bgpd/bgp_evpn_mh.h

   1 /* EVPN header for multihoming procedures
   2  *
   3  * Copyright (C) 2019 Cumulus Networks
   4  * Anuradha Karuppiah
   5  *
   6  * This file is part of FRRouting.
   7  *
   8  * FRRouting is free software; you can redistribute it and/or modify it
   9  * under the terms of the GNU General Public License as published by the
  10  * Free Software Foundation; either version 2, or (at your option) any
  11  * later version.
  12  *
  13  * FRRouting is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * General Public License for more details.
  17  *
  18  */
  19
  20 #ifndef _FRR_BGP_EVPN_MH_H
  21 #define _FRR_BGP_EVPN_MH_H
  22
  23 #include "vxlan.h"
  24 #include "bgpd.h"
  25 #include "bgp_evpn.h"
  26 #include "bgp_evpn_private.h"
  27
  28 #define BGP_EVPN_AD_ES_ETH_TAG 0xffffffff
  29 #define BGP_EVPN_AD_EVI_ETH_TAG 0
  30
  31 #define BGP_EVPNES_INCONS_STR_SZ 80
  32 #define BGP_EVPN_VTEPS_FLAG_STR_SZ (BGP_EVPN_FLAG_STR_SZ * ES_VTEP_MAX_CNT)
  33
  34 #define BGP_EVPN_CONS_CHECK_INTERVAL 60
  35
  36 #define BGP_EVPN_MH_USE_ES_L3NHG_DEF true
  37
  38 /* Ethernet Segment entry -
  39  * - Local and remote ESs are maintained in a global RB tree,
  40  * bgp_mh_info->es_rb_tree using ESI as key
  41  * - Local ESs are received from zebra (BGP_EVPNES_LOCAL)
  42  * - Remotes ESs are implicitly created (by reference) by a remote ES-EVI
  43  *   (BGP_EVPNES_REMOTE)
  44  * - An ES can be simulatenously LOCAL and REMOTE; infact all LOCAL ESs are
  45  *   expected to have REMOTE ES peers.
  46  */
  47 struct bgp_evpn_es {
  48         /* Ethernet Segment Identifier */
  49         esi_t esi;
  50         char esi_str[ESI_STR_LEN];
  51
  52         /* es flags */
  53         uint32_t flags;
  54         /* created via zebra config */
  55 #define BGP_EVPNES_LOCAL           (1 << 0)
  56         /* created implicitly by a remote ES-EVI reference */
  57 #define BGP_EVPNES_REMOTE          (1 << 1)
  58         /* local ES link is oper-up */
  59 #define BGP_EVPNES_OPER_UP         (1 << 2)
  60         /* enable generation of EAD-EVI routes */
  61 #define BGP_EVPNES_ADV_EVI         (1 << 3)
  62         /* consistency checks pending */
  63 #define BGP_EVPNES_CONS_CHECK_PEND (1 << 4)
  64         /* ES is in LACP bypass mode - don't advertise EAD-ES or ESR */
  65 #define BGP_EVPNES_BYPASS (1 << 5)
  66         /* bits needed for printing the flags + null */
  67 #define BGP_EVPN_FLAG_STR_SZ 7
  68
  69         /* memory used for adding the es to bgp->es_rb_tree */
  70         RB_ENTRY(bgp_evpn_es) rb_node;
  71
  72         /* [EVPNES_LOCAL] memory used for linking the es to
  73          * bgp_mh_info->local_es_list
  74          */
  75         struct listnode es_listnode;
  76
  77         /* memory used for linking the es to "processing" pending list
  78          * bgp_mh_info->pend_es_list
  79          */
  80         struct listnode pend_es_listnode;
  81
  82         /* [EVPNES_LOCAL] Id for deriving the RD automatically for this ESI */
  83         uint16_t rd_id;
  84
  85         /* [EVPNES_LOCAL] RD for this ES */
  86         struct prefix_rd prd;
  87
  88         /* [EVPNES_LOCAL] originator ip address  */
  89         struct in_addr originator_ip;
  90
  91         /* [EVPNES_LOCAL] Route table for EVPN routes for this ESI-
  92          * - Type-4 local and remote routes
  93          * - Type-1 local routes
  94          */
  95         struct bgp_table *route_table;
  96
  97         /* list of PEs (bgp_evpn_es_vtep) attached to the ES */
  98         struct list *es_vtep_list;
  99
 100         /* List of ES-EVIs associated with this ES */
 101         struct list *es_evi_list;
 102
 103         /* List of ES-VRFs associated with this ES */
 104         struct list *es_vrf_list;
 105
 106         /* List of MAC-IP VNI paths using this ES as destination -
 107          * element is bgp_path_info_extra->es_info
 108          * Note: Only local/zebra-added MACIP paths in the VNI
 109          * routing table are linked to this list
 110          */
 111         struct list *macip_evi_path_list;
 112
 113         /* List of MAC-IP paths in the global routing table using this
 114          * ES as destination - data is bgp_path_info_extra->es_info
 115          * Note: Only non-local/imported MACIP paths in the global
 116          * routing table are linked to this list
 117          */
 118         struct list *macip_global_path_list;
 119
 120         /* Number of remote VNIs referencing this ES */
 121         uint32_t remote_es_evi_cnt;
 122
 123         uint32_t inconsistencies;
 124         /* there are one or more EVIs whose VTEP list doesn't match
 125          * with the ES's VTEP list
 126          */
 127 #define BGP_EVPNES_INCONS_VTEP_LIST (1 << 0)
 128
 129         /* number of es-evi entries whose VTEP list doesn't match
 130          * with the ES's
 131          */
 132         uint32_t incons_evi_vtep_cnt;
 133
 134         /* preference config for BUM-DF election. advertised via the ESR. */
 135         uint16_t df_pref;
 136
 137         QOBJ_FIELDS;
 138 };
 139 DECLARE_QOBJ_TYPE(bgp_evpn_es);
 140 RB_HEAD(bgp_es_rb_head, bgp_evpn_es);
 141 RB_PROTOTYPE(bgp_es_rb_head, bgp_evpn_es, rb_node, bgp_es_rb_cmp);
 142
 143 /* PE attached to an ES */
 144 struct bgp_evpn_es_vtep {
 145         struct bgp_evpn_es *es; /* parent ES */
 146         struct in_addr vtep_ip;
 147
 148         char vtep_str[INET6_ADDRSTRLEN];
 149
 150         uint32_t flags;
 151         /* Rxed a Type4 route from this PE */
 152 #define BGP_EVPNES_VTEP_ESR        (1 << 0)
 153         /* Active (rxed EAD-ES and EAD-EVI) and can be included as
 154          * a nexthop
 155          */
 156 #define BGP_EVPNES_VTEP_ACTIVE (1 << 1)
 157
 158         uint32_t evi_cnt; /* es_evis referencing this vtep as an active path */
 159
 160         /* Algorithm and preference for DF election. Rxed via the ESR */
 161         uint8_t df_alg;
 162         uint16_t df_pref;
 163
 164         /* memory used for adding the entry to es->es_vtep_list */
 165         struct listnode es_listnode;
 166 };
 167
 168 /* ES-VRF element needed for managing L3 NHGs. It is implicitly created
 169  * when an ES-EVI is associated with a tenant VRF
 170  */
 171 struct bgp_evpn_es_vrf {
 172         struct bgp_evpn_es *es;
 173         struct bgp *bgp_vrf;
 174
 175         uint32_t flags;
 176 /* NHG can only be activated if there are active VTEPs in the ES and
 177  * there is a valid L3-VNI associated with the VRF
 178  */
 179 #define BGP_EVPNES_VRF_NHG_ACTIVE (1 << 0)
 180
 181         /* memory used for adding the es_vrf to
 182          * es_vrf->bgp_vrf->es_vrf_rb_tree
 183          */
 184         RB_ENTRY(bgp_evpn_es_vrf) rb_node;
 185
 186         /* memory used for linking the es_vrf to es_vrf->es->es_vrf_list */
 187         struct listnode es_listnode;
 188
 189         uint32_t nhg_id;
 190         uint32_t v6_nhg_id;
 191
 192         /* Number of ES-EVI entries associated with this ES-VRF */
 193         uint32_t ref_cnt;
 194 };
 195
 196 /* ES per-EVI info
 197  * - ES-EVIs are maintained per-L2-VNI (vpn->es_evi_rb_tree)
 198  * - ES-EVIs are also linked to the parent ES (es->es_evi_list)
 199  * - Local ES-EVIs are created by zebra (via config). They are linked to a
 200  *   per-VNI list (vpn->local_es_evi_list) for quick access
 201  * - Remote ES-EVIs are created implicitly when a bgp_evpn_es_evi_vtep
 202  *   references it.
 203  */
 204 struct bgp_evpn_es_evi {
 205         struct bgp_evpn_es *es;
 206         struct bgpevpn *vpn;
 207
 208         /* ES-EVI flags */
 209         uint32_t flags;
 210 /* local ES-EVI, created by zebra */
 211 #define BGP_EVPNES_EVI_LOCAL            (1 << 0)
 212 /* created via a remote VTEP imported by BGP */
 213 #define BGP_EVPNES_EVI_REMOTE           (1 << 1)
 214 #define BGP_EVPNES_EVI_INCONS_VTEP_LIST (1 << 2)
 215
 216         /* memory used for adding the es_evi to es_evi->vpn->es_evi_rb_tree */
 217         RB_ENTRY(bgp_evpn_es_evi) rb_node;
 218         /* memory used for linking the es_evi to
 219          * es_evi->vpn->local_es_evi_list
 220          */
 221         struct listnode l2vni_listnode;
 222         /* memory used for linking the es_evi to
 223          * es_evi->es->es_evi_list
 224          */
 225         struct listnode es_listnode;
 226
 227         /* list of PEs (bgp_evpn_es_evi_vtep) attached to the ES for this VNI */
 228         struct list *es_evi_vtep_list;
 229
 230         struct bgp_evpn_es_vrf *es_vrf;
 231 };
 232
 233 /* PE attached to an ES for a VNI. This entry is created when an EAD-per-ES
 234  * or EAD-per-EVI Type1 route is imported into the VNI.
 235  */
 236 struct bgp_evpn_es_evi_vtep {
 237         struct bgp_evpn_es_evi *es_evi; /* parent ES-EVI */
 238         struct in_addr vtep_ip;
 239
 240         uint32_t flags;
 241         /* Rxed an EAD-per-ES route from the PE */
 242 #define BGP_EVPN_EVI_VTEP_EAD_PER_ES  (1 << 0) /* rxed EAD-per-ES */
 243         /* Rxed an EAD-per-EVI route from the PE */
 244 #define BGP_EVPN_EVI_VTEP_EAD_PER_EVI (1 << 1) /* rxed EAD-per-EVI */
 245         /* VTEP is active i.e. will result in the creation of an es-vtep */
 246 #define BGP_EVPN_EVI_VTEP_ACTIVE      (1 << 2)
 247 #define BGP_EVPN_EVI_VTEP_EAD         (BGP_EVPN_EVI_VTEP_EAD_PER_ES |\
 248                 BGP_EVPN_EVI_VTEP_EAD_PER_EVI)
 249
 250         /* memory used for adding the entry to es_evi->es_evi_vtep_list */
 251         struct listnode es_evi_listnode;
 252         struct bgp_evpn_es_vtep *es_vtep;
 253 };
 254
 255 /* A nexthop is created when a path (imported from an EVPN type-2 route)
 256  * is added to the VRF route table using that nexthop.
 257  * It is added on first pi reference and removed on last pi deref.
 258  */
 259 struct bgp_evpn_nh {
 260         /* backpointer to the VRF */
 261         struct bgp *bgp_vrf;
 262         /* nexthop/VTEP IP */
 263         struct ipaddr ip;
 264         /* description for easy logging */
 265         char nh_str[INET6_ADDRSTRLEN];
 266         struct ethaddr rmac;
 267         /* pi from which we are pulling the nh RMAC */
 268         struct bgp_path_info *ref_pi;
 269         /* List of VRF paths using this nexthop */
 270         struct list *pi_list;
 271         uint8_t flags;
 272 #define BGP_EVPN_NH_READY_FOR_ZEBRA (1 << 0)
 273 };
 274
 275 /* multihoming information stored in bgp_master */
 276 #define bgp_mh_info (bm->mh_info)
 277 struct bgp_evpn_mh_info {
 278         /* RB tree of Ethernet segments (used for EVPN-MH)  */
 279         struct bgp_es_rb_head es_rb_tree;
 280         /* List of local ESs */
 281         struct list *local_es_list;
 282         /* List of ESs with pending/periodic processing */
 283         struct list *pend_es_list;
 284         /* periodic timer for running background consistency checks */
 285         struct thread *t_cons_check;
 286
 287         /* config knobs for optimizing or interop */
 288         /* Generate EAD-EVI routes even if the ES is oper-down. This can be
 289          * enabled as an optimization to avoid a storm of updates when an ES
 290          * link flaps.
 291          */
 292         bool ead_evi_adv_for_down_links;
 293         /* Enable ES consistency checking */
 294         bool consistency_checking;
 295         /* Use L3 NHGs for host routes in symmetric IRB */
 296         bool install_l3nhg;
 297         bool host_routes_use_l3nhg;
 298         /* Some vendors are not generating the EAD-per-EVI route. This knob
 299          * can be turned off to activate a remote ES-PE when the EAD-per-ES
 300          * route is rxed i.e. not wait on the EAD-per-EVI route
 301          */
 302         bool ead_evi_rx;
 303 #define BGP_EVPN_MH_EAD_EVI_RX_DEF true
 304         /* Skip EAD-EVI advertisements by turning off this knob */
 305         bool ead_evi_tx;
 306 #define BGP_EVPN_MH_EAD_EVI_TX_DEF true
 307         /* If the Local ES is inactive we advertise the MAC-IP without the
 308          * L3 ecomm
 309          */
 310         bool suppress_l3_ecomm_on_inactive_es;
 311         /* Setup EVPN PE nexthops and their RMAC in bgpd */
 312         bool bgp_evpn_nh_setup;
 313 };
 314
 315 /****************************************************************************/
 316 static inline int bgp_evpn_is_es_local(struct bgp_evpn_es *es)
 317 {
 318         return CHECK_FLAG(es->flags, BGP_EVPNES_LOCAL) ? 1 : 0;
 319 }
 320
 321 extern esi_t *zero_esi;
 322 static inline bool bgp_evpn_is_esi_valid(esi_t *esi)
 323 {
 324         return !!memcmp(esi, zero_esi, sizeof(esi_t));
 325 }
 326
 327 static inline esi_t *bgp_evpn_attr_get_esi(struct attr *attr)
 328 {
 329         return attr ? &attr->esi : zero_esi;
 330 }
 331
 332 static inline bool bgp_evpn_attr_is_sync(struct attr *attr)
 333 {
 334         return attr ? !!(attr->es_flags &
 335                 (ATTR_ES_PEER_PROXY | ATTR_ES_PEER_ACTIVE)) : false;
 336 }
 337
 338 static inline uint32_t bgp_evpn_attr_get_sync_seq(struct attr *attr)
 339 {
 340         return attr ?  attr->mm_sync_seqnum : 0;
 341 }
 342
 343 static inline bool bgp_evpn_attr_is_active_on_peer(struct attr *attr)
 344 {
 345         return attr ?
 346                 !!(attr->es_flags & ATTR_ES_PEER_ACTIVE) : false;
 347 }
 348
 349 static inline bool bgp_evpn_attr_is_router_on_peer(struct attr *attr)
 350 {
 351         return attr ?
 352                 !!(attr->es_flags & ATTR_ES_PEER_ROUTER) : false;
 353 }
 354
 355 static inline bool bgp_evpn_attr_is_proxy(struct attr *attr)
 356 {
 357         return attr ? !!(attr->es_flags & ATTR_ES_PROXY_ADVERT) : false;
 358 }
 359
 360 static inline bool bgp_evpn_attr_is_local_es(struct attr *attr)
 361 {
 362         return attr ? !!(attr->es_flags & ATTR_ES_IS_LOCAL) : false;
 363 }
 364
 365 static inline uint32_t bgp_evpn_attr_get_df_pref(struct attr *attr)
 366 {
 367         return (attr) ? attr->df_pref : 0;
 368 }
 369
 370 static inline bool bgp_evpn_local_es_is_active(struct bgp_evpn_es *es)
 371 {
 372         return (es->flags & BGP_EVPNES_OPER_UP)
 373                && !(es->flags & BGP_EVPNES_BYPASS);
 374 }
 375
 376 /****************************************************************************/
 377 extern int bgp_evpn_es_route_install_uninstall(struct bgp *bgp,
 378                 struct bgp_evpn_es *es, afi_t afi, safi_t safi,
 379                 struct prefix_evpn *evp, struct bgp_path_info *pi,
 380                 int install);
 381 extern void update_type1_routes_for_evi(struct bgp *bgp, struct bgpevpn *vpn);
 382 extern int delete_global_ead_evi_routes(struct bgp *bgp, struct bgpevpn *vpn);
 383 extern int bgp_evpn_mh_route_update(struct bgp *bgp, struct bgp_evpn_es *es,
 384                                     struct bgpevpn *vpn, afi_t afi, safi_t safi,
 385                                     struct bgp_dest *dest, struct attr *attr,
 386                                     int add, struct bgp_path_info **ri,
 387                                     int *route_changed);
 388 int bgp_evpn_type1_route_process(struct peer *peer, afi_t afi, safi_t safi,
 389                 struct attr *attr, uint8_t *pfx, int psize,
 390                 uint32_t addpath_id);
 391 int bgp_evpn_type4_route_process(struct peer *peer, afi_t afi, safi_t safi,
 392                 struct attr *attr, uint8_t *pfx, int psize,
 393                 uint32_t addpath_id);
 394 extern int bgp_evpn_local_es_add(struct bgp *bgp, esi_t *esi,
 395                                  struct in_addr originator_ip, bool oper_up,
 396                                  uint16_t df_pref, bool bypass);
 397 extern int bgp_evpn_local_es_del(struct bgp *bgp, esi_t *esi);
 398 extern int bgp_evpn_local_es_evi_add(struct bgp *bgp, esi_t *esi, vni_t vni);
 399 extern int bgp_evpn_local_es_evi_del(struct bgp *bgp, esi_t *esi, vni_t vni);
 400 extern int bgp_evpn_remote_es_evi_add(struct bgp *bgp, struct bgpevpn *vpn,
 401                 const struct prefix_evpn *p);
 402 extern int bgp_evpn_remote_es_evi_del(struct bgp *bgp, struct bgpevpn *vpn,
 403                 const struct prefix_evpn *p);
 404 extern void bgp_evpn_mh_init(void);
 405 extern void bgp_evpn_mh_finish(void);
 406 void bgp_evpn_vni_es_init(struct bgpevpn *vpn);
 407 void bgp_evpn_vni_es_cleanup(struct bgpevpn *vpn);
 408 void bgp_evpn_es_show_esi(struct vty *vty, esi_t *esi, bool uj);
 409 void bgp_evpn_es_show(struct vty *vty, bool uj, bool detail);
 410 void bgp_evpn_es_evi_show_vni(struct vty *vty, vni_t vni,
 411                 bool uj, bool detail);
 412 void bgp_evpn_es_evi_show(struct vty *vty, bool uj, bool detail);
 413 struct bgp_evpn_es *bgp_evpn_es_find(const esi_t *esi);
 414 extern void bgp_evpn_vrf_es_init(struct bgp *bgp_vrf);
 415 extern bool bgp_evpn_is_esi_local_and_non_bypass(esi_t *esi);
 416 extern void bgp_evpn_es_vrf_deref(struct bgp_evpn_es_evi *es_evi);
 417 extern void bgp_evpn_es_vrf_ref(struct bgp_evpn_es_evi *es_evi,
 418                                 struct bgp *bgp_vrf);
 419 extern void bgp_evpn_path_mh_info_free(struct bgp_path_mh_info *mh_info);
 420 extern void bgp_evpn_path_es_link(struct bgp_path_info *pi, vni_t vni,
 421                                   esi_t *esi);
 422 extern bool bgp_evpn_path_es_use_nhg(struct bgp *bgp_vrf,
 423                                      struct bgp_path_info *pi, uint32_t *nhg_p);
 424 extern void bgp_evpn_es_vrf_show(struct vty *vty, bool uj,
 425                                  struct bgp_evpn_es *es);
 426 extern void bgp_evpn_es_vrf_show_esi(struct vty *vty, esi_t *esi, bool uj);
 427 extern void bgp_evpn_switch_ead_evi_rx(void);
 428 extern bool bgp_evpn_es_add_l3_ecomm_ok(esi_t *esi);
 429 extern void bgp_evpn_es_vrf_use_nhg(struct bgp *bgp_vrf, esi_t *esi,
 430                                     bool *use_l3nhg, bool *is_l3nhg_active,
 431                                     struct bgp_evpn_es_vrf **es_vrf_p);
 432 extern void bgp_evpn_nh_init(struct bgp *bgp_vrf);
 433 extern void bgp_evpn_nh_finish(struct bgp *bgp_vrf);
 434 extern void bgp_evpn_nh_show(struct vty *vty, bool uj);
 435 extern void bgp_evpn_path_nh_add(struct bgp *bgp_vrf, struct bgp_path_info *pi);
 436 extern void bgp_evpn_path_nh_del(struct bgp *bgp_vrf, struct bgp_path_info *pi);
 437
 438 #endif /* _FRR_BGP_EVPN_MH_H */