bgp.c

   1 /*
   2  * BGPv4
   3  * Used to advertise routes for upstream (l2tp port, rather than gratiutious
   4  * arp) and downstream--allowing routers to load-balance both.
   5  *
   6  * Implementation limitations:
   7  * - We never listen for incoming connections (session always initiated by us).
   8  * - Any routes advertised by the peer are accepted, but ignored.
   9  * - No password support; neither RFC1771 (which no-one seems to do anyway)
  10  *   nor RFC2385 (which requires a kernel patch on 2.4 kernels).
  11  */
  12
  13 #include <stdlib.h>
  14 #include <unistd.h>
  15 #include <string.h>
  16 #include <time.h>
  17 #include <errno.h>
  18 #include <sys/socket.h>
  19 #include <netinet/in.h>
  20 #include <arpa/inet.h>
  21 #include <netdb.h>
  22 #include <fcntl.h>
  23 #include <linux/rtnetlink.h>
  24
  25 #include "l2tpns.h"
  26 #include "bgp.h"
  27 #include "util.h"
  28
  29 static void bgp_clear(struct bgp_peer *peer);
  30 static void bgp_set_retry(struct bgp_peer *peer);
  31 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
  32     struct bgp_route_list *new);
  33 static struct bgp_route6_list *bgp_insert_route6(struct bgp_route6_list *head,
  34     struct bgp_route6_list *new);
  35
  36 static void bgp_process_timers(struct bgp_peer *peer);
  37 static void bgp_free_routes(struct bgp_route_list *routes);
  38 static void bgp_free_routes6(struct bgp_route6_list *routes);
  39 static char const *bgp_msg_type_str(uint8_t type);
  40 static int bgp_connect(struct bgp_peer *peer);
  41 static int bgp_handle_connect(struct bgp_peer *peer);
  42 static int bgp_write(struct bgp_peer *peer);
  43 static int bgp_read(struct bgp_peer *peer);
  44 static int bgp_handle_input(struct bgp_peer *peer);
  45 static int bgp_send_open(struct bgp_peer *peer);
  46 static int bgp_send_keepalive(struct bgp_peer *peer);
  47 static int bgp_send_update(struct bgp_peer *peer);
  48 static int bgp_send_update6(struct bgp_peer *peer);
  49 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
  50     uint8_t subcode);
  51
  52 static uint16_t our_as;
  53 static struct bgp_route_list *bgp_routes = 0;
  54 static struct bgp_route6_list *bgp_routes6 = 0;
  55
  56 int bgp_configured = 0;
  57 struct bgp_peer *bgp_peers = 0;
  58
  59 /* prepare peer structure, globals */
  60 int bgp_setup(int as)
  61 {
  62     int i;
  63     struct bgp_peer *peer;
  64
  65     for (i = 0; i < BGP_NUM_PEERS; i++)
  66     {
  67         peer = &bgp_peers[i];
  68         memset(peer, 0, sizeof(*peer));
  69
  70         peer->addr = INADDR_NONE;
  71         peer->sock = -1;
  72         peer->state = peer->next_state = Disabled;
  73
  74         if (!((peer->outbuf = malloc(sizeof(*peer->outbuf)))
  75             && (peer->inbuf = malloc(sizeof(*peer->inbuf)))))
  76         {
  77             LOG(0, 0, 0, "Can't allocate buffers for bgp peer (%s)\n",
  78                 strerror(errno));
  79
  80             return 0;
  81         }
  82
  83         peer->edata.type = FD_TYPE_BGP;
  84         peer->edata.index = i;
  85         peer->events = 0;
  86     }
  87
  88     if (as < 1)
  89         as = 0;
  90
  91     if ((our_as = as))
  92         return 0;
  93
  94     bgp_routes = 0;
  95     bgp_routes6 = 0;
  96     bgp_configured = 0; /* set by bgp_start */
  97
  98     return 1;
  99 }
 100
 101 /* start connection with a peer */
 102 int bgp_start(struct bgp_peer *peer, char *name, int as, int keepalive,
 103     int hold, struct in_addr update_source, int enable)
 104 {
 105     struct hostent *h;
 106     int ibgp;
 107     int i;
 108     struct bgp_path_attr a;
 109     char path_attrs[64];
 110     char *p = path_attrs;
 111     in_addr_t ip;
 112     uint32_t metric = htonl(BGP_METRIC);
 113     uint32_t no_export = htonl(BGP_COMMUNITY_NO_EXPORT);
 114
 115     if (!our_as)
 116         return 0;
 117
 118     if (peer->state != Disabled)
 119         bgp_halt(peer);
 120
 121     snprintf(peer->name, sizeof(peer->name), "%s", name);
 122
 123     if (!(h = gethostbyname(name)) || h->h_addrtype != AF_INET)
 124     {
 125         LOG(0, 0, 0, "Can't get address for BGP peer %s (%s)\n",
 126             name, h ? "no address" : hstrerror(h_errno));
 127
 128         return 0;
 129     }
 130
 131     memcpy(&peer->addr, h->h_addr, sizeof(peer->addr));
 132     peer->source_addr = update_source.s_addr;
 133     peer->as = as > 0 ? as : our_as;
 134     ibgp = peer->as == our_as;
 135
 136     /* set initial timer values */
 137     peer->init_keepalive = keepalive == -1 ? BGP_KEEPALIVE_TIME : keepalive;
 138     peer->init_hold = hold == -1 ? BGP_HOLD_TIME : hold;
 139
 140     if (peer->init_hold < 3)
 141         peer->init_hold = 3;
 142
 143     if (peer->init_keepalive * 3 > peer->init_hold)
 144         peer->init_keepalive = peer->init_hold / 3;
 145
 146     /* clear buffers, go to Idle state */
 147     peer->next_state = Idle;
 148     bgp_clear(peer);
 149
 150     /* set initial routing state */
 151     peer->routing = enable;
 152
 153     /* all our routes use the same attributes, so prepare it in advance */
 154     if (peer->path_attrs)
 155         free(peer->path_attrs);
 156
 157     peer->path_attr_len = 0;
 158
 159     /* ORIGIN */
 160     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 161     a.code = BGP_PATH_ATTR_CODE_ORIGIN;
 162     a.data.s.len = 1;
 163     a.data.s.value[0] = BGP_PATH_ATTR_CODE_ORIGIN_IGP;
 164
 165 #define ADD_ATTRIBUTE()         do { \
 166     i = BGP_PATH_ATTR_SIZE(a);  \
 167     memcpy(p, &a, i);           \
 168     p += i;                     \
 169     peer->path_attr_len += i;   } while (0)
 170
 171     ADD_ATTRIBUTE();
 172
 173     /* AS_PATH */
 174     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 175     a.code = BGP_PATH_ATTR_CODE_AS_PATH;
 176     if (ibgp)
 177     {
 178         /* empty path */
 179         a.data.s.len = 0;
 180     }
 181     else
 182     {
 183         /* just our AS */
 184         struct {
 185             uint8_t type;
 186             uint8_t len;
 187             uint16_t value;
 188         } as_path = {
 189             BGP_PATH_ATTR_CODE_AS_PATH_AS_SEQUENCE,
 190             1,
 191             htons(our_as),
 192         };
 193
 194         a.data.s.len = sizeof(as_path);
 195         memcpy(&a.data.s.value, &as_path, sizeof(as_path));
 196     }
 197
 198     ADD_ATTRIBUTE();
 199
 200     /* MULTI_EXIT_DISC */
 201     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL;
 202     a.code = BGP_PATH_ATTR_CODE_MULTI_EXIT_DISC;
 203     a.data.s.len = sizeof(metric);
 204     memcpy(a.data.s.value, &metric, sizeof(metric));
 205
 206     ADD_ATTRIBUTE();
 207
 208     if (ibgp)
 209     {
 210         uint32_t local_pref = htonl(BGP_LOCAL_PREF);
 211
 212         /* LOCAL_PREF */
 213         a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 214         a.code = BGP_PATH_ATTR_CODE_LOCAL_PREF;
 215         a.data.s.len = sizeof(local_pref);
 216         memcpy(a.data.s.value, &local_pref, sizeof(local_pref));
 217
 218         ADD_ATTRIBUTE();
 219     }
 220
 221     /* COMMUNITIES */
 222     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL | BGP_PATH_ATTR_FLAG_TRANS;
 223     a.code = BGP_PATH_ATTR_CODE_COMMUNITIES;
 224     a.data.s.len = sizeof(no_export);
 225     memcpy(a.data.s.value, &no_export, sizeof(no_export));
 226
 227     ADD_ATTRIBUTE();
 228
 229     /* remember the len before adding NEXT_HOP */
 230     peer->path_attr_len_without_nexthop = peer->path_attr_len;
 231
 232     /* NEXT_HOP */
 233     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 234     a.code = BGP_PATH_ATTR_CODE_NEXT_HOP;
 235     if (config->nexthop_address)
 236     {
 237         ip = config->nexthop_address;
 238     }
 239     else
 240     {
 241         ip = my_address; /* we're it */
 242     }
 243     a.data.s.len = sizeof(ip);
 244     memcpy(a.data.s.value, &ip, sizeof(ip));
 245
 246     ADD_ATTRIBUTE();
 247
 248     if (!(peer->path_attrs = malloc(peer->path_attr_len)))
 249     {
 250         LOG(0, 0, 0, "Can't allocate path_attrs for %s (%s)\n",
 251             name, strerror(errno));
 252
 253         return 0;
 254     }
 255
 256     memcpy(peer->path_attrs, path_attrs, peer->path_attr_len);
 257
 258     /* multiprotocol attributes initialization */
 259     if (config->ipv6_prefix.s6_addr[0])
 260     {
 261         struct bgp_attr_mp_reach_nlri_partial mp_reach_nlri_partial;
 262         struct bgp_attr_mp_unreach_nlri_partial mp_unreach_nlri_partial;
 263
 264         a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL;
 265         a.code = BGP_PATH_ATTR_CODE_MP_REACH_NLRI;
 266         a.data.s.len = 0; /* will be set on UPDATE */
 267
 268         mp_reach_nlri_partial.afi = htons(BGP_MP_AFI_IPv6);
 269         mp_reach_nlri_partial.safi = BGP_MP_SAFI_UNICAST;
 270         mp_reach_nlri_partial.reserved = 0;
 271         mp_reach_nlri_partial.next_hop_len = 16;
 272
 273         /* use the defined nexthop6, or our address in ipv6_prefix */
 274         if (config->nexthop6_address.s6_addr[0])
 275             memcpy(&mp_reach_nlri_partial.next_hop,
 276                     &config->nexthop6_address.s6_addr, 16);
 277         else
 278         {
 279             /* our address is ipv6prefix::1 */
 280             memcpy(&mp_reach_nlri_partial.next_hop,
 281                     &config->ipv6_prefix.s6_addr, 16);
 282             mp_reach_nlri_partial.next_hop[15] = 1;
 283         }
 284
 285         memcpy(&a.data.s.value, &mp_reach_nlri_partial,
 286                 sizeof(struct bgp_attr_mp_reach_nlri_partial));
 287         memcpy(&peer->mp_reach_nlri_partial, &a,
 288                 BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE);
 289
 290         a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL | BGP_PATH_ATTR_FLAG_EXTLEN;
 291         a.code = BGP_PATH_ATTR_CODE_MP_UNREACH_NLRI;
 292         a.data.e.len = 0; /* will be set on UPDATE */
 293
 294         mp_unreach_nlri_partial.afi = htons(BGP_MP_AFI_IPv6);
 295         mp_unreach_nlri_partial.safi = BGP_MP_SAFI_UNICAST;
 296
 297         memcpy(&a.data.e.value, &mp_unreach_nlri_partial,
 298                 sizeof(struct bgp_attr_mp_unreach_nlri_partial));
 299         memcpy(&peer->mp_unreach_nlri_partial, &a,
 300                 BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE);
 301     }
 302
 303     peer->mp_handling = HandlingUnknown;
 304
 305     LOG(4, 0, 0, "Initiating BGP connection to %s (routing %s)\n",
 306         name, enable ? "enabled" : "suspended");
 307
 308     /* we have at least one peer configured */
 309     bgp_configured = 1;
 310
 311     /* connect */
 312     return bgp_connect(peer);
 313 }
 314
 315 /* clear counters, timers, routes and buffers; close socket; move to
 316    next_state, which may be Disabled or Idle */
 317 static void bgp_clear(struct bgp_peer *peer)
 318 {
 319     if (peer->sock != -1)
 320     {
 321         close(peer->sock);
 322         peer->sock = -1;
 323     }
 324
 325     peer->keepalive_time = 0;
 326     peer->expire_time = 0;
 327
 328     peer->keepalive = peer->init_keepalive;
 329     peer->hold = peer->init_hold;
 330
 331     bgp_free_routes(peer->routes);
 332     peer->routes = 0;
 333     bgp_free_routes6(peer->routes6);
 334     peer->routes6 = 0;
 335
 336     peer->outbuf->packet.header.len = 0;
 337     peer->outbuf->done = 0;
 338     peer->inbuf->packet.header.len = 0;
 339     peer->inbuf->done = 0;
 340
 341     peer->cli_flag = 0;
 342     peer->events = 0;
 343
 344     if (peer->state != peer->next_state)
 345     {
 346         peer->state = peer->next_state;
 347         peer->state_time = time_now;
 348
 349         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
 350             bgp_state_str(peer->next_state));
 351     }
 352 }
 353
 354 /* initiate a clean shutdown */
 355 void bgp_stop(struct bgp_peer *peer)
 356 {
 357     LOG(4, 0, 0, "Terminating BGP connection to %s\n", peer->name);
 358     bgp_send_notification(peer, BGP_ERR_CEASE, 0);
 359 }
 360
 361 /* drop connection (if any) and set state to Disabled */
 362 void bgp_halt(struct bgp_peer *peer)
 363 {
 364     LOG(4, 0, 0, "Aborting BGP connection to %s\n", peer->name);
 365     peer->next_state = Disabled;
 366     bgp_clear(peer);
 367 }
 368
 369 /* drop connection (if any) and set to Idle for connection retry */
 370 int bgp_restart(struct bgp_peer *peer)
 371 {
 372     peer->next_state = Idle;
 373     bgp_clear(peer);
 374
 375     /* restart now */
 376     peer->retry_time = time_now;
 377     peer->retry_count = 0;
 378
 379     /* connect */
 380     return bgp_connect(peer);
 381 }
 382
 383 static void bgp_set_retry(struct bgp_peer *peer)
 384 {
 385     if (peer->retry_count++ < BGP_MAX_RETRY)
 386     {
 387         peer->retry_time = time_now + (BGP_RETRY_BACKOFF * peer->retry_count);
 388         peer->next_state = Idle;
 389         bgp_clear(peer);
 390     }
 391     else
 392         bgp_halt(peer); /* give up */
 393 }
 394
 395 /* insert route into list; sorted */
 396 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
 397     struct bgp_route_list *new)
 398 {
 399     struct bgp_route_list *p = head;
 400     struct bgp_route_list *e = 0;
 401
 402     while (p && memcmp(&p->dest, &new->dest, sizeof(p->dest)) < 0)
 403     {
 404         e = p;
 405         p = p->next;
 406     }
 407
 408     if (e)
 409     {
 410         new->next = e->next;
 411         e->next = new;
 412     }
 413     else
 414     {
 415         new->next = head;
 416         head = new;
 417     }
 418
 419     return head;
 420 }
 421
 422 /* insert route6 into list; sorted */
 423 static struct bgp_route6_list *bgp_insert_route6(struct bgp_route6_list *head,
 424     struct bgp_route6_list *new)
 425 {
 426     struct bgp_route6_list *p = head;
 427     struct bgp_route6_list *e = 0;
 428
 429     while (p && memcmp(&p->dest, &new->dest, sizeof(p->dest)) < 0)
 430     {
 431         e = p;
 432         p = p->next;
 433     }
 434
 435     if (e)
 436     {
 437         new->next = e->next;
 438         e->next = new;
 439     }
 440     else
 441     {
 442         new->next = head;
 443         head = new;
 444     }
 445
 446     return head;
 447 }
 448
 449 /* add route to list for peers */
 450 /*
 451  * Note:  this doesn't do route aggregation, nor drop routes if a less
 452  * specific match already exists (partly because I'm lazy, but also so
 453  * that if that route is later deleted we don't have to be concerned
 454  * about adding back the more specific one).
 455  */
 456 int bgp_add_route(in_addr_t ip, int prefixlen)
 457 {
 458     struct bgp_route_list *r = bgp_routes;
 459     struct bgp_route_list add;
 460     int i;
 461
 462     add.dest.prefix = ip;
 463     add.dest.len = prefixlen;
 464     add.next = 0;
 465
 466     /* check for duplicate */
 467     while (r)
 468     {
 469         i = memcmp(&r->dest, &add.dest, sizeof(r->dest));
 470         if (!i)
 471             return 1; /* already covered */
 472
 473         if (i > 0)
 474             break;
 475
 476         r = r->next;
 477     }
 478
 479     /* insert into route list; sorted */
 480     if (!(r = malloc(sizeof(*r))))
 481     {
 482         LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
 483             fmtaddr(add.dest.prefix, 0), add.dest.len, strerror(errno));
 484
 485         return 0;
 486     }
 487
 488     memcpy(r, &add, sizeof(*r));
 489     bgp_routes = bgp_insert_route(bgp_routes, r);
 490
 491     /* flag established peers for update */
 492     for (i = 0; i < BGP_NUM_PEERS; i++)
 493         if (bgp_peers[i].state == Established)
 494             bgp_peers[i].update_routes = 1;
 495
 496     LOG(4, 0, 0, "Registered BGP route %s/%d\n",
 497         fmtaddr(add.dest.prefix, 0), add.dest.len);
 498
 499     return 1;
 500 }
 501
 502 /* add route to list for peers */
 503 /*
 504  * Note: same provisions as above
 505  */
 506 int bgp_add_route6(struct in6_addr ip, int prefixlen)
 507 {
 508     struct bgp_route6_list *r = bgp_routes6;
 509     struct bgp_route6_list add;
 510     int i;
 511     char ipv6addr[INET6_ADDRSTRLEN];
 512
 513     memcpy(&add.dest.prefix, &ip.s6_addr, 16);
 514     add.dest.len = prefixlen;
 515     add.next = 0;
 516
 517     /* check for duplicate */
 518     while (r)
 519     {
 520         i = memcmp(&r->dest, &add.dest, sizeof(r->dest));
 521         if (!i)
 522             return 1; /* already covered */
 523
 524         if (i > 0)
 525             break;
 526
 527         r = r->next;
 528     }
 529
 530     /* insert into route list; sorted */
 531     if (!(r = malloc(sizeof(*r))))
 532     {
 533         LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
 534             inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN), add.dest.len,
 535             strerror(errno));
 536
 537         return 0;
 538     }
 539
 540     memcpy(r, &add, sizeof(*r));
 541     bgp_routes6 = bgp_insert_route6(bgp_routes6, r);
 542
 543     /* flag established peers for update */
 544     for (i = 0; i < BGP_NUM_PEERS; i++)
 545         if (bgp_peers[i].state == Established
 546                 && bgp_peers[i].mp_handling == HandleIPv6Routes)
 547             bgp_peers[i].update_routes6 = 1;
 548
 549     LOG(4, 0, 0, "Registered BGP route %s/%d\n",
 550         inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN), add.dest.len);
 551
 552     return 1;
 553 }
 554
 555 /* remove route from list for peers */
 556 int bgp_del_route(in_addr_t ip, int prefixlen)
 557 {
 558     struct bgp_route_list *r = bgp_routes;
 559     struct bgp_route_list *e = 0;
 560     struct bgp_route_list del;
 561     int i;
 562
 563     del.dest.prefix = ip;
 564     del.dest.len = prefixlen;
 565     del.next = 0;
 566
 567     /* find entry in routes list and remove */
 568     while (r)
 569     {
 570         i = memcmp(&r->dest, &del.dest, sizeof(r->dest));
 571         if (!i)
 572         {
 573             if (e)
 574                 e->next = r->next;
 575             else
 576                 bgp_routes = r->next;
 577
 578             free(r);
 579             break;
 580         }
 581
 582         e = r;
 583
 584         if (i > 0)
 585             r = 0; /* stop */
 586         else
 587             r = r->next;
 588     }
 589
 590     /* not found */
 591     if (!r)
 592         return 1;
 593
 594     /* flag established peers for update */
 595     for (i = 0; i < BGP_NUM_PEERS; i++)
 596         if (bgp_peers[i].state == Established)
 597             bgp_peers[i].update_routes = 1;
 598
 599     LOG(4, 0, 0, "Removed BGP route %s/%d\n",
 600         fmtaddr(del.dest.prefix, 0), del.dest.len);
 601
 602     return 1;
 603 }
 604
 605 /* remove route from list for peers */
 606 int bgp_del_route6(struct in6_addr ip, int prefixlen)
 607 {
 608     struct bgp_route6_list *r = bgp_routes6;
 609     struct bgp_route6_list *e = 0;
 610     struct bgp_route6_list del;
 611     int i;
 612     char ipv6addr[INET6_ADDRSTRLEN];
 613
 614     memcpy(&del.dest.prefix, &ip.s6_addr, 16);
 615     del.dest.len = prefixlen;
 616     del.next = 0;
 617
 618     /* find entry in routes list and remove */
 619     while (r)
 620     {
 621         i = memcmp(&r->dest, &del.dest, sizeof(r->dest));
 622         if (!i)
 623         {
 624             if (e)
 625                 e->next = r->next;
 626             else
 627                 bgp_routes6 = r->next;
 628
 629             free(r);
 630             break;
 631         }
 632
 633         e = r;
 634
 635         if (i > 0)
 636             r = 0; /* stop */
 637         else
 638             r = r->next;
 639     }
 640
 641     /* not found */
 642     if (!r)
 643         return 1;
 644
 645     /* flag established peers for update */
 646     for (i = 0; i < BGP_NUM_PEERS; i++)
 647         if (bgp_peers[i].state == Established
 648                 && bgp_peers[i].mp_handling == HandleIPv6Routes)
 649             bgp_peers[i].update_routes6 = 1;
 650
 651     LOG(4, 0, 0, "Removed BGP route %s/%d\n",
 652         inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN), del.dest.len);
 653
 654     return 1;
 655 }
 656
 657 /* enable or disable routing */
 658 void bgp_enable_routing(int enable)
 659 {
 660     int i;
 661
 662     for (i = 0; i < BGP_NUM_PEERS; i++)
 663     {
 664         bgp_peers[i].routing = enable;
 665
 666         /* flag established peers for update */
 667         if (bgp_peers[i].state == Established)
 668             bgp_peers[i].update_routes = 1;
 669     }
 670
 671     LOG(4, 0, 0, "%s BGP routing\n", enable ? "Enabled" : "Suspended");
 672 }
 673
 674 #ifdef HAVE_EPOLL
 675 # include <sys/epoll.h>
 676 #else
 677 # include "fake_epoll.h"
 678 #endif
 679
 680 /* return a bitmask of the events required to poll this peer's fd */
 681 int bgp_set_poll()
 682 {
 683     int i;
 684
 685     if (!bgp_configured)
 686         return 0;
 687
 688     for (i = 0; i < BGP_NUM_PEERS; i++)
 689     {
 690         struct bgp_peer *peer = &bgp_peers[i];
 691         int events = 0;
 692
 693         if (peer->state == Disabled || peer->state == Idle)
 694             continue;
 695
 696         if (peer->inbuf->done < BGP_MAX_PACKET_SIZE)
 697             events |= EPOLLIN;
 698
 699         if (peer->state == Connect ||           /* connection in progress */
 700             peer->update_routes ||              /* routing updates */
 701             peer->outbuf->packet.header.len)    /* pending output */
 702             events |= EPOLLOUT;
 703
 704         if (peer->events != events)
 705         {
 706             struct epoll_event ev;
 707
 708             ev.events = peer->events = events;
 709             ev.data.ptr = &peer->edata;
 710             epoll_ctl(epollfd, EPOLL_CTL_MOD, peer->sock, &ev);
 711         }
 712     }
 713
 714     return 1;
 715 }
 716
 717 /* process bgp events/timers */
 718 int bgp_process(uint32_t events[])
 719 {
 720     int i;
 721
 722     if (!bgp_configured)
 723         return 0;
 724
 725     for (i = 0; i < BGP_NUM_PEERS; i++)
 726     {
 727         struct bgp_peer *peer = &bgp_peers[i];
 728
 729         if (*peer->name && peer->cli_flag == BGP_CLI_RESTART)
 730         {
 731             bgp_restart(peer);
 732             continue;
 733         }
 734
 735         if (peer->state == Disabled)
 736             continue;
 737
 738         if (peer->cli_flag)
 739         {
 740             switch (peer->cli_flag)
 741             {
 742             case BGP_CLI_SUSPEND:
 743                 if (peer->routing)
 744                 {
 745                     peer->routing = 0;
 746                     if (peer->state == Established)
 747                         peer->update_routes = 1;
 748                 }
 749
 750                 break;
 751
 752             case BGP_CLI_ENABLE:
 753                 if (!peer->routing)
 754                 {
 755                     peer->routing = 1;
 756                     if (peer->state == Established)
 757                         peer->update_routes = 1;
 758                 }
 759
 760                 break;
 761             }
 762
 763             peer->cli_flag = 0;
 764         }
 765
 766         /* handle empty/fill of buffers */
 767         if (events[i] & EPOLLOUT)
 768         {
 769             int r = 1;
 770             if (peer->state == Connect)
 771                 r = bgp_handle_connect(peer);
 772             else if (peer->outbuf->packet.header.len)
 773                 r = bgp_write(peer);
 774
 775             if (!r)
 776                 continue;
 777         }
 778
 779         if (events[i] & (EPOLLIN|EPOLLHUP))
 780         {
 781             if (!bgp_read(peer))
 782                 continue;
 783         }
 784
 785         /* process input buffer contents */
 786         while (peer->inbuf->done >= sizeof(peer->inbuf->packet.header)
 787             && !peer->outbuf->packet.header.len) /* may need to queue a response */
 788         {
 789             if (bgp_handle_input(peer) < 0)
 790                 continue;
 791         }
 792
 793         /* process pending updates */
 794         if (peer->update_routes
 795             && !peer->outbuf->packet.header.len) /* ditto */
 796         {
 797             if (!bgp_send_update(peer))
 798                 continue;
 799         }
 800
 801         /* process pending IPv6 updates */
 802         if (peer->update_routes6
 803             && !peer->outbuf->packet.header.len) /* ditto */
 804         {
 805             if (!bgp_send_update6(peer))
 806                 continue;
 807         }
 808
 809         /* process timers */
 810         bgp_process_timers(peer);
 811     }
 812
 813     return 1;
 814 }
 815
 816 /* process bgp timers only */
 817 void bgp_process_peers_timers()
 818 {
 819     int i;
 820
 821     if (!bgp_configured)
 822         return;
 823
 824     for (i = 0; i < BGP_NUM_PEERS; i++)
 825     {
 826         struct bgp_peer *peer = &bgp_peers[i];
 827
 828         if (peer->state == Disabled)
 829             continue;
 830
 831         bgp_process_timers(peer);
 832     }
 833 }
 834
 835 static void bgp_process_timers(struct bgp_peer *peer)
 836 {
 837     if (peer->state == Established)
 838     {
 839         if (time_now > peer->expire_time)
 840         {
 841             LOG(1, 0, 0, "No message from BGP peer %s in %ds\n",
 842                 peer->name, peer->hold);
 843
 844             bgp_send_notification(peer, BGP_ERR_HOLD_TIMER_EXP, 0);
 845             return;
 846         }
 847
 848         if (time_now > peer->keepalive_time && !peer->outbuf->packet.header.len)
 849             bgp_send_keepalive(peer);
 850     }
 851     else if (peer->state == Idle)
 852     {
 853         if (time_now > peer->retry_time)
 854             bgp_connect(peer);
 855     }
 856     else if (time_now > peer->state_time + BGP_STATE_TIME)
 857     {
 858         LOG(1, 0, 0, "%s timer expired for BGP peer %s\n",
 859             bgp_state_str(peer->state), peer->name);
 860
 861         bgp_restart(peer);
 862     }
 863 }
 864
 865 static void bgp_free_routes(struct bgp_route_list *routes)
 866 {
 867     struct bgp_route_list *tmp;
 868
 869     while ((tmp = routes))
 870     {
 871         routes = tmp->next;
 872         free(tmp);
 873     }
 874 }
 875
 876 static void bgp_free_routes6(struct bgp_route6_list *routes)
 877 {
 878     struct bgp_route6_list *tmp;
 879
 880     while ((tmp = routes))
 881     {
 882         routes = tmp->next;
 883         free(tmp);
 884     }
 885 }
 886
 887 char const *bgp_state_str(enum bgp_state state)
 888 {
 889     switch (state)
 890     {
 891     case Disabled:      return "Disabled";
 892     case Idle:          return "Idle";
 893     case Connect:       return "Connect";
 894     case Active:        return "Active";
 895     case OpenSent:      return "OpenSent";
 896     case OpenConfirm:   return "OpenConfirm";
 897     case Established:   return "Established";
 898     }
 899
 900     return "?";
 901 }
 902
 903 static char const *bgp_msg_type_str(uint8_t type)
 904 {
 905     switch (type)
 906     {
 907     case BGP_MSG_OPEN:          return "OPEN";
 908     case BGP_MSG_UPDATE:        return "UPDATE";
 909     case BGP_MSG_NOTIFICATION:  return "NOTIFICATION";
 910     case BGP_MSG_KEEPALIVE:     return "KEEPALIVE";
 911     }
 912
 913     return "?";
 914 }
 915
 916 /* attempt to connect to peer */
 917 static int bgp_connect(struct bgp_peer *peer)
 918 {
 919     static int bgp_port = 0;
 920     struct sockaddr_in addr;
 921     struct sockaddr_in source_addr;
 922     struct epoll_event ev;
 923
 924     if (!bgp_port)
 925     {
 926         struct servent *serv;
 927         if (!(serv = getservbyname("bgp", "tcp")))
 928         {
 929             LOG(0, 0, 0, "Can't get bgp service (%s)\n", strerror(errno));
 930             return 0;
 931         }
 932
 933         bgp_port = serv->s_port;
 934     }
 935
 936     if ((peer->sock = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
 937     {
 938         LOG(0, 0, 0, "Can't create a socket for BGP peer %s (%s)\n",
 939             peer->name, strerror(errno));
 940
 941         peer->state = peer->next_state = Disabled;
 942         return 0;
 943     }
 944
 945     /* add to poll set */
 946     ev.events = peer->events = EPOLLOUT;
 947     ev.data.ptr = &peer->edata;
 948     epoll_ctl(epollfd, EPOLL_CTL_ADD, peer->sock, &ev);
 949
 950     /* set to non-blocking */
 951     fcntl(peer->sock, F_SETFL, fcntl(peer->sock, F_GETFL, 0) | O_NONBLOCK);
 952
 953     /* set source address */
 954     memset(&source_addr, 0, sizeof(source_addr));
 955     source_addr.sin_family = AF_INET;
 956     source_addr.sin_addr.s_addr = peer->source_addr; /* defaults to INADDR_ANY */
 957     if (bind(peer->sock, (struct sockaddr *) &source_addr, sizeof(source_addr)) < 0)
 958     {
 959         LOG(1, 0, 0, "Can't set source address to %s: %s\n",
 960             inet_ntoa(source_addr.sin_addr), strerror(errno));
 961
 962         bgp_set_retry(peer);
 963         return 0;
 964     }
 965
 966     /* try connect */
 967     memset(&addr, 0, sizeof(addr));
 968     addr.sin_family = AF_INET;
 969     addr.sin_port = bgp_port;
 970     addr.sin_addr.s_addr = peer->addr;
 971
 972     while (connect(peer->sock, (struct sockaddr *) &addr, sizeof(addr)) == -1)
 973     {
 974         if (errno == EINTR) /* SIGALARM handler */
 975             continue;
 976
 977         if (errno != EINPROGRESS)
 978         {
 979             LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n",
 980                 inet_ntoa(addr.sin_addr), strerror(errno));
 981
 982             bgp_set_retry(peer);
 983             return 0;
 984         }
 985
 986         peer->state = Connect;
 987         peer->state_time = time_now;
 988
 989         LOG(4, 0, 0, "BGP peer %s: state Connect\n", peer->name);
 990         return 1;
 991     }
 992
 993     peer->state = Active;
 994     peer->state_time = time_now;
 995     peer->retry_time = peer->retry_count = 0;
 996
 997     LOG(4, 0, 0, "BGP peer %s: state Active\n", inet_ntoa(addr.sin_addr));
 998
 999     return bgp_send_open(peer);
1000 }
1001
1002 /* complete partial connection (state = Connect) */
1003 static int bgp_handle_connect(struct bgp_peer *peer)
1004 {
1005     int err = 0;
1006     socklen_t len = sizeof(int);
1007     getsockopt(peer->sock, SOL_SOCKET, SO_ERROR, &err, &len);
1008     if (err)
1009     {
1010         LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n", peer->name,
1011             strerror(err));
1012
1013         bgp_set_retry(peer);
1014         return 0;
1015     }
1016
1017     peer->state = Active;
1018     peer->state_time = time_now;
1019
1020     LOG(4, 0, 0, "BGP peer %s: state Active\n", peer->name);
1021
1022     return bgp_send_open(peer);
1023 }
1024
1025 /* initiate a write */
1026 static int bgp_write(struct bgp_peer *peer)
1027 {
1028     int len = htons(peer->outbuf->packet.header.len);
1029     int r;
1030
1031     while ((r = write(peer->sock, &peer->outbuf->packet + peer->outbuf->done,
1032         len - peer->outbuf->done)) == -1)
1033     {
1034         if (errno == EINTR)
1035             continue;
1036
1037         if (errno == EAGAIN)
1038             return 1;
1039
1040         if (errno == EPIPE)
1041             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
1042         else
1043             LOG(1, 0, 0, "Can't write to BGP peer %s (%s)\n", peer->name,
1044                 strerror(errno));
1045
1046         bgp_set_retry(peer);
1047         return 0;
1048     }
1049
1050     if (r < len)
1051     {
1052         peer->outbuf->done += r;
1053         return 1;
1054     }
1055
1056     LOG(4, 0, 0, "Sent %s to BGP peer %s\n",
1057         bgp_msg_type_str(peer->outbuf->packet.header.type), peer->name);
1058
1059     peer->outbuf->packet.header.len = 0;
1060     peer->outbuf->done = 0;
1061
1062     if (peer->state == Established)
1063         peer->keepalive_time = time_now + peer->keepalive;
1064
1065     if (peer->state != peer->next_state)
1066     {
1067         if (peer->next_state == Disabled || peer->next_state == Idle)
1068         {
1069             bgp_clear(peer);
1070             return 0;
1071         }
1072
1073         peer->state = peer->next_state;
1074         peer->state_time = time_now;
1075
1076         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
1077             bgp_state_str(peer->state));
1078     }
1079
1080     return 1;
1081 }
1082
1083 /* initiate a read */
1084 static int bgp_read(struct bgp_peer *peer)
1085 {
1086     int r;
1087
1088     while ((r = read(peer->sock, &peer->inbuf->packet + peer->inbuf->done,
1089         BGP_MAX_PACKET_SIZE - peer->inbuf->done)) < 1)
1090     {
1091         if (!r)
1092         {
1093             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
1094         }
1095         else
1096         {
1097             if (errno == EINTR)
1098                 continue;
1099
1100             if (errno == EAGAIN)
1101                 return 1;
1102
1103             LOG(1, 0, 0, "Can't read from BGP peer %s (%s)\n", peer->name,
1104                 strerror(errno));
1105         }
1106
1107         bgp_set_retry(peer);
1108         return 0;
1109     }
1110
1111     peer->inbuf->done += r;
1112     return 1;
1113 }
1114
1115 /* process buffered packets */
1116 static int bgp_handle_input(struct bgp_peer *peer)
1117 {
1118     struct bgp_packet *p = &peer->inbuf->packet;
1119     int len = ntohs(p->header.len);
1120
1121     if (len > BGP_MAX_PACKET_SIZE)
1122     {
1123         LOG(1, 0, 0, "Bad header length from BGP %s\n", peer->name);
1124         bgp_send_notification(peer, BGP_ERR_HEADER, BGP_ERR_HDR_BAD_LEN);
1125         return 0;
1126     }
1127
1128     if (peer->inbuf->done < len)
1129         return 0;
1130
1131     LOG(4, 0, 0, "Received %s from BGP peer %s\n",
1132         bgp_msg_type_str(p->header.type), peer->name);
1133
1134     switch (p->header.type)
1135     {
1136     case BGP_MSG_OPEN:
1137         {
1138             struct bgp_data_open data;
1139             int hold;
1140             int i;
1141             off_t param_offset, capability_offset;
1142             struct bgp_opt_param *param;
1143             uint8_t capabilities_len;
1144             char *capabilities = NULL;
1145             struct bgp_capability *capability;
1146             struct bgp_mp_cap_param *mp_cap;
1147
1148             for (i = 0; i < sizeof(p->header.marker); i++)
1149             {
1150                 if ((unsigned char) p->header.marker[i] != 0xff)
1151                 {
1152                     LOG(1, 0, 0, "Invalid marker from BGP peer %s\n",
1153                         peer->name);
1154
1155                     bgp_send_notification(peer, BGP_ERR_HEADER,
1156                         BGP_ERR_HDR_NOT_SYNC);
1157
1158                     return 0;
1159                 }
1160             }
1161
1162             if (peer->state != OpenSent)
1163             {
1164                 LOG(1, 0, 0, "OPEN from BGP peer %s in %s state\n",
1165                     peer->name, bgp_state_str(peer->state));
1166
1167                 bgp_send_notification(peer, BGP_ERR_FSM, 0);
1168                 return 0;
1169             }
1170
1171             memcpy(&data, p->data, len - sizeof(p->header));
1172
1173             if (data.version != BGP_VERSION)
1174             {
1175                 LOG(1, 0, 0, "Bad version (%d) sent by BGP peer %s\n",
1176                     (int) data.version, peer->name);
1177
1178                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_VERSION);
1179                 return 0;
1180             }
1181
1182             if (ntohs(data.as) != peer->as)
1183             {
1184                 LOG(1, 0, 0, "Bad AS sent by BGP peer %s (got %d, "
1185                     "expected %d)\n", peer->name, (int) htons(data.as),
1186                     (int) peer->as);
1187
1188                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_BAD_AS);
1189                 return 0;
1190             }
1191
1192             if ((hold = ntohs(data.hold_time)) < 3)
1193             {
1194                 LOG(1, 0, 0, "Bad hold time (%d) from BGP peer %s\n",
1195                     hold, peer->name);
1196
1197                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_HOLD_TIME);
1198                 return 0;
1199             }
1200
1201             /* pick lowest hold time */
1202             if (hold < peer->hold)
1203                 peer->hold = hold;
1204
1205             /* adjust our keepalive based on negotiated hold value */
1206             if (peer->keepalive * 3 > peer->hold)
1207                 peer->keepalive = peer->hold / 3;
1208
1209             /* check for optional parameters */
1210             /* 2 is for the size of type + len (both uint8_t) */
1211             for (param_offset = 0;
1212                     param_offset < data.opt_len;
1213                     param_offset += 2 + param->len)
1214             {
1215                 param = (struct bgp_opt_param *)((char *)&data.opt_params + param_offset);
1216
1217                 /* sensible check */
1218                 if (data.opt_len - param_offset < 2
1219                         || param->len > data.opt_len - param_offset - 2)
1220                 {
1221                     LOG(1, 0, 0, "Malformed Optional Parameter list from BGP peer %s\n",
1222                         peer->name);
1223
1224                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
1225                     return 0;
1226                 }
1227
1228                 /* we know only one parameter type */
1229                 if (param->type != BGP_PARAM_TYPE_CAPABILITY)
1230                 {
1231                     LOG(1, 0, 0, "Unsupported Optional Parameter type %d from BGP peer %s\n",
1232                         param->type, peer->name);
1233
1234                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_PARAM);
1235                     return 0;
1236                 }
1237
1238                 capabilities_len = param->len;
1239                 capabilities = (char *)&param->value;
1240
1241                 /* look for BGP multiprotocol capability */
1242                 for (capability_offset = 0;
1243                         capability_offset < capabilities_len;
1244                         capability_offset += 2 + capability->len)
1245                 {
1246                     capability = (struct bgp_capability *)(capabilities + capability_offset);
1247
1248                     /* sensible check */
1249                     if (capabilities_len - capability_offset < 2
1250                             || capability->len > capabilities_len - capability_offset - 2)
1251                     {
1252                         LOG(1, 0, 0, "Malformed Capabilities list from BGP peer %s\n",
1253                             peer->name);
1254
1255                         bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
1256                         return 0;
1257                     }
1258
1259                     /* we only know one capability code */
1260                     if (capability->code != BGP_CAP_CODE_MP
1261                             && capability->len != sizeof(struct bgp_mp_cap_param))
1262                     {
1263                         LOG(4, 0, 0, "Unsupported Capability code %d from BGP peer %s\n",
1264                             capability->code, peer->name);
1265
1266                         /* we don't terminate, still; we just jump to the next one */
1267                         continue;
1268                     }
1269
1270                     mp_cap = (struct bgp_mp_cap_param *)&capability->value;
1271                     /* the only <AFI, SAFI> tuple we support */
1272                     if (ntohs(mp_cap->afi) != BGP_MP_AFI_IPv6 && mp_cap->safi != BGP_MP_SAFI_UNICAST)
1273                     {
1274                         LOG(4, 0, 0, "Unsupported multiprotocol AFI %d and SAFI %d from BGP peer %s\n",
1275                             mp_cap->afi, mp_cap->safi, peer->name);
1276
1277                         /* we don't terminate, still; we just jump to the next one */
1278                         continue;
1279                     }
1280
1281                     /* yes it can! */
1282                     peer->mp_handling = HandleIPv6Routes;
1283                 }
1284             }
1285
1286             if (peer->mp_handling != HandleIPv6Routes)
1287             {
1288                 peer->mp_handling = DoesntHandleIPv6Routes;
1289                 if (config->ipv6_prefix.s6_addr[0])
1290                     LOG(1, 0, 0, "Warning: BGP peer %s doesn't handle IPv6 prefixes updates\n",
1291                             peer->name);
1292             }
1293
1294             /* next transition requires an exchange of keepalives */
1295             bgp_send_keepalive(peer);
1296         }
1297
1298         break;
1299
1300     case BGP_MSG_KEEPALIVE:
1301         if (peer->state == OpenConfirm)
1302         {
1303             peer->state = peer->next_state = Established;
1304             peer->state_time = time_now;
1305             peer->keepalive_time = time_now + peer->keepalive;
1306             peer->update_routes = 1;
1307             peer->retry_count = 0;
1308             peer->retry_time = 0;
1309
1310             LOG(4, 0, 0, "BGP peer %s: state Established\n", peer->name);
1311         }
1312
1313         break;
1314
1315     case BGP_MSG_NOTIFICATION:
1316         if (len > sizeof(p->header))
1317         {
1318             struct bgp_data_notification *notification =
1319                 (struct bgp_data_notification *) p->data;
1320
1321             if (notification->error_code == BGP_ERR_CEASE)
1322             {
1323                 LOG(4, 0, 0, "BGP peer %s sent CEASE\n", peer->name);
1324                 bgp_set_retry(peer);
1325                 return 0;
1326             }
1327
1328             if (notification->error_code == BGP_ERR_OPEN
1329                     && notification->error_subcode == BGP_ERR_OPN_UNSUP_PARAM)
1330             {
1331                 LOG(4, 0, 0, "BGP peer %s doesn't support BGP Capabilities\n", peer->name);
1332                 peer->mp_handling = DoesntHandleIPv6Routes;
1333                 bgp_set_retry(peer);
1334                 return 0;
1335             }
1336
1337             if (notification->error_code == BGP_ERR_OPEN
1338                     && notification->error_subcode == BGP_ERR_OPN_UNSUP_CAP)
1339             {
1340                 /* the only capability we advertise is this one, so upon receiving
1341                    an "unsupported capability" message, we disable IPv6 routes for
1342                    this peer */
1343                 LOG(4, 0, 0, "BGP peer %s doesn't support IPv6 routes advertisement\n", peer->name);
1344                 peer->mp_handling = DoesntHandleIPv6Routes;
1345                 break;
1346             }
1347
1348             /* FIXME: should handle more notifications */
1349             LOG(4, 0, 0, "BGP peer %s sent unhandled NOTIFICATION %d\n",
1350                 peer->name, (int) notification->error_code);
1351         }
1352
1353         break;
1354     }
1355
1356     /* reset timer */
1357     peer->expire_time = time_now + peer->hold;
1358
1359     /* see if there's another message in the same packet/buffer */
1360     if (peer->inbuf->done > len)
1361     {
1362         peer->inbuf->done -= len;
1363         memmove(p, (char *) p + len, peer->inbuf->done);
1364     }
1365     else
1366     {
1367         peer->inbuf->packet.header.len = 0;
1368         peer->inbuf->done = 0;
1369     }
1370
1371     return peer->inbuf->done;
1372 }
1373
1374 /* send/buffer OPEN message */
1375 static int bgp_send_open(struct bgp_peer *peer)
1376 {
1377     struct bgp_data_open data;
1378     struct bgp_mp_cap_param mp_ipv6 = { htons(BGP_MP_AFI_IPv6), 0, BGP_MP_SAFI_UNICAST };
1379     struct bgp_capability cap_mp_ipv6;
1380     struct bgp_opt_param param_cap_mp_ipv6;
1381     uint16_t len = sizeof(peer->outbuf->packet.header);
1382
1383     memset(peer->outbuf->packet.header.marker, 0xff,
1384         sizeof(peer->outbuf->packet.header.marker));
1385
1386     peer->outbuf->packet.header.type = BGP_MSG_OPEN;
1387
1388     data.version = BGP_VERSION;
1389     data.as = htons(our_as);
1390     data.hold_time = htons(peer->hold);
1391     /* use the source IP we use as identifier, if available */
1392     if (peer->source_addr != INADDR_ANY)
1393         data.identifier = peer->source_addr;
1394     else
1395         data.identifier = my_address;
1396
1397     /* if we know peer doesn't support MP (mp_handling == DoesntHandleIPv6Routes)
1398        then don't add this parameter */
1399     if (config->ipv6_prefix.s6_addr[0]
1400             && (peer->mp_handling == HandlingUnknown
1401                 || peer->mp_handling == HandleIPv6Routes))
1402     {
1403         /* construct the param and capability */
1404         cap_mp_ipv6.code = BGP_CAP_CODE_MP;
1405         cap_mp_ipv6.len = sizeof(mp_ipv6);
1406         memcpy(&cap_mp_ipv6.value, &mp_ipv6, cap_mp_ipv6.len);
1407
1408         param_cap_mp_ipv6.type = BGP_PARAM_TYPE_CAPABILITY;
1409         param_cap_mp_ipv6.len = 2 + sizeof(mp_ipv6);
1410         memcpy(&param_cap_mp_ipv6.value, &cap_mp_ipv6, param_cap_mp_ipv6.len);
1411
1412         data.opt_len = 2 + param_cap_mp_ipv6.len;
1413         memcpy(&data.opt_params, &param_cap_mp_ipv6, data.opt_len);
1414     }
1415     else
1416         data.opt_len = 0;
1417
1418     memcpy(peer->outbuf->packet.data, &data, BGP_DATA_OPEN_SIZE + data.opt_len);
1419     len += BGP_DATA_OPEN_SIZE + data.opt_len;
1420
1421     peer->outbuf->packet.header.len = htons(len);
1422     peer->outbuf->done = 0;
1423     peer->next_state = OpenSent;
1424
1425     return bgp_write(peer);
1426 }
1427
1428 /* send/buffer KEEPALIVE message */
1429 static int bgp_send_keepalive(struct bgp_peer *peer)
1430 {
1431     memset(peer->outbuf->packet.header.marker, 0xff,
1432         sizeof(peer->outbuf->packet.header.marker));
1433
1434     peer->outbuf->packet.header.type = BGP_MSG_KEEPALIVE;
1435     peer->outbuf->packet.header.len =
1436         htons(sizeof(peer->outbuf->packet.header));
1437
1438     peer->outbuf->done = 0;
1439     peer->next_state = (peer->state == OpenSent) ? OpenConfirm : peer->state;
1440
1441     return bgp_write(peer);
1442 }
1443
1444 /* send/buffer UPDATE message */
1445 static int bgp_send_update(struct bgp_peer *peer)
1446 {
1447     uint16_t unf_len = 0;
1448     uint16_t attr_len;
1449     uint16_t len = sizeof(peer->outbuf->packet.header);
1450     struct bgp_route_list *have = peer->routes;
1451     struct bgp_route_list *want = peer->routing ? bgp_routes : 0;
1452     struct bgp_route_list *e = 0;
1453     struct bgp_route_list *add = 0;
1454     int s;
1455
1456     char *data = (char *) &peer->outbuf->packet.data;
1457
1458     /* need leave room for attr_len, bgp_path_attrs and one prefix */
1459     char *max = (char *) &peer->outbuf->packet.data
1460         + sizeof(peer->outbuf->packet.data)
1461         - sizeof(attr_len) - peer->path_attr_len - sizeof(struct bgp_ip_prefix);
1462
1463     /* skip over unf_len */
1464     data += sizeof(unf_len);
1465     len += sizeof(unf_len);
1466
1467     memset(peer->outbuf->packet.header.marker, 0xff,
1468         sizeof(peer->outbuf->packet.header.marker));
1469
1470     peer->outbuf->packet.header.type = BGP_MSG_UPDATE;
1471
1472     peer->update_routes = 0; /* tentatively clear */
1473
1474     /* find differences */
1475     while ((have || want) && data < (max - sizeof(struct bgp_ip_prefix)))
1476     {
1477         if (have)
1478             s = want
1479                 ? memcmp(&have->dest, &want->dest, sizeof(have->dest))
1480                 : -1;
1481         else
1482             s = 1;
1483
1484         if (s < 0) /* found one to delete */
1485         {
1486             struct bgp_route_list *tmp = have;
1487             have = have->next;
1488
1489             s = BGP_IP_PREFIX_SIZE(tmp->dest);
1490             memcpy(data, &tmp->dest, s);
1491             data += s;
1492             unf_len += s;
1493             len += s;
1494
1495             LOG(5, 0, 0, "Withdrawing route %s/%d from BGP peer %s\n",
1496                 fmtaddr(tmp->dest.prefix, 0), tmp->dest.len, peer->name);
1497
1498             free(tmp);
1499
1500             if (e)
1501                 e->next = have;
1502             else
1503                 peer->routes = have;
1504         }
1505         else
1506         {
1507             if (!s) /* same */
1508             {
1509                 e = have; /* stash the last found to relink above */
1510                 have = have->next;
1511                 want = want->next;
1512             }
1513             else if (s > 0) /* addition reqd. */
1514             {
1515                 if (add)
1516                 {
1517                     peer->update_routes = 1; /* only one add per packet */
1518                     if (!have)
1519                         break;
1520                 }
1521                 else
1522                     add = want;
1523
1524                 if (want)
1525                     want = want->next;
1526             }
1527         }
1528     }
1529
1530     if (have || want)
1531         peer->update_routes = 1; /* more to do */
1532
1533     /* anything changed? */
1534     if (!(unf_len || add))
1535         return 1;
1536
1537     /* go back and insert unf_len */
1538     unf_len = htons(unf_len);
1539     memcpy(&peer->outbuf->packet.data, &unf_len, sizeof(unf_len));
1540
1541     if (add)
1542     {
1543         if (!(e = malloc(sizeof(*e))))
1544         {
1545             LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
1546                 fmtaddr(add->dest.prefix, 0), add->dest.len, strerror(errno));
1547
1548             return 0;
1549         }
1550
1551         memcpy(e, add, sizeof(*e));
1552         e->next = 0;
1553         peer->routes = bgp_insert_route(peer->routes, e);
1554
1555         attr_len = htons(peer->path_attr_len);
1556         memcpy(data, &attr_len, sizeof(attr_len));
1557         data += sizeof(attr_len);
1558         len += sizeof(attr_len);
1559
1560         memcpy(data, peer->path_attrs, peer->path_attr_len);
1561         data += peer->path_attr_len;
1562         len += peer->path_attr_len;
1563
1564         s = BGP_IP_PREFIX_SIZE(add->dest);
1565         memcpy(data, &add->dest, s);
1566         data += s;
1567         len += s;
1568
1569         LOG(5, 0, 0, "Advertising route %s/%d to BGP peer %s\n",
1570             fmtaddr(add->dest.prefix, 0), add->dest.len, peer->name);
1571     }
1572     else
1573     {
1574         attr_len = 0;
1575         memcpy(data, &attr_len, sizeof(attr_len));
1576         data += sizeof(attr_len);
1577         len += sizeof(attr_len);
1578     }
1579
1580     peer->outbuf->packet.header.len = htons(len);
1581     peer->outbuf->done = 0;
1582
1583     return bgp_write(peer);
1584 }
1585
1586 /* send/buffer UPDATE message for IPv6 routes */
1587 static int bgp_send_update6(struct bgp_peer *peer)
1588 {
1589     uint16_t attr_len;
1590     uint16_t unreach_len = 0;
1591     char *unreach_len_pos;
1592     uint8_t reach_len;
1593     uint16_t len = sizeof(peer->outbuf->packet.header);
1594     struct bgp_route6_list *have = peer->routes6;
1595     struct bgp_route6_list *want = peer->routing ? bgp_routes6 : 0;
1596     struct bgp_route6_list *e = 0;
1597     struct bgp_route6_list *add = 0;
1598     int s;
1599     char ipv6addr[INET6_ADDRSTRLEN];
1600
1601     char *data = (char *) &peer->outbuf->packet.data;
1602
1603     /* need leave room for attr_len, bgp_path_attrs and one prefix */
1604     char *max = (char *) &peer->outbuf->packet.data
1605         + sizeof(peer->outbuf->packet.data)
1606         - sizeof(attr_len) - peer->path_attr_len_without_nexthop
1607         - BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE - sizeof(struct bgp_ip6_prefix);
1608
1609     memset(peer->outbuf->packet.header.marker, 0xff,
1610         sizeof(peer->outbuf->packet.header.marker));
1611
1612     peer->outbuf->packet.header.type = BGP_MSG_UPDATE;
1613
1614     /* insert non-MP unfeasible routes length */
1615     memcpy(data, &unreach_len, sizeof(unreach_len));
1616     /* skip over it and attr_len too; it will be filled when known */
1617     data += sizeof(unreach_len) + sizeof(attr_len);
1618     len += sizeof(unreach_len) + sizeof(attr_len);
1619
1620     /* copy usual attributes */
1621     memcpy(data, peer->path_attrs, peer->path_attr_len_without_nexthop);
1622     data += peer->path_attr_len_without_nexthop;
1623     attr_len = peer->path_attr_len_without_nexthop;
1624
1625     /* copy MP unreachable NLRI heading */
1626     memcpy(data, peer->mp_unreach_nlri_partial,
1627             BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE);
1628     /* remember where to update this attr len */
1629     unreach_len_pos = data + 2;
1630     data += BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1631     attr_len += BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1632
1633     peer->update_routes6 = 0; /* tentatively clear */
1634
1635     /* find differences */
1636     while ((have || want) && data < (max - sizeof(struct bgp_ip6_prefix)))
1637     {
1638         if (have)
1639             s = want
1640                 ? memcmp(&have->dest, &want->dest, sizeof(have->dest))
1641                 : -1;
1642         else
1643             s = 1;
1644
1645         if (s < 0) /* found one to delete */
1646         {
1647             struct bgp_route6_list *tmp = have;
1648             have = have->next;
1649
1650             s = BGP_IP_PREFIX_SIZE(tmp->dest);
1651             memcpy(data, &tmp->dest, s);
1652             data += s;
1653             unreach_len += s;
1654             attr_len += s;
1655
1656             LOG(5, 0, 0, "Withdrawing route %s/%d from BGP peer %s\n",
1657                 inet_ntop(AF_INET6, &tmp->dest.prefix, ipv6addr, INET6_ADDRSTRLEN),
1658                 tmp->dest.len, peer->name);
1659
1660             free(tmp);
1661
1662             if (e)
1663                 e->next = have;
1664             else
1665                 peer->routes6 = have;
1666         }
1667         else
1668         {
1669             if (!s) /* same */
1670             {
1671                 e = have; /* stash the last found to relink above */
1672                 have = have->next;
1673                 want = want->next;
1674             }
1675             else if (s > 0) /* addition reqd. */
1676             {
1677                 if (add)
1678                 {
1679                     peer->update_routes6 = 1; /* only one add per packet */
1680                     if (!have)
1681                         break;
1682                 }
1683                 else
1684                     add = want;
1685
1686                 if (want)
1687                     want = want->next;
1688             }
1689         }
1690     }
1691
1692     if (have || want)
1693         peer->update_routes6 = 1; /* more to do */
1694
1695     /* anything changed? */
1696     if (!(unreach_len || add))
1697         return 1;
1698
1699     if (unreach_len)
1700     {
1701         /* go back and insert MP unreach_len */
1702         unreach_len += sizeof(struct bgp_attr_mp_unreach_nlri_partial);
1703         unreach_len = htons(unreach_len);
1704         memcpy(unreach_len_pos, &unreach_len, sizeof(unreach_len));
1705     }
1706     else
1707     {
1708         /* we can remove this attribute, then */
1709         data -= BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1710         attr_len -= BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1711     }
1712
1713     if (add)
1714     {
1715         if (!(e = malloc(sizeof(*e))))
1716         {
1717             LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
1718                 inet_ntop(AF_INET6, &add->dest.prefix, ipv6addr, INET6_ADDRSTRLEN),
1719                 add->dest.len, strerror(errno));
1720
1721             return 0;
1722         }
1723
1724         memcpy(e, add, sizeof(*e));
1725         e->next = 0;
1726         peer->routes6 = bgp_insert_route6(peer->routes6, e);
1727
1728         /* copy MP reachable NLRI heading */
1729         memcpy(data, peer->mp_reach_nlri_partial,
1730                 BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE);
1731         /* with proper len */
1732         reach_len = BGP_IP_PREFIX_SIZE(add->dest);
1733         data[2] = sizeof(struct bgp_attr_mp_reach_nlri_partial) + reach_len;
1734         data += BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE;
1735         attr_len += BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE;
1736
1737         memcpy(data, &add->dest, reach_len);
1738         data += reach_len;
1739         attr_len += reach_len;
1740
1741         LOG(5, 0, 0, "Advertising route %s/%d to BGP peer %s\n",
1742             inet_ntop(AF_INET6, &add->dest.prefix, ipv6addr, INET6_ADDRSTRLEN),
1743             add->dest.len, peer->name);
1744     }
1745
1746     /* update len with attributes we added */
1747     len += attr_len;
1748
1749     /* go back and insert attr_len */
1750     attr_len = htons(attr_len);
1751     memcpy((char *)&peer->outbuf->packet.data + 2, &attr_len, sizeof(attr_len));
1752
1753     peer->outbuf->packet.header.len = htons(len);
1754     peer->outbuf->done = 0;
1755
1756     return bgp_write(peer);
1757 }
1758
1759 /* send/buffer NOTIFICATION message */
1760 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
1761     uint8_t subcode)
1762 {
1763     struct bgp_data_notification data;
1764     uint16_t len = 0;
1765
1766     data.error_code = code;
1767     len += sizeof(data.error_code);
1768
1769     data.error_subcode = subcode;
1770     len += sizeof(data.error_code);
1771
1772     memset(peer->outbuf->packet.header.marker, 0xff,
1773         sizeof(peer->outbuf->packet.header.marker));
1774
1775     peer->outbuf->packet.header.type = BGP_MSG_NOTIFICATION;
1776     peer->outbuf->packet.header.len =
1777         htons(sizeof(peer->outbuf->packet.header) + len);
1778
1779     memcpy(peer->outbuf->packet.data, &data, len);
1780
1781     peer->outbuf->done = 0;
1782     peer->next_state = code == BGP_ERR_CEASE ? Disabled : Idle;
1783
1784     /* we're dying; ignore any pending input */
1785     peer->inbuf->packet.header.len = 0;
1786     peer->inbuf->done = 0;
1787
1788     return bgp_write(peer);
1789 }