diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-09 13:16:35 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-09 13:16:35 +0000 |
commit | e2bbf175a2184bd76f6c54ccf8456babeb1a46fc (patch) | |
tree | f0b76550d6e6f500ada964a3a4ee933a45e5a6f1 /bgpd/bgp_evpn_private.h | |
parent | Initial commit. (diff) | |
download | frr-e2bbf175a2184bd76f6c54ccf8456babeb1a46fc.tar.xz frr-e2bbf175a2184bd76f6c54ccf8456babeb1a46fc.zip |
Adding upstream version 9.1.upstream/9.1
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'bgpd/bgp_evpn_private.h')
-rw-r--r-- | bgpd/bgp_evpn_private.h | 763 |
1 files changed, 763 insertions, 0 deletions
diff --git a/bgpd/bgp_evpn_private.h b/bgpd/bgp_evpn_private.h new file mode 100644 index 0000000..5af99af --- /dev/null +++ b/bgpd/bgp_evpn_private.h @@ -0,0 +1,763 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* BGP EVPN internal definitions + * Copyright (C) 2017 Cumulus Networks, Inc. + */ + +#ifndef _BGP_EVPN_PRIVATE_H +#define _BGP_EVPN_PRIVATE_H + +#include "vxlan.h" +#include "zebra.h" + +#include "bgpd/bgpd.h" +#include "bgpd/bgp_ecommunity.h" + +#define RT_ADDRSTRLEN 28 + +/* EVPN prefix lengths. This represents the sizeof struct evpn_addr + * in bits */ +#define EVPN_ROUTE_PREFIXLEN (sizeof(struct evpn_addr) * 8) + +/* EVPN route RD buffer length */ +#define BGP_EVPN_PREFIX_RD_LEN 100 + +/* packet sizes for EVPN routes */ +/* Type-1 route should be 25 bytes + * RD (8), ESI (10), eth-tag (4), vni (3) + */ +#define BGP_EVPN_TYPE1_PSIZE 25 +/* Type-4 route should be either 23 or 35 bytes + * RD (8), ESI (10), ip-len (1), ip (4 or 16) + */ +#define BGP_EVPN_TYPE4_V4_PSIZE 23 +#define BGP_EVPN_TYPE4_V6_PSIZE 34 + +RB_HEAD(bgp_es_evi_rb_head, bgp_evpn_es_evi); +RB_PROTOTYPE(bgp_es_evi_rb_head, bgp_evpn_es_evi, rb_node, + bgp_es_evi_rb_cmp); +/* + * Hash table of EVIs. Right now, the only type of EVI supported is with + * VxLAN encapsulation, hence each EVI corresponds to a L2 VNI. + * The VNIs are not "created" through BGP but through some other interface + * on the system. This table stores VNIs that BGP comes to know as present + * on the system (through interaction with zebra) as well as pre-configured + * VNIs (which need to be defined in the system to become "live"). + */ +struct bgpevpn { + vni_t vni; + vrf_id_t tenant_vrf_id; + ifindex_t svi_ifindex; + uint32_t flags; +#define VNI_FLAG_CFGD 0x1 /* VNI is user configured */ +#define VNI_FLAG_LIVE 0x2 /* VNI is "live" */ +#define VNI_FLAG_RD_CFGD 0x4 /* RD is user configured. */ +#define VNI_FLAG_IMPRT_CFGD 0x8 /* Import RT is user configured */ +#define VNI_FLAG_EXPRT_CFGD 0x10 /* Export RT is user configured */ +#define VNI_FLAG_USE_TWO_LABELS 0x20 /* Attach both L2-VNI and L3-VNI if + needed for this VPN */ + + struct bgp *bgp_vrf; /* back pointer to the vrf instance */ + + /* Flag to indicate if we are + * advertising the g/w mac ip for + * this VNI*/ + uint8_t advertise_gw_macip; + + /* Flag to indicate if we are + * advertising subnet for this VNI */ + uint8_t advertise_subnet; + + /* Flag to indicate if we are advertising the svi mac ip for this VNI*/ + uint8_t advertise_svi_macip; + + /* Id for deriving the RD + * automatically for this VNI */ + uint16_t rd_id; + + /* RD for this VNI. */ + struct prefix_rd prd; + char *prd_pretty; + + /* Route type 3 field */ + struct in_addr originator_ip; + + /* PIM-SM MDT group for BUM flooding */ + struct in_addr mcast_grp; + + /* Import and Export RTs. */ + struct list *import_rtl; + struct list *export_rtl; + + /* + * EVPN route that uses gateway IP overlay index as its nexthop + * needs to do a recursive lookup. + * A remote MAC/IP entry should be present for the gateway IP. + * Maintain a hash of the addresses received via remote MAC/IP routes + * for efficient gateway IP recursive lookup in this EVI + */ + struct hash *remote_ip_hash; + + /* Route tables for EVPN routes for + * this VNI. */ + struct bgp_table *ip_table; + struct bgp_table *mac_table; + + /* RB tree of ES-EVIs */ + struct bgp_es_evi_rb_head es_evi_rb_tree; + + /* List of local ESs */ + struct list *local_es_evi_list; + + QOBJ_FIELDS; +}; + +DECLARE_QOBJ_TYPE(bgpevpn); + +/* Mapping of Import RT to VNIs. + * The Import RTs of all VNIs are maintained in a hash table with each + * RT linking to all VNIs that will import routes matching this RT. + */ +struct irt_node { + /* RT */ + struct ecommunity_val rt; + + /* List of VNIs importing routes matching this RT. */ + struct list *vnis; +}; + +/* Mapping of Import RT to VRFs. + * The Import RTs of all VRFss are maintained in a hash table with each + * RT linking to all VRFs that will import routes matching this RT. + */ +struct vrf_irt_node { + /* RT */ + struct ecommunity_val rt; + + /* List of VNIs importing routes matching this RT. */ + struct list *vrfs; +}; + + +#define RT_TYPE_IMPORT 1 +#define RT_TYPE_EXPORT 2 +#define RT_TYPE_BOTH 3 + +#define EVPN_DAD_DEFAULT_TIME 180 /* secs */ +#define EVPN_DAD_DEFAULT_MAX_MOVES 5 /* default from RFC 7432 */ +#define EVPN_DAD_DEFAULT_AUTO_RECOVERY_TIME 1800 /* secs */ + +struct bgp_evpn_info { + /* enable disable dup detect */ + bool dup_addr_detect; + + /* Detection time(M) */ + int dad_time; + /* Detection max moves(N) */ + uint32_t dad_max_moves; + /* Permanent freeze */ + bool dad_freeze; + /* Recovery time */ + uint32_t dad_freeze_time; + + /* EVPN enable - advertise svi macip routes */ + int advertise_svi_macip; + + /* MAC-VRF Site-of-Origin + * - added to all routes exported from L2VNI + * - Type-2/3 routes with matching SoO not imported into L2VNI + * - Type-2/5 routes with matching SoO not imported into L3VNI + */ + struct ecommunity *soo; + + /* PIP feature knob */ + bool advertise_pip; + /* PIP IP (sys ip) */ + struct in_addr pip_ip; + struct in_addr pip_ip_static; + /* PIP MAC (sys MAC) */ + struct ethaddr pip_rmac; + struct ethaddr pip_rmac_static; + struct ethaddr pip_rmac_zebra; + bool is_anycast_mac; +}; + +/* This structure defines an entry in remote_ip_hash */ +struct evpn_remote_ip { + struct ipaddr addr; + struct list *macip_path_list; +}; + +/* + * Wrapper struct for l3 RT's + */ +struct vrf_route_target { + /* flags based on config to determine how RTs are handled */ + uint8_t flags; +#define BGP_VRF_RT_AUTO (1 << 0) +#define BGP_VRF_RT_WILD (1 << 1) + + struct ecommunity *ecom; +}; + +static inline int is_vrf_rd_configured(struct bgp *bgp_vrf) +{ + return (CHECK_FLAG(bgp_vrf->vrf_flags, BGP_VRF_RD_CFGD)); +} + +static inline int bgp_evpn_vrf_rd_matches_existing(struct bgp *bgp_vrf, + struct prefix_rd *prd) +{ + return (memcmp(&bgp_vrf->vrf_prd.val, prd->val, ECOMMUNITY_SIZE) == 0); +} + +static inline vni_t bgpevpn_get_l3vni(struct bgpevpn *vpn) +{ + return vpn->bgp_vrf ? vpn->bgp_vrf->l3vni : 0; +} + +static inline void bgpevpn_get_rmac(struct bgpevpn *vpn, struct ethaddr *rmac) +{ + memset(rmac, 0, sizeof(struct ethaddr)); + if (!vpn->bgp_vrf) + return; + memcpy(rmac, &vpn->bgp_vrf->rmac, sizeof(struct ethaddr)); +} + +static inline struct list *bgpevpn_get_vrf_export_rtl(struct bgpevpn *vpn) +{ + if (!vpn->bgp_vrf) + return NULL; + + return vpn->bgp_vrf->vrf_export_rtl; +} + +static inline struct list *bgpevpn_get_vrf_import_rtl(struct bgpevpn *vpn) +{ + if (!vpn->bgp_vrf) + return NULL; + + return vpn->bgp_vrf->vrf_import_rtl; +} + +extern void bgp_evpn_es_evi_vrf_ref(struct bgpevpn *vpn); +extern void bgp_evpn_es_evi_vrf_deref(struct bgpevpn *vpn); + +static inline void bgpevpn_unlink_from_l3vni(struct bgpevpn *vpn) +{ + /* bail if vpn is not associated to bgp_vrf */ + if (!vpn->bgp_vrf) + return; + + UNSET_FLAG(vpn->flags, VNI_FLAG_USE_TWO_LABELS); + listnode_delete(vpn->bgp_vrf->l2vnis, vpn); + + bgp_evpn_es_evi_vrf_deref(vpn); + + /* remove the backpointer to the vrf instance */ + bgp_unlock(vpn->bgp_vrf); + vpn->bgp_vrf = NULL; +} + +static inline void bgpevpn_link_to_l3vni(struct bgpevpn *vpn) +{ + struct bgp *bgp_vrf = NULL; + + /* bail if vpn is already associated to vrf */ + if (vpn->bgp_vrf) + return; + + bgp_vrf = bgp_lookup_by_vrf_id(vpn->tenant_vrf_id); + if (!bgp_vrf) + return; + + /* associate the vpn to the bgp_vrf instance */ + vpn->bgp_vrf = bgp_lock(bgp_vrf); + listnode_add_sort(bgp_vrf->l2vnis, vpn); + + /* + * If L3VNI is configured, + * check if we are advertising two labels for this vpn + */ + if (bgp_vrf->l3vni && + !CHECK_FLAG(bgp_vrf->vrf_flags, BGP_VRF_L3VNI_PREFIX_ROUTES_ONLY)) + SET_FLAG(vpn->flags, VNI_FLAG_USE_TWO_LABELS); + + bgp_evpn_es_evi_vrf_ref(vpn); +} + +static inline int is_vni_configured(struct bgpevpn *vpn) +{ + return (CHECK_FLAG(vpn->flags, VNI_FLAG_CFGD)); +} + +static inline int is_vni_live(struct bgpevpn *vpn) +{ + return (CHECK_FLAG(vpn->flags, VNI_FLAG_LIVE)); +} + +static inline int is_l3vni_live(struct bgp *bgp_vrf) +{ + return (bgp_vrf->l3vni && bgp_vrf->l3vni_svi_ifindex); +} + +static inline int is_rd_configured(struct bgpevpn *vpn) +{ + return (CHECK_FLAG(vpn->flags, VNI_FLAG_RD_CFGD)); +} + +static inline int bgp_evpn_rd_matches_existing(struct bgpevpn *vpn, + struct prefix_rd *prd) +{ + return (memcmp(&vpn->prd.val, prd->val, ECOMMUNITY_SIZE) == 0); +} + +static inline int is_import_rt_configured(struct bgpevpn *vpn) +{ + return (CHECK_FLAG(vpn->flags, VNI_FLAG_IMPRT_CFGD)); +} + +static inline int is_export_rt_configured(struct bgpevpn *vpn) +{ + return (CHECK_FLAG(vpn->flags, VNI_FLAG_EXPRT_CFGD)); +} + +static inline void encode_es_rt_extcomm(struct ecommunity_val *eval, + struct ethaddr *mac) +{ + memset(eval, 0, sizeof(struct ecommunity_val)); + eval->val[0] = ECOMMUNITY_ENCODE_EVPN; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_ES_IMPORT_RT; + memcpy(&eval->val[2], mac, ETH_ALEN); +} + +static inline void encode_df_elect_extcomm(struct ecommunity_val *eval, + uint16_t pref) +{ + memset(eval, 0, sizeof(*eval)); + eval->val[0] = ECOMMUNITY_ENCODE_EVPN; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_DF_ELECTION; + eval->val[2] = EVPN_MH_DF_ALG_PREF; + eval->val[6] = (pref >> 8) & 0xff; + eval->val[7] = pref & 0xff; +} + +static inline void encode_esi_label_extcomm(struct ecommunity_val *eval, + bool single_active) +{ + memset(eval, 0, sizeof(struct ecommunity_val)); + eval->val[0] = ECOMMUNITY_ENCODE_EVPN; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_ESI_LABEL; + if (single_active) + eval->val[2] |= (1 << 0); +} + +static inline void encode_rmac_extcomm(struct ecommunity_val *eval, + struct ethaddr *rmac) +{ + memset(eval, 0, sizeof(*eval)); + eval->val[0] = ECOMMUNITY_ENCODE_EVPN; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_ROUTERMAC; + memcpy(&eval->val[2], rmac, ETH_ALEN); +} + +static inline void encode_default_gw_extcomm(struct ecommunity_val *eval) +{ + memset(eval, 0, sizeof(*eval)); + eval->val[0] = ECOMMUNITY_ENCODE_OPAQUE; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_DEF_GW; +} + +static inline void encode_mac_mobility_extcomm(int static_mac, uint32_t seq, + struct ecommunity_val *eval) +{ + memset(eval, 0, sizeof(*eval)); + eval->val[0] = ECOMMUNITY_ENCODE_EVPN; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_MACMOBILITY; + if (static_mac) + eval->val[2] = ECOMMUNITY_EVPN_SUBTYPE_MACMOBILITY_FLAG_STICKY; + eval->val[4] = (seq >> 24) & 0xff; + eval->val[5] = (seq >> 16) & 0xff; + eval->val[6] = (seq >> 8) & 0xff; + eval->val[7] = seq & 0xff; +} + +static inline void encode_na_flag_extcomm(struct ecommunity_val *eval, + uint8_t na_flag, bool proxy) +{ + memset(eval, 0, sizeof(*eval)); + eval->val[0] = ECOMMUNITY_ENCODE_EVPN; + eval->val[1] = ECOMMUNITY_EVPN_SUBTYPE_ND; + if (na_flag) + eval->val[2] |= ECOMMUNITY_EVPN_SUBTYPE_ND_ROUTER_FLAG; + if (proxy) + eval->val[2] |= ECOMMUNITY_EVPN_SUBTYPE_PROXY_FLAG; +} + +static inline void ip_prefix_from_type5_prefix(const struct prefix_evpn *evp, + struct prefix *ip) +{ + memset(ip, 0, sizeof(struct prefix)); + if (is_evpn_prefix_ipaddr_v4(evp)) { + ip->family = AF_INET; + ip->prefixlen = evp->prefix.prefix_addr.ip_prefix_length; + memcpy(&(ip->u.prefix4), &(evp->prefix.prefix_addr.ip.ip), + IPV4_MAX_BYTELEN); + } else if (is_evpn_prefix_ipaddr_v6(evp)) { + ip->family = AF_INET6; + ip->prefixlen = evp->prefix.prefix_addr.ip_prefix_length; + memcpy(&(ip->u.prefix6), &(evp->prefix.prefix_addr.ip.ip), + IPV6_MAX_BYTELEN); + } +} + +static inline int is_evpn_prefix_default(const struct prefix *evp) +{ + if (evp->family != AF_EVPN) + return 0; + + return ((evp->u.prefix_evpn.prefix_addr.ip_prefix_length == 0) ? + 1 : 0); +} + +static inline void ip_prefix_from_type2_prefix(const struct prefix_evpn *evp, + struct prefix *ip) +{ + memset(ip, 0, sizeof(struct prefix)); + if (is_evpn_prefix_ipaddr_v4(evp)) { + ip->family = AF_INET; + ip->prefixlen = IPV4_MAX_BITLEN; + memcpy(&(ip->u.prefix4), &(evp->prefix.macip_addr.ip.ip), + IPV4_MAX_BYTELEN); + } else if (is_evpn_prefix_ipaddr_v6(evp)) { + ip->family = AF_INET6; + ip->prefixlen = IPV6_MAX_BITLEN; + memcpy(&(ip->u.prefix6), &(evp->prefix.macip_addr.ip.ip), + IPV6_MAX_BYTELEN); + } +} + +static inline void ip_prefix_from_evpn_prefix(const struct prefix_evpn *evp, + struct prefix *ip) +{ + if (evp->prefix.route_type == BGP_EVPN_MAC_IP_ROUTE) + ip_prefix_from_type2_prefix(evp, ip); + else if (evp->prefix.route_type == BGP_EVPN_IP_PREFIX_ROUTE) + ip_prefix_from_type5_prefix(evp, ip); +} + +static inline void build_evpn_type2_prefix(struct prefix_evpn *p, + struct ethaddr *mac, + struct ipaddr *ip) +{ + memset(p, 0, sizeof(struct prefix_evpn)); + p->family = AF_EVPN; + p->prefixlen = EVPN_ROUTE_PREFIXLEN; + p->prefix.route_type = BGP_EVPN_MAC_IP_ROUTE; + memcpy(&p->prefix.macip_addr.mac.octet, mac->octet, ETH_ALEN); + p->prefix.macip_addr.ip.ipa_type = IPADDR_NONE; + memcpy(&p->prefix.macip_addr.ip, ip, sizeof(*ip)); +} + +static inline void +build_type5_prefix_from_ip_prefix(struct prefix_evpn *evp, + const struct prefix *ip_prefix) +{ + struct ipaddr ip; + + memset(&ip, 0, sizeof(struct ipaddr)); + if (ip_prefix->family == AF_INET) { + ip.ipa_type = IPADDR_V4; + memcpy(&ip.ipaddr_v4, &ip_prefix->u.prefix4, + sizeof(struct in_addr)); + } else { + ip.ipa_type = IPADDR_V6; + memcpy(&ip.ipaddr_v6, &ip_prefix->u.prefix6, + sizeof(struct in6_addr)); + } + + memset(evp, 0, sizeof(struct prefix_evpn)); + evp->family = AF_EVPN; + evp->prefixlen = EVPN_ROUTE_PREFIXLEN; + evp->prefix.route_type = BGP_EVPN_IP_PREFIX_ROUTE; + evp->prefix.prefix_addr.ip_prefix_length = ip_prefix->prefixlen; + evp->prefix.prefix_addr.ip.ipa_type = ip.ipa_type; + memcpy(&evp->prefix.prefix_addr.ip, &ip, sizeof(struct ipaddr)); +} + +static inline void build_evpn_type3_prefix(struct prefix_evpn *p, + struct in_addr originator_ip) +{ + memset(p, 0, sizeof(struct prefix_evpn)); + p->family = AF_EVPN; + p->prefixlen = EVPN_ROUTE_PREFIXLEN; + p->prefix.route_type = BGP_EVPN_IMET_ROUTE; + p->prefix.imet_addr.ip.ipa_type = IPADDR_V4; + p->prefix.imet_addr.ip.ipaddr_v4 = originator_ip; +} + +static inline void build_evpn_type4_prefix(struct prefix_evpn *p, + esi_t *esi, + struct in_addr originator_ip) +{ + memset(p, 0, sizeof(struct prefix_evpn)); + p->family = AF_EVPN; + p->prefixlen = EVPN_ROUTE_PREFIXLEN; + p->prefix.route_type = BGP_EVPN_ES_ROUTE; + p->prefix.es_addr.ip_prefix_length = IPV4_MAX_BITLEN; + p->prefix.es_addr.ip.ipa_type = IPADDR_V4; + p->prefix.es_addr.ip.ipaddr_v4 = originator_ip; + memcpy(&p->prefix.es_addr.esi, esi, sizeof(esi_t)); +} + +static inline void build_evpn_type1_prefix(struct prefix_evpn *p, + uint32_t eth_tag, + esi_t *esi, + struct in_addr originator_ip) +{ + memset(p, 0, sizeof(struct prefix_evpn)); + p->family = AF_EVPN; + p->prefixlen = EVPN_ROUTE_PREFIXLEN; + p->prefix.route_type = BGP_EVPN_AD_ROUTE; + p->prefix.ead_addr.eth_tag = eth_tag; + p->prefix.ead_addr.ip.ipa_type = IPADDR_V4; + p->prefix.ead_addr.ip.ipaddr_v4 = originator_ip; + memcpy(&p->prefix.ead_addr.esi, esi, sizeof(esi_t)); +} + +static inline void evpn_type1_prefix_global_copy(struct prefix_evpn *global_p, + const struct prefix_evpn *vni_p) +{ + memcpy(global_p, vni_p, sizeof(*global_p)); + global_p->prefix.ead_addr.ip.ipa_type = 0; + global_p->prefix.ead_addr.ip.ipaddr_v4.s_addr = INADDR_ANY; + global_p->prefix.ead_addr.frag_id = 0; +} + +/* EAD prefix in the global table doesn't include the VTEP-IP so + * we need to create a different copy for the VNI + */ +static inline struct prefix_evpn * +evpn_type1_prefix_vni_ip_copy(struct prefix_evpn *vni_p, + const struct prefix_evpn *global_p, + struct in_addr originator_ip) +{ + memcpy(vni_p, global_p, sizeof(*vni_p)); + vni_p->prefix.ead_addr.ip.ipa_type = IPADDR_V4; + vni_p->prefix.ead_addr.ip.ipaddr_v4 = originator_ip; + + return vni_p; +} + +static inline void evpn_type2_prefix_global_copy( + struct prefix_evpn *global_p, const struct prefix_evpn *vni_p, + const struct ethaddr *mac, const struct ipaddr *ip) +{ + memcpy(global_p, vni_p, sizeof(*global_p)); + + if (mac) + global_p->prefix.macip_addr.mac = *mac; + + if (ip) + global_p->prefix.macip_addr.ip = *ip; +} + +static inline void +evpn_type2_prefix_vni_ip_copy(struct prefix_evpn *vni_p, + const struct prefix_evpn *global_p) +{ + memcpy(vni_p, global_p, sizeof(*vni_p)); + memset(&vni_p->prefix.macip_addr.mac, 0, sizeof(struct ethaddr)); +} + +static inline void +evpn_type2_prefix_vni_mac_copy(struct prefix_evpn *vni_p, + const struct prefix_evpn *global_p) +{ + memcpy(vni_p, global_p, sizeof(*vni_p)); + memset(&vni_p->prefix.macip_addr.ip, 0, sizeof(struct ipaddr)); +} + +/* Get MAC of path_info prefix */ +static inline struct ethaddr * +evpn_type2_path_info_get_mac(const struct bgp_path_info *local_pi) +{ + assert(local_pi->extra && local_pi->extra->evpn); + return &local_pi->extra->evpn->vni_info.mac; +} + +/* Get IP of path_info prefix */ +static inline struct ipaddr * +evpn_type2_path_info_get_ip(const struct bgp_path_info *local_pi) +{ + assert(local_pi->extra && local_pi->extra->evpn); + return &local_pi->extra->evpn->vni_info.ip; +} + +/* Set MAC of path_info prefix */ +static inline void evpn_type2_path_info_set_mac(struct bgp_path_info *local_pi, + const struct ethaddr mac) +{ + assert(local_pi->extra && local_pi->extra->evpn); + local_pi->extra->evpn->vni_info.mac = mac; +} + +/* Set IP of path_info prefix */ +static inline void evpn_type2_path_info_set_ip(struct bgp_path_info *local_pi, + const struct ipaddr ip) +{ + assert(local_pi->extra && local_pi->extra->evpn); + local_pi->extra->evpn->vni_info.ip = ip; +} + +/* Is the IP empty for the RT's dest? */ +static inline bool is_evpn_type2_dest_ipaddr_none(const struct bgp_dest *dest) +{ + const struct prefix_evpn *evp = + (const struct prefix_evpn *)bgp_dest_get_prefix(dest); + + assert(evp->prefix.route_type == BGP_EVPN_MAC_IP_ROUTE); + return is_evpn_prefix_ipaddr_none(evp); +} + +static inline int evpn_default_originate_set(struct bgp *bgp, afi_t afi, + safi_t safi) +{ + if (afi == AFI_IP && + CHECK_FLAG(bgp->af_flags[AFI_L2VPN][SAFI_EVPN], + BGP_L2VPN_EVPN_DEFAULT_ORIGINATE_IPV4)) + return 1; + else if (afi == AFI_IP6 && + CHECK_FLAG(bgp->af_flags[AFI_L2VPN][SAFI_EVPN], + BGP_L2VPN_EVPN_DEFAULT_ORIGINATE_IPV6)) + return 1; + return 0; +} + +static inline void es_get_system_mac(esi_t *esi, + struct ethaddr *mac) +{ + /* + * for type-1 and type-3 ESIs, + * the system mac starts at val[1] + */ + memcpy(mac, &esi->val[1], ETH_ALEN); +} + +static inline bool bgp_evpn_is_svi_macip_enabled(struct bgpevpn *vpn) +{ + struct bgp *bgp_evpn = NULL; + + bgp_evpn = bgp_get_evpn(); + + return (bgp_evpn->evpn_info->advertise_svi_macip || + vpn->advertise_svi_macip); +} + +static inline bool bgp_evpn_is_path_local(struct bgp *bgp, + struct bgp_path_info *pi) +{ + return (pi->peer == bgp->peer_self + && pi->type == ZEBRA_ROUTE_BGP + && pi->sub_type == BGP_ROUTE_STATIC); +} + +extern struct zclient *zclient; + +extern void bgp_evpn_install_uninstall_default_route(struct bgp *bgp_vrf, + afi_t afi, safi_t safi, + bool add); +extern void evpn_rt_delete_auto(struct bgp *bgp, vni_t vni, struct list *rtl, + bool is_l3); +extern void bgp_evpn_configure_export_rt_for_vrf(struct bgp *bgp_vrf, + struct ecommunity *ecomadd); +extern void bgp_evpn_configure_export_auto_rt_for_vrf(struct bgp *bgp_vrf); +extern void bgp_evpn_unconfigure_export_rt_for_vrf(struct bgp *bgp_vrf, + struct ecommunity *ecomdel); +extern void bgp_evpn_unconfigure_export_auto_rt_for_vrf(struct bgp *bgp_vrf); +extern void bgp_evpn_configure_import_rt_for_vrf(struct bgp *bgp_vrf, + struct ecommunity *ecomadd, + bool is_wildcard); +extern void bgp_evpn_configure_import_auto_rt_for_vrf(struct bgp *bgp_vrf); +extern void bgp_evpn_unconfigure_import_rt_for_vrf(struct bgp *bgp_vrf, + struct ecommunity *ecomdel); +extern void bgp_evpn_unconfigure_import_auto_rt_for_vrf(struct bgp *bgp_vrf); +extern int bgp_evpn_handle_export_rt_change(struct bgp *bgp, + struct bgpevpn *vpn); +extern void bgp_evpn_handle_autort_change(struct bgp *bgp); +extern void bgp_evpn_handle_vrf_rd_change(struct bgp *bgp_vrf, int withdraw); +extern void bgp_evpn_handle_rd_change(struct bgp *bgp, struct bgpevpn *vpn, + int withdraw); +void bgp_evpn_handle_global_macvrf_soo_change(struct bgp *bgp, + struct ecommunity *new_soo); +extern int bgp_evpn_install_routes(struct bgp *bgp, struct bgpevpn *vpn); +extern int bgp_evpn_uninstall_routes(struct bgp *bgp, struct bgpevpn *vpn); +extern void bgp_evpn_map_vrf_to_its_rts(struct bgp *bgp_vrf); +extern void bgp_evpn_unmap_vrf_from_its_rts(struct bgp *bgp_vrf); +extern void bgp_evpn_map_vni_to_its_rts(struct bgp *bgp, struct bgpevpn *vpn); +extern void bgp_evpn_unmap_vni_from_its_rts(struct bgp *bgp, + struct bgpevpn *vpn); +extern void bgp_evpn_derive_auto_rt_import(struct bgp *bgp, + struct bgpevpn *vpn); +extern void bgp_evpn_derive_auto_rt_export(struct bgp *bgp, + struct bgpevpn *vpn); +extern void bgp_evpn_derive_auto_rd(struct bgp *bgp, struct bgpevpn *vpn); +extern void bgp_evpn_derive_auto_rd_for_vrf(struct bgp *bgp); +extern struct bgpevpn *bgp_evpn_lookup_vni(struct bgp *bgp, vni_t vni); +extern struct bgpevpn *bgp_evpn_new(struct bgp *bgp, vni_t vni, + struct in_addr originator_ip, + vrf_id_t tenant_vrf_id, + struct in_addr mcast_grp, + ifindex_t svi_ifindex); +extern void bgp_evpn_free(struct bgp *bgp, struct bgpevpn *vpn); +extern bool bgp_evpn_lookup_l3vni_l2vni_table(vni_t vni); +extern int update_routes_for_vni(struct bgp *bgp, struct bgpevpn *vpn); +extern void delete_evpn_route_entry(struct bgp *bgp, afi_t afi, safi_t safi, + struct bgp_dest *dest, + struct bgp_path_info **pi); +int vni_list_cmp(void *p1, void *p2); +extern int evpn_route_select_install(struct bgp *bgp, struct bgpevpn *vpn, + struct bgp_dest *dest); +extern struct bgp_dest * +bgp_evpn_global_node_get(struct bgp_table *table, afi_t afi, safi_t safi, + const struct prefix_evpn *evp, struct prefix_rd *prd, + const struct bgp_path_info *local_pi); +extern struct bgp_dest *bgp_evpn_global_node_lookup( + struct bgp_table *table, safi_t safi, const struct prefix_evpn *evp, + struct prefix_rd *prd, const struct bgp_path_info *local_pi); +extern struct bgp_dest * +bgp_evpn_vni_ip_node_get(struct bgp_table *const table, + const struct prefix_evpn *evp, + const struct bgp_path_info *parent_pi); +extern struct bgp_dest * +bgp_evpn_vni_ip_node_lookup(const struct bgp_table *const table, + const struct prefix_evpn *evp, + const struct bgp_path_info *parent_pi); +extern struct bgp_dest * +bgp_evpn_vni_mac_node_get(struct bgp_table *const table, + const struct prefix_evpn *evp, + const struct bgp_path_info *parent_pi); +extern struct bgp_dest * +bgp_evpn_vni_mac_node_lookup(const struct bgp_table *const table, + const struct prefix_evpn *evp, + const struct bgp_path_info *parent_pi); +extern struct bgp_dest * +bgp_evpn_vni_node_get(struct bgpevpn *vpn, const struct prefix_evpn *p, + const struct bgp_path_info *parent_pi); +extern struct bgp_dest * +bgp_evpn_vni_node_lookup(const struct bgpevpn *vpn, const struct prefix_evpn *p, + const struct bgp_path_info *parent_pi); + +extern void bgp_evpn_import_route_in_vrfs(struct bgp_path_info *pi, int import); +extern void bgp_evpn_update_type2_route_entry(struct bgp *bgp, + struct bgpevpn *vpn, + struct bgp_dest *rn, + struct bgp_path_info *local_pi, + const char *caller); +extern int bgp_evpn_route_entry_install_if_vrf_match(struct bgp *bgp_vrf, + struct bgp_path_info *pi, + int install); +extern void bgp_evpn_import_type2_route(struct bgp_path_info *pi, int import); +extern void bgp_evpn_xxport_delete_ecomm(void *val); +extern int bgp_evpn_route_target_cmp(struct ecommunity *ecom1, + struct ecommunity *ecom2); +#endif /* _BGP_EVPN_PRIVATE_H */ |