[mirror_frr.git] / pimd / pim_vxlan.c

/* PIM support for VxLAN BUM flooding
 *
 * Copyright (C) 2019 Cumulus Networks, Inc.
 *
 * This file is part of FRR.
 *
 * FRR is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the
 * Free Software Foundation; either version 2, or (at your option) any
 * later version.
 *
 * FRR is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 */

#include <zebra.h>

#include <hash.h>
#include <jhash.h>
#include <log.h>
#include <prefix.h>
#include <vrf.h>

#include "pimd.h"
#include "pim_iface.h"
#include "pim_memory.h"
#include "pim_oil.h"
#include "pim_register.h"
#include "pim_str.h"
#include "pim_upstream.h"
#include "pim_ifchannel.h"
#include "pim_nht.h"
#include "pim_zebra.h"
#include "pim_vxlan.h"

/* pim-vxlan global info */
struct pim_vxlan vxlan_info, *pim_vxlan_p = &vxlan_info;

static void pim_vxlan_work_timer_setup(bool start);
static void pim_vxlan_set_peerlink_rif(struct pim_instance *pim,
                        struct interface *ifp);

/*************************** vxlan work list **********************************
 * A work list is maintained for staggered generation of pim null register
 * messages for vxlan SG entries that are in a reg_join state.
 *
 * A max of 500 NULL registers are generated at one shot. If paused reg
 * generation continues on the next second and so on till all register
 * messages have been sent out. And the process is restarted every 60s.
 *
 * purpose of this null register generation is to setup the SPT and maintain
 * independent of the presence of overlay BUM traffic.
 ****************************************************************************/
static void pim_vxlan_do_reg_work(void)
{
        struct listnode *listnode;
        int work_cnt = 0;
        struct pim_vxlan_sg *vxlan_sg;
        static int sec_count;

        ++sec_count;

        if (sec_count > PIM_VXLAN_NULL_REG_INTERVAL) {
                sec_count = 0;
                listnode = vxlan_info.next_work ?
                                        vxlan_info.next_work :
                                        vxlan_info.work_list->head;
                if (PIM_DEBUG_VXLAN && listnode)
                        zlog_debug("vxlan SG work %s",
                                vxlan_info.next_work ? "continues" : "starts");
        } else {
                listnode = vxlan_info.next_work;
        }

        for (; listnode; listnode = listnode->next) {
                vxlan_sg = (struct pim_vxlan_sg *)listnode->data;
                if (vxlan_sg->up && (vxlan_sg->up->reg_state == PIM_REG_JOIN)) {
                        if (PIM_DEBUG_VXLAN)
                                zlog_debug("vxlan SG %s periodic NULL register",
                                                vxlan_sg->sg_str);
                        pim_null_register_send(vxlan_sg->up);
                        ++work_cnt;
                }

                if (work_cnt > vxlan_info.max_work_cnt) {
                        vxlan_info.next_work = listnode->next;
                        if (PIM_DEBUG_VXLAN)
                                zlog_debug("vxlan SG %d work items proc and pause",
                                        work_cnt);
                        return;
                }
        }

        if (work_cnt) {
                if (PIM_DEBUG_VXLAN)
                        zlog_debug("vxlan SG %d work items proc", work_cnt);
        }
        vxlan_info.next_work = NULL;
}

/* Staggered work related info is initialized when the first work comes
 * along
 */
static void pim_vxlan_init_work(void)
{
        if (vxlan_info.flags & PIM_VXLANF_WORK_INITED)
                return;

        vxlan_info.max_work_cnt = PIM_VXLAN_WORK_MAX;
        vxlan_info.flags |= PIM_VXLANF_WORK_INITED;
        vxlan_info.work_list = list_new();
        pim_vxlan_work_timer_setup(TRUE /* start */);
}

static void pim_vxlan_add_work(struct pim_vxlan_sg *vxlan_sg)
{
        if (vxlan_sg->flags & PIM_VXLAN_SGF_DEL_IN_PROG) {
                if (PIM_DEBUG_VXLAN)
                        zlog_debug("vxlan SG %s skip work list; del-in-prog",
                                        vxlan_sg->sg_str);
                return;
        }

        pim_vxlan_init_work();

        /* already a part of the work list */
        if (vxlan_sg->work_node)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s work list add",
                                vxlan_sg->sg_str);
        vxlan_sg->work_node = listnode_add(vxlan_info.work_list, vxlan_sg);
        /* XXX: adjust max_work_cnt if needed */
}

static void pim_vxlan_del_work(struct pim_vxlan_sg *vxlan_sg)
{
        if (!vxlan_sg->work_node)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s work list del",
                                vxlan_sg->sg_str);

        if (vxlan_sg->work_node == vxlan_info.next_work)
                vxlan_info.next_work = vxlan_sg->work_node->next;

        list_delete_node(vxlan_info.work_list, vxlan_sg->work_node);
        vxlan_sg->work_node = NULL;
}

void pim_vxlan_update_sg_reg_state(struct pim_instance *pim,
                struct pim_upstream *up, bool reg_join)
{
        struct pim_vxlan_sg *vxlan_sg;

        vxlan_sg = pim_vxlan_sg_find(pim, &up->sg);
        if (!vxlan_sg)
                return;

        /* add the vxlan sg entry to a work list for periodic reg joins.
         * the entry will stay in the list as long as the register state is
         * PIM_REG_JOIN
         */
        if (reg_join)
                pim_vxlan_add_work(vxlan_sg);
        else
                pim_vxlan_del_work(vxlan_sg);
}

static int pim_vxlan_work_timer_cb(struct thread *t)
{
        pim_vxlan_do_reg_work();
        pim_vxlan_work_timer_setup(true /* start */);
        return 0;
}

/* global 1second timer used for periodic processing */
static void pim_vxlan_work_timer_setup(bool start)
{
        THREAD_OFF(vxlan_info.work_timer);
        if (start)
                thread_add_timer(router->master, pim_vxlan_work_timer_cb, NULL,
                        PIM_VXLAN_WORK_TIME, &vxlan_info.work_timer);
}

/**************************** vxlan origination mroutes ***********************
 * For every (local-vtep-ip, bum-mcast-grp) registered by evpn an origination
 * mroute is setup by pimd. The purpose of this mroute is to forward vxlan
 * encapsulated BUM (broadcast, unknown-unicast and unknown-multicast packets
 * over the underlay.)
 *
 * Sample mroute (single VTEP):
 * (27.0.0.7, 239.1.1.100)     Iif: lo      Oifs: uplink-1
 *
 * Sample mroute (anycast VTEP):
 * (36.0.0.9, 239.1.1.100)          Iif: peerlink-3.4094\
 *                                       Oifs: peerlink-3.4094 uplink-1
 ***************************************************************************/
static void pim_vxlan_orig_mr_up_del(struct pim_vxlan_sg *vxlan_sg)
{
        struct pim_upstream *up = vxlan_sg->up;

        if (!up)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s orig mroute-up del",
                        vxlan_sg->sg_str);

        vxlan_sg->up = NULL;
        if (up->flags & PIM_UPSTREAM_FLAG_MASK_SRC_VXLAN_ORIG) {
                /* clear out all the vxlan properties */
                up->flags &= ~(PIM_UPSTREAM_FLAG_MASK_SRC_VXLAN_ORIG |
                        PIM_UPSTREAM_FLAG_MASK_STATIC_IIF |
                        PIM_UPSTREAM_FLAG_MASK_DISABLE_KAT_EXPIRY |
                        PIM_UPSTREAM_FLAG_MASK_FORCE_PIMREG |
                        PIM_UPSTREAM_FLAG_MASK_NO_PIMREG_DATA |
                        PIM_UPSTREAM_FLAG_MASK_ALLOW_IIF_IN_OIL);

                /* We bring things to a grinding halt by force expirying
                 * the kat. Doing this will also remove the reference we
                 * created as a "vxlan" source and delete the upstream entry
                 * if there are no other references.
                 */
                if (PIM_UPSTREAM_FLAG_TEST_SRC_STREAM(up->flags)) {
                        THREAD_OFF(up->t_ka_timer);
                        up = pim_upstream_keep_alive_timer_proc(up);
                } else {
                        /* this is really unexpected as we force vxlan
                         * origination mroutes active sources but just in
                         * case
                         */
                        up = pim_upstream_del(vxlan_sg->pim, up,
                                __PRETTY_FUNCTION__);
                }
                /* if there are other references register the source
                 * for nht
                 */
                if (up)
                        pim_rpf_update(vxlan_sg->pim, up, NULL, 1 /* is_new */);
        }
}

static void pim_vxlan_orig_mr_up_iif_update(struct pim_vxlan_sg *vxlan_sg)
{
        int vif_index;

        /* update MFC with the new IIF */
        pim_upstream_fill_static_iif(vxlan_sg->up, vxlan_sg->iif);
        vif_index = pim_if_find_vifindex_by_ifindex(vxlan_sg->pim,
                        vxlan_sg->iif->ifindex);
        if (vif_index > 0)
                pim_scan_individual_oil(vxlan_sg->up->channel_oil,
                                vif_index);

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s orig mroute-up updated with iif %s vifi %d",
                        vxlan_sg->sg_str,
                        vxlan_sg->iif?vxlan_sg->iif->name:"-", vif_index);

}

/* For every VxLAN BUM multicast group we setup a SG-up that has the following
 * "forced properties" -
 * 1. Directly connected on a DR interface i.e. we must act as an FHR
 * 2. We prime the pump i.e. no multicast data is needed to register this
 *    source with the FHR. To do that we send periodic null registers if
 *    the SG entry is in a register-join state. We also prevent expiry of
 *    KAT.
 * 3. As this SG is setup without data there is no need to register encapsulate
 *    data traffic. This encapsulation is explicitly skipped for the following
 *    reasons -
 *    a) Many levels of encapsulation are needed creating MTU disc challenges.
 *       Overlay BUM is encapsulated in a vxlan/UDP/IP header and then
 *       encapsulated again in a pim-register header.
 *    b) On a vxlan-aa setup both switches rx a copy of each BUM packet. if
 *       they both reg encapsulated traffic the RP will accept the duplicates
 *       as there are no RPF checks for this encapsulated data.
 *    a), b) can be workarounded if needed, but there is really no need because
 *    of (2) i.e. the pump is primed without data.
 */
static void pim_vxlan_orig_mr_up_add(struct pim_vxlan_sg *vxlan_sg)
{
        struct pim_upstream *up;
        int flags = 0;
        struct prefix nht_p;

        if (vxlan_sg->up) {
                /* nothing to do */
                return;
        }

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s orig mroute-up add with iif %s",
                        vxlan_sg->sg_str,
                        vxlan_sg->iif?vxlan_sg->iif->name:"-");

        PIM_UPSTREAM_FLAG_SET_SRC_VXLAN_ORIG(flags);
        /* pin the IIF to lo or peerlink-subinterface and disable NHT */
        PIM_UPSTREAM_FLAG_SET_STATIC_IIF(flags);
        /* Fake traffic by setting SRC_STREAM and starting KAT */
        /* We intentionally skip updating ref count for SRC_STREAM/FHR.
         * Setting SRC_VXLAN should have already created a reference
         * preventing the entry from being deleted
         */
        PIM_UPSTREAM_FLAG_SET_FHR(flags);
        PIM_UPSTREAM_FLAG_SET_SRC_STREAM(flags);
        /* Force pimreg even if non-DR. This is needed on a MLAG setup for
         * VxLAN AA
         */
        PIM_UPSTREAM_FLAG_SET_FORCE_PIMREG(flags);
        /* prevent KAT expiry. we want the MDT setup even if there is no BUM
         * traffic
         */
        PIM_UPSTREAM_FLAG_SET_DISABLE_KAT_EXPIRY(flags);
        /* SPT for vxlan BUM groups is primed and maintained via NULL
         * registers so there is no need to reg-encapsulate
         * vxlan-encapsulated overlay data traffic
         */
        PIM_UPSTREAM_FLAG_SET_NO_PIMREG_DATA(flags);
        /* On a MLAG setup we force a copy to the MLAG peer while also
         * accepting traffic from the peer. To do this we set peerlink-rif as
         * the IIF and also add it to the OIL
         */
        PIM_UPSTREAM_FLAG_SET_ALLOW_IIF_IN_OIL(flags);

        /* XXX: todo: defer pim_upstream add if pim is not enabled on the iif */
        up = pim_upstream_find(vxlan_sg->pim, &vxlan_sg->sg);
        if (up) {
                /* if the iif is set to something other than the vxlan_sg->iif
                 * we must dereg the old nexthop and force to new "static"
                 * iif
                 */
                if (!PIM_UPSTREAM_FLAG_TEST_STATIC_IIF(up->flags)) {
                        nht_p.family = AF_INET;
                        nht_p.prefixlen = IPV4_MAX_BITLEN;
                        nht_p.u.prefix4 = up->upstream_addr;
                        pim_delete_tracked_nexthop(vxlan_sg->pim,
                                &nht_p, up, NULL);
                }
                pim_upstream_ref(up, flags, __PRETTY_FUNCTION__);
                vxlan_sg->up = up;
                pim_vxlan_orig_mr_up_iif_update(vxlan_sg);
        } else {
                up = pim_upstream_add(vxlan_sg->pim, &vxlan_sg->sg,
                                vxlan_sg->iif, flags,
                                __PRETTY_FUNCTION__, NULL);
                vxlan_sg->up = up;
        }

        if (!up) {
                if (PIM_DEBUG_VXLAN)
                        zlog_debug("vxlan SG %s orig mroute-up add failed",
                                        vxlan_sg->sg_str);
                return;
        }

        pim_upstream_keep_alive_timer_start(up, vxlan_sg->pim->keep_alive_time);

        /* register the source with the RP */
        if (up->reg_state == PIM_REG_NOINFO) {
                pim_register_join(up);
                pim_null_register_send(up);
        }

        /* update the inherited OIL */
        pim_upstream_inherited_olist(vxlan_sg->pim, up);
}

static void pim_vxlan_orig_mr_oif_add(struct pim_vxlan_sg *vxlan_sg)
{
        if (!vxlan_sg->up || !vxlan_sg->orig_oif)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s oif %s add",
                        vxlan_sg->sg_str, vxlan_sg->orig_oif->name);

        vxlan_sg->flags |= PIM_VXLAN_SGF_OIF_INSTALLED;
        pim_channel_add_oif(vxlan_sg->up->channel_oil,
                vxlan_sg->orig_oif, PIM_OIF_FLAG_PROTO_VXLAN);
}

static void pim_vxlan_orig_mr_oif_del(struct pim_vxlan_sg *vxlan_sg)
{
        struct interface *orig_oif;

        orig_oif = vxlan_sg->orig_oif;
        vxlan_sg->orig_oif = NULL;

        if (!(vxlan_sg->flags & PIM_VXLAN_SGF_OIF_INSTALLED))
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s oif %s del",
                        vxlan_sg->sg_str, orig_oif->name);

        vxlan_sg->flags &= ~PIM_VXLAN_SGF_OIF_INSTALLED;
        pim_channel_del_oif(vxlan_sg->up->channel_oil,
                orig_oif, PIM_OIF_FLAG_PROTO_VXLAN);
}

static inline struct interface *pim_vxlan_orig_mr_oif_get(
                struct pim_instance *pim)
{
        return (vxlan_mlag.flags & PIM_VXLAN_MLAGF_ENABLED) ?
                pim->vxlan.peerlink_rif : NULL;
}

/* Single VTEPs: IIF for the vxlan-origination-mroutes is lo or vrf-dev (if
 * the mroute is in a non-default vrf).
 * Anycast VTEPs: IIF is the MLAG ISL/peerlink.
 */
static inline struct interface *pim_vxlan_orig_mr_iif_get(
                struct pim_instance *pim)
{
        return ((vxlan_mlag.flags & PIM_VXLAN_MLAGF_ENABLED) &&
                        pim->vxlan.peerlink_rif) ?
                pim->vxlan.peerlink_rif : pim->vxlan.default_iif;
}

static bool pim_vxlan_orig_mr_add_is_ok(struct pim_vxlan_sg *vxlan_sg)
{
        struct pim_interface *pim_ifp;

        vxlan_sg->iif = pim_vxlan_orig_mr_iif_get(vxlan_sg->pim);
        if (!vxlan_sg->iif)
                return false;

        pim_ifp = (struct pim_interface *)vxlan_sg->iif->info;
        if (!pim_ifp || (pim_ifp->mroute_vif_index < 0))
                return false;

        return true;
}

static void pim_vxlan_orig_mr_install(struct pim_vxlan_sg *vxlan_sg)
{
        pim_vxlan_orig_mr_up_add(vxlan_sg);

        vxlan_sg->orig_oif = pim_vxlan_orig_mr_oif_get(vxlan_sg->pim);
        pim_vxlan_orig_mr_oif_add(vxlan_sg);
}

static void pim_vxlan_orig_mr_add(struct pim_vxlan_sg *vxlan_sg)
{
        if (!pim_vxlan_orig_mr_add_is_ok(vxlan_sg))
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s orig-mr add", vxlan_sg->sg_str);

        pim_vxlan_orig_mr_install(vxlan_sg);
}

static void pim_vxlan_orig_mr_del(struct pim_vxlan_sg *vxlan_sg)
{
        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s orig-mr del", vxlan_sg->sg_str);

        pim_vxlan_orig_mr_oif_del(vxlan_sg);
        pim_vxlan_orig_mr_up_del(vxlan_sg);
}

static void pim_vxlan_orig_mr_iif_update(struct hash_backet *backet, void *arg)
{
        struct interface *ifp = (struct interface *)arg;
        struct pim_vxlan_sg *vxlan_sg = (struct pim_vxlan_sg *)backet->data;
        struct interface *old_iif = vxlan_sg->iif;

        if (!pim_vxlan_is_orig_mroute(vxlan_sg))
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s iif changed from %s to %s",
                                vxlan_sg->sg_str,
                                old_iif ? old_iif->name : "-",
                                ifp ? ifp->name : "-");

        if (pim_vxlan_orig_mr_add_is_ok(vxlan_sg)) {
                if (vxlan_sg->up) {
                        /* upstream exists but iif changed */
                        pim_vxlan_orig_mr_up_iif_update(vxlan_sg);
                } else {
                        /* install mroute */
                        pim_vxlan_orig_mr_install(vxlan_sg);
                }
        } else {
                pim_vxlan_orig_mr_del(vxlan_sg);
        }
}

/**************************** vxlan termination mroutes ***********************
 * For every bum-mcast-grp registered by evpn a *G termination
 * mroute is setup by pimd. The purpose of this mroute is to pull down vxlan
 * packets with the bum-mcast-grp dip from the underlay and terminate the
 * tunnel. This is done by including the vxlan termination device (ipmr-lo) in
 * its OIL. The vxlan de-capsulated packets are subject to subsequent overlay
 * bridging.
 *
 * Sample mroute:
 * (0.0.0.0, 239.1.1.100)     Iif: uplink-1      Oifs: ipmr-lo, uplink-1
 *****************************************************************************/
struct pim_interface *pim_vxlan_get_term_ifp(struct pim_instance *pim)
{
        return pim->vxlan.term_if ?
                (struct pim_interface *)pim->vxlan.term_if->info : NULL;
}

static void pim_vxlan_term_mr_oif_add(struct pim_vxlan_sg *vxlan_sg)
{
        if (vxlan_sg->flags & PIM_VXLAN_SGF_OIF_INSTALLED)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s term-oif %s add",
                        vxlan_sg->sg_str, vxlan_sg->term_oif->name);

        if (pim_ifchannel_local_membership_add(vxlan_sg->term_oif,
                                &vxlan_sg->sg)) {
                vxlan_sg->flags |= PIM_VXLAN_SGF_OIF_INSTALLED;
        } else {
                zlog_warn("vxlan SG %s term-oif %s add failed",
                        vxlan_sg->sg_str, vxlan_sg->term_oif->name);
        }
}

static void pim_vxlan_term_mr_oif_del(struct pim_vxlan_sg *vxlan_sg)
{
        if (!(vxlan_sg->flags & PIM_VXLAN_SGF_OIF_INSTALLED))
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s oif %s del",
                        vxlan_sg->sg_str, vxlan_sg->term_oif->name);

        vxlan_sg->flags &= ~PIM_VXLAN_SGF_OIF_INSTALLED;
        pim_ifchannel_local_membership_del(vxlan_sg->term_oif, &vxlan_sg->sg);
}

static void pim_vxlan_term_mr_up_add(struct pim_vxlan_sg *vxlan_sg)
{
        struct pim_upstream *up;
        int flags = 0;

        if (vxlan_sg->up) {
                /* nothing to do */
                return;
        }

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s term mroute-up add",
                        vxlan_sg->sg_str);

        PIM_UPSTREAM_FLAG_SET_SRC_VXLAN_TERM(flags);
        /* enable MLAG designated-forwarder election on termination mroutes */
        PIM_UPSTREAM_FLAG_SET_MLAG_VXLAN(flags);

        up = pim_upstream_add(vxlan_sg->pim, &vxlan_sg->sg,
                        NULL /* iif */, flags,
                        __PRETTY_FUNCTION__, NULL);
        vxlan_sg->up = up;

        if (!up) {
                zlog_warn("vxlan SG %s term mroute-up add failed",
                        vxlan_sg->sg_str);
        }
}

static void pim_vxlan_term_mr_up_del(struct pim_vxlan_sg *vxlan_sg)
{
        struct pim_upstream *up = vxlan_sg->up;

        if (!up)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s term mroute-up del",
                        vxlan_sg->sg_str);
        vxlan_sg->up = NULL;
        if (up->flags & PIM_UPSTREAM_FLAG_MASK_SRC_VXLAN_TERM) {
                /* clear out all the vxlan related flags */
                up->flags &= ~(PIM_UPSTREAM_FLAG_MASK_SRC_VXLAN_TERM |
                        PIM_UPSTREAM_FLAG_MASK_MLAG_VXLAN);

                pim_upstream_del(vxlan_sg->pim, up,
                                __PRETTY_FUNCTION__);
        }
}

static void pim_vxlan_term_mr_add(struct pim_vxlan_sg *vxlan_sg)
{
        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s term mroute add", vxlan_sg->sg_str);

        vxlan_sg->term_oif = vxlan_sg->pim->vxlan.term_if;
        if (!vxlan_sg->term_oif)
                /* defer termination mroute till we have a termination device */
                return;

        pim_vxlan_term_mr_up_add(vxlan_sg);
        /* set up local membership for the term-oif */
        pim_vxlan_term_mr_oif_add(vxlan_sg);
}

static void pim_vxlan_term_mr_del(struct pim_vxlan_sg *vxlan_sg)
{
        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s term mroute del", vxlan_sg->sg_str);

        /* remove local membership associated with the term oif */
        pim_vxlan_term_mr_oif_del(vxlan_sg);
        /* remove references to the upstream entry */
        pim_vxlan_term_mr_up_del(vxlan_sg);
}

/************************** vxlan SG cache management ************************/
static unsigned int pim_vxlan_sg_hash_key_make(void *p)
{
        struct pim_vxlan_sg *vxlan_sg = p;

        return (jhash_2words(vxlan_sg->sg.src.s_addr,
                                vxlan_sg->sg.grp.s_addr, 0));
}

static bool pim_vxlan_sg_hash_eq(const void *p1, const void *p2)
{
        const struct pim_vxlan_sg *sg1 = p1;
        const struct pim_vxlan_sg *sg2 = p2;

        return ((sg1->sg.src.s_addr == sg2->sg.src.s_addr)
                        && (sg1->sg.grp.s_addr == sg2->sg.grp.s_addr));
}

static struct pim_vxlan_sg *pim_vxlan_sg_new(struct pim_instance *pim,
                struct prefix_sg *sg)
{
        struct pim_vxlan_sg *vxlan_sg;

        vxlan_sg = XCALLOC(MTYPE_PIM_VXLAN_SG, sizeof(*vxlan_sg));

        vxlan_sg->pim = pim;
        vxlan_sg->sg = *sg;
        pim_str_sg_set(sg, vxlan_sg->sg_str);

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s alloc", vxlan_sg->sg_str);

        vxlan_sg = hash_get(pim->vxlan.sg_hash, vxlan_sg, hash_alloc_intern);

        return vxlan_sg;
}

struct pim_vxlan_sg *pim_vxlan_sg_find(struct pim_instance *pim,
                struct prefix_sg *sg)
{
        struct pim_vxlan_sg lookup;

        lookup.sg = *sg;
        return hash_lookup(pim->vxlan.sg_hash, &lookup);
}

struct pim_vxlan_sg *pim_vxlan_sg_add(struct pim_instance *pim,
                struct prefix_sg *sg)
{
        struct pim_vxlan_sg *vxlan_sg;

        vxlan_sg = pim_vxlan_sg_find(pim, sg);
        if (vxlan_sg)
                return vxlan_sg;

        vxlan_sg = pim_vxlan_sg_new(pim, sg);

        if (pim_vxlan_is_orig_mroute(vxlan_sg))
                pim_vxlan_orig_mr_add(vxlan_sg);
        else
                pim_vxlan_term_mr_add(vxlan_sg);

        return vxlan_sg;
}

void pim_vxlan_sg_del(struct pim_instance *pim, struct prefix_sg *sg)
{
        struct pim_vxlan_sg *vxlan_sg;

        vxlan_sg = pim_vxlan_sg_find(pim, sg);
        if (!vxlan_sg)
                return;

        vxlan_sg->flags |= PIM_VXLAN_SGF_DEL_IN_PROG;

        pim_vxlan_del_work(vxlan_sg);

        if (pim_vxlan_is_orig_mroute(vxlan_sg))
                pim_vxlan_orig_mr_del(vxlan_sg);
        else
                pim_vxlan_term_mr_del(vxlan_sg);

        hash_release(vxlan_sg->pim->vxlan.sg_hash, vxlan_sg);

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s free", vxlan_sg->sg_str);

        XFREE(MTYPE_PIM_VXLAN_SG, vxlan_sg);
}

/******************************* MLAG handling *******************************/
/* The peerlink sub-interface is added as an OIF to the origination-mroute.
 * This is done to send a copy of the multicast-vxlan encapsulated traffic
 * to the MLAG peer which may mroute it over the underlay if there are any
 * interested receivers.
 */
static void pim_vxlan_sg_peerlink_update(struct hash_backet *backet, void *arg)
{
        struct interface *new_oif = (struct interface *)arg;
        struct pim_vxlan_sg *vxlan_sg = (struct pim_vxlan_sg *)backet->data;

        if (!pim_vxlan_is_orig_mroute(vxlan_sg))
                return;

        if (vxlan_sg->orig_oif == new_oif)
                return;

        pim_vxlan_orig_mr_oif_del(vxlan_sg);

        vxlan_sg->orig_oif = new_oif;
        pim_vxlan_orig_mr_oif_add(vxlan_sg);
}

/* In the case of anycast VTEPs the VTEP-PIP must be used as the
 * register source.
 */
bool pim_vxlan_get_register_src(struct pim_instance *pim,
                struct pim_upstream *up, struct in_addr *src_p)
{
        if (!(vxlan_mlag.flags & PIM_VXLAN_MLAGF_ENABLED))
                return true;

        /* if address is not available suppress the pim-register */
        if (vxlan_mlag.reg_addr.s_addr == INADDR_ANY)
                return false;

        *src_p = vxlan_mlag.reg_addr;
        return true;
}

void pim_vxlan_mlag_update(bool enable, bool peer_state, uint32_t role,
                                struct interface *peerlink_rif,
                                struct in_addr *reg_addr)
{
        struct pim_instance *pim;
        struct interface *old_oif;
        struct interface *new_oif;
        char addr_buf[INET_ADDRSTRLEN];
        struct pim_interface *pim_ifp = NULL;

        if (PIM_DEBUG_VXLAN) {
                inet_ntop(AF_INET, reg_addr,
                                addr_buf, INET_ADDRSTRLEN);
                zlog_debug("vxlan MLAG update %s state %s role %d rif %s addr %s",
                                enable ? "enable" : "disable",
                                peer_state ? "up" : "down",
                                role,
                                peerlink_rif ? peerlink_rif->name : "-",
                                addr_buf);
        }

        /* XXX: for now vxlan termination is only possible in the default VRF
         * when that changes this will need to change to iterate all VRFs
         */
        pim = pim_get_pim_instance(VRF_DEFAULT);

        old_oif = pim_vxlan_orig_mr_oif_get(pim);

        if (enable)
                vxlan_mlag.flags |= PIM_VXLAN_MLAGF_ENABLED;
        else
                vxlan_mlag.flags &= ~PIM_VXLAN_MLAGF_ENABLED;

        if (vxlan_mlag.peerlink_rif != peerlink_rif)
                vxlan_mlag.peerlink_rif = peerlink_rif;

        vxlan_mlag.reg_addr = *reg_addr;
        vxlan_mlag.peer_state = peer_state;
        vxlan_mlag.role = role;

        /* process changes */
        if (vxlan_mlag.peerlink_rif)
                pim_ifp = (struct pim_interface *)vxlan_mlag.peerlink_rif->info;
        if ((vxlan_mlag.flags & PIM_VXLAN_MLAGF_ENABLED) &&
                        pim_ifp && (pim_ifp->mroute_vif_index > 0))
                pim_vxlan_set_peerlink_rif(pim, peerlink_rif);
        else
                pim_vxlan_set_peerlink_rif(pim, NULL);

        new_oif = pim_vxlan_orig_mr_oif_get(pim);
        if (old_oif != new_oif)
                hash_iterate(pim->vxlan.sg_hash, pim_vxlan_sg_peerlink_update,
                        new_oif);
}

/****************************** misc callbacks *******************************/
void pim_vxlan_config_write(struct vty *vty, char *spaces, int *writes)
{
        char addr_buf[INET_ADDRSTRLEN];

        if ((vxlan_mlag.flags & PIM_VXLAN_MLAGF_ENABLED) &&
                        vxlan_mlag.peerlink_rif) {

                inet_ntop(AF_INET, &vxlan_mlag.reg_addr,
                                addr_buf, sizeof(addr_buf));
                vty_out(vty,
                        "%sip pim mlag %s role %s state %s addr %s\n",
                        spaces,
                        vxlan_mlag.peerlink_rif->name,
                        (vxlan_mlag.role == PIM_VXLAN_MLAG_ROLE_PRIMARY) ?
                                "primary":"secondary",
                        vxlan_mlag.peer_state ? "up" : "down",
                        addr_buf);
                *writes += 1;
        }
}

static void pim_vxlan_set_default_iif(struct pim_instance *pim,
                                struct interface *ifp)
{
        struct interface *old_iif;

        if (pim->vxlan.default_iif == ifp)
                return;

        old_iif = pim->vxlan.default_iif;
        if (PIM_DEBUG_VXLAN)
                zlog_debug("%s: vxlan default iif changed from %s to %s",
                        __PRETTY_FUNCTION__,
                        old_iif ? old_iif->name : "-",
                        ifp ? ifp->name : "-");

        old_iif = pim_vxlan_orig_mr_iif_get(pim);
        pim->vxlan.default_iif = ifp;
        ifp = pim_vxlan_orig_mr_iif_get(pim);
        if (old_iif == ifp)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("%s: vxlan orig iif changed from %s to %s",
                        __PRETTY_FUNCTION__, old_iif ? old_iif->name : "-",
                        ifp ? ifp->name : "-");

        /* add/del upstream entries for the existing vxlan SG when the
         * interface becomes available
         */
        if (pim->vxlan.sg_hash)
                hash_iterate(pim->vxlan.sg_hash,
                                pim_vxlan_orig_mr_iif_update, ifp);
}

static void pim_vxlan_set_peerlink_rif(struct pim_instance *pim,
                        struct interface *ifp)
{
        struct interface *old_iif;

        if (pim->vxlan.peerlink_rif == ifp)
                return;

        old_iif = pim->vxlan.peerlink_rif;
        if (PIM_DEBUG_VXLAN)
                zlog_debug("%s: vxlan peerlink_rif changed from %s to %s",
                        __PRETTY_FUNCTION__, old_iif ? old_iif->name : "-",
                        ifp ? ifp->name : "-");

        old_iif = pim_vxlan_orig_mr_iif_get(pim);
        pim->vxlan.peerlink_rif = ifp;
        ifp = pim_vxlan_orig_mr_iif_get(pim);
        if (old_iif == ifp)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("%s: vxlan orig iif changed from %s to %s",
                        __PRETTY_FUNCTION__, old_iif ? old_iif->name : "-",
                        ifp ? ifp->name : "-");

        /* add/del upstream entries for the existing vxlan SG when the
         * interface becomes available
         */
        if (pim->vxlan.sg_hash)
                hash_iterate(pim->vxlan.sg_hash,
                                pim_vxlan_orig_mr_iif_update, ifp);
}

void pim_vxlan_add_vif(struct interface *ifp)
{
        struct pim_interface *pim_ifp = ifp->info;
        struct pim_instance *pim = pim_ifp->pim;

        if (pim->vrf_id != VRF_DEFAULT)
                return;

        if (if_is_loopback_or_vrf(ifp))
                pim_vxlan_set_default_iif(pim, ifp);

        if (vxlan_mlag.flags & PIM_VXLAN_MLAGF_ENABLED &&
                        (ifp == vxlan_mlag.peerlink_rif))
                pim_vxlan_set_peerlink_rif(pim, ifp);
}

void pim_vxlan_del_vif(struct interface *ifp)
{
        struct pim_interface *pim_ifp = ifp->info;
        struct pim_instance *pim = pim_ifp->pim;

        if (pim->vrf_id != VRF_DEFAULT)
                return;

        if (pim->vxlan.default_iif == ifp)
                pim_vxlan_set_default_iif(pim, NULL);

        if (pim->vxlan.peerlink_rif == ifp)
                pim_vxlan_set_peerlink_rif(pim, NULL);
}

static void pim_vxlan_term_mr_oif_update(struct hash_backet *backet, void *arg)
{
        struct interface *ifp = (struct interface *)arg;
        struct pim_vxlan_sg *vxlan_sg = (struct pim_vxlan_sg *)backet->data;

        if (pim_vxlan_is_orig_mroute(vxlan_sg))
                return;

        if (vxlan_sg->term_oif == ifp)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan SG %s term oif changed from %s to %s",
                        vxlan_sg->sg_str,
                        vxlan_sg->term_oif ? vxlan_sg->term_oif->name : "-",
                        ifp ? ifp->name : "-");

        pim_vxlan_term_mr_del(vxlan_sg);
        vxlan_sg->term_oif = ifp;
        pim_vxlan_term_mr_add(vxlan_sg);
}

void pim_vxlan_add_term_dev(struct pim_instance *pim,
                struct interface *ifp)
{
        struct pim_interface *pim_ifp;

        if (pim->vxlan.term_if == ifp)
                return;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan term oif changed from %s to %s",
                        pim->vxlan.term_if ? pim->vxlan.term_if->name : "-",
                        ifp->name);

        /* enable pim on the term ifp */
        pim_ifp = (struct pim_interface *)ifp->info;
        if (pim_ifp) {
                PIM_IF_DO_PIM(pim_ifp->options);
        } else {
                pim_ifp = pim_if_new(ifp, false /*igmp*/, true /*pim*/,
                                false /*pimreg*/, true /*vxlan_term*/);
                /* ensure that pimreg existss before using the newly created
                 * vxlan termination device
                 */
                pim_if_create_pimreg(pim);
        }

        pim->vxlan.term_if = ifp;

        if (pim->vxlan.sg_hash)
                hash_iterate(pim_ifp->pim->vxlan.sg_hash,
                                pim_vxlan_term_mr_oif_update, ifp);
}

void pim_vxlan_del_term_dev(struct pim_instance *pim)
{
        struct interface *ifp = pim->vxlan.term_if;
        struct pim_interface *pim_ifp;

        if (PIM_DEBUG_VXLAN)
                zlog_debug("vxlan term oif changed from %s to -", ifp->name);

        pim->vxlan.term_if = NULL;

        if (pim->vxlan.sg_hash)
                hash_iterate(pim->vxlan.sg_hash,
                                pim_vxlan_term_mr_oif_update, NULL);

        pim_ifp = (struct pim_interface *)ifp->info;
        if (pim_ifp) {
                PIM_IF_DONT_PIM(pim_ifp->options);
                if (!PIM_IF_TEST_IGMP(pim_ifp->options))
                        pim_if_delete(ifp);
        }

}

void pim_vxlan_init(struct pim_instance *pim)
{
        char hash_name[64];

        snprintf(hash_name, sizeof(hash_name),
                "PIM %s vxlan SG hash", pim->vrf->name);
        pim->vxlan.sg_hash = hash_create(pim_vxlan_sg_hash_key_make,
                        pim_vxlan_sg_hash_eq, hash_name);
}

void pim_vxlan_exit(struct pim_instance *pim)
{
        if (pim->vxlan.sg_hash) {
                hash_clean(pim->vxlan.sg_hash, NULL);
                hash_free(pim->vxlan.sg_hash);
                pim->vxlan.sg_hash = NULL;
        }
}