ifnet: Delete INET address should not fail; panic upon failure.

[dragonfly.git] / sys / net / if.c

/*
 * Copyright (c) 1980, 1986, 1993
 *      The Regents of the University of California.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *      @(#)if.c        8.3 (Berkeley) 1/4/94
 * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
 */

#include "opt_inet6.h"
#include "opt_inet.h"
#include "opt_ifpoll.h"

#include <sys/param.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/priv.h>
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/socketops.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/mutex.h>
#include <sys/sockio.h>
#include <sys/syslog.h>
#include <sys/sysctl.h>
#include <sys/domain.h>
#include <sys/thread.h>
#include <sys/serialize.h>
#include <sys/bus.h>

#include <sys/thread2.h>
#include <sys/msgport2.h>
#include <sys/mutex2.h>

#include <net/if.h>
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <net/if_var.h>
#include <net/if_ringmap.h>
#include <net/ifq_var.h>
#include <net/radix.h>
#include <net/route.h>
#include <net/if_clone.h>
#include <net/netisr2.h>
#include <net/netmsg2.h>

#include <machine/atomic.h>
#include <machine/stdarg.h>
#include <machine/smp.h>

#if defined(INET) || defined(INET6)
/*XXX*/
#include <netinet/in.h>
#include <netinet/in_var.h>
#include <netinet/if_ether.h>
#ifdef INET6
#include <netinet6/in6_var.h>
#include <netinet6/in6_ifattach.h>
#endif
#endif

struct netmsg_ifaddr {
        struct netmsg_base base;
        struct ifaddr   *ifa;
        struct ifnet    *ifp;
        int             tail;
};

struct ifsubq_stage_head {
        TAILQ_HEAD(, ifsubq_stage)      stg_head;
} __cachealign;

struct if_ringmap {
        int             rm_cnt;
        int             rm_grid;
        int             rm_cpumap[];
};

#define RINGMAP_FLAG_NONE               0x0
#define RINGMAP_FLAG_POWEROF2           0x1

/*
 * System initialization
 */
static void     if_attachdomain(void *);
static void     if_attachdomain1(struct ifnet *);
static int      ifconf(u_long, caddr_t, struct ucred *);
static void     ifinit(void *);
static void     ifnetinit(void *);
static void     if_slowtimo(void *);
static void     link_rtrequest(int, struct rtentry *);
static int      if_rtdel(struct radix_node *, void *);
static void     if_slowtimo_dispatch(netmsg_t);

/* Helper functions */
static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
static int      if_delmulti_serialized(struct ifnet *, struct sockaddr *);
static struct ifnet_array *ifnet_array_alloc(int);
static void     ifnet_array_free(struct ifnet_array *);
static struct ifnet_array *ifnet_array_add(struct ifnet *,
                    const struct ifnet_array *);
static struct ifnet_array *ifnet_array_del(struct ifnet *,
                    const struct ifnet_array *);

#ifdef INET6
/*
 * XXX: declare here to avoid to include many inet6 related files..
 * should be more generalized?
 */
extern void     nd6_setmtu(struct ifnet *);
#endif

SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");

static int ifsq_stage_cntmax = 4;
TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
    &ifsq_stage_cntmax, 0, "ifq staging packet count max");

static int if_stats_compat = 0;
SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
    &if_stats_compat, 0, "Compat the old ifnet stats");

static int if_ringmap_dumprdr = 0;
SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
    &if_ringmap_dumprdr, 0, "dump redirect table");

SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);

static  if_com_alloc_t *if_com_alloc[256];
static  if_com_free_t *if_com_free[256];

MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");

int                     ifqmaxlen = IFQ_MAXLEN;
struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);

static struct ifnet_array       ifnet_array0;
static struct ifnet_array       *ifnet_array = &ifnet_array0;

static struct callout           if_slowtimo_timer;
static struct netmsg_base       if_slowtimo_netmsg;

int                     if_index = 0;
struct ifnet            **ifindex2ifnet = NULL;
static struct mtx       ifnet_mtx = MTX_INITIALIZER("ifnet");

static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];

#ifdef notyet
#define IFQ_KTR_STRING          "ifq=%p"
#define IFQ_KTR_ARGS    struct ifaltq *ifq
#ifndef KTR_IFQ
#define KTR_IFQ                 KTR_ALL
#endif
KTR_INFO_MASTER(ifq);
KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
#define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)

#define IF_START_KTR_STRING     "ifp=%p"
#define IF_START_KTR_ARGS       struct ifnet *ifp
#ifndef KTR_IF_START
#define KTR_IF_START            KTR_ALL
#endif
KTR_INFO_MASTER(if_start);
KTR_INFO(KTR_IF_START, if_start, run, 0,
         IF_START_KTR_STRING, IF_START_KTR_ARGS);
KTR_INFO(KTR_IF_START, if_start, sched, 1,
         IF_START_KTR_STRING, IF_START_KTR_ARGS);
KTR_INFO(KTR_IF_START, if_start, avoid, 2,
         IF_START_KTR_STRING, IF_START_KTR_ARGS);
KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
         IF_START_KTR_STRING, IF_START_KTR_ARGS);
KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
         IF_START_KTR_STRING, IF_START_KTR_ARGS);
#define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
#endif

TAILQ_HEAD(, ifg_group) ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);

/*
 * Network interface utility routines.
 *
 * Routines with ifa_ifwith* names take sockaddr *'s as
 * parameters.
 */
/* ARGSUSED*/
static void
ifinit(void *dummy)
{
        struct ifnet *ifp;

        callout_init_mp(&if_slowtimo_timer);
        netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
            MSGF_PRIORITY, if_slowtimo_dispatch);

        /* XXX is this necessary? */
        ifnet_lock();
        TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
                if (ifp->if_snd.altq_maxlen == 0) {
                        if_printf(ifp, "XXX: driver didn't set altq_maxlen\n");
                        ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
                }
        }
        ifnet_unlock();

        /* Start if_slowtimo */
        lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
}

static void
ifsq_ifstart_ipifunc(void *arg)
{
        struct ifaltq_subque *ifsq = arg;
        struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);

        crit_enter();
        if (lmsg->ms_flags & MSGF_DONE)
                lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
        crit_exit();
}

static __inline void
ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
{
        KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
        TAILQ_REMOVE(&head->stg_head, stage, stg_link);
        stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
        stage->stg_cnt = 0;
        stage->stg_len = 0;
}

static __inline void
ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
{
        KKASSERT((stage->stg_flags &
            (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
        stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
        TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
}

/*
 * Schedule ifnet.if_start on the subqueue owner CPU
 */
static void
ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
{
        int cpu;

        if (!force && curthread->td_type == TD_TYPE_NETISR &&
            ifsq_stage_cntmax > 0) {
                struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);

                stage->stg_cnt = 0;
                stage->stg_len = 0;
                if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
                        ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
                stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
                return;
        }

        cpu = ifsq_get_cpuid(ifsq);
        if (cpu != mycpuid)
                lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
        else
                ifsq_ifstart_ipifunc(ifsq);
}

/*
 * NOTE:
 * This function will release ifnet.if_start subqueue interlock,
 * if ifnet.if_start for the subqueue does not need to be scheduled
 */
static __inline int
ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
{
        if (!running || ifsq_is_empty(ifsq)
#ifdef ALTQ
            || ifsq->ifsq_altq->altq_tbr != NULL
#endif
        ) {
                ALTQ_SQ_LOCK(ifsq);
                /*
                 * ifnet.if_start subqueue interlock is released, if:
                 * 1) Hardware can not take any packets, due to
                 *    o  interface is marked down
                 *    o  hardware queue is full (ifsq_is_oactive)
                 *    Under the second situation, hardware interrupt
                 *    or polling(4) will call/schedule ifnet.if_start
                 *    on the subqueue when hardware queue is ready
                 * 2) There is no packet in the subqueue.
                 *    Further ifq_dispatch or ifq_handoff will call/
                 *    schedule ifnet.if_start on the subqueue.
                 * 3) TBR is used and it does not allow further
                 *    dequeueing.
                 *    TBR callout will call ifnet.if_start on the
                 *    subqueue.
                 */
                if (!running || !ifsq_data_ready(ifsq)) {
                        ifsq_clr_started(ifsq);
                        ALTQ_SQ_UNLOCK(ifsq);
                        return 0;
                }
                ALTQ_SQ_UNLOCK(ifsq);
        }
        return 1;
}

static void
ifsq_ifstart_dispatch(netmsg_t msg)
{
        struct lwkt_msg *lmsg = &msg->base.lmsg;
        struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
        struct ifnet *ifp = ifsq_get_ifp(ifsq);
        struct globaldata *gd = mycpu;
        int running = 0, need_sched;

        crit_enter_gd(gd);

        lwkt_replymsg(lmsg, 0); /* reply ASAP */

        if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
                /*
                 * We need to chase the subqueue owner CPU change.
                 */
                ifsq_ifstart_schedule(ifsq, 1);
                crit_exit_gd(gd);
                return;
        }

        ifsq_serialize_hw(ifsq);
        if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
                ifp->if_start(ifp, ifsq);
                if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
                        running = 1;
        }
        need_sched = ifsq_ifstart_need_schedule(ifsq, running);
        ifsq_deserialize_hw(ifsq);

        if (need_sched) {
                /*
                 * More data need to be transmitted, ifnet.if_start is
                 * scheduled on the subqueue owner CPU, and we keep going.
                 * NOTE: ifnet.if_start subqueue interlock is not released.
                 */
                ifsq_ifstart_schedule(ifsq, 0);
        }

        crit_exit_gd(gd);
}

/* Device driver ifnet.if_start helper function */
void
ifsq_devstart(struct ifaltq_subque *ifsq)
{
        struct ifnet *ifp = ifsq_get_ifp(ifsq);
        int running = 0;

        ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);

        ALTQ_SQ_LOCK(ifsq);
        if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
                ALTQ_SQ_UNLOCK(ifsq);
                return;
        }
        ifsq_set_started(ifsq);
        ALTQ_SQ_UNLOCK(ifsq);

        ifp->if_start(ifp, ifsq);

        if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
                running = 1;

        if (ifsq_ifstart_need_schedule(ifsq, running)) {
                /*
                 * More data need to be transmitted, ifnet.if_start is
                 * scheduled on ifnet's CPU, and we keep going.
                 * NOTE: ifnet.if_start interlock is not released.
                 */
                ifsq_ifstart_schedule(ifsq, 0);
        }
}

void
if_devstart(struct ifnet *ifp)
{
        ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
}

/* Device driver ifnet.if_start schedule helper function */
void
ifsq_devstart_sched(struct ifaltq_subque *ifsq)
{
        ifsq_ifstart_schedule(ifsq, 1);
}

void
if_devstart_sched(struct ifnet *ifp)
{
        ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
}

static void
if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
{
        lwkt_serialize_enter(ifp->if_serializer);
}

static void
if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
{
        lwkt_serialize_exit(ifp->if_serializer);
}

static int
if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
{
        return lwkt_serialize_try(ifp->if_serializer);
}

#ifdef INVARIANTS
static void
if_default_serialize_assert(struct ifnet *ifp,
                            enum ifnet_serialize slz __unused,
                            boolean_t serialized)
{
        if (serialized)
                ASSERT_SERIALIZED(ifp->if_serializer);
        else
                ASSERT_NOT_SERIALIZED(ifp->if_serializer);
}
#endif

/*
 * Attach an interface to the list of "active" interfaces.
 *
 * The serializer is optional.
 */
void
if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
{
        unsigned socksize;
        int namelen, masklen;
        struct sockaddr_dl *sdl, *sdl_addr;
        struct ifaddr *ifa;
        struct ifaltq *ifq;
        struct ifnet **old_ifindex2ifnet = NULL;
        struct ifnet_array *old_ifnet_array;
        int i, q;

        static int if_indexlim = 8;

        if (ifp->if_serialize != NULL) {
                KASSERT(ifp->if_deserialize != NULL &&
                        ifp->if_tryserialize != NULL &&
                        ifp->if_serialize_assert != NULL,
                        ("serialize functions are partially setup"));

                /*
                 * If the device supplies serialize functions,
                 * then clear if_serializer to catch any invalid
                 * usage of this field.
                 */
                KASSERT(serializer == NULL,
                        ("both serialize functions and default serializer "
                         "are supplied"));
                ifp->if_serializer = NULL;
        } else {
                KASSERT(ifp->if_deserialize == NULL &&
                        ifp->if_tryserialize == NULL &&
                        ifp->if_serialize_assert == NULL,
                        ("serialize functions are partially setup"));
                ifp->if_serialize = if_default_serialize;
                ifp->if_deserialize = if_default_deserialize;
                ifp->if_tryserialize = if_default_tryserialize;
#ifdef INVARIANTS
                ifp->if_serialize_assert = if_default_serialize_assert;
#endif

                /*
                 * The serializer can be passed in from the device,
                 * allowing the same serializer to be used for both
                 * the interrupt interlock and the device queue.
                 * If not specified, the netif structure will use an
                 * embedded serializer.
                 */
                if (serializer == NULL) {
                        serializer = &ifp->if_default_serializer;
                        lwkt_serialize_init(serializer);
                }
                ifp->if_serializer = serializer;
        }

        /*
         * XXX -
         * The old code would work if the interface passed a pre-existing
         * chain of ifaddrs to this code.  We don't trust our callers to
         * properly initialize the tailq, however, so we no longer allow
         * this unlikely case.
         */
        ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
                                    M_IFADDR, M_WAITOK | M_ZERO);
        for (i = 0; i < ncpus; ++i)
                TAILQ_INIT(&ifp->if_addrheads[i]);

        TAILQ_INIT(&ifp->if_multiaddrs);
        TAILQ_INIT(&ifp->if_groups);
        getmicrotime(&ifp->if_lastchange);

        /*
         * create a Link Level name for this device
         */
        namelen = strlen(ifp->if_xname);
        masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
        socksize = masklen + ifp->if_addrlen;
        if (socksize < sizeof(*sdl))
                socksize = sizeof(*sdl);
        socksize = RT_ROUNDUP(socksize);
        ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
        sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
        sdl->sdl_len = socksize;
        sdl->sdl_family = AF_LINK;
        bcopy(ifp->if_xname, sdl->sdl_data, namelen);
        sdl->sdl_nlen = namelen;
        sdl->sdl_type = ifp->if_type;
        ifp->if_lladdr = ifa;
        ifa->ifa_ifp = ifp;
        ifa->ifa_rtrequest = link_rtrequest;
        ifa->ifa_addr = (struct sockaddr *)sdl;
        sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
        ifa->ifa_netmask = (struct sockaddr *)sdl;
        sdl->sdl_len = masklen;
        while (namelen != 0)
                sdl->sdl_data[--namelen] = 0xff;
        ifa_iflink(ifa, ifp, 0 /* Insert head */);

        ifp->if_data_pcpu = kmalloc_cachealign(
            ncpus * sizeof(struct ifdata_pcpu), M_DEVBUF, M_WAITOK | M_ZERO);

        if (ifp->if_mapsubq == NULL)
                ifp->if_mapsubq = ifq_mapsubq_default;

        ifq = &ifp->if_snd;
        ifq->altq_type = 0;
        ifq->altq_disc = NULL;
        ifq->altq_flags &= ALTQF_CANTCHANGE;
        ifq->altq_tbr = NULL;
        ifq->altq_ifp = ifp;

        if (ifq->altq_subq_cnt <= 0)
                ifq->altq_subq_cnt = 1;
        ifq->altq_subq = kmalloc_cachealign(
            ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
            M_DEVBUF, M_WAITOK | M_ZERO);

        if (ifq->altq_maxlen == 0) {
                if_printf(ifp, "driver didn't set altq_maxlen\n");
                ifq_set_maxlen(ifq, ifqmaxlen);
        }

        for (q = 0; q < ifq->altq_subq_cnt; ++q) {
                struct ifaltq_subque *ifsq = &ifq->altq_subq[q];

                ALTQ_SQ_LOCK_INIT(ifsq);
                ifsq->ifsq_index = q;

                ifsq->ifsq_altq = ifq;
                ifsq->ifsq_ifp = ifp;

                ifsq->ifsq_maxlen = ifq->altq_maxlen;
                ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
                ifsq->ifsq_prepended = NULL;
                ifsq->ifsq_started = 0;
                ifsq->ifsq_hw_oactive = 0;
                ifsq_set_cpuid(ifsq, 0);
                if (ifp->if_serializer != NULL)
                        ifsq_set_hw_serialize(ifsq, ifp->if_serializer);

                ifsq->ifsq_stage =
                    kmalloc_cachealign(ncpus * sizeof(struct ifsubq_stage),
                    M_DEVBUF, M_WAITOK | M_ZERO);
                for (i = 0; i < ncpus; ++i)
                        ifsq->ifsq_stage[i].stg_subq = ifsq;

                ifsq->ifsq_ifstart_nmsg =
                    kmalloc(ncpus * sizeof(struct netmsg_base),
                    M_LWKTMSG, M_WAITOK);
                for (i = 0; i < ncpus; ++i) {
                        netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
                            &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
                        ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
                }
        }
        ifq_set_classic(ifq);

        /*
         * Increase mbuf cluster/jcluster limits for the mbufs that
         * could sit on the device queues for quite some time.
         */
        if (ifp->if_nmbclusters > 0)
                mcl_inclimit(ifp->if_nmbclusters);
        if (ifp->if_nmbjclusters > 0)
                mjcl_inclimit(ifp->if_nmbjclusters);

        /*
         * Install this ifp into ifindex2inet, ifnet queue and ifnet
         * array after it is setup.
         *
         * Protect ifindex2ifnet, ifnet queue and ifnet array changes
         * by ifnet lock, so that non-netisr threads could get a
         * consistent view.
         */
        ifnet_lock();

        /* Don't update if_index until ifindex2ifnet is setup */
        ifp->if_index = if_index + 1;
        sdl_addr->sdl_index = ifp->if_index;

        /*
         * Install this ifp into ifindex2ifnet
         */
        if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
                unsigned int n;
                struct ifnet **q;

                /*
                 * Grow ifindex2ifnet
                 */
                if_indexlim <<= 1;
                n = if_indexlim * sizeof(*q);
                q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
                if (ifindex2ifnet != NULL) {
                        bcopy(ifindex2ifnet, q, n/2);
                        /* Free old ifindex2ifnet after sync all netisrs */
                        old_ifindex2ifnet = ifindex2ifnet;
                }
                ifindex2ifnet = q;
        }
        ifindex2ifnet[ifp->if_index] = ifp;
        /*
         * Update if_index after this ifp is installed into ifindex2ifnet,
         * so that netisrs could get a consistent view of ifindex2ifnet.
         */
        cpu_sfence();
        if_index = ifp->if_index;

        /*
         * Install this ifp into ifnet array.
         */
        /* Free old ifnet array after sync all netisrs */
        old_ifnet_array = ifnet_array;
        ifnet_array = ifnet_array_add(ifp, old_ifnet_array);

        /*
         * Install this ifp into ifnet queue.
         */
        TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);

        ifnet_unlock();

        /*
         * Sync all netisrs so that the old ifindex2ifnet and ifnet array
         * are no longer accessed and we can free them safely later on.
         */
        netmsg_service_sync();
        if (old_ifindex2ifnet != NULL)
                kfree(old_ifindex2ifnet, M_IFADDR);
        ifnet_array_free(old_ifnet_array);

        if (!SLIST_EMPTY(&domains))
                if_attachdomain1(ifp);

        /* Announce the interface. */
        EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
        devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
        rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
}

static void
if_attachdomain(void *dummy)
{
        struct ifnet *ifp;

        ifnet_lock();
        TAILQ_FOREACH(ifp, &ifnetlist, if_list)
                if_attachdomain1(ifp);
        ifnet_unlock();
}
SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
        if_attachdomain, NULL);

static void
if_attachdomain1(struct ifnet *ifp)
{
        struct domain *dp;

        crit_enter();

        /* address family dependent data region */
        bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
        SLIST_FOREACH(dp, &domains, dom_next)
                if (dp->dom_ifattach)
                        ifp->if_afdata[dp->dom_family] =
                                (*dp->dom_ifattach)(ifp);
        crit_exit();
}

/*
 * Purge all addresses whose type is _not_ AF_LINK
 */
static void
if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
{
        struct lwkt_msg *lmsg = &nmsg->lmsg;
        struct ifnet *ifp = lmsg->u.ms_resultp;
        struct ifaddr_container *ifac, *next;

        ASSERT_IN_NETISR(0);

        /*
         * The ifaddr processing in the following loop will block,
         * however, this function is called in netisr0, in which
         * ifaddr list changes happen, so we don't care about the
         * blockness of the ifaddr processing here.
         */
        TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
                              ifa_link, next) {
                struct ifaddr *ifa = ifac->ifa;

                /* Ignore marker */
                if (ifa->ifa_addr->sa_family == AF_UNSPEC)
                        continue;

                /* Leave link ifaddr as it is */
                if (ifa->ifa_addr->sa_family == AF_LINK)
                        continue;
#ifdef INET
                /* XXX: Ugly!! ad hoc just for INET */
                if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET) {
                        struct ifaliasreq ifr;
                        struct sockaddr_in saved_addr, saved_dst;
#ifdef IFADDR_DEBUG_VERBOSE
                        int i;

                        kprintf("purge in4 addr %p: ", ifa);
                        for (i = 0; i < ncpus; ++i)
                                kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
                        kprintf("\n");
#endif

                        /* Save information for panic. */
                        memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
                        if (ifa->ifa_dstaddr != NULL) {
                                memcpy(&saved_dst, ifa->ifa_dstaddr,
                                    sizeof(saved_dst));
                        } else {
                                memset(&saved_dst, 0, sizeof(saved_dst));
                        }

                        bzero(&ifr, sizeof ifr);
                        ifr.ifra_addr = *ifa->ifa_addr;
                        if (ifa->ifa_dstaddr)
                                ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
                        if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
                                       NULL) == 0)
                                continue;

                        /* MUST NOT HAPPEN */
                        panic("%s: in_control failed %x, dst %x", ifp->if_xname,
                            ntohl(saved_addr.sin_addr.s_addr),
                            ntohl(saved_dst.sin_addr.s_addr));
                }
#endif /* INET */
#ifdef INET6
                if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET6) {
#ifdef IFADDR_DEBUG_VERBOSE
                        int i;

                        kprintf("purge in6 addr %p: ", ifa);
                        for (i = 0; i < ncpus; ++i)
                                kprintf("%d ", ifa->ifa_containers[i].ifa_refcnt);
                        kprintf("\n");
#endif

                        in6_purgeaddr(ifa);
                        /* ifp_addrhead is already updated */
                        continue;
                }
#endif /* INET6 */
                ifa_ifunlink(ifa, ifp);
                ifa_destroy(ifa);
        }

        lwkt_replymsg(lmsg, 0);
}

void
if_purgeaddrs_nolink(struct ifnet *ifp)
{
        struct netmsg_base nmsg;
        struct lwkt_msg *lmsg = &nmsg.lmsg;

        ASSERT_CANDOMSG_NETISR0(curthread);

        netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
            if_purgeaddrs_nolink_dispatch);
        lmsg->u.ms_resultp = ifp;
        lwkt_domsg(netisr_cpuport(0), lmsg, 0);
}

static void
ifq_stage_detach_handler(netmsg_t nmsg)
{
        struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
        int q;

        for (q = 0; q < ifq->altq_subq_cnt; ++q) {
                struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
                struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);

                if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
                        ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
        }
        lwkt_replymsg(&nmsg->lmsg, 0);
}

static void
ifq_stage_detach(struct ifaltq *ifq)
{
        struct netmsg_base base;
        int cpu;

        netmsg_init(&base, NULL, &curthread->td_msgport, 0,
            ifq_stage_detach_handler);
        base.lmsg.u.ms_resultp = ifq;

        for (cpu = 0; cpu < ncpus; ++cpu)
                lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
}

struct netmsg_if_rtdel {
        struct netmsg_base      base;
        struct ifnet            *ifp;
};

static void
if_rtdel_dispatch(netmsg_t msg)
{
        struct netmsg_if_rtdel *rmsg = (void *)msg;
        int i, nextcpu, cpu;

        cpu = mycpuid;
        for (i = 1; i <= AF_MAX; i++) {
                struct radix_node_head  *rnh;

                if ((rnh = rt_tables[cpu][i]) == NULL)
                        continue;
                rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
        }

        nextcpu = cpu + 1;
        if (nextcpu < ncpus)
                lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg);
        else
                lwkt_replymsg(&rmsg->base.lmsg, 0);
}

/*
 * Detach an interface, removing it from the
 * list of "active" interfaces.
 */
void
if_detach(struct ifnet *ifp)
{
        struct ifnet_array *old_ifnet_array;
        struct netmsg_if_rtdel msg;
        struct domain *dp;
        int q;

        /* Announce that the interface is gone. */
        EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
        rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
        devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);

        /*
         * Remove this ifp from ifindex2inet, ifnet queue and ifnet
         * array before it is whacked.
         *
         * Protect ifindex2ifnet, ifnet queue and ifnet array changes
         * by ifnet lock, so that non-netisr threads could get a
         * consistent view.
         */
        ifnet_lock();

        /*
         * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
         */
        ifindex2ifnet[ifp->if_index] = NULL;
        while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
                if_index--;

        /*
         * Remove this ifp from ifnet queue.
         */
        TAILQ_REMOVE(&ifnetlist, ifp, if_link);

        /*
         * Remove this ifp from ifnet array.
         */
        /* Free old ifnet array after sync all netisrs */
        old_ifnet_array = ifnet_array;
        ifnet_array = ifnet_array_del(ifp, old_ifnet_array);

        ifnet_unlock();

        /*
         * Sync all netisrs so that the old ifnet array is no longer
         * accessed and we can free it safely later on.
         */
        netmsg_service_sync();
        ifnet_array_free(old_ifnet_array);

        /*
         * Remove routes and flush queues.
         */
        crit_enter();
#ifdef IFPOLL_ENABLE
        if (ifp->if_flags & IFF_NPOLLING)
                ifpoll_deregister(ifp);
#endif
        if_down(ifp);

        /* Decrease the mbuf clusters/jclusters limits increased by us */
        if (ifp->if_nmbclusters > 0)
                mcl_inclimit(-ifp->if_nmbclusters);
        if (ifp->if_nmbjclusters > 0)
                mjcl_inclimit(-ifp->if_nmbjclusters);

#ifdef ALTQ
        if (ifq_is_enabled(&ifp->if_snd))
                altq_disable(&ifp->if_snd);
        if (ifq_is_attached(&ifp->if_snd))
                altq_detach(&ifp->if_snd);
#endif

        /*
         * Clean up all addresses.
         */
        ifp->if_lladdr = NULL;

        if_purgeaddrs_nolink(ifp);
        if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
                struct ifaddr *ifa;

                ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
                KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
                        ("non-link ifaddr is left on if_addrheads"));

                ifa_ifunlink(ifa, ifp);
                ifa_destroy(ifa);
                KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
                        ("there are still ifaddrs left on if_addrheads"));
        }

#ifdef INET
        /*
         * Remove all IPv4 kernel structures related to ifp.
         */
        in_ifdetach(ifp);
#endif

#ifdef INET6
        /*
         * Remove all IPv6 kernel structs related to ifp.  This should be done
         * before removing routing entries below, since IPv6 interface direct
         * routes are expected to be removed by the IPv6-specific kernel API.
         * Otherwise, the kernel will detect some inconsistency and bark it.
         */
        in6_ifdetach(ifp);
#endif

        /*
         * Delete all remaining routes using this interface
         */
        netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
            if_rtdel_dispatch);
        msg.ifp = ifp;
        rt_domsg_global(&msg.base);

        SLIST_FOREACH(dp, &domains, dom_next)
                if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
                        (*dp->dom_ifdetach)(ifp,
                                ifp->if_afdata[dp->dom_family]);

        kfree(ifp->if_addrheads, M_IFADDR);

        lwkt_synchronize_ipiqs("if_detach");
        ifq_stage_detach(&ifp->if_snd);

        for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
                struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];

                kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
                kfree(ifsq->ifsq_stage, M_DEVBUF);
        }
        kfree(ifp->if_snd.altq_subq, M_DEVBUF);

        kfree(ifp->if_data_pcpu, M_DEVBUF);

        crit_exit();
}

/*
 * Create interface group without members
 */
struct ifg_group *
if_creategroup(const char *groupname)
{
        struct ifg_group        *ifg = NULL;

        if ((ifg = (struct ifg_group *)kmalloc(sizeof(struct ifg_group),
            M_TEMP, M_NOWAIT)) == NULL)
                return (NULL);

        strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
        ifg->ifg_refcnt = 0;
        ifg->ifg_carp_demoted = 0;
        TAILQ_INIT(&ifg->ifg_members);
#if NPF > 0
        pfi_attach_ifgroup(ifg);
#endif
        TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);

        return (ifg);
}

/*
 * Add a group to an interface
 */
int
if_addgroup(struct ifnet *ifp, const char *groupname)
{
        struct ifg_list         *ifgl;
        struct ifg_group        *ifg = NULL;
        struct ifg_member       *ifgm;

        if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
            groupname[strlen(groupname) - 1] <= '9')
                return (EINVAL);

        TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
                if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
                        return (EEXIST);

        if ((ifgl = kmalloc(sizeof(*ifgl), M_TEMP, M_NOWAIT)) == NULL)
                return (ENOMEM);

        if ((ifgm = kmalloc(sizeof(*ifgm), M_TEMP, M_NOWAIT)) == NULL) {
                kfree(ifgl, M_TEMP);
                return (ENOMEM);
        }

        TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
                if (!strcmp(ifg->ifg_group, groupname))
                        break;

        if (ifg == NULL && (ifg = if_creategroup(groupname)) == NULL) {
                kfree(ifgl, M_TEMP);
                kfree(ifgm, M_TEMP);
                return (ENOMEM);
        }

        ifg->ifg_refcnt++;
        ifgl->ifgl_group = ifg;
        ifgm->ifgm_ifp = ifp;

        TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
        TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);

#if NPF > 0
        pfi_group_change(groupname);
#endif

        return (0);
}

/*
 * Remove a group from an interface
 */
int
if_delgroup(struct ifnet *ifp, const char *groupname)
{
        struct ifg_list         *ifgl;
        struct ifg_member       *ifgm;

        TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
                if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
                        break;
        if (ifgl == NULL)
                return (ENOENT);

        TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);

        TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
                if (ifgm->ifgm_ifp == ifp)
                        break;

        if (ifgm != NULL) {
                TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
                kfree(ifgm, M_TEMP);
        }

        if (--ifgl->ifgl_group->ifg_refcnt == 0) {
                TAILQ_REMOVE(&ifg_head, ifgl->ifgl_group, ifg_next);
#if NPF > 0
                pfi_detach_ifgroup(ifgl->ifgl_group);
#endif
                kfree(ifgl->ifgl_group, M_TEMP);
        }

        kfree(ifgl, M_TEMP);

#if NPF > 0
        pfi_group_change(groupname);
#endif

        return (0);
}

/*
 * Stores all groups from an interface in memory pointed
 * to by data
 */
int
if_getgroup(caddr_t data, struct ifnet *ifp)
{
        int                      len, error;
        struct ifg_list         *ifgl;
        struct ifg_req           ifgrq, *ifgp;
        struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;

        if (ifgr->ifgr_len == 0) {
                TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
                        ifgr->ifgr_len += sizeof(struct ifg_req);
                return (0);
        }

        len = ifgr->ifgr_len;
        ifgp = ifgr->ifgr_groups;
        TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
                if (len < sizeof(ifgrq))
                        return (EINVAL);
                bzero(&ifgrq, sizeof ifgrq);
                strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
                    sizeof(ifgrq.ifgrq_group));
                if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
                    sizeof(struct ifg_req))))
                        return (error);
                len -= sizeof(ifgrq);
                ifgp++;
        }

        return (0);
}

/*
 * Stores all members of a group in memory pointed to by data
 */
int
if_getgroupmembers(caddr_t data)
{
        struct ifgroupreq       *ifgr = (struct ifgroupreq *)data;
        struct ifg_group        *ifg;
        struct ifg_member       *ifgm;
        struct ifg_req           ifgrq, *ifgp;
        int                      len, error;

        TAILQ_FOREACH(ifg, &ifg_head, ifg_next)
                if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
                        break;
        if (ifg == NULL)
                return (ENOENT);

        if (ifgr->ifgr_len == 0) {
                TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
                        ifgr->ifgr_len += sizeof(ifgrq);
                return (0);
        }

        len = ifgr->ifgr_len;
        ifgp = ifgr->ifgr_groups;
        TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
                if (len < sizeof(ifgrq))
                        return (EINVAL);
                bzero(&ifgrq, sizeof ifgrq);
                strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
                    sizeof(ifgrq.ifgrq_member));
                if ((error = copyout((caddr_t)&ifgrq, (caddr_t)ifgp,
                    sizeof(struct ifg_req))))
                        return (error);
                len -= sizeof(ifgrq);
                ifgp++;
        }

        return (0);
}

/*
 * Delete Routes for a Network Interface
 *
 * Called for each routing entry via the rnh->rnh_walktree() call above
 * to delete all route entries referencing a detaching network interface.
 *
 * Arguments:
 *      rn      pointer to node in the routing table
 *      arg     argument passed to rnh->rnh_walktree() - detaching interface
 *
 * Returns:
 *      0       successful
 *      errno   failed - reason indicated
 *
 */
static int
if_rtdel(struct radix_node *rn, void *arg)
{
        struct rtentry  *rt = (struct rtentry *)rn;
        struct ifnet    *ifp = arg;
        int             err;

        if (rt->rt_ifp == ifp) {

                /*
                 * Protect (sorta) against walktree recursion problems
                 * with cloned routes
                 */
                if (!(rt->rt_flags & RTF_UP))
                        return (0);

                err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
                                rt_mask(rt), rt->rt_flags,
                                NULL);
                if (err) {
                        log(LOG_WARNING, "if_rtdel: error %d\n", err);
                }
        }

        return (0);
}

static __inline boolean_t
ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
{
        if (old_ifa == NULL)
                return TRUE;

        if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
            (cur_ifa->ifa_ifp->if_flags & IFF_UP))
                return TRUE;
        if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
            (cur_ifa->ifa_flags & IFA_ROUTE))
                return TRUE;
        return FALSE;
}

/*
 * Locate an interface based on a complete address.
 */
struct ifaddr *
ifa_ifwithaddr(struct sockaddr *addr)
{
        const struct ifnet_array *arr;
        int i;

        arr = ifnet_array_get();
        for (i = 0; i < arr->ifnet_count; ++i) {
                struct ifnet *ifp = arr->ifnet_arr[i];
                struct ifaddr_container *ifac;

                TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                        struct ifaddr *ifa = ifac->ifa;

                        if (ifa->ifa_addr->sa_family != addr->sa_family)
                                continue;
                        if (sa_equal(addr, ifa->ifa_addr))
                                return (ifa);
                        if ((ifp->if_flags & IFF_BROADCAST) &&
                            ifa->ifa_broadaddr &&
                            /* IPv6 doesn't have broadcast */
                            ifa->ifa_broadaddr->sa_len != 0 &&
                            sa_equal(ifa->ifa_broadaddr, addr))
                                return (ifa);
                }
        }
        return (NULL);
}

/*
 * Locate the point to point interface with a given destination address.
 */
struct ifaddr *
ifa_ifwithdstaddr(struct sockaddr *addr)
{
        const struct ifnet_array *arr;
        int i;

        arr = ifnet_array_get();
        for (i = 0; i < arr->ifnet_count; ++i) {
                struct ifnet *ifp = arr->ifnet_arr[i];
                struct ifaddr_container *ifac;

                if (!(ifp->if_flags & IFF_POINTOPOINT))
                        continue;

                TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                        struct ifaddr *ifa = ifac->ifa;

                        if (ifa->ifa_addr->sa_family != addr->sa_family)
                                continue;
                        if (ifa->ifa_dstaddr &&
                            sa_equal(addr, ifa->ifa_dstaddr))
                                return (ifa);
                }
        }
        return (NULL);
}

/*
 * Find an interface on a specific network.  If many, choice
 * is most specific found.
 */
struct ifaddr *
ifa_ifwithnet(struct sockaddr *addr)
{
        struct ifaddr *ifa_maybe = NULL;
        u_int af = addr->sa_family;
        char *addr_data = addr->sa_data, *cplim;
        const struct ifnet_array *arr;
        int i;

        /*
         * AF_LINK addresses can be looked up directly by their index number,
         * so do that if we can.
         */
        if (af == AF_LINK) {
                struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;

                if (sdl->sdl_index && sdl->sdl_index <= if_index)
                        return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
        }

        /*
         * Scan though each interface, looking for ones that have
         * addresses in this address family.
         */
        arr = ifnet_array_get();
        for (i = 0; i < arr->ifnet_count; ++i) {
                struct ifnet *ifp = arr->ifnet_arr[i];
                struct ifaddr_container *ifac;

                TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                        struct ifaddr *ifa = ifac->ifa;
                        char *cp, *cp2, *cp3;

                        if (ifa->ifa_addr->sa_family != af)
next:                           continue;
                        if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
                                /*
                                 * This is a bit broken as it doesn't
                                 * take into account that the remote end may
                                 * be a single node in the network we are
                                 * looking for.
                                 * The trouble is that we don't know the
                                 * netmask for the remote end.
                                 */
                                if (ifa->ifa_dstaddr != NULL &&
                                    sa_equal(addr, ifa->ifa_dstaddr))
                                        return (ifa);
                        } else {
                                /*
                                 * if we have a special address handler,
                                 * then use it instead of the generic one.
                                 */
                                if (ifa->ifa_claim_addr) {
                                        if ((*ifa->ifa_claim_addr)(ifa, addr)) {
                                                return (ifa);
                                        } else {
                                                continue;
                                        }
                                }

                                /*
                                 * Scan all the bits in the ifa's address.
                                 * If a bit dissagrees with what we are
                                 * looking for, mask it with the netmask
                                 * to see if it really matters.
                                 * (A byte at a time)
                                 */
                                if (ifa->ifa_netmask == 0)
                                        continue;
                                cp = addr_data;
                                cp2 = ifa->ifa_addr->sa_data;
                                cp3 = ifa->ifa_netmask->sa_data;
                                cplim = ifa->ifa_netmask->sa_len +
                                        (char *)ifa->ifa_netmask;
                                while (cp3 < cplim)
                                        if ((*cp++ ^ *cp2++) & *cp3++)
                                                goto next; /* next address! */
                                /*
                                 * If the netmask of what we just found
                                 * is more specific than what we had before
                                 * (if we had one) then remember the new one
                                 * before continuing to search for an even
                                 * better one.  If the netmasks are equal,
                                 * we prefer the this ifa based on the result
                                 * of ifa_prefer().
                                 */
                                if (ifa_maybe == NULL ||
                                    rn_refines((char *)ifa->ifa_netmask,
                                        (char *)ifa_maybe->ifa_netmask) ||
                                    (sa_equal(ifa_maybe->ifa_netmask,
                                        ifa->ifa_netmask) &&
                                     ifa_prefer(ifa, ifa_maybe)))
                                        ifa_maybe = ifa;
                        }
                }
        }
        return (ifa_maybe);
}

/*
 * Find an interface address specific to an interface best matching
 * a given address.
 */
struct ifaddr *
ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
{
        struct ifaddr_container *ifac;
        char *cp, *cp2, *cp3;
        char *cplim;
        struct ifaddr *ifa_maybe = NULL;
        u_int af = addr->sa_family;

        if (af >= AF_MAX)
                return (0);
        TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                struct ifaddr *ifa = ifac->ifa;

                if (ifa->ifa_addr->sa_family != af)
                        continue;
                if (ifa_maybe == NULL)
                        ifa_maybe = ifa;
                if (ifa->ifa_netmask == NULL) {
                        if (sa_equal(addr, ifa->ifa_addr) ||
                            (ifa->ifa_dstaddr != NULL &&
                             sa_equal(addr, ifa->ifa_dstaddr)))
                                return (ifa);
                        continue;
                }
                if (ifp->if_flags & IFF_POINTOPOINT) {
                        if (sa_equal(addr, ifa->ifa_dstaddr))
                                return (ifa);
                } else {
                        cp = addr->sa_data;
                        cp2 = ifa->ifa_addr->sa_data;
                        cp3 = ifa->ifa_netmask->sa_data;
                        cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
                        for (; cp3 < cplim; cp3++)
                                if ((*cp++ ^ *cp2++) & *cp3)
                                        break;
                        if (cp3 == cplim)
                                return (ifa);
                }
        }
        return (ifa_maybe);
}

/*
 * Default action when installing a route with a Link Level gateway.
 * Lookup an appropriate real ifa to point to.
 * This should be moved to /sys/net/link.c eventually.
 */
static void
link_rtrequest(int cmd, struct rtentry *rt)
{
        struct ifaddr *ifa;
        struct sockaddr *dst;
        struct ifnet *ifp;

        if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
            (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
                return;
        ifa = ifaof_ifpforaddr(dst, ifp);
        if (ifa != NULL) {
                IFAFREE(rt->rt_ifa);
                IFAREF(ifa);
                rt->rt_ifa = ifa;
                if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
                        ifa->ifa_rtrequest(cmd, rt);
        }
}

struct netmsg_ifroute {
        struct netmsg_base      base;
        struct ifnet            *ifp;
        int                     flag;
        int                     fam;
};

/*
 * Mark an interface down and notify protocols of the transition.
 */
static void
if_unroute_dispatch(netmsg_t nmsg)
{
        struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
        struct ifnet *ifp = msg->ifp;
        int flag = msg->flag, fam = msg->fam;
        struct ifaddr_container *ifac;

        ifp->if_flags &= ~flag;
        getmicrotime(&ifp->if_lastchange);
        /*
         * The ifaddr processing in the following loop will block,
         * however, this function is called in netisr0, in which
         * ifaddr list changes happen, so we don't care about the
         * blockness of the ifaddr processing here.
         */
        TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                struct ifaddr *ifa = ifac->ifa;

                /* Ignore marker */
                if (ifa->ifa_addr->sa_family == AF_UNSPEC)
                        continue;

                if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
                        kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
        }
        ifq_purge_all(&ifp->if_snd);
        rt_ifmsg(ifp);

        lwkt_replymsg(&nmsg->lmsg, 0);
}

void
if_unroute(struct ifnet *ifp, int flag, int fam)
{
        struct netmsg_ifroute msg;

        ASSERT_CANDOMSG_NETISR0(curthread);

        netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
            if_unroute_dispatch);
        msg.ifp = ifp;
        msg.flag = flag;
        msg.fam = fam;
        lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
}

/*
 * Mark an interface up and notify protocols of the transition.
 */
static void
if_route_dispatch(netmsg_t nmsg)
{
        struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
        struct ifnet *ifp = msg->ifp;
        int flag = msg->flag, fam = msg->fam;
        struct ifaddr_container *ifac;

        ifq_purge_all(&ifp->if_snd);
        ifp->if_flags |= flag;
        getmicrotime(&ifp->if_lastchange);
        /*
         * The ifaddr processing in the following loop will block,
         * however, this function is called in netisr0, in which
         * ifaddr list changes happen, so we don't care about the
         * blockness of the ifaddr processing here.
         */
        TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                struct ifaddr *ifa = ifac->ifa;

                /* Ignore marker */
                if (ifa->ifa_addr->sa_family == AF_UNSPEC)
                        continue;

                if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
                        kpfctlinput(PRC_IFUP, ifa->ifa_addr);
        }
        rt_ifmsg(ifp);
#ifdef INET6
        in6_if_up(ifp);
#endif

        lwkt_replymsg(&nmsg->lmsg, 0);
}

void
if_route(struct ifnet *ifp, int flag, int fam)
{
        struct netmsg_ifroute msg;

        ASSERT_CANDOMSG_NETISR0(curthread);

        netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
            if_route_dispatch);
        msg.ifp = ifp;
        msg.flag = flag;
        msg.fam = fam;
        lwkt_domsg(netisr_cpuport(0), &msg.base.lmsg, 0);
}

/*
 * Mark an interface down and notify protocols of the transition.  An
 * interface going down is also considered to be a synchronizing event.
 * We must ensure that all packet processing related to the interface
 * has completed before we return so e.g. the caller can free the ifnet
 * structure that the mbufs may be referencing.
 *
 * NOTE: must be called at splnet or eqivalent.
 */
void
if_down(struct ifnet *ifp)
{
        if_unroute(ifp, IFF_UP, AF_UNSPEC);
        netmsg_service_sync();
}

/*
 * Mark an interface up and notify protocols of
 * the transition.
 * NOTE: must be called at splnet or eqivalent.
 */
void
if_up(struct ifnet *ifp)
{
        if_route(ifp, IFF_UP, AF_UNSPEC);
}

/*
 * Process a link state change.
 * NOTE: must be called at splsoftnet or equivalent.
 */
void
if_link_state_change(struct ifnet *ifp)
{
        int link_state = ifp->if_link_state;

        rt_ifmsg(ifp);
        devctl_notify("IFNET", ifp->if_xname,
            (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
}

/*
 * Handle interface watchdog timer routines.  Called
 * from softclock, we decrement timers (if set) and
 * call the appropriate interface routine on expiration.
 */
static void
if_slowtimo_dispatch(netmsg_t nmsg)
{
        struct globaldata *gd = mycpu;
        const struct ifnet_array *arr;
        int i;

        ASSERT_IN_NETISR(0);

        crit_enter_gd(gd);
        lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
        crit_exit_gd(gd);

        arr = ifnet_array_get();
        for (i = 0; i < arr->ifnet_count; ++i) {
                struct ifnet *ifp = arr->ifnet_arr[i];

                crit_enter_gd(gd);

                if (if_stats_compat) {
                        IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
                        IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
                        IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
                        IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
                        IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
                        IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
                        IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
                        IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
                        IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
                        IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
                        IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
                        IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
                }

                if (ifp->if_timer == 0 || --ifp->if_timer) {
                        crit_exit_gd(gd);
                        continue;
                }
                if (ifp->if_watchdog) {
                        if (ifnet_tryserialize_all(ifp)) {
                                (*ifp->if_watchdog)(ifp);
                                ifnet_deserialize_all(ifp);
                        } else {
                                /* try again next timeout */
                                ++ifp->if_timer;
                        }
                }

                crit_exit_gd(gd);
        }

        callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
}

static void
if_slowtimo(void *arg __unused)
{
        struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;

        KASSERT(mycpuid == 0, ("not on cpu0"));
        crit_enter();
        if (lmsg->ms_flags & MSGF_DONE)
                lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
        crit_exit();
}

/*
 * Map interface name to
 * interface structure pointer.
 */
struct ifnet *
ifunit(const char *name)
{
        struct ifnet *ifp;

        /*
         * Search all the interfaces for this name/number
         */
        KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));

        TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
                if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
                        break;
        }
        return (ifp);
}

struct ifnet *
ifunit_netisr(const char *name)
{
        const struct ifnet_array *arr;
        int i;

        /*
         * Search all the interfaces for this name/number
         */

        arr = ifnet_array_get();
        for (i = 0; i < arr->ifnet_count; ++i) {
                struct ifnet *ifp = arr->ifnet_arr[i];

                if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
                        return ifp;
        }
        return NULL;
}

/*
 * Interface ioctls.
 */
int
ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
{
        struct ifnet *ifp;
        struct ifreq *ifr;
        struct ifstat *ifs;
        int error, do_ifup = 0;
        short oif_flags;
        int new_flags;
        size_t namelen, onamelen;
        char new_name[IFNAMSIZ];
        struct ifaddr *ifa;
        struct sockaddr_dl *sdl;

        switch (cmd) {
        case SIOCGIFCONF:
        case OSIOCGIFCONF:
                return (ifconf(cmd, data, cred));
        default:
                break;
        }

        ifr = (struct ifreq *)data;

        switch (cmd) {
        case SIOCIFCREATE:
        case SIOCIFCREATE2:
                if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
                        return (error);
                return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
                        cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
        case SIOCIFDESTROY:
                if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
                        return (error);
                return (if_clone_destroy(ifr->ifr_name));
        case SIOCIFGCLONERS:
                return (if_clone_list((struct if_clonereq *)data));
        default:
                break;
        }

        /*
         * Nominal ioctl through interface, lookup the ifp and obtain a
         * lock to serialize the ifconfig ioctl operation.
         */
        ifnet_lock();

        ifp = ifunit(ifr->ifr_name);
        if (ifp == NULL) {
                ifnet_unlock();
                return (ENXIO);
        }
        error = 0;

        switch (cmd) {
        case SIOCGIFINDEX:
                ifr->ifr_index = ifp->if_index;
                break;

        case SIOCGIFFLAGS:
                ifr->ifr_flags = ifp->if_flags;
                ifr->ifr_flagshigh = ifp->if_flags >> 16;
                break;

        case SIOCGIFCAP:
                ifr->ifr_reqcap = ifp->if_capabilities;
                ifr->ifr_curcap = ifp->if_capenable;
                break;

        case SIOCGIFMETRIC:
                ifr->ifr_metric = ifp->if_metric;
                break;

        case SIOCGIFMTU:
                ifr->ifr_mtu = ifp->if_mtu;
                break;

        case SIOCGIFTSOLEN:
                ifr->ifr_tsolen = ifp->if_tsolen;
                break;

        case SIOCGIFDATA:
                error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
                                sizeof(ifp->if_data));
                break;

        case SIOCGIFPHYS:
                ifr->ifr_phys = ifp->if_physical;
                break;

        case SIOCGIFPOLLCPU:
                ifr->ifr_pollcpu = -1;
                break;

        case SIOCSIFPOLLCPU:
                break;

        case SIOCSIFFLAGS:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                new_flags = (ifr->ifr_flags & 0xffff) |
                    (ifr->ifr_flagshigh << 16);
                if (ifp->if_flags & IFF_SMART) {
                        /* Smart drivers twiddle their own routes */
                } else if (ifp->if_flags & IFF_UP &&
                    (new_flags & IFF_UP) == 0) {
                        if_down(ifp);
                } else if (new_flags & IFF_UP &&
                    (ifp->if_flags & IFF_UP) == 0) {
                        do_ifup = 1;
                }

#ifdef IFPOLL_ENABLE
                if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
                        if (new_flags & IFF_NPOLLING)
                                ifpoll_register(ifp);
                        else
                                ifpoll_deregister(ifp);
                }
#endif

                ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
                        (new_flags &~ IFF_CANTCHANGE);
                if (new_flags & IFF_PPROMISC) {
                        /* Permanently promiscuous mode requested */
                        ifp->if_flags |= IFF_PROMISC;
                } else if (ifp->if_pcount == 0) {
                        ifp->if_flags &= ~IFF_PROMISC;
                }
                if (ifp->if_ioctl) {
                        ifnet_serialize_all(ifp);
                        ifp->if_ioctl(ifp, cmd, data, cred);
                        ifnet_deserialize_all(ifp);
                }
                if (do_ifup)
                        if_up(ifp);
                getmicrotime(&ifp->if_lastchange);
                break;

        case SIOCSIFCAP:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
                        error = EINVAL;
                        break;
                }
                ifnet_serialize_all(ifp);
                ifp->if_ioctl(ifp, cmd, data, cred);
                ifnet_deserialize_all(ifp);
                break;

        case SIOCSIFNAME:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
                if (error)
                        break;
                if (new_name[0] == '\0') {
                        error = EINVAL;
                        break;
                }
                if (ifunit(new_name) != NULL) {
                        error = EEXIST;
                        break;
                }

                EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);

                /* Announce the departure of the interface. */
                rt_ifannouncemsg(ifp, IFAN_DEPARTURE);

                strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
                ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
                sdl = (struct sockaddr_dl *)ifa->ifa_addr;
                namelen = strlen(new_name);
                onamelen = sdl->sdl_nlen;
                /*
                 * Move the address if needed.  This is safe because we
                 * allocate space for a name of length IFNAMSIZ when we
                 * create this in if_attach().
                 */
                if (namelen != onamelen) {
                        bcopy(sdl->sdl_data + onamelen,
                            sdl->sdl_data + namelen, sdl->sdl_alen);
                }
                bcopy(new_name, sdl->sdl_data, namelen);
                sdl->sdl_nlen = namelen;
                sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
                bzero(sdl->sdl_data, onamelen);
                while (namelen != 0)
                        sdl->sdl_data[--namelen] = 0xff;

                EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);

                /* Announce the return of the interface. */
                rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
                break;

        case SIOCSIFMETRIC:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                ifp->if_metric = ifr->ifr_metric;
                getmicrotime(&ifp->if_lastchange);
                break;

        case SIOCSIFPHYS:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                if (ifp->if_ioctl == NULL) {
                        error = EOPNOTSUPP;
                        break;
                }
                ifnet_serialize_all(ifp);
                error = ifp->if_ioctl(ifp, cmd, data, cred);
                ifnet_deserialize_all(ifp);
                if (error == 0)
                        getmicrotime(&ifp->if_lastchange);
                break;

        case SIOCSIFMTU:
        {
                u_long oldmtu = ifp->if_mtu;

                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                if (ifp->if_ioctl == NULL) {
                        error = EOPNOTSUPP;
                        break;
                }
                if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
                        error = EINVAL;
                        break;
                }
                ifnet_serialize_all(ifp);
                error = ifp->if_ioctl(ifp, cmd, data, cred);
                ifnet_deserialize_all(ifp);
                if (error == 0) {
                        getmicrotime(&ifp->if_lastchange);
                        rt_ifmsg(ifp);
                }
                /*
                 * If the link MTU changed, do network layer specific procedure.
                 */
                if (ifp->if_mtu != oldmtu) {
#ifdef INET6
                        nd6_setmtu(ifp);
#endif
                }
                break;
        }

        case SIOCSIFTSOLEN:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;

                /* XXX need driver supplied upper limit */
                if (ifr->ifr_tsolen <= 0) {
                        error = EINVAL;
                        break;
                }
                ifp->if_tsolen = ifr->ifr_tsolen;
                break;

        case SIOCADDMULTI:
        case SIOCDELMULTI:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;

                /* Don't allow group membership on non-multicast interfaces. */
                if ((ifp->if_flags & IFF_MULTICAST) == 0) {
                        error = EOPNOTSUPP;
                        break;
                }

                /* Don't let users screw up protocols' entries. */
                if (ifr->ifr_addr.sa_family != AF_LINK) {
                        error = EINVAL;
                        break;
                }

                if (cmd == SIOCADDMULTI) {
                        struct ifmultiaddr *ifma;
                        error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
                } else {
                        error = if_delmulti(ifp, &ifr->ifr_addr);
                }
                if (error == 0)
                        getmicrotime(&ifp->if_lastchange);
                break;

        case SIOCSIFPHYADDR:
        case SIOCDIFPHYADDR:
#ifdef INET6
        case SIOCSIFPHYADDR_IN6:
#endif
        case SIOCSLIFPHYADDR:
        case SIOCSIFMEDIA:
        case SIOCSIFGENERIC:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                if (ifp->if_ioctl == 0) {
                        error = EOPNOTSUPP;
                        break;
                }
                ifnet_serialize_all(ifp);
                error = ifp->if_ioctl(ifp, cmd, data, cred);
                ifnet_deserialize_all(ifp);
                if (error == 0)
                        getmicrotime(&ifp->if_lastchange);
                break;

        case SIOCGIFSTATUS:
                ifs = (struct ifstat *)data;
                ifs->ascii[0] = '\0';
                /* fall through */
        case SIOCGIFPSRCADDR:
        case SIOCGIFPDSTADDR:
        case SIOCGLIFPHYADDR:
        case SIOCGIFMEDIA:
        case SIOCGIFGENERIC:
                if (ifp->if_ioctl == NULL) {
                        error = EOPNOTSUPP;
                        break;
                }
                ifnet_serialize_all(ifp);
                error = ifp->if_ioctl(ifp, cmd, data, cred);
                ifnet_deserialize_all(ifp);
                break;

        case SIOCSIFLLADDR:
                error = priv_check_cred(cred, PRIV_ROOT, 0);
                if (error)
                        break;
                error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
                                     ifr->ifr_addr.sa_len);
                EVENTHANDLER_INVOKE(iflladdr_event, ifp);
                break;

        default:
                oif_flags = ifp->if_flags;
                if (so->so_proto == 0) {
                        error = EOPNOTSUPP;
                        break;
                }
                error = so_pru_control_direct(so, cmd, data, ifp);

                if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
#ifdef INET6
                        DELAY(100);/* XXX: temporary workaround for fxp issue*/
                        if (ifp->if_flags & IFF_UP) {
                                crit_enter();
                                in6_if_up(ifp);
                                crit_exit();
                        }
#endif
                }
                break;
        }

        ifnet_unlock();
        return (error);
}

/*
 * Set/clear promiscuous mode on interface ifp based on the truth value
 * of pswitch.  The calls are reference counted so that only the first
 * "on" request actually has an effect, as does the final "off" request.
 * Results are undefined if the "off" and "on" requests are not matched.
 */
int
ifpromisc(struct ifnet *ifp, int pswitch)
{
        struct ifreq ifr;
        int error;
        int oldflags;

        oldflags = ifp->if_flags;
        if (ifp->if_flags & IFF_PPROMISC) {
                /* Do nothing if device is in permanently promiscuous mode */
                ifp->if_pcount += pswitch ? 1 : -1;
                return (0);
        }
        if (pswitch) {
                /*
                 * If the device is not configured up, we cannot put it in
                 * promiscuous mode.
                 */
                if ((ifp->if_flags & IFF_UP) == 0)
                        return (ENETDOWN);
                if (ifp->if_pcount++ != 0)
                        return (0);
                ifp->if_flags |= IFF_PROMISC;
                log(LOG_INFO, "%s: promiscuous mode enabled\n",
                    ifp->if_xname);
        } else {
                if (--ifp->if_pcount > 0)
                        return (0);
                ifp->if_flags &= ~IFF_PROMISC;
                log(LOG_INFO, "%s: promiscuous mode disabled\n",
                    ifp->if_xname);
        }
        ifr.ifr_flags = ifp->if_flags;
        ifr.ifr_flagshigh = ifp->if_flags >> 16;
        ifnet_serialize_all(ifp);
        error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
        ifnet_deserialize_all(ifp);
        if (error == 0)
                rt_ifmsg(ifp);
        else
                ifp->if_flags = oldflags;
        return error;
}

/*
 * Return interface configuration
 * of system.  List may be used
 * in later ioctl's (above) to get
 * other information.
 */
static int
ifconf(u_long cmd, caddr_t data, struct ucred *cred)
{
        struct ifconf *ifc = (struct ifconf *)data;
        struct ifnet *ifp;
        struct sockaddr *sa;
        struct ifreq ifr, *ifrp;
        int space = ifc->ifc_len, error = 0;

        ifrp = ifc->ifc_req;

        ifnet_lock();
        TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
                struct ifaddr_container *ifac, *ifac_mark;
                struct ifaddr_marker mark;
                struct ifaddrhead *head;
                int addrs;

                if (space <= sizeof ifr)
                        break;

                /*
                 * Zero the stack declared structure first to prevent
                 * memory disclosure.
                 */
                bzero(&ifr, sizeof(ifr));
                if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
                    >= sizeof(ifr.ifr_name)) {
                        error = ENAMETOOLONG;
                        break;
                }

                /*
                 * Add a marker, since copyout() could block and during that
                 * period the list could be changed.  Inserting the marker to
                 * the header of the list will not cause trouble for the code
                 * assuming that the first element of the list is AF_LINK; the
                 * marker will be moved to the next position w/o blocking.
                 */
                ifa_marker_init(&mark, ifp);
                ifac_mark = &mark.ifac;
                head = &ifp->if_addrheads[mycpuid];

                addrs = 0;
                TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
                while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
                        struct ifaddr *ifa = ifac->ifa;

                        TAILQ_REMOVE(head, ifac_mark, ifa_link);
                        TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);

                        /* Ignore marker */
                        if (ifa->ifa_addr->sa_family == AF_UNSPEC)
                                continue;

                        if (space <= sizeof ifr)
                                break;
                        sa = ifa->ifa_addr;
                        if (cred->cr_prison &&
                            prison_if(cred, sa))
                                continue;
                        addrs++;
                        /*
                         * Keep a reference on this ifaddr, so that it will
                         * not be destroyed when its address is copied to
                         * the userland, which could block.
                         */
                        IFAREF(ifa);
                        if (sa->sa_len <= sizeof(*sa)) {
                                ifr.ifr_addr = *sa;
                                error = copyout(&ifr, ifrp, sizeof ifr);
                                ifrp++;
                        } else {
                                if (space < (sizeof ifr) + sa->sa_len -
                                            sizeof(*sa)) {
                                        IFAFREE(ifa);
                                        break;
                                }
                                space -= sa->sa_len - sizeof(*sa);
                                error = copyout(&ifr, ifrp,
                                                sizeof ifr.ifr_name);
                                if (error == 0)
                                        error = copyout(sa, &ifrp->ifr_addr,
                                                        sa->sa_len);
                                ifrp = (struct ifreq *)
                                        (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
                        }
                        IFAFREE(ifa);
                        if (error)
                                break;
                        space -= sizeof ifr;
                }
                TAILQ_REMOVE(head, ifac_mark, ifa_link);
                if (error)
                        break;
                if (!addrs) {
                        bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
                        error = copyout(&ifr, ifrp, sizeof ifr);
                        if (error)
                                break;
                        space -= sizeof ifr;
                        ifrp++;
                }
        }
        ifnet_unlock();

        ifc->ifc_len -= space;
        return (error);
}

/*
 * Just like if_promisc(), but for all-multicast-reception mode.
 */
int
if_allmulti(struct ifnet *ifp, int onswitch)
{
        int error = 0;
        struct ifreq ifr;

        crit_enter();

        if (onswitch) {
                if (ifp->if_amcount++ == 0) {
                        ifp->if_flags |= IFF_ALLMULTI;
                        ifr.ifr_flags = ifp->if_flags;
                        ifr.ifr_flagshigh = ifp->if_flags >> 16;
                        ifnet_serialize_all(ifp);
                        error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
                                              NULL);
                        ifnet_deserialize_all(ifp);
                }
        } else {
                if (ifp->if_amcount > 1) {
                        ifp->if_amcount--;
                } else {
                        ifp->if_amcount = 0;
                        ifp->if_flags &= ~IFF_ALLMULTI;
                        ifr.ifr_flags = ifp->if_flags;
                        ifr.ifr_flagshigh = ifp->if_flags >> 16;
                        ifnet_serialize_all(ifp);
                        error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
                                              NULL);
                        ifnet_deserialize_all(ifp);
                }
        }

        crit_exit();

        if (error == 0)
                rt_ifmsg(ifp);
        return error;
}

/*
 * Add a multicast listenership to the interface in question.
 * The link layer provides a routine which converts
 */
int
if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
    struct ifmultiaddr **retifma)
{
        struct sockaddr *llsa, *dupsa;
        int error;
        struct ifmultiaddr *ifma;

        ASSERT_IFNET_SERIALIZED_ALL(ifp);

        /*
         * If the matching multicast address already exists
         * then don't add a new one, just add a reference
         */
        TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
                if (sa_equal(sa, ifma->ifma_addr)) {
                        ifma->ifma_refcount++;
                        if (retifma)
                                *retifma = ifma;
                        return 0;
                }
        }

        /*
         * Give the link layer a chance to accept/reject it, and also
         * find out which AF_LINK address this maps to, if it isn't one
         * already.
         */
        if (ifp->if_resolvemulti) {
                error = ifp->if_resolvemulti(ifp, &llsa, sa);
                if (error)
                        return error;
        } else {
                llsa = NULL;
        }

        ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
        dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
        bcopy(sa, dupsa, sa->sa_len);

        ifma->ifma_addr = dupsa;
        ifma->ifma_lladdr = llsa;
        ifma->ifma_ifp = ifp;
        ifma->ifma_refcount = 1;
        ifma->ifma_protospec = NULL;
        rt_newmaddrmsg(RTM_NEWMADDR, ifma);

        TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
        if (retifma)
                *retifma = ifma;

        if (llsa != NULL) {
                TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
                        if (sa_equal(ifma->ifma_addr, llsa))
                                break;
                }
                if (ifma) {
                        ifma->ifma_refcount++;
                } else {
                        ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
                        dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
                        bcopy(llsa, dupsa, llsa->sa_len);
                        ifma->ifma_addr = dupsa;
                        ifma->ifma_ifp = ifp;
                        ifma->ifma_refcount = 1;
                        TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
                }
        }
        /*
         * We are certain we have added something, so call down to the
         * interface to let them know about it.
         */
        if (ifp->if_ioctl)
                ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);

        return 0;
}

int
if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
    struct ifmultiaddr **retifma)
{
        int error;

        ifnet_serialize_all(ifp);
        error = if_addmulti_serialized(ifp, sa, retifma);
        ifnet_deserialize_all(ifp);

        return error;
}

/*
 * Remove a reference to a multicast address on this interface.  Yell
 * if the request does not match an existing membership.
 */
static int
if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
{
        struct ifmultiaddr *ifma;

        ASSERT_IFNET_SERIALIZED_ALL(ifp);

        TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
                if (sa_equal(sa, ifma->ifma_addr))
                        break;
        if (ifma == NULL)
                return ENOENT;

        if (ifma->ifma_refcount > 1) {
                ifma->ifma_refcount--;
                return 0;
        }

        rt_newmaddrmsg(RTM_DELMADDR, ifma);
        sa = ifma->ifma_lladdr;
        TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
        /*
         * Make sure the interface driver is notified
         * in the case of a link layer mcast group being left.
         */
        if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
                ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
        kfree(ifma->ifma_addr, M_IFMADDR);
        kfree(ifma, M_IFMADDR);
        if (sa == NULL)
                return 0;

        /*
         * Now look for the link-layer address which corresponds to
         * this network address.  It had been squirreled away in
         * ifma->ifma_lladdr for this purpose (so we don't have
         * to call ifp->if_resolvemulti() again), and we saved that
         * value in sa above.  If some nasty deleted the
         * link-layer address out from underneath us, we can deal because
         * the address we stored was is not the same as the one which was
         * in the record for the link-layer address.  (So we don't complain
         * in that case.)
         */
        TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
                if (sa_equal(sa, ifma->ifma_addr))
                        break;
        if (ifma == NULL)
                return 0;

        if (ifma->ifma_refcount > 1) {
                ifma->ifma_refcount--;
                return 0;
        }

        TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
        ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
        kfree(ifma->ifma_addr, M_IFMADDR);
        kfree(sa, M_IFMADDR);
        kfree(ifma, M_IFMADDR);

        return 0;
}

int
if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
{
        int error;

        ifnet_serialize_all(ifp);
        error = if_delmulti_serialized(ifp, sa);
        ifnet_deserialize_all(ifp);

        return error;
}

/*
 * Delete all multicast group membership for an interface.
 * Should be used to quickly flush all multicast filters.
 */
void
if_delallmulti_serialized(struct ifnet *ifp)
{
        struct ifmultiaddr *ifma, mark;
        struct sockaddr sa;

        ASSERT_IFNET_SERIALIZED_ALL(ifp);

        bzero(&sa, sizeof(sa));
        sa.sa_family = AF_UNSPEC;
        sa.sa_len = sizeof(sa);

        bzero(&mark, sizeof(mark));
        mark.ifma_addr = &sa;

        TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
        while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
                TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
                TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
                    ifma_link);

                if (ifma->ifma_addr->sa_family == AF_UNSPEC)
                        continue;

                if_delmulti_serialized(ifp, ifma->ifma_addr);
        }
        TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
}


/*
 * Set the link layer address on an interface.
 *
 * At this time we only support certain types of interfaces,
 * and we don't allow the length of the address to change.
 */
int
if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
{
        struct sockaddr_dl *sdl;
        struct ifreq ifr;

        sdl = IF_LLSOCKADDR(ifp);
        if (sdl == NULL)
                return (EINVAL);
        if (len != sdl->sdl_alen)       /* don't allow length to change */
                return (EINVAL);
        switch (ifp->if_type) {
        case IFT_ETHER:                 /* these types use struct arpcom */
        case IFT_XETHER:
        case IFT_L2VLAN:
        case IFT_IEEE8023ADLAG:
                bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
                bcopy(lladdr, LLADDR(sdl), len);
                break;
        default:
                return (ENODEV);
        }
        /*
         * If the interface is already up, we need
         * to re-init it in order to reprogram its
         * address filter.
         */
        ifnet_serialize_all(ifp);
        if ((ifp->if_flags & IFF_UP) != 0) {
#ifdef INET
                struct ifaddr_container *ifac;
#endif

                ifp->if_flags &= ~IFF_UP;
                ifr.ifr_flags = ifp->if_flags;
                ifr.ifr_flagshigh = ifp->if_flags >> 16;
                ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
                              NULL);
                ifp->if_flags |= IFF_UP;
                ifr.ifr_flags = ifp->if_flags;
                ifr.ifr_flagshigh = ifp->if_flags >> 16;
                ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
                                 NULL);
#ifdef INET
                /*
                 * Also send gratuitous ARPs to notify other nodes about
                 * the address change.
                 */
                TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
                        struct ifaddr *ifa = ifac->ifa;

                        if (ifa->ifa_addr != NULL &&
                            ifa->ifa_addr->sa_family == AF_INET)
                                arp_gratuitous(ifp, ifa);
                }
#endif
        }
        ifnet_deserialize_all(ifp);
        return (0);
}

struct ifmultiaddr *
ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
{
        struct ifmultiaddr *ifma;

        /* TODO: need ifnet_serialize_main */
        ifnet_serialize_all(ifp);
        TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
                if (sa_equal(ifma->ifma_addr, sa))
                        break;
        ifnet_deserialize_all(ifp);

        return ifma;
}

/*
 * This function locates the first real ethernet MAC from a network
 * card and loads it into node, returning 0 on success or ENOENT if
 * no suitable interfaces were found.  It is used by the uuid code to
 * generate a unique 6-byte number.
 */
int
if_getanyethermac(uint16_t *node, int minlen)
{
        struct ifnet *ifp;
        struct sockaddr_dl *sdl;

        ifnet_lock();
        TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
                if (ifp->if_type != IFT_ETHER)
                        continue;
                sdl = IF_LLSOCKADDR(ifp);
                if (sdl->sdl_alen < minlen)
                        continue;
                bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
                      minlen);
                ifnet_unlock();
                return(0);
        }
        ifnet_unlock();
        return (ENOENT);
}

/*
 * The name argument must be a pointer to storage which will last as
 * long as the interface does.  For physical devices, the result of
 * device_get_name(dev) is a good choice and for pseudo-devices a
 * static string works well.
 */
void
if_initname(struct ifnet *ifp, const char *name, int unit)
{
        ifp->if_dname = name;
        ifp->if_dunit = unit;
        if (unit != IF_DUNIT_NONE)
                ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
        else
                strlcpy(ifp->if_xname, name, IFNAMSIZ);
}

int
if_printf(struct ifnet *ifp, const char *fmt, ...)
{
        __va_list ap;
        int retval;

        retval = kprintf("%s: ", ifp->if_xname);
        __va_start(ap, fmt);
        retval += kvprintf(fmt, ap);
        __va_end(ap);
        return (retval);
}

struct ifnet *
if_alloc(uint8_t type)
{
        struct ifnet *ifp;
        size_t size;

        /*
         * XXX temporary hack until arpcom is setup in if_l2com
         */
        if (type == IFT_ETHER)
                size = sizeof(struct arpcom);
        else
                size = sizeof(struct ifnet);

        ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);

        ifp->if_type = type;

        if (if_com_alloc[type] != NULL) {
                ifp->if_l2com = if_com_alloc[type](type, ifp);
                if (ifp->if_l2com == NULL) {
                        kfree(ifp, M_IFNET);
                        return (NULL);
                }
        }
        return (ifp);
}

void
if_free(struct ifnet *ifp)
{
        kfree(ifp, M_IFNET);
}

void
ifq_set_classic(struct ifaltq *ifq)
{
        ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
            ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
}

void
ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
    ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
{
        int q;

        KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
        KASSERT(enqueue != NULL, ("enqueue is not specified"));
        KASSERT(dequeue != NULL, ("dequeue is not specified"));
        KASSERT(request != NULL, ("request is not specified"));

        ifq->altq_mapsubq = mapsubq;
        for (q = 0; q < ifq->altq_subq_cnt; ++q) {
                struct ifaltq_subque *ifsq = &ifq->altq_subq[q];

                ifsq->ifsq_enqueue = enqueue;
                ifsq->ifsq_dequeue = dequeue;
                ifsq->ifsq_request = request;
        }
}

static void
ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
{

        classq_add(&ifsq->ifsq_norm, m);
        ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
}

static void
ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
{

        classq_add(&ifsq->ifsq_prio, m);
        ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
        ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
}

static struct mbuf *
ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
{
        struct mbuf *m;

        m = classq_get(&ifsq->ifsq_norm);
        if (m != NULL)
                ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
        return (m);
}

static struct mbuf *
ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
{
        struct mbuf *m;

        m = classq_get(&ifsq->ifsq_prio);
        if (m != NULL) {
                ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
                ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
        }
        return (m);
}

int
ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
    struct altq_pktattr *pa __unused)
{

        M_ASSERTPKTHDR(m);
again:
        if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
            ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
                struct mbuf *m_drop;

                if (m->m_flags & M_PRIO) {
                        m_drop = NULL;
                        if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
                            ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
                                /* Try dropping some from normal queue. */
                                m_drop = ifsq_norm_dequeue(ifsq);
                        }
                        if (m_drop == NULL)
                                m_drop = ifsq_prio_dequeue(ifsq);
                } else {
                        m_drop = ifsq_norm_dequeue(ifsq);
                }
                if (m_drop != NULL) {
                        IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
                        m_freem(m_drop);
                        goto again;
                }
                /*
                 * No old packets could be dropped!
                 * NOTE: Caller increases oqdrops.
                 */
                m_freem(m);
                return (ENOBUFS);
        } else {
                if (m->m_flags & M_PRIO)
                        ifsq_prio_enqueue(ifsq, m);
                else
                        ifsq_norm_enqueue(ifsq, m);
                return (0);
        }
}

struct mbuf *
ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
{
        struct mbuf *m;

        switch (op) {
        case ALTDQ_POLL:
                m = classq_head(&ifsq->ifsq_prio);
                if (m == NULL)
                        m = classq_head(&ifsq->ifsq_norm);
                break;

        case ALTDQ_REMOVE:
                m = ifsq_prio_dequeue(ifsq);
                if (m == NULL)
                        m = ifsq_norm_dequeue(ifsq);
                break;

        default:
                panic("unsupported ALTQ dequeue op: %d", op);
        }
        return m;
}

int
ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
{
        switch (req) {
        case ALTRQ_PURGE:
                for (;;) {
                        struct mbuf *m;

                        m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
                        if (m == NULL)
                                break;
                        m_freem(m);
                }
                break;

        default:
                panic("unsupported ALTQ request: %d", req);
        }
        return 0;
}

static void
ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
{
        struct ifnet *ifp = ifsq_get_ifp(ifsq);
        int running = 0, need_sched;

        /*
         * Try to do direct ifnet.if_start on the subqueue first, if there is
         * contention on the subqueue hardware serializer, ifnet.if_start on
         * the subqueue will be scheduled on the subqueue owner CPU.
         */
        if (!ifsq_tryserialize_hw(ifsq)) {
                /*
                 * Subqueue hardware serializer contention happened,
                 * ifnet.if_start on the subqueue is scheduled on
                 * the subqueue owner CPU, and we keep going.
                 */
                ifsq_ifstart_schedule(ifsq, 1);
                return;
        }

        if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
                ifp->if_start(ifp, ifsq);
                if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
                        running = 1;
        }
        need_sched = ifsq_ifstart_need_schedule(ifsq, running);

        ifsq_deserialize_hw(ifsq);

        if (need_sched) {
                /*
                 * More data need to be transmitted, ifnet.if_start on the
                 * subqueue is scheduled on the subqueue owner CPU, and we
                 * keep going.
                 * NOTE: ifnet.if_start subqueue interlock is not released.
                 */
                ifsq_ifstart_schedule(ifsq, force_sched);
        }
}

/*
 * Subqeue packets staging mechanism:
 *
 * The packets enqueued into the subqueue are staged to a certain amount
 * before the ifnet.if_start on the subqueue is called.  In this way, the
 * driver could avoid writing to hardware registers upon every packet,
 * instead, hardware registers could be written when certain amount of
 * packets are put onto hardware TX ring.  The measurement on several modern
 * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
 * registers writing aggregation could save ~20% CPU time when 18bytes UDP
 * datagrams are transmitted at 1.48Mpps.  The performance improvement by
 * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
 * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
 *
 * Subqueue packets staging is performed for two entry points into drivers'
 * transmission function:
 * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
 * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
 *
 * Subqueue packets staging will be stopped upon any of the following
 * conditions:
 * - If the count of packets enqueued on the current CPU is great than or
 *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
 * - If the total length of packets enqueued on the current CPU is great
 *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
 *   cut from the hardware's MTU mainly bacause a full TCP segment's size
 *   is usually less than hardware's MTU.
 * - ifsq_ifstart_schedule() is not pending on the current CPU and
 *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
 *   released.
 * - The if_start_rollup(), which is registered as low priority netisr
 *   rollup function, is called; probably because no more work is pending
 *   for netisr.
 *
 * NOTE:
 * Currently subqueue packet staging is only performed in netisr threads.
 */
int
ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
{
        struct ifaltq *ifq = &ifp->if_snd;
        struct ifaltq_subque *ifsq;
        int error, start = 0, len, mcast = 0, avoid_start = 0;
        struct ifsubq_stage_head *head = NULL;
        struct ifsubq_stage *stage = NULL;
        struct globaldata *gd = mycpu;
        struct thread *td = gd->gd_curthread;

        crit_enter_quick(td);

        ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
        ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);

        len = m->m_pkthdr.len;
        if (m->m_flags & M_MCAST)
                mcast = 1;

        if (td->td_type == TD_TYPE_NETISR) {
                head = &ifsubq_stage_heads[mycpuid];
                stage = ifsq_get_stage(ifsq, mycpuid);

                stage->stg_cnt++;
                stage->stg_len += len;
                if (stage->stg_cnt < ifsq_stage_cntmax &&
                    stage->stg_len < (ifp->if_mtu - max_protohdr))
                        avoid_start = 1;
        }

        ALTQ_SQ_LOCK(ifsq);
        error = ifsq_enqueue_locked(ifsq, m, pa);
        if (error) {
                IFNET_STAT_INC(ifp, oqdrops, 1);
                if (!ifsq_data_ready(ifsq)) {
                        ALTQ_SQ_UNLOCK(ifsq);
                        crit_exit_quick(td);
                        return error;
                }
                avoid_start = 0;
        }
        if (!ifsq_is_started(ifsq)) {
                if (avoid_start) {
                        ALTQ_SQ_UNLOCK(ifsq);

                        KKASSERT(!error);
                        if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
                                ifsq_stage_insert(head, stage);

                        IFNET_STAT_INC(ifp, obytes, len);
                        if (mcast)
                                IFNET_STAT_INC(ifp, omcasts, 1);
                        crit_exit_quick(td);
                        return error;
                }

                /*
                 * Hold the subqueue interlock of ifnet.if_start
                 */
                ifsq_set_started(ifsq);
                start = 1;
        }
        ALTQ_SQ_UNLOCK(ifsq);

        if (!error) {
                IFNET_STAT_INC(ifp, obytes, len);
                if (mcast)
                        IFNET_STAT_INC(ifp, omcasts, 1);
        }

        if (stage != NULL) {
                if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
                        KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
                        if (!avoid_start) {
                                ifsq_stage_remove(head, stage);
                                ifsq_ifstart_schedule(ifsq, 1);
                        }
                        crit_exit_quick(td);
                        return error;
                }

                if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
                        ifsq_stage_remove(head, stage);
                } else {
                        stage->stg_cnt = 0;
                        stage->stg_len = 0;
                }
        }

        if (!start) {
                crit_exit_quick(td);
                return error;
        }

        ifsq_ifstart_try(ifsq, 0);

        crit_exit_quick(td);
        return error;
}

void *
ifa_create(int size)
{
        struct ifaddr *ifa;
        int i;

        KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));

        ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
        ifa->ifa_containers =
            kmalloc_cachealign(ncpus * sizeof(struct ifaddr_container),
                M_IFADDR, M_INTWAIT | M_ZERO);

        ifa->ifa_ncnt = ncpus;
        for (i = 0; i < ncpus; ++i) {
                struct ifaddr_container *ifac = &ifa->ifa_containers[i];

                ifac->ifa_magic = IFA_CONTAINER_MAGIC;
                ifac->ifa = ifa;
                ifac->ifa_refcnt = 1;
        }
#ifdef IFADDR_DEBUG
        kprintf("alloc ifa %p %d\n", ifa, size);
#endif
        return ifa;
}

void
ifac_free(struct ifaddr_container *ifac, int cpu_id)
{
        struct ifaddr *ifa = ifac->ifa;

        KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
        KKASSERT(ifac->ifa_refcnt == 0);
        KASSERT(ifac->ifa_listmask == 0,
                ("ifa is still on %#x lists", ifac->ifa_listmask));

        ifac->ifa_magic = IFA_CONTAINER_DEAD;

#ifdef IFADDR_DEBUG_VERBOSE
        kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
#endif

        KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
                ("invalid # of ifac, %d", ifa->ifa_ncnt));
        if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
#ifdef IFADDR_DEBUG
                kprintf("free ifa %p\n", ifa);
#endif
                kfree(ifa->ifa_containers, M_IFADDR);
                kfree(ifa, M_IFADDR);
        }
}

static void
ifa_iflink_dispatch(netmsg_t nmsg)
{
        struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
        struct ifaddr *ifa = msg->ifa;
        struct ifnet *ifp = msg->ifp;
        int cpu = mycpuid;
        struct ifaddr_container *ifac;

        crit_enter();

        ifac = &ifa->ifa_containers[cpu];
        ASSERT_IFAC_VALID(ifac);
        KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
                ("ifaddr is on if_addrheads"));

        ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
        if (msg->tail)
                TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
        else
                TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);

        crit_exit();

        netisr_forwardmsg(&nmsg->base, cpu + 1);
}

void
ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
{
        struct netmsg_ifaddr msg;

        netmsg_init(&msg.base, NULL, &curthread->td_msgport,
                    0, ifa_iflink_dispatch);
        msg.ifa = ifa;
        msg.ifp = ifp;
        msg.tail = tail;

        netisr_domsg(&msg.base, 0);
}

static void
ifa_ifunlink_dispatch(netmsg_t nmsg)
{
        struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
        struct ifaddr *ifa = msg->ifa;
        struct ifnet *ifp = msg->ifp;
        int cpu = mycpuid;
        struct ifaddr_container *ifac;

        crit_enter();

        ifac = &ifa->ifa_containers[cpu];
        ASSERT_IFAC_VALID(ifac);
        KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
                ("ifaddr is not on if_addrhead"));

        TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
        ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;

        crit_exit();

        netisr_forwardmsg(&nmsg->base, cpu + 1);
}

void
ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
{
        struct netmsg_ifaddr msg;

        netmsg_init(&msg.base, NULL, &curthread->td_msgport,
                    0, ifa_ifunlink_dispatch);
        msg.ifa = ifa;
        msg.ifp = ifp;

        netisr_domsg(&msg.base, 0);
}

static void
ifa_destroy_dispatch(netmsg_t nmsg)
{
        struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;

        IFAFREE(msg->ifa);
        netisr_forwardmsg(&nmsg->base, mycpuid + 1);
}

void
ifa_destroy(struct ifaddr *ifa)
{
        struct netmsg_ifaddr msg;

        netmsg_init(&msg.base, NULL, &curthread->td_msgport,
                    0, ifa_destroy_dispatch);
        msg.ifa = ifa;

        netisr_domsg(&msg.base, 0);
}

static void
if_start_rollup(void)
{
        struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
        struct ifsubq_stage *stage;

        crit_enter();

        while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
                struct ifaltq_subque *ifsq = stage->stg_subq;
                int is_sched = 0;

                if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
                        is_sched = 1;
                ifsq_stage_remove(head, stage);

                if (is_sched) {
                        ifsq_ifstart_schedule(ifsq, 1);
                } else {
                        int start = 0;

                        ALTQ_SQ_LOCK(ifsq);
                        if (!ifsq_is_started(ifsq)) {
                                /*
                                 * Hold the subqueue interlock of
                                 * ifnet.if_start
                                 */
                                ifsq_set_started(ifsq);
                                start = 1;
                        }
                        ALTQ_SQ_UNLOCK(ifsq);

                        if (start)
                                ifsq_ifstart_try(ifsq, 1);
                }
                KKASSERT((stage->stg_flags &
                    (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
        }

        crit_exit();
}

static void
ifnetinit(void *dummy __unused)
{
        int i;

        for (i = 0; i < ncpus; ++i)
                TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
        netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
}

void
if_register_com_alloc(u_char type,
    if_com_alloc_t *a, if_com_free_t *f)
{

        KASSERT(if_com_alloc[type] == NULL,
            ("if_register_com_alloc: %d already registered", type));
        KASSERT(if_com_free[type] == NULL,
            ("if_register_com_alloc: %d free already registered", type));

        if_com_alloc[type] = a;
        if_com_free[type] = f;
}

void
if_deregister_com_alloc(u_char type)
{

        KASSERT(if_com_alloc[type] != NULL,
            ("if_deregister_com_alloc: %d not registered", type));
        KASSERT(if_com_free[type] != NULL,
            ("if_deregister_com_alloc: %d free not registered", type));
        if_com_alloc[type] = NULL;
        if_com_free[type] = NULL;
}

void
ifq_set_maxlen(struct ifaltq *ifq, int len)
{
        ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
}

int
ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
{
        return ALTQ_SUBQ_INDEX_DEFAULT;
}

int
ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
{

        return (cpuid % ifq->altq_subq_mappriv);
}

static void
ifsq_watchdog(void *arg)
{
        struct ifsubq_watchdog *wd = arg;
        struct ifnet *ifp;

        if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
                goto done;

        ifp = ifsq_get_ifp(wd->wd_subq);
        if (ifnet_tryserialize_all(ifp)) {
                wd->wd_watchdog(wd->wd_subq);
                ifnet_deserialize_all(ifp);
        } else {
                /* try again next timeout */
                wd->wd_timer = 1;
        }
done:
        ifsq_watchdog_reset(wd);
}

static void
ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
{
        callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
            ifsq_get_cpuid(wd->wd_subq));
}

void
ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
    ifsq_watchdog_t watchdog)
{
        callout_init_mp(&wd->wd_callout);
        wd->wd_timer = 0;
        wd->wd_subq = ifsq;
        wd->wd_watchdog = watchdog;
}

void
ifsq_watchdog_start(struct ifsubq_watchdog *wd)
{
        wd->wd_timer = 0;
        ifsq_watchdog_reset(wd);
}

void
ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
{
        wd->wd_timer = 0;
        callout_stop(&wd->wd_callout);
}

void
ifnet_lock(void)
{
        KASSERT(curthread->td_type != TD_TYPE_NETISR,
            ("try holding ifnet lock in netisr"));
        mtx_lock(&ifnet_mtx);
}

void
ifnet_unlock(void)
{
        KASSERT(curthread->td_type != TD_TYPE_NETISR,
            ("try holding ifnet lock in netisr"));
        mtx_unlock(&ifnet_mtx);
}

static struct ifnet_array *
ifnet_array_alloc(int count)
{
        struct ifnet_array *arr;

        arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
            M_IFNET, M_WAITOK);
        arr->ifnet_count = count;

        return arr;
}

static void
ifnet_array_free(struct ifnet_array *arr)
{
        if (arr == &ifnet_array0)
                return;
        kfree(arr, M_IFNET);
}

static struct ifnet_array *
ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
{
        struct ifnet_array *arr;
        int count, i;

        KASSERT(old_arr->ifnet_count >= 0,
            ("invalid ifnet array count %d", old_arr->ifnet_count));
        count = old_arr->ifnet_count + 1;
        arr = ifnet_array_alloc(count);

        /*
         * Save the old ifnet array and append this ifp to the end of
         * the new ifnet array.
         */
        for (i = 0; i < old_arr->ifnet_count; ++i) {
                KASSERT(old_arr->ifnet_arr[i] != ifp,
                    ("%s is already in ifnet array", ifp->if_xname));
                arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
        }
        KASSERT(i == count - 1,
            ("add %s, ifnet array index mismatch, should be %d, but got %d",
             ifp->if_xname, count - 1, i));
        arr->ifnet_arr[i] = ifp;

        return arr;
}

static struct ifnet_array *
ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
{
        struct ifnet_array *arr;
        int count, i, idx, found = 0;

        KASSERT(old_arr->ifnet_count > 0,
            ("invalid ifnet array count %d", old_arr->ifnet_count));
        count = old_arr->ifnet_count - 1;
        arr = ifnet_array_alloc(count);

        /*
         * Save the old ifnet array, but skip this ifp.
         */
        idx = 0;
        for (i = 0; i < old_arr->ifnet_count; ++i) {
                if (old_arr->ifnet_arr[i] == ifp) {
                        KASSERT(!found,
                            ("dup %s is in ifnet array", ifp->if_xname));
                        found = 1;
                        continue;
                }
                KASSERT(idx < count,
                    ("invalid ifnet array index %d, count %d", idx, count));
                arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
                ++idx;
        }
        KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
        KASSERT(idx == count,
            ("del %s, ifnet array count mismatch, should be %d, but got %d ",
             ifp->if_xname, count, idx));

        return arr;
}

const struct ifnet_array *
ifnet_array_get(void)
{
        const struct ifnet_array *ret;

        KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
        ret = ifnet_array;
        /* Make sure 'ret' is really used. */
        cpu_ccfence();
        return (ret);
}

int
ifnet_array_isempty(void)
{
        KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
        if (ifnet_array->ifnet_count == 0)
                return 1;
        else
                return 0;
}

void
ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
{
        struct ifaddr *ifa;

        memset(mark, 0, sizeof(*mark));
        ifa = &mark->ifa;

        mark->ifac.ifa = ifa;

        ifa->ifa_addr = &mark->addr;
        ifa->ifa_dstaddr = &mark->dstaddr;
        ifa->ifa_netmask = &mark->netmask;
        ifa->ifa_ifp = ifp;
}

static int
if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
{

        KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));

        if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
                ring_cnt = ring_cntmax;
        if (ring_cnt > netisr_ncpus)
                ring_cnt = netisr_ncpus;
        return (ring_cnt);
}

static void
if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
{
        int i, offset;

        KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
        KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
            grid, rm->rm_cnt));
        rm->rm_grid = grid;

        offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
        for (i = 0; i < rm->rm_cnt; ++i) {
                rm->rm_cpumap[i] = offset + i;
                KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
                    ("invalid cpumap[%d] = %d, offset %d", i,
                     rm->rm_cpumap[i], offset));
        }
}

static struct if_ringmap *
if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
    uint32_t flags)
{
        struct if_ringmap *rm;
        int i, grid = 0, prev_grid;

        ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
        rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
            M_DEVBUF, M_WAITOK | M_ZERO);

        rm->rm_cnt = ring_cnt;
        if (flags & RINGMAP_FLAG_POWEROF2)
                rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);

        prev_grid = netisr_ncpus;
        for (i = 0; i < netisr_ncpus; ++i) {
                if (netisr_ncpus % (i + 1) != 0)
                        continue;

                grid = netisr_ncpus / (i + 1);
                if (rm->rm_cnt > grid) {
                        grid = prev_grid;
                        break;
                }

                if (rm->rm_cnt > netisr_ncpus / (i + 2))
                        break;
                prev_grid = grid;
        }
        if_ringmap_set_grid(dev, rm, grid);

        return (rm);
}

struct if_ringmap *
if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
{

        return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
            RINGMAP_FLAG_NONE));
}

struct if_ringmap *
if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
{

        return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
            RINGMAP_FLAG_POWEROF2));
}

void
if_ringmap_free(struct if_ringmap *rm)
{

        kfree(rm, M_DEVBUF);
}

/*
 * Align the two ringmaps.
 *
 * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
 *
 * Before:
 *
 * CPU      0  1  2  3   4  5  6  7
 * NIC_RX               n0 n1 n2 n3
 * NIC_TX        N0 N1
 *
 * After:
 *
 * CPU      0  1  2  3   4  5  6  7
 * NIC_RX               n0 n1 n2 n3
 * NIC_TX               N0 N1
 */
void
if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
{

        if (rm0->rm_grid > rm1->rm_grid)
                if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
        else if (rm0->rm_grid < rm1->rm_grid)
                if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
}

void
if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
{
        int subset_grid, cnt, divisor, mod, offset, i;
        struct if_ringmap *subset_rm, *rm;
        int old_rm0_grid, old_rm1_grid;

        if (rm0->rm_grid == rm1->rm_grid)
                return;

        /* Save grid for later use */
        old_rm0_grid = rm0->rm_grid;
        old_rm1_grid = rm1->rm_grid;

        if_ringmap_align(dev, rm0, rm1);

        /*
         * Re-shuffle rings to get more even distribution.
         *
         * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
         *
         * CPU       0  1  2  3   4  5  6  7   8  9 10 11
         *
         * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
         * NIC_TX   A0 A1        B0 B1        C0 C1
         *
         * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
         * NIC_TX         D0 D1        E0 E1        F0 F1
         */

        if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
                cnt = rm0->rm_cnt;
                subset_grid = old_rm1_grid;
                subset_rm = rm1;
                rm = rm0;
        } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
                cnt = rm1->rm_cnt;
                subset_grid = old_rm0_grid;
                subset_rm = rm0;
                rm = rm1;
        } else {
                /* No space to shuffle. */
                return;
        }

        mod = cnt / subset_grid;
        KKASSERT(mod >= 2);
        divisor = netisr_ncpus / rm->rm_grid;
        offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;

        for (i = 0; i < subset_rm->rm_cnt; ++i) {
                subset_rm->rm_cpumap[i] += offset;
                KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
                    ("match: invalid cpumap[%d] = %d, offset %d",
                     i, subset_rm->rm_cpumap[i], offset));
        }
#ifdef INVARIANTS
        for (i = 0; i < subset_rm->rm_cnt; ++i) {
                int j;

                for (j = 0; j < rm->rm_cnt; ++j) {
                        if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
                                break;
                }
                KASSERT(j < rm->rm_cnt,
                    ("subset cpumap[%d] = %d not found in superset",
                     i, subset_rm->rm_cpumap[i]));
        }
#endif
}

int
if_ringmap_count(const struct if_ringmap *rm)
{

        return (rm->rm_cnt);
}

int
if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
{

        KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
        return (rm->rm_cpumap[ring]);
}

void
if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
{
        int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;

        KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
            ("invalid redirect table entries %d", table_nent));

        grid_idx = 0;
        for (i = 0; i < NETISR_CPUMAX; ++i) {
                table[i] = grid_idx++ % rm->rm_cnt;

                if (grid_idx == rm->rm_grid)
                        grid_idx = 0;
        }

        /*
         * Make the ring distributed more evenly for the remainder
         * of each grid.
         *
         * e.g. 12 netisrs, rm contains 8 rings.
         *
         * Redirect table before:
         *
         *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
         *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
         *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
         *  ....
         *
         * Redirect table after being patched (pX, patched entries):
         *
         *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
         *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
         * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
         *  ....
         */
        patch_cnt = rm->rm_grid % rm->rm_cnt;
        if (patch_cnt == 0)
                goto done;
        patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);

        grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
        grid_idx = 0;
        for (i = 0; i < grid_cnt; ++i) {
                int j;

                for (j = 0; j < patch_cnt; ++j) {
                        int fix_idx;

                        fix_idx = (i * rm->rm_grid) + patch_off + j;
                        if (fix_idx >= NETISR_CPUMAX)
                                goto done;
                        table[fix_idx] = grid_idx++ % rm->rm_cnt;
                }
        }
done:
        /*
         * If the device supports larger redirect table, duplicate
         * the first NETISR_CPUMAX entries to the rest of the table,
         * so that it matches upper layer's expectation:
         * (hash & NETISR_CPUMASK) % netisr_ncpus
         */
        ncopy = table_nent / NETISR_CPUMAX;
        for (i = 1; i < ncopy; ++i) {
                memcpy(&table[i * NETISR_CPUMAX], table,
                    NETISR_CPUMAX * sizeof(table[0]));
        }
        if (if_ringmap_dumprdr) {
                for (i = 0; i < table_nent; ++i) {
                        if (i != 0 && i % 16 == 0)
                                kprintf("\n");
                        kprintf("%03d ", table[i]);
                }
                kprintf("\n");
        }
}

int
if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
{
        struct if_ringmap *rm = arg1;
        int i, error = 0;

        for (i = 0; i < rm->rm_cnt; ++i) {
                int cpu = rm->rm_cpumap[i];

                error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
                if (error)
                        break;
        }
        return (error);
}