nce_resolve_src(ncec, &sender6);
	if (src_ill == NULL) {
		/* Make sure we try again later */
		ms = ncec->ncec_ill->ill_reachable_retrans_time;
		nce_restart_timer(ncec, (clock_t)ms);
		return;
	}
	if (ncec->ncec_ipversion == IPV4_VERSION)
		IN6_V4MAPPED_TO_IPADDR(&sender6, sender4);
	mutex_enter(&ncec->ncec_lock);
	if (ncec->ncec_ipversion == IPV6_VERSION)
		ms = ndp_solicit(ncec, sender6, src_ill);
	else
		ms = arp_request(ncec, sender4, src_ill);
	mutex_exit(&ncec->ncec_lock);
	if (ms == 0) {
		if (ncec->ncec_state != ND_REACHABLE) {
			if (ncec->ncec_ipversion == IPV6_VERSION)
				ndp_resolv_failed(ncec);
			else
				arp_resolv_failed(ncec);
			ASSERT((ncec->ncec_flags & NCE_F_STATIC) == 0);
			nce_make_unreachable(ncec);
			ncec_delete(ncec);
		}
	} else {
		nce_restart_timer(ncec, (clock_t)ms);
	}
done:
	ill_refrele(src_ill);
}

/*
 * Send an IPv6 neighbor solicitation.
 * Returns number of milliseconds after which we should either rexmit or abort.
 * Return of zero means we should abort.
 * The caller holds the ncec_lock to protect ncec_qd_mp and ncec_rcnt.
 * The optional source address is used as a hint to ndp_solicit for
 * which source to use in the packet.
 *
 * NOTE: This routine drops ncec_lock (and later reacquires it) when sending
 * the packet.
 */
uint32_t
ndp_solicit(ncec_t *ncec, in6_addr_t src, ill_t *ill)
{
	in6_addr_t	dst;
	boolean_t	dropped = B_FALSE;

	ASSERT(ncec->ncec_ipversion == IPV6_VERSION);
	ASSERT(MUTEX_HELD(&ncec->ncec_lock));

	if (ncec->ncec_rcnt == 0)
		return (0);

	dst = ncec->ncec_addr;
	ncec->ncec_rcnt--;
	mutex_exit(&ncec->ncec_lock);
	dropped = ndp_xmit(ill, ND_NEIGHBOR_SOLICIT, ill->ill_phys_addr,
	    ill->ill_phys_addr_length, &src, &dst, 0);
	mutex_enter(&ncec->ncec_lock);
	if (dropped)
		ncec->ncec_rcnt++;
	return (ncec->ncec_ill->ill_reachable_retrans_time);
}

/*
 * Attempt to recover an address on an interface that's been marked as a
 * duplicate.  Because NCEs are destroyed when the interface goes down, there's
 * no easy way to just probe the address and have the right thing happen if
 * it's no longer in use.  Instead, we just bring it up normally and allow the
 * regular interface start-up logic to probe for a remaining duplicate and take
 * us back down if necessary.
 * Neither DHCP nor temporary addresses arrive here; they're excluded by
 * ip_ndp_excl.
 */
/* ARGSUSED */
void
ip_addr_recover(ipsq_t *ipsq, queue_t *rq, mblk_t *mp, void *dummy_arg)
{
	ill_t	*ill = rq->q_ptr;
	ipif_t	*ipif;
	in6_addr_t *addr6 = (in6_addr_t *)mp->b_rptr;
	in_addr_t *addr4 = (in_addr_t *)mp->b_rptr;
	boolean_t addr_equal;

	for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
		/*
		 * We do not support recovery of proxy ARP'd interfaces,
		 * because the system lacks a complete proxy ARP mechanism.
		 */
		if (ill->ill_isv6) {
			addr_equal = IN6_ARE_ADDR_EQUAL(&ipif->ipif_v6lcl_addr,
			    addr6);
		} else {
			addr_equal = (ipif->ipif_lcl_addr == *addr4);
		}

		if ((ipif->ipif_flags & IPIF_POINTOPOINT) || !addr_equal)
			continue;

		/*
		 * If we have already recovered or if the interface is going
		 * away, then ignore.
		 */
		mutex_enter(&ill->ill_lock);
		if (!(ipif->ipif_flags & IPIF_DUPLICATE) ||
		    (ipif->ipif_state_flags & IPIF_CONDEMNED)) {
			mutex_exit(&ill->ill_lock);
			continue;
		}

		ipif->ipif_flags &= ~IPIF_DUPLICATE;
		ill->ill_ipif_dup_count--;
		mutex_exit(&ill->ill_lock);
		ipif->ipif_was_dup = B_TRUE;

		if (ill->ill_isv6) {
			VERIFY(ipif_ndp_up(ipif, B_TRUE) != EINPROGRESS);
			(void) ipif_up_done_v6(ipif);
		} else {
			VERIFY(ipif_arp_up(ipif, Res_act_initial, B_TRUE) !=
			    EINPROGRESS);
			(void) ipif_up_done(ipif);
		}
	}
	freeb(mp);
}

/*
 * Attempt to recover an IPv6 interface that's been shut down as a duplicate.
 * As long as someone else holds the address, the interface will stay down.
 * When that conflict goes away, the interface is brought back up.  This is
 * done so that accidental shutdowns of addresses aren't made permanent.  Your
 * server will recover from a failure.
 *
 * For DHCP and temporary addresses, recovery is not done in the kernel.
 * Instead, it's handled by user space processes (dhcpagent and in.ndpd).
 *
 * This function is entered on a timer expiry; the ID is in ipif_recovery_id.
 */
void
ipif_dup_recovery(void *arg)
{
	ipif_t *ipif = arg;

	ipif->ipif_recovery_id = 0;
	if (!(ipif->ipif_flags & IPIF_DUPLICATE))
		return;

	/*
	 * No lock, because this is just an optimization.
	 */
	if (ipif->ipif_state_flags & IPIF_CONDEMNED)
		return;

	/* If the link is down, we'll retry this later */
	if (!(ipif->ipif_ill->ill_phyint->phyint_flags & PHYI_RUNNING))
		return;

	ipif_do_recovery(ipif);
}

/*
 * Perform interface recovery by forcing the duplicate interfaces up and
 * allowing the system to determine which ones should stay up.
 *
 * Called both by recovery timer expiry and link-up notification.
 */
void
ipif_do_recovery(ipif_t *ipif)
{
	ill_t *ill = ipif->ipif_ill;
	mblk_t *mp;
	ip_stack_t *ipst = ill->ill_ipst;
	size_t mp_size;

	if (ipif->ipif_isv6)
		mp_size = sizeof (ipif->ipif_v6lcl_addr);
	else
		mp_size = sizeof (ipif->ipif_lcl_addr);
	mp = allocb(mp_size, BPRI_MED);
	if (mp == NULL) {
		mutex_enter(&ill->ill_lock);
		if (ipst->ips_ip_dup_recovery > 0 &&
		    ipif->ipif_recovery_id == 0 &&
		    !(ipif->ipif_state_flags & IPIF_CONDEMNED)) {
			ipif->ipif_recovery_id = timeout(ipif_dup_recovery,
			    ipif, MSEC_TO_TICK(ipst->ips_ip_dup_recovery));
		}
		mutex_exit(&ill->ill_lock);
	} else {
		/*
		 * A recovery timer may still be running if we got here from
		 * ill_restart_dad(); cancel that timer.
		 */
		if (ipif->ipif_recovery_id != 0)
			(void) untimeout(ipif->ipif_recovery_id);
		ipif->ipif_recovery_id = 0;

		if (ipif->ipif_isv6) {
			bcopy(&ipif->ipif_v6lcl_addr, mp->b_rptr,
			    sizeof (ipif->ipif_v6lcl_addr));
		} else  {
			bcopy(&ipif->ipif_lcl_addr, mp->b_rptr,
			    sizeof (ipif->ipif_lcl_addr));
		}
		ill_refhold(ill);
		qwriter_ip(ill, ill->ill_rq, mp, ip_addr_recover, NEW_OP,
		    B_FALSE);
	}
}

/*
 * Find the MAC and IP addresses in an NA/NS message.
 */
static void
ip_ndp_find_addresses(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill,
    in6_addr_t *targp, uchar_t **haddr, uint_t *haddrlenp)
{
	icmp6_t *icmp6 = (icmp6_t *)(mp->b_rptr + IPV6_HDR_LEN);
	nd_neighbor_solicit_t *ns = (nd_neighbor_solicit_t *)icmp6;
	uchar_t *addr;
	int alen;

	/* icmp_inbound_v6 ensures this */
	ASSERT(ira->ira_flags & IRAF_L2SRC_SET);

	addr = ira->ira_l2src;
	alen = ill->ill_phys_addr_length;
	if (alen > 0) {
		*haddr = addr;
		*haddrlenp = alen;
	} else {
		*haddr = NULL;
		*haddrlenp = 0;
	}

	/* nd_ns_target and nd_na_target are at the same offset, so we cheat */
	*targp = ns->nd_ns_target;
}

/*
 * This is for exclusive changes due to NDP duplicate address detection
 * failure.
 */
/* ARGSUSED */
static void
ip_ndp_excl(ipsq_t *ipsq, queue_t *rq, mblk_t *mp, void *dummy_arg)
{
	ill_t	*ill = rq->q_ptr;
	ipif_t	*ipif;
	uchar_t	*haddr;
	uint_t	haddrlen;
	ip_stack_t *ipst = ill->ill_ipst;
	in6_addr_t targ;
	ip_recv_attr_t iras;
	mblk_t	*attrmp;

	attrmp = mp;
	mp = mp->b_cont;
	attrmp->b_cont = NULL;
	if (!ip_recv_attr_from_mblk(attrmp, &iras)) {
		/* The ill or ip_stack_t disappeared on us */
		BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
		ip_drop_input("ip_recv_attr_from_mblk", mp, ill);
		freemsg(mp);
		ira_cleanup(&iras, B_TRUE);
		return;
	}

	ASSERT(ill == iras.ira_rill);

	ip_ndp_find_addresses(mp, &iras, ill, &targ, &haddr, &haddrlen);
	if (haddr != NULL && haddrlen == ill->ill_phys_addr_length) {
		/*
		 * Ignore conflicts generated by misbehaving switches that
		 * just reflect our own messages back to us.  For IPMP, we may
		 * see reflections across any ill in the illgrp.
		 *
		 * RFC2462 and revisions tried to detect both the case
		 * when a statically configured IPv6 address is a duplicate,
		 * and the case when the L2 address itself is a duplicate. The
		 * later is important because, with stateles address autoconf,
		 * if the L2 address is a duplicate, the resulting IPv6
		 * address(es) would also be duplicates. We rely on DAD of the
		 * IPv6 address itself to detect the latter case.
		 */
		/* For an under ill_grp can change under lock */
		rw_enter(&ipst->ips_ill_g_lock, RW_READER);
		if (bcmp(haddr, ill->ill_phys_addr, haddrlen) == 0 ||
		    IS_UNDER_IPMP(ill) &&
		    ipmp_illgrp_find_ill(ill->ill_grp, haddr,
		    haddrlen) != NULL) {
			rw_exit(&ipst->ips_ill_g_lock);
			goto ignore_conflict;
		}
		rw_exit(&ipst->ips_ill_g_lock);
	}

	/*
	 * Look up the appropriate ipif.
	 */
	ipif = ipif_lookup_addr_v6(&targ, ill, ALL_ZONES, ipst);
	if (ipif == NULL)
		goto ignore_conflict;

	/* Reload the ill to match the ipif */
	ill = ipif->ipif_ill;

	/* If it's already duplicate or ineligible, then don't do anything. */
	if (ipif->ipif_flags & (IPIF_POINTOPOINT|IPIF_DUPLICATE)) {
		ipif_refrele(ipif);
		goto ignore_conflict;
	}

	/*
	 * If this is a failure during duplicate recovery, then don't
	 * complain.  It may take a long time to recover.
	 */
	if (!ipif->ipif_was_dup) {
		char ibuf[LIFNAMSIZ];
		char hbuf[MAC_STR_LEN];
		char sbuf[INET6_ADDRSTRLEN];

		ipif_get_name(ipif, ibuf, sizeof (ibuf));
		cmn_err(CE_WARN, "%s has duplicate address %s (in use by %s);"
		    " disabled", ibuf,
		    inet_ntop(AF_INET6, &targ, sbuf, sizeof (sbuf)),
		    mac_colon_addr(haddr, haddrlen, hbuf, sizeof (hbuf)));
	}
	mutex_enter(&ill->ill_lock);
	ASSERT(!(ipif->ipif_flags & IPIF_DUPLICATE));
	ipif->ipif_flags |= IPIF_DUPLICATE;
	ill->ill_ipif_dup_count++;
	mutex_exit(&ill->ill_lock);
	(void) ipif_down(ipif, NULL, NULL);
	(void) ipif_down_tail(ipif);
	mutex_enter(&ill->ill_lock);
	if (!(ipif->ipif_flags & (IPIF_DHCPRUNNING|IPIF_TEMPORARY)) &&
	    ill->ill_net_type == IRE_IF_RESOLVER &&
	    !(ipif->ipif_state_flags & IPIF_CONDEMNED) &&
	    ipst->ips_ip_dup_recovery > 0) {
		ASSERT(ipif->ipif_recovery_id == 0);
		ipif->ipif_recovery_id = timeout(ipif_dup_recovery,
		    ipif, MSEC_TO_TICK(ipst->ips_ip_dup_recovery));
	}
	mutex_exit(&ill->ill_lock);
	ipif_refrele(ipif);

ignore_conflict:
	freemsg(mp);
	ira_cleanup(&iras, B_TRUE);
}

/*
 * Handle failure by tearing down the ipifs with the specified address.  Note
 * that tearing down the ipif also means deleting the ncec through ipif_down, so
 * it's not possible to do recovery by just restarting the ncec timer.  Instead,
 * we start a timer on the ipif.
 * Caller has to free mp;
 */
static void
ndp_failure(mblk_t *mp, ip_recv_attr_t *ira)
{
	const uchar_t	*haddr;
	ill_t		*ill = ira->ira_rill;

	/*
	 * Ignore conflicts generated by misbehaving switches that just
	 * reflect our own messages back to us.
	 */

	/* icmp_inbound_v6 ensures this */
	ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
	haddr = ira->ira_l2src;
	if (haddr != NULL &&
	    bcmp(haddr, ill->ill_phys_addr, ill->ill_phys_addr_length) == 0) {
		return;
	}

	if ((mp = copymsg(mp)) != NULL) {
		mblk_t	*attrmp;

		attrmp = ip_recv_attr_to_mblk(ira);
		if (attrmp == NULL) {
			BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
			ip_drop_input("ipIfStatsInDiscards", mp, ill);
			freemsg(mp);
		} else {
			ASSERT(attrmp->b_cont == NULL);
			attrmp->b_cont = mp;
			mp = attrmp;
			ill_refhold(ill);
			qwriter_ip(ill, ill->ill_rq, mp, ip_ndp_excl, NEW_OP,
			    B_FALSE);
		}
	}
}

/*
 * Handle a discovered conflict: some other system is advertising that it owns
 * one of our IP addresses.  We need to defend ourselves, or just shut down the
 * interface.
 *
 * Handles both IPv4 and IPv6
 */
boolean_t
ip_nce_conflict(mblk_t *mp, ip_recv_attr_t *ira, ncec_t *ncec)
{
	ipif_t		*ipif;
	clock_t		now;
	uint_t		maxdefense;
	uint_t		defs;
	ill_t		*ill = ira->ira_ill;
	ip_stack_t	*ipst = ill->ill_ipst;
	uint32_t	elapsed;
	boolean_t	isv6 = ill->ill_isv6;
	ipaddr_t	ncec_addr;

	if (isv6) {
		ipif = ipif_lookup_addr_v6(&ncec->ncec_addr, ill, ALL_ZONES,
		    ipst);
	} else {
		if (arp_no_defense) {
			/*
			 * Yes, there is a conflict, but no, we do not
			 * defend ourself.
			 */
			return (B_TRUE);
		}
		IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
		ipif = ipif_lookup_addr(ncec_addr, ill, ALL_ZONES,
		    ipst);
	}
	if (ipif == NULL)
		return (B_FALSE);

	/*
	 * First, figure out if this address is disposable.
	 */
	if (ipif->ipif_flags & (IPIF_DHCPRUNNING | IPIF_TEMPORARY))
		maxdefense = ipst->ips_ip_max_temp_defend;
	else
		maxdefense = ipst->ips_ip_max_defend;

	/*
	 * Now figure out how many times we've defended ourselves.  Ignore
	 * defenses that happened long in the past.
	 */
	now = ddi_get_lbolt();
	elapsed = (drv_hztousec(now - ncec->ncec_last_time_defended))/1000000;
	mutex_enter(&ncec->ncec_lock);
	if ((defs = ncec->ncec_defense_count) > 0 &&
	    elapsed > ipst->ips_ip_defend_interval) {
		/*
		 * ip_defend_interval has elapsed.
		 * reset the defense count.
		 */
		ncec->ncec_defense_count = defs = 0;
	}
	ncec->ncec_defense_count++;
	ncec->ncec_last_time_defended = now;
	mutex_exit(&ncec->ncec_lock);
	ipif_refrele(ipif);

	/*
	 * If we've defended ourselves too many times already, then give up and
	 * tear down the interface(s) using this address.
	 * Otherwise, caller has to defend by sending out an announce.
	 */
	if (defs >= maxdefense) {
		if (isv6)
			ndp_failure(mp, ira);
		else
			arp_failure(mp, ira);
	} else {
		return (B_TRUE); /* caller must defend this address */
	}
	return (B_FALSE);
}

/*
 * Handle reception of Neighbor Solicitation messages.
 */
static void
ndp_input_solicit(mblk_t *mp, ip_recv_attr_t *ira)
{
	ill_t		*ill = ira->ira_ill, *under_ill;
	nd_neighbor_solicit_t *ns;
	uint32_t	hlen = ill->ill_phys_addr_length;
	uchar_t		*haddr = NULL;
	icmp6_t		*icmp_nd;
	ip6_t		*ip6h;
	ncec_t		*our_ncec = NULL;
	in6_addr_t	target;
	in6_addr_t	src;
	int		len;
	int		flag = 0;
	nd_opt_hdr_t	*opt = NULL;
	boolean_t	bad_solicit = B_FALSE;
	mib2_ipv6IfIcmpEntry_t	*mib = ill->ill_icmp6_mib;
	boolean_t	need_ill_refrele = B_FALSE;

	ip6h = (ip6_t *)mp->b_rptr;
	icmp_nd = (icmp6_t *)(mp->b_rptr + IPV6_HDR_LEN);
	len = mp->b_wptr - mp->b_rptr - IPV6_HDR_LEN;
	src = ip6h->ip6_src;
	ns = (nd_neighbor_solicit_t *)icmp_nd;
	target = ns->nd_ns_target;
	if (IN6_IS_ADDR_MULTICAST(&target) || IN6_IS_ADDR_V4MAPPED(&target) ||
	    IN6_IS_ADDR_LOOPBACK(&target)) {
		if (ip_debug > 2) {
			/* ip1dbg */
			pr_addr_dbg("ndp_input_solicit: Martian Target %s\n",
			    AF_INET6, &target);
		}
		bad_solicit = B_TRUE;
		goto done;
	}