proto/bgp/bgp.h

   1 /*
   2  *      BIRD -- The Border Gateway Protocol
   3  *
   4  *      (c) 2000 Martin Mares <mj@ucw.cz>
   5  *
   6  *      Can be freely distributed and used under the terms of the GNU GPL.
   7  */
   8
   9 #ifndef _BIRD_BGP_H_
  10 #define _BIRD_BGP_H_
  11
  12 #include <stdint.h>
  13 #include "nest/route.h"
  14
  15 struct linpool;
  16 struct eattr;
  17
  18 struct bgp_config {
  19   struct proto_config c;
  20   u32 local_as, remote_as;
  21   ip_addr remote_ip;
  22   ip_addr source_addr;                  /* Source address to use */
  23   struct iface *iface;                  /* Interface for link-local addresses */
  24   int multihop;                         /* Number of hops if multihop */
  25   int ttl_security;                     /* Enable TTL security [RFC5082] */
  26   int next_hop_self;                    /* Always set next hop to local IP address */
  27   int next_hop_keep;                    /* Do not touch next hop attribute */
  28   int missing_lladdr;                   /* What we will do when we don' know link-local addr, see MLL_* */
  29   int gw_mode;                          /* How we compute route gateway from next_hop attr, see GW_* */
  30   int compare_path_lengths;             /* Use path lengths when selecting best route */
  31   int med_metric;                       /* Compare MULTI_EXIT_DISC even between routes from differen ASes */
  32   int igp_metric;                       /* Use IGP metrics when selecting best route */
  33   int prefer_older;                     /* Prefer older routes according to RFC 5004 */
  34   int deterministic_med;                /* Use more complicated algo to have strict RFC 4271 MED comparison */
  35   u32 default_local_pref;               /* Default value for LOCAL_PREF attribute */
  36   u32 default_med;                      /* Default value for MULTI_EXIT_DISC attribute */
  37   int capabilities;                     /* Enable capability handshake [RFC3392] */
  38   int enable_refresh;                   /* Enable local support for route refresh [RFC2918] */
  39   int enable_as4;                       /* Enable local support for 4B AS numbers [RFC4893] */
  40   u32 rr_cluster_id;                    /* Route reflector cluster ID, if different from local ID */
  41   int rr_client;                        /* Whether neighbor is RR client of me */
  42   int rs_client;                        /* Whether neighbor is RS client of me */
  43   int advertise_ipv4;                   /* Whether we should add IPv4 capability advertisement to OPEN message */
  44   int passive;                          /* Do not initiate outgoing connection */
  45   int interpret_communities;            /* Hardwired handling of well-known communities */
  46   int secondary;                        /* Accept also non-best routes (i.e. RA_ACCEPTED) */
  47   unsigned connect_retry_time;
  48   unsigned hold_time, initial_hold_time;
  49   unsigned keepalive_time;
  50   unsigned start_delay_time;            /* Minimum delay between connects */
  51   unsigned error_amnesia_time;          /* Errors are forgotten after */
  52   unsigned error_delay_time_min;        /* Time to wait after an error is detected */
  53   unsigned error_delay_time_max;
  54   unsigned disable_after_error;         /* Disable the protocol when error is detected */
  55   char *password;                       /* Password used for MD5 authentication */
  56   struct rtable_config *igp_table;      /* Table used for recursive next hop lookups */
  57 };
  58
  59 #define MLL_SELF 1
  60 #define MLL_DROP 2
  61 #define MLL_IGNORE 3
  62
  63 #define GW_DIRECT 1
  64 #define GW_RECURSIVE 2
  65
  66 struct bgp_conn {
  67   struct bgp_proto *bgp;
  68   struct birdsock *sk;
  69   unsigned int state;                   /* State of connection state machine */
  70   struct timer *connect_retry_timer;
  71   struct timer *hold_timer;
  72   struct timer *keepalive_timer;
  73   struct event *tx_ev;
  74   int packets_to_send;                  /* Bitmap of packet types to be sent */
  75   int notify_code, notify_subcode, notify_size;
  76   byte *notify_data;
  77   u32 advertised_as;                    /* Temporary value for AS number received */
  78   int start_state;                      /* protocol start_state snapshot when connection established */
  79   int want_as4_support;                 /* Connection tries to establish AS4 session */
  80   int peer_as4_support;                 /* Peer supports 4B AS numbers [RFC4893] */
  81   int peer_refresh_support;             /* Peer supports route refresh [RFC2918] */
  82   unsigned hold_time, keepalive_time;   /* Times calculated from my and neighbor's requirements */
  83 };
  84
  85 struct bgp_proto {
  86   struct proto p;
  87   struct bgp_config *cf;                /* Shortcut to BGP configuration */
  88   u32 local_as, remote_as;
  89   int start_state;                      /* Substates that partitions BS_START */
  90   int is_internal;                      /* Internal BGP connection (local_as == remote_as) */
  91   int as4_session;                      /* Session uses 4B AS numbers in AS_PATH (both sides support it) */
  92   u32 local_id;                         /* BGP identifier of this router */
  93   u32 remote_id;                        /* BGP identifier of the neighbor */
  94   u32 rr_cluster_id;                    /* Route reflector cluster ID */
  95   int rr_client;                        /* Whether neighbor is RR client of me */
  96   int rs_client;                        /* Whether neighbor is RS client of me */
  97   struct bgp_conn *conn;                /* Connection we have established */
  98   struct bgp_conn outgoing_conn;        /* Outgoing connection we're working with */
  99   struct bgp_conn incoming_conn;        /* Incoming connection we have neither accepted nor rejected yet */
 100   struct object_lock *lock;             /* Lock for neighbor connection */
 101   struct neighbor *neigh;               /* Neighbor entry corresponding to remote ip, NULL if multihop */
 102   ip_addr source_addr;                  /* Local address used as an advertised next hop */
 103   rtable *igp_table;                    /* Table used for recursive next hop lookups */
 104   struct event *event;                  /* Event for respawning and shutting process */
 105   struct timer *startup_timer;          /* Timer used to delay protocol startup due to previous errors (startup_delay) */
 106   struct bgp_bucket **bucket_hash;      /* Hash table of attribute buckets */
 107   unsigned int hash_size, hash_count, hash_limit;
 108   struct fib prefix_fib;                /* Prefixes to be sent */
 109   list bucket_queue;                    /* Queue of buckets to send */
 110   struct bgp_bucket *withdraw_bucket;   /* Withdrawn routes */
 111   unsigned startup_delay;               /* Time to delay protocol startup by due to errors */
 112   bird_clock_t last_proto_error;        /* Time of last error that leads to protocol stop */
 113   u8 last_error_class;                  /* Error class of last error */
 114   u32 last_error_code;                  /* Error code of last error. BGP protocol errors
 115                                            are encoded as (bgp_err_code << 16 | bgp_err_subcode) */
 116 #ifdef IPV6
 117   byte *mp_reach_start, *mp_unreach_start; /* Multiprotocol BGP attribute notes */
 118   unsigned mp_reach_len, mp_unreach_len;
 119   ip_addr local_link;                   /* Link-level version of source_addr */
 120 #endif
 121 };
 122
 123 struct bgp_prefix {
 124   struct fib_node n;                    /* Node in prefix fib */
 125   node bucket_node;                     /* Node in per-bucket list */
 126 };
 127
 128 struct bgp_bucket {
 129   node send_node;                       /* Node in send queue */
 130   struct bgp_bucket *hash_next, *hash_prev;     /* Node in bucket hash table */
 131   unsigned hash;                        /* Hash over extended attributes */
 132   list prefixes;                        /* Prefixes in this buckets */
 133   ea_list eattrs[0];                    /* Per-bucket extended attributes */
 134 };
 135
 136 #define BGP_PORT                179
 137 #define BGP_VERSION             4
 138 #define BGP_HEADER_LENGTH       19
 139 #define BGP_MAX_PACKET_LENGTH   4096
 140 #define BGP_RX_BUFFER_SIZE      4096
 141 #define BGP_TX_BUFFER_SIZE      BGP_MAX_PACKET_LENGTH
 142
 143 extern struct linpool *bgp_linpool;
 144
 145
 146 void bgp_start_timer(struct timer *t, int value);
 147 void bgp_check_config(struct bgp_config *c);
 148 void bgp_error(struct bgp_conn *c, unsigned code, unsigned subcode, byte *data, int len);
 149 void bgp_close_conn(struct bgp_conn *c);
 150 void bgp_update_startup_delay(struct bgp_proto *p);
 151 void bgp_conn_enter_openconfirm_state(struct bgp_conn *conn);
 152 void bgp_conn_enter_established_state(struct bgp_conn *conn);
 153 void bgp_conn_enter_close_state(struct bgp_conn *conn);
 154 void bgp_conn_enter_idle_state(struct bgp_conn *conn);
 155 void bgp_store_error(struct bgp_proto *p, struct bgp_conn *c, u8 class, u32 code);
 156 void bgp_stop(struct bgp_proto *p, unsigned subcode);
 157
 158
 159
 160 #ifdef LOCAL_DEBUG
 161 #define BGP_FORCE_DEBUG 1
 162 #else
 163 #define BGP_FORCE_DEBUG 0
 164 #endif
 165 #define BGP_TRACE(flags, msg, args...) do { if ((p->p.debug & flags) || BGP_FORCE_DEBUG) \
 166         log(L_TRACE "%s: " msg, p->p.name , ## args ); } while(0)
 167
 168 #define BGP_TRACE_RL(rl, flags, msg, args...) do { if ((p->p.debug & flags) || BGP_FORCE_DEBUG) \
 169         log_rl(rl, L_TRACE "%s: " msg, p->p.name , ## args ); } while(0)
 170
 171
 172 /* attrs.c */
 173
 174 /* Hack: although BA_NEXT_HOP attribute has type EAF_TYPE_IP_ADDRESS, in IPv6
 175  * we store two addesses in it - a global address and a link local address.
 176  */
 177 #ifdef IPV6
 178 #define NEXT_HOP_LENGTH (2*sizeof(ip_addr))
 179 static inline void set_next_hop(byte *b, ip_addr addr) { ((ip_addr *) b)[0] = addr; ((ip_addr *) b)[1] = IPA_NONE; }
 180 #else
 181 #define NEXT_HOP_LENGTH sizeof(ip_addr)
 182 static inline void set_next_hop(byte *b, ip_addr addr) { ((ip_addr *) b)[0] = addr; }
 183 #endif
 184
 185 void bgp_attach_attr(struct ea_list **to, struct linpool *pool, unsigned attr, uintptr_t val);
 186 byte *bgp_attach_attr_wa(struct ea_list **to, struct linpool *pool, unsigned attr, unsigned len);
 187 struct rta *bgp_decode_attrs(struct bgp_conn *conn, byte *a, unsigned int len, struct linpool *pool, int mandatory);
 188 int bgp_get_attr(struct eattr *e, byte *buf, int buflen);
 189 int bgp_rte_better(struct rte *, struct rte *);
 190 int bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best);
 191 void bgp_rt_notify(struct proto *P, rtable *tbl UNUSED, net *n, rte *new, rte *old UNUSED, ea_list *attrs);
 192 int bgp_import_control(struct proto *, struct rte **, struct ea_list **, struct linpool *);
 193 void bgp_attr_init(struct bgp_proto *);
 194 unsigned int bgp_encode_attrs(struct bgp_proto *p, byte *w, ea_list *attrs, int remains);
 195 void bgp_free_bucket(struct bgp_proto *p, struct bgp_bucket *buck);
 196 void bgp_get_route_info(struct rte *, byte *buf, struct ea_list *attrs);
 197
 198 inline static void bgp_attach_attr_ip(struct ea_list **to, struct linpool *pool, unsigned attr, ip_addr a)
 199 { *(ip_addr *) bgp_attach_attr_wa(to, pool, attr, sizeof(ip_addr)) = a; }
 200
 201 /* packets.c */
 202
 203 void mrt_dump_bgp_state_change(struct bgp_conn *conn, unsigned old, unsigned new);
 204 void bgp_schedule_packet(struct bgp_conn *conn, int type);
 205 void bgp_kick_tx(void *vconn);
 206 void bgp_tx(struct birdsock *sk);
 207 int bgp_rx(struct birdsock *sk, int size);
 208 const char * bgp_error_dsc(unsigned code, unsigned subcode);
 209 void bgp_log_error(struct bgp_proto *p, u8 class, char *msg, unsigned code, unsigned subcode, byte *data, unsigned len);
 210
 211 /* Packet types */
 212
 213 #define PKT_OPEN                0x01
 214 #define PKT_UPDATE              0x02
 215 #define PKT_NOTIFICATION        0x03
 216 #define PKT_KEEPALIVE           0x04
 217 #define PKT_ROUTE_REFRESH       0x05
 218 #define PKT_SCHEDULE_CLOSE      0x1f    /* Used internally to schedule socket close */
 219
 220 /* Attributes */
 221
 222 #define BAF_OPTIONAL            0x80
 223 #define BAF_TRANSITIVE          0x40
 224 #define BAF_PARTIAL             0x20
 225 #define BAF_EXT_LEN             0x10
 226
 227 #define BA_ORIGIN               0x01    /* [RFC1771] */         /* WM */
 228 #define BA_AS_PATH              0x02                            /* WM */
 229 #define BA_NEXT_HOP             0x03                            /* WM */
 230 #define BA_MULTI_EXIT_DISC      0x04                            /* ON */
 231 #define BA_LOCAL_PREF           0x05                            /* WD */
 232 #define BA_ATOMIC_AGGR          0x06                            /* WD */
 233 #define BA_AGGREGATOR           0x07                            /* OT */
 234 #define BA_COMMUNITY            0x08    /* [RFC1997] */         /* OT */
 235 #define BA_ORIGINATOR_ID        0x09    /* [RFC1966] */         /* ON */
 236 #define BA_CLUSTER_LIST         0x0a                            /* ON */
 237 /* We don't support these: */
 238 #define BA_DPA                  0x0b    /* ??? */
 239 #define BA_ADVERTISER           0x0c    /* [RFC1863] */
 240 #define BA_RCID_PATH            0x0d
 241 #define BA_MP_REACH_NLRI        0x0e    /* [RFC2283] */
 242 #define BA_MP_UNREACH_NLRI      0x0f
 243 #define BA_EXT_COMMUNITY        0x10    /* [RFC4360] */
 244 #define BA_AS4_PATH             0x11    /* [RFC4893] */
 245 #define BA_AS4_AGGREGATOR       0x12
 246
 247 /* BGP connection states */
 248
 249 #define BS_IDLE                 0
 250 #define BS_CONNECT              1       /* Attempting to connect */
 251 #define BS_ACTIVE               2       /* Waiting for connection retry & listening */
 252 #define BS_OPENSENT             3
 253 #define BS_OPENCONFIRM          4
 254 #define BS_ESTABLISHED          5
 255 #define BS_CLOSE                6       /* Used during transition to BS_IDLE */
 256
 257 /* BGP start states
 258  *
 259  * Used in PS_START for fine-grained specification of starting state.
 260  *
 261  * When BGP protocol is started by core, it goes to BSS_PREPARE. When BGP protocol
 262  * done what is neccessary to start itself (like acquiring the lock), it goes to BSS_CONNECT.
 263  * When some connection attempt failed because of option or capability error, it goes to
 264  * BSS_CONNECT_NOCAP.
 265  */
 266
 267 #define BSS_PREPARE             0       /* Used before ordinary BGP started, i. e. waiting for lock */
 268 #define BSS_DELAY               1       /* Startup delay due to previous errors */
 269 #define BSS_CONNECT             2       /* Ordinary BGP connecting */
 270 #define BSS_CONNECT_NOCAP       3       /* Legacy BGP connecting (without capabilities) */
 271
 272 /* Error classes */
 273
 274 #define BE_NONE                 0
 275 #define BE_MISC                 1       /* Miscellaneous error */
 276 #define BE_SOCKET               2       /* Socket error */
 277 #define BE_BGP_RX               3       /* BGP protocol error notification received */
 278 #define BE_BGP_TX               4       /* BGP protocol error notification sent */
 279 #define BE_AUTO_DOWN            5       /* Automatic shutdown */
 280 #define BE_MAN_DOWN             6       /* Manual shutdown */
 281
 282 /* Misc error codes */
 283
 284 #define BEM_NEIGHBOR_LOST       1
 285 #define BEM_INVALID_NEXT_HOP    2
 286 #define BEM_INVALID_MD5         3       /* MD5 authentication kernel request failed (possibly not supported) */
 287 #define BEM_NO_SOCKET           4
 288
 289 /* Automatic shutdown error codes */
 290
 291 #define BEA_ROUTE_LIMIT_EXCEEDED 1
 292
 293 /* Well-known communities */
 294
 295 #define BGP_COMM_NO_EXPORT              0xffffff01      /* Don't export outside local AS / confed. */
 296 #define BGP_COMM_NO_ADVERTISE           0xffffff02      /* Don't export at all */
 297 #define BGP_COMM_NO_EXPORT_SUBCONFED    0xffffff03      /* NO_EXPORT even in local confederation */
 298
 299 /* Origins */
 300
 301 #define ORIGIN_IGP              0
 302 #define ORIGIN_EGP              1
 303 #define ORIGIN_INCOMPLETE       2
 304
 305 /* Address families */
 306
 307 #define BGP_AF_IPV4             1
 308 #define BGP_AF_IPV6             2
 309
 310 #ifdef IPV6
 311 #define BGP_AF BGP_AF_IPV6
 312 #else
 313 #define BGP_AF BGP_AF_IPV4
 314 #endif
 315
 316 #endif