proto/bgp/attrs.c

   1 /*
   2  *      BIRD -- BGP Attributes
   3  *
   4  *      (c) 2000 Martin Mares <mj@ucw.cz>
   5  *      (c) 2008--2016 Ondrej Zajicek <santiago@crfreenet.org>
   6  *      (c) 2008--2016 CZ.NIC z.s.p.o.
   7  *
   8  *      Can be freely distributed and used under the terms of the GNU GPL.
   9  */
  10
  11 #undef LOCAL_DEBUG
  12
  13 #include <stdlib.h>
  14
  15 #include "nest/bird.h"
  16 #include "nest/iface.h"
  17 #include "nest/protocol.h"
  18 #include "nest/route.h"
  19 #include "nest/attrs.h"
  20 #include "conf/conf.h"
  21 #include "lib/resource.h"
  22 #include "lib/string.h"
  23 #include "lib/unaligned.h"
  24
  25 #include "bgp.h"
  26
  27 /*
  28  *   UPDATE message error handling
  29  *
  30  * All checks from RFC 4271 6.3 are done as specified with these exceptions:
  31  *  - The semantic check of an IP address from NEXT_HOP attribute is missing.
  32  *  - Checks of some optional attribute values are missing.
  33  *  - Syntactic and semantic checks of NLRIs (done in DECODE_PREFIX())
  34  *    are probably inadequate.
  35  *
  36  * Loop detection based on AS_PATH causes updates to be withdrawn. RFC
  37  * 4271 does not explicitly specifiy the behavior in that case.
  38  *
  39  * Loop detection related to route reflection (based on ORIGINATOR_ID
  40  * and CLUSTER_LIST) causes updates to be withdrawn. RFC 4456 8
  41  * specifies that such updates should be ignored, but that is generally
  42  * a bad idea.
  43  *
  44  * BGP attribute table has several hooks:
  45  *
  46  * export - Hook that validates and normalizes attribute during export phase.
  47  * Receives eattr, may modify it (e.g., sort community lists for canonical
  48  * representation), UNSET() it (e.g., skip empty lists), or WITHDRAW() it if
  49  * necessary. May assume that eattr has value valid w.r.t. its type, but may be
  50  * invalid w.r.t. BGP constraints. Optional.
  51  *
  52  * encode - Hook that converts internal representation to external one during
  53  * packet writing. Receives eattr and puts it in the buffer (including attribute
  54  * header). Returns number of bytes, or -1 if not enough space. May assume that
  55  * eattr has value valid w.r.t. its type and validated by export hook. Mandatory
  56  * for all known attributes that exist internally after export phase (i.e., all
  57  * except pseudoattributes MP_(UN)REACH_NLRI).
  58  *
  59  * decode - Hook that converts external representation to internal one during
  60  * packet parsing. Receives attribute data in buffer, validates it and adds
  61  * attribute to ea_list. If data are invalid, steps DISCARD(), WITHDRAW() or
  62  * bgp_parse_error() may be used to escape. Mandatory for all known attributes.
  63  *
  64  * format - Optional hook that converts eattr to textual representation.
  65  */
  66
  67
  68 struct bgp_attr_desc {
  69   const char *name;
  70   uint type;
  71   uint flags;
  72   void (*export)(struct bgp_export_state *s, eattr *a);
  73   int  (*encode)(struct bgp_write_state *s, eattr *a, byte *buf, uint size);
  74   void (*decode)(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to);
  75   void (*format)(eattr *ea, byte *buf, uint size);
  76 };
  77
  78 static const struct bgp_attr_desc bgp_attr_table[];
  79
  80 static inline int bgp_attr_known(uint code);
  81
  82 eattr *
  83 bgp_set_attr(ea_list **attrs, struct linpool *pool, uint code, uint flags, uintptr_t val)
  84 {
  85   ASSERT(bgp_attr_known(code));
  86
  87   ea_list *a = lp_alloc(pool, sizeof(ea_list) + sizeof(eattr));
  88   eattr *e = &a->attrs[0];
  89
  90   a->flags = EALF_SORTED;
  91   a->count = 1;
  92   a->next = *attrs;
  93   *attrs = a;
  94
  95   e->id = EA_CODE(EAP_BGP, code);
  96   e->type = bgp_attr_table[code].type;
  97   e->flags = flags;
  98
  99   if (e->type & EAF_EMBEDDED)
 100     e->u.data = (u32) val;
 101   else
 102     e->u.ptr = (struct adata *) val;
 103
 104   return e;
 105 }
 106
 107
 108
 109 #define REPORT(msg, args...) \
 110   ({ log(L_REMOTE "%s: " msg, s->proto->p.name, ## args); })
 111
 112 #define DISCARD(msg, args...) \
 113   ({ REPORT(msg, ## args); return; })
 114
 115 #define WITHDRAW(msg, args...) \
 116   ({ REPORT(msg, ## args); s->err_withdraw = 1; return; })
 117
 118 #define UNSET(a) \
 119   ({ a->type = EAF_TYPE_UNDEF; return; })
 120
 121 #define NEW_BGP         "Discarding %s attribute received from AS4-aware neighbor"
 122 #define BAD_EBGP        "Discarding %s attribute received from EBGP neighbor"
 123 #define BAD_LENGTH      "Malformed %s attribute - invalid length (%u)"
 124 #define BAD_VALUE       "Malformed %s attribute - invalid value (%u)"
 125 #define NO_MANDATORY    "Missing mandatory %s attribute"
 126
 127
 128 static inline int
 129 bgp_put_attr_hdr3(byte *buf, uint code, uint flags, uint len)
 130 {
 131   *buf++ = flags;
 132   *buf++ = code;
 133   *buf++ = len;
 134   return 3;
 135 }
 136
 137 static inline int
 138 bgp_put_attr_hdr4(byte *buf, uint code, uint flags, uint len)
 139 {
 140   *buf++ = flags | BAF_EXT_LEN;
 141   *buf++ = code;
 142   put_u16(buf, len);
 143   return 4;
 144 }
 145
 146 static inline int
 147 bgp_put_attr_hdr(byte *buf, uint code, uint flags, uint len)
 148 {
 149   if (len < 256)
 150     return bgp_put_attr_hdr3(buf, code, flags, len);
 151   else
 152     return bgp_put_attr_hdr4(buf, code, flags, len);
 153 }
 154
 155 static int
 156 bgp_encode_u8(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size)
 157 {
 158   if (size < (3+1))
 159     return -1;
 160
 161   bgp_put_attr_hdr3(buf, EA_ID(a->id), a->flags, 1);
 162   buf[3] = a->u.data;
 163
 164   return 3+1;
 165 }
 166
 167 static int
 168 bgp_encode_u32(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size)
 169 {
 170   if (size < (3+4))
 171     return -1;
 172
 173   bgp_put_attr_hdr3(buf, EA_ID(a->id), a->flags, 4);
 174   put_u32(buf+3, a->u.data);
 175
 176   return 3+4;
 177 }
 178
 179 static int
 180 bgp_encode_u32s(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size)
 181 {
 182   uint len = a->u.ptr->length;
 183
 184   if (size < (4+len))
 185     return -1;
 186
 187   uint hdr = bgp_put_attr_hdr(buf, EA_ID(a->id), a->flags, len);
 188   put_u32s(buf + hdr, (u32 *) a->u.ptr->data, len / 4);
 189
 190   return hdr + len;
 191 }
 192
 193 static int
 194 bgp_put_attr(byte *buf, uint size, uint code, uint flags, byte *data, uint len)
 195 {
 196   if (size < (4+len))
 197     return -1;
 198
 199   uint hdr = bgp_put_attr_hdr(buf, code, flags, len);
 200   memcpy(buf + hdr, data, len);
 201
 202   return hdr + len;
 203 }
 204
 205 static int
 206 bgp_encode_raw(struct bgp_write_state *s UNUSED, eattr *a, byte *buf, uint size)
 207 {
 208   return bgp_put_attr(buf, size, EA_ID(a->id), a->flags, a->u.ptr->data, a->u.ptr->length);
 209 }
 210
 211
 212 /*
 213  *      Attribute hooks
 214  */
 215
 216 static void
 217 bgp_export_origin(struct bgp_export_state *s, eattr *a)
 218 {
 219   if (a->u.data > 2)
 220     WITHDRAW(BAD_VALUE, "ORIGIN", a->u.data);
 221 }
 222
 223 static void
 224 bgp_decode_origin(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 225 {
 226   if (len != 1)
 227     WITHDRAW(BAD_LENGTH, "ORIGIN", len);
 228
 229   if (data[0] > 2)
 230     WITHDRAW(BAD_VALUE, "ORIGIN", data[0]);
 231
 232   bgp_set_attr_u32(to, s->pool, BA_ORIGIN, flags, data[0]);
 233 }
 234
 235 static void
 236 bgp_format_origin(eattr *a, byte *buf, uint size UNUSED)
 237 {
 238   static const char *bgp_origin_names[] = { "IGP", "EGP", "Incomplete" };
 239
 240   bsprintf(buf, (a->u.data <= 2) ? bgp_origin_names[a->u.data] : "?");
 241 }
 242
 243
 244 static int
 245 bgp_encode_as_path(struct bgp_write_state *s, eattr *a, byte *buf, uint size)
 246 {
 247   byte *data = a->u.ptr->data;
 248   uint len = a->u.ptr->length;
 249
 250   if (!s->as4_session)
 251   {
 252     /* Prepare 16-bit AS_PATH (from 32-bit one) in a temporary buffer */
 253     byte *src = data;
 254     data = alloca(len);
 255     len = as_path_32to16(data, src, len);
 256   }
 257
 258   return bgp_put_attr(buf, size, BA_AS_PATH, a->flags, data, len);
 259 }
 260
 261 static void
 262 bgp_decode_as_path(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 263 {
 264   struct bgp_proto *p = s->proto;
 265   int as_length = s->as4_session ? 4 : 2;
 266   int as_confed = p->cf->confederation && p->is_interior;
 267   char err[128];
 268
 269   if (!as_path_valid(data, len, as_length, as_confed, err, sizeof(err)))
 270     WITHDRAW("Malformed AS_PATH attribute - %s", err);
 271
 272   /* In some circumstances check for initial AS_CONFED_SEQUENCE; RFC 5065 5.0 */
 273   if (p->is_interior && !p->is_internal &&
 274       ((len < 2) || (data[0] != AS_PATH_CONFED_SEQUENCE)))
 275     WITHDRAW("Malformed AS_PATH attribute - %s", "missing initial AS_CONFED_SEQUENCE");
 276
 277   if (!s->as4_session)
 278   {
 279     /* Prepare 32-bit AS_PATH (from 16-bit one) in a temporary buffer */
 280     byte *src = data;
 281     data = alloca(2*len);
 282     len = as_path_16to32(data, src, len);
 283   }
 284
 285   bgp_set_attr_data(to, s->pool, BA_AS_PATH, flags, data, len);
 286 }
 287
 288
 289 static int
 290 bgp_encode_next_hop(struct bgp_write_state *s, eattr *a, byte *buf, uint size)
 291 {
 292   /*
 293    * The NEXT_HOP attribute is used only in traditional (IPv4) BGP. In MP-BGP,
 294    * the next hop is encoded as a part of the MP_REACH_NLRI attribute, so we
 295    * store it and encode it later by AFI-specific hooks.
 296    */
 297
 298   if (s->channel->afi == BGP_AF_IPV4)
 299   {
 300     ASSERT(a->u.ptr->length == sizeof(ip_addr));
 301
 302     if (size < (3+4))
 303       return -1;
 304
 305     bgp_put_attr_hdr3(buf, BA_NEXT_HOP, a->flags, 4);
 306     put_ip4(buf+3, ipa_to_ip4( *(ip_addr *) a->u.ptr->data ));
 307
 308     return 3+4;
 309   }
 310   else
 311   {
 312     s->mp_next_hop = a;
 313     return 0;
 314   }
 315 }
 316
 317 static void
 318 bgp_decode_next_hop(struct bgp_parse_state *s, uint code UNUSED, uint flags UNUSED, byte *data, uint len, ea_list **to UNUSED)
 319 {
 320   if (len != 4)
 321     WITHDRAW(BAD_LENGTH, "NEXT_HOP", len);
 322
 323   /* Semantic checks are done later */
 324   s->ip_next_hop_len = len;
 325   s->ip_next_hop_data = data;
 326 }
 327
 328 /* TODO: This function should use AF-specific hook */
 329 static void
 330 bgp_format_next_hop(eattr *a, byte *buf, uint size UNUSED)
 331 {
 332   ip_addr *nh = (void *) a->u.ptr->data;
 333   uint len = a->u.ptr->length;
 334
 335   ASSERT((len == 16) || (len == 32));
 336
 337   /* in IPv6, we may have two addresses in NEXT HOP */
 338   if ((len == 16) || ipa_zero(nh[1]))
 339     bsprintf(buf, "%I", nh[0]);
 340   else
 341     bsprintf(buf, "%I %I", nh[0], nh[1]);
 342 }
 343
 344
 345 static void
 346 bgp_decode_med(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 347 {
 348   if (len != 4)
 349     WITHDRAW(BAD_LENGTH, "MULTI_EXIT_DISC", len);
 350
 351   u32 val = get_u32(data);
 352   bgp_set_attr_u32(to, s->pool, BA_MULTI_EXIT_DISC, flags, val);
 353 }
 354
 355
 356 static void
 357 bgp_export_local_pref(struct bgp_export_state *s, eattr *a)
 358 {
 359   if (!s->proto->is_interior && !s->proto->cf->allow_local_pref)
 360     UNSET(a);
 361 }
 362
 363 static void
 364 bgp_decode_local_pref(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 365 {
 366   if (!s->proto->is_interior && !s->proto->cf->allow_local_pref)
 367     DISCARD(BAD_EBGP, "LOCAL_PREF");
 368
 369   if (len != 4)
 370     WITHDRAW(BAD_LENGTH, "LOCAL_PREF", len);
 371
 372   u32 val = get_u32(data);
 373   bgp_set_attr_u32(to, s->pool, BA_LOCAL_PREF, flags, val);
 374 }
 375
 376
 377 static void
 378 bgp_decode_atomic_aggr(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data UNUSED, uint len, ea_list **to)
 379 {
 380   if (len != 0)
 381     DISCARD(BAD_LENGTH, "ATOMIC_AGGR", len);
 382
 383   bgp_set_attr_data(to, s->pool, BA_ATOMIC_AGGR, flags, NULL, 0);
 384 }
 385
 386 static int
 387 bgp_encode_aggregator(struct bgp_write_state *s, eattr *a, byte *buf, uint size)
 388 {
 389   byte *data = a->u.ptr->data;
 390   uint len = a->u.ptr->length;
 391
 392   if (!s->as4_session)
 393   {
 394     /* Prepare 16-bit AGGREGATOR (from 32-bit one) in a temporary buffer */
 395     byte *src = data;
 396     data = alloca(6);
 397     len = aggregator_32to16(data, src);
 398   }
 399
 400   return bgp_put_attr(buf, size, BA_AGGREGATOR, a->flags, data, len);
 401 }
 402
 403 static void
 404 bgp_decode_aggregator(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 405 {
 406   if (len != (s->as4_session ? 8 : 6))
 407     DISCARD(BAD_LENGTH, "AGGREGATOR", len);
 408
 409   if (!s->as4_session)
 410   {
 411     /* Prepare 32-bit AGGREGATOR (from 16-bit one) in a temporary buffer */
 412     byte *src = data;
 413     data = alloca(8);
 414     len = aggregator_16to32(data, src);
 415   }
 416
 417   bgp_set_attr_data(to, s->pool, BA_AGGREGATOR, flags, data, len);
 418 }
 419
 420 static void
 421 bgp_format_aggregator(eattr *a, byte *buf, uint size UNUSED)
 422 {
 423   byte *data = a->u.ptr->data;
 424
 425   bsprintf(buf, "%I4 AS%u", get_ip4(data+4), get_u32(data+0));
 426 }
 427
 428
 429 static void
 430 bgp_export_community(struct bgp_export_state *s, eattr *a)
 431 {
 432   if (a->u.ptr->length == 0)
 433     UNSET(a);
 434
 435   a->u.ptr = int_set_sort(s->pool, a->u.ptr);
 436 }
 437
 438 static void
 439 bgp_decode_community(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 440 {
 441   if (!len || (len % 4))
 442     WITHDRAW(BAD_LENGTH, "COMMUNITY", len);
 443
 444   struct adata *ad = lp_alloc_adata(s->pool, len);
 445   get_u32s(data, (u32 *) ad->data, len / 4);
 446   bgp_set_attr_ptr(to, s->pool, BA_COMMUNITY, flags, ad);
 447 }
 448
 449
 450 static void
 451 bgp_export_originator_id(struct bgp_export_state *s, eattr *a)
 452 {
 453   if (!s->proto->is_internal)
 454     UNSET(a);
 455 }
 456
 457 static void
 458 bgp_decode_originator_id(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 459 {
 460   if (!s->proto->is_internal)
 461     DISCARD(BAD_EBGP, "ORIGINATOR_ID");
 462
 463   if (len != 4)
 464     WITHDRAW(BAD_LENGTH, "ORIGINATOR_ID", len);
 465
 466   u32 val = get_u32(data);
 467   bgp_set_attr_u32(to, s->pool, BA_ORIGINATOR_ID, flags, val);
 468 }
 469
 470
 471 static void
 472 bgp_export_cluster_list(struct bgp_export_state *s UNUSED, eattr *a)
 473 {
 474   if (!s->proto->is_internal)
 475     UNSET(a);
 476
 477   if (a->u.ptr->length == 0)
 478     UNSET(a);
 479 }
 480
 481 static void
 482 bgp_decode_cluster_list(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 483 {
 484   if (!s->proto->is_internal)
 485     DISCARD(BAD_EBGP, "CLUSTER_LIST");
 486
 487   if (!len || (len % 4))
 488     WITHDRAW(BAD_LENGTH, "CLUSTER_LIST", len);
 489
 490   struct adata *ad = lp_alloc_adata(s->pool, len);
 491   get_u32s(data, (u32 *) ad->data, len / 4);
 492   bgp_set_attr_ptr(to, s->pool, BA_CLUSTER_LIST, flags, ad);
 493 }
 494
 495 static void
 496 bgp_format_cluster_list(eattr *a, byte *buf, uint size)
 497 {
 498   /* Truncates cluster lists larger than buflen, probably not a problem */
 499   int_set_format(a->u.ptr, 0, -1, buf, size);
 500 }
 501
 502
 503 static inline u32
 504 get_af3(byte *buf)
 505 {
 506   return (get_u16(buf) << 16) | buf[2];
 507 }
 508
 509 static void
 510 bgp_decode_mp_reach_nlri(struct bgp_parse_state *s, uint code UNUSED, uint flags UNUSED, byte *data, uint len, ea_list **to UNUSED)
 511 {
 512   /*
 513    *    2 B     MP_REACH_NLRI data - Address Family Identifier
 514    *    1 B     MP_REACH_NLRI data - Subsequent Address Family Identifier
 515    *    1 B     MP_REACH_NLRI data - Length of Next Hop Network Address
 516    *    var     MP_REACH_NLRI data - Network Address of Next Hop
 517    *    1 B     MP_REACH_NLRI data - Reserved (zero)
 518    *    var     MP_REACH_NLRI data - Network Layer Reachability Information
 519    */
 520
 521   if ((len < 5) || (len < (5 + (uint) data[3])))
 522     bgp_parse_error(s, 9);
 523
 524   s->mp_reach_af = get_af3(data);
 525   s->mp_next_hop_len = data[3];
 526   s->mp_next_hop_data = data + 4;
 527   s->mp_reach_len = len - 5 - s->mp_next_hop_len;
 528   s->mp_reach_nlri = data + 5 + s->mp_next_hop_len;
 529 }
 530
 531
 532 static void
 533 bgp_decode_mp_unreach_nlri(struct bgp_parse_state *s, uint code UNUSED, uint flags UNUSED, byte *data, uint len, ea_list **to UNUSED)
 534 {
 535   /*
 536    *    2 B     MP_UNREACH_NLRI data - Address Family Identifier
 537    *    1 B     MP_UNREACH_NLRI data - Subsequent Address Family Identifier
 538    *    var     MP_UNREACH_NLRI data - Network Layer Reachability Information
 539    */
 540
 541   if (len < 3)
 542     bgp_parse_error(s, 9);
 543
 544   s->mp_unreach_af = get_af3(data);
 545   s->mp_unreach_len = len - 3;
 546   s->mp_unreach_nlri = data + 3;
 547 }
 548
 549
 550 static void
 551 bgp_export_ext_community(struct bgp_export_state *s, eattr *a)
 552 {
 553   a->u.ptr = ec_set_del_nontrans(s->pool, a->u.ptr);
 554
 555   if (a->u.ptr->length == 0)
 556     UNSET(a);
 557
 558   ec_set_sort_x(a->u.ptr);
 559 }
 560
 561 static void
 562 bgp_decode_ext_community(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 563 {
 564   if (!len || (len % 8))
 565     WITHDRAW(BAD_LENGTH, "EXT_COMMUNITY", len);
 566
 567   struct adata *ad = lp_alloc_adata(s->pool, len);
 568   get_u32s(data, (u32 *) ad->data, len / 4);
 569   bgp_set_attr_ptr(to, s->pool, BA_EXT_COMMUNITY, flags, ad);
 570 }
 571
 572
 573 static void
 574 bgp_decode_as4_aggregator(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 575 {
 576   if (s->as4_session)
 577     DISCARD(NEW_BGP, "AS4_AGGREGATOR");
 578
 579   if (len != 8)
 580     DISCARD(BAD_LENGTH, "AS4_AGGREGATOR", len);
 581
 582   bgp_set_attr_data(to, s->pool, BA_AS4_AGGREGATOR, flags, data, len);
 583 }
 584
 585 static void
 586 bgp_decode_as4_path(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 587 {
 588   char err[128];
 589
 590   if (s->as4_session)
 591     DISCARD(NEW_BGP, "AS4_PATH");
 592
 593   if (len < 6)
 594     DISCARD(BAD_LENGTH, "AS4_PATH", len);
 595
 596   if (!as_path_valid(data, len, 4, 1, err, sizeof(err)))
 597     DISCARD("Malformed AS4_PATH attribute - %s", err);
 598
 599   struct adata *a = lp_alloc_adata(s->pool, len);
 600   memcpy(a->data, data, len);
 601
 602   /* AS_CONFED* segments are invalid in AS4_PATH; RFC 6793 6 */
 603   if (as_path_contains_confed(a))
 604   {
 605     REPORT("Discarding AS_CONFED* segment from AS4_PATH attribute");
 606     a = as_path_strip_confed(s->pool, a);
 607   }
 608
 609   bgp_set_attr_ptr(to, s->pool, BA_AS4_PATH, flags, a);
 610 }
 611
 612 static void
 613 bgp_export_large_community(struct bgp_export_state *s, eattr *a)
 614 {
 615   if (a->u.ptr->length == 0)
 616     UNSET(a);
 617
 618   a->u.ptr = lc_set_sort(s->pool, a->u.ptr);
 619 }
 620
 621 static void
 622 bgp_decode_large_community(struct bgp_parse_state *s, uint code UNUSED, uint flags, byte *data, uint len, ea_list **to)
 623 {
 624   if (!len || (len % 12))
 625     WITHDRAW(BAD_LENGTH, "LARGE_COMMUNITY", len);
 626
 627   struct adata *ad = lp_alloc_adata(s->pool, len);
 628   get_u32s(data, (u32 *) ad->data, len / 4);
 629   bgp_set_attr_ptr(to, s->pool, BA_LARGE_COMMUNITY, flags, ad);
 630 }
 631
 632 static void
 633 bgp_export_mpls_label_stack(struct bgp_export_state *s, eattr *a)
 634 {
 635   net_addr *n = s->route->net->n.addr;
 636   u32 *labels = (u32 *) a->u.ptr->data;
 637   uint lnum = a->u.ptr->length / 4;
 638
 639   /* Perhaps we should just ignore it? */
 640   if (!s->mpls)
 641     WITHDRAW("Unexpected MPLS stack");
 642
 643   /* Empty MPLS stack is not allowed */
 644   if (!lnum)
 645     WITHDRAW("Malformed MPLS stack - empty");
 646
 647   /* This is ugly, but we must ensure that labels fit into NLRI field */
 648   if ((24*lnum + (net_is_vpn(n) ? 64 : 0) + net_pxlen(n)) > 255)
 649     WITHDRAW("Malformed MPLS stack - too many labels (%u)", lnum);
 650
 651   for (uint i = 0; i < lnum; i++)
 652   {
 653     if (labels[i] > 0xfffff)
 654       WITHDRAW("Malformed MPLS stack - invalid label (%u)", labels[i]);
 655
 656     /* TODO: Check for special-purpose label values? */
 657   }
 658 }
 659
 660 static int
 661 bgp_encode_mpls_label_stack(struct bgp_write_state *s, eattr *a, byte *buf UNUSED, uint size UNUSED)
 662 {
 663   /*
 664    * MPLS labels are encoded as a part of the NLRI in MP_REACH_NLRI attribute,
 665    * so we store MPLS_LABEL_STACK and encode it later by AFI-specific hooks.
 666    */
 667
 668   s->mpls_labels = a->u.ptr;
 669   return 0;
 670 }
 671
 672 static void
 673 bgp_decode_mpls_label_stack(struct bgp_parse_state *s, uint code UNUSED, uint flags UNUSED, byte *data UNUSED, uint len UNUSED, ea_list **to UNUSED)
 674 {
 675   DISCARD("Discarding received attribute #0");
 676 }
 677
 678 static void
 679 bgp_format_mpls_label_stack(eattr *a, byte *buf, uint size)
 680 {
 681   u32 *labels = (u32 *) a->u.ptr->data;
 682   uint lnum = a->u.ptr->length / 4;
 683   char *pos = buf;
 684
 685   for (uint i = 0; i < lnum; i++)
 686   {
 687     if (size < 20)
 688     {
 689       bsprintf(pos, "...");
 690       return;
 691     }
 692
 693     uint l = bsprintf(pos, "%d/", labels[i]);
 694     ADVANCE(pos, size, l);
 695   }
 696
 697   /* Clear last slash or terminate empty string */
 698   pos[lnum ? -1 : 0] = 0;
 699 }
 700
 701 static inline void
 702 bgp_decode_unknown(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to)
 703 {
 704   bgp_set_attr_data(to, s->pool, code, flags, data, len);
 705 }
 706
 707
 708 /*
 709  *      Attribute table
 710  */
 711
 712 static const struct bgp_attr_desc bgp_attr_table[] = {
 713   [BA_ORIGIN] = {
 714     .name = "origin",
 715     .type = EAF_TYPE_INT,
 716     .flags = BAF_TRANSITIVE,
 717     .export = bgp_export_origin,
 718     .encode = bgp_encode_u8,
 719     .decode = bgp_decode_origin,
 720     .format = bgp_format_origin,
 721   },
 722   [BA_AS_PATH] = {
 723     .name = "as_path",
 724     .type = EAF_TYPE_AS_PATH,
 725     .flags = BAF_TRANSITIVE,
 726     .encode = bgp_encode_as_path,
 727     .decode = bgp_decode_as_path,
 728   },
 729   [BA_NEXT_HOP] = {
 730     .name = "next_hop",
 731     .type = EAF_TYPE_IP_ADDRESS,
 732     .flags = BAF_TRANSITIVE,
 733     .encode = bgp_encode_next_hop,
 734     .decode = bgp_decode_next_hop,
 735     .format = bgp_format_next_hop,
 736   },
 737   [BA_MULTI_EXIT_DISC] = {
 738     .name = "med",
 739     .type = EAF_TYPE_INT,
 740     .flags = BAF_OPTIONAL,
 741     .encode = bgp_encode_u32,
 742     .decode = bgp_decode_med,
 743   },
 744   [BA_LOCAL_PREF] = {
 745     .name = "local_pref",
 746     .type = EAF_TYPE_INT,
 747     .flags = BAF_TRANSITIVE,
 748     .export = bgp_export_local_pref,
 749     .encode = bgp_encode_u32,
 750     .decode = bgp_decode_local_pref,
 751   },
 752   [BA_ATOMIC_AGGR] = {
 753     .name = "atomic_aggr",
 754     .type = EAF_TYPE_OPAQUE,
 755     .flags = BAF_TRANSITIVE,
 756     .encode = bgp_encode_raw,
 757     .decode = bgp_decode_atomic_aggr,
 758   },
 759   [BA_AGGREGATOR] = {
 760     .name = "aggregator",
 761     .type = EAF_TYPE_OPAQUE,
 762     .flags = BAF_OPTIONAL | BAF_TRANSITIVE,
 763     .encode = bgp_encode_aggregator,
 764     .decode = bgp_decode_aggregator,
 765     .format = bgp_format_aggregator,
 766   },
 767   [BA_COMMUNITY] = {
 768     .name = "community",
 769     .type = EAF_TYPE_INT_SET,
 770     .flags = BAF_OPTIONAL | BAF_TRANSITIVE,
 771     .export = bgp_export_community,
 772     .encode = bgp_encode_u32s,
 773     .decode = bgp_decode_community,
 774   },
 775   [BA_ORIGINATOR_ID] = {
 776     .name = "originator_id",
 777     .type = EAF_TYPE_ROUTER_ID,
 778     .flags = BAF_OPTIONAL,
 779     .export = bgp_export_originator_id,
 780     .encode = bgp_encode_u32,
 781     .decode = bgp_decode_originator_id,
 782   },
 783   [BA_CLUSTER_LIST] = {
 784     .name = "cluster_list",
 785     .type = EAF_TYPE_INT_SET,
 786     .flags = BAF_OPTIONAL,
 787     .export = bgp_export_cluster_list,
 788     .encode = bgp_encode_u32s,
 789     .decode = bgp_decode_cluster_list,
 790     .format = bgp_format_cluster_list,
 791   },
 792   [BA_MP_REACH_NLRI] = {
 793     .name = "mp_reach_nlri",
 794     .type = EAF_TYPE_OPAQUE,
 795     .flags = BAF_OPTIONAL,
 796     .decode = bgp_decode_mp_reach_nlri,
 797   },
 798   [BA_MP_UNREACH_NLRI] = {
 799     .name = "mp_unreach_nlri",
 800     .type = EAF_TYPE_OPAQUE,
 801     .flags = BAF_OPTIONAL,
 802     .decode = bgp_decode_mp_unreach_nlri,
 803   },
 804   [BA_EXT_COMMUNITY] = {
 805     .name = "ext_community",
 806     .type = EAF_TYPE_EC_SET,
 807     .flags = BAF_OPTIONAL | BAF_TRANSITIVE,
 808     .export = bgp_export_ext_community,
 809     .encode = bgp_encode_u32s,
 810     .decode = bgp_decode_ext_community,
 811   },
 812   [BA_AS4_PATH] = {
 813     .name = "as4_path",
 814     .type = EAF_TYPE_AS_PATH,
 815     .flags = BAF_OPTIONAL | BAF_TRANSITIVE,
 816     .encode = bgp_encode_raw,
 817     .decode = bgp_decode_as4_path,
 818   },
 819   [BA_AS4_AGGREGATOR] = {
 820     .name = "as4_aggregator",
 821     .type = EAF_TYPE_OPAQUE,
 822     .flags = BAF_OPTIONAL | BAF_TRANSITIVE,
 823     .encode = bgp_encode_raw,
 824     .decode = bgp_decode_as4_aggregator,
 825     .format = bgp_format_aggregator,
 826   },
 827   [BA_LARGE_COMMUNITY] = {
 828     .name = "large_community",
 829     .type = EAF_TYPE_LC_SET,
 830     .flags = BAF_OPTIONAL | BAF_TRANSITIVE,
 831     .export = bgp_export_large_community,
 832     .encode = bgp_encode_u32s,
 833     .decode = bgp_decode_large_community,
 834   },
 835   [BA_MPLS_LABEL_STACK] = {
 836     .name = "mpls_label_stack",
 837     .type = EAF_TYPE_INT_SET,
 838     .export = bgp_export_mpls_label_stack,
 839     .encode = bgp_encode_mpls_label_stack,
 840     .decode = bgp_decode_mpls_label_stack,
 841     .format = bgp_format_mpls_label_stack,
 842   },
 843 };
 844
 845 static inline int
 846 bgp_attr_known(uint code)
 847 {
 848   return (code < ARRAY_SIZE(bgp_attr_table)) && bgp_attr_table[code].name;
 849 }
 850
 851
 852 /*
 853  *      Attribute export
 854  */
 855
 856 static inline void
 857 bgp_export_attr(struct bgp_export_state *s, eattr *a, ea_list *to)
 858 {
 859   if (EA_PROTO(a->id) != EAP_BGP)
 860     return;
 861
 862   uint code = EA_ID(a->id);
 863
 864   if (bgp_attr_known(code))
 865   {
 866     const struct bgp_attr_desc *desc = &bgp_attr_table[code];
 867
 868     /* The flags might have been zero if the attr was added by filters */
 869     a->flags = (a->flags & BAF_PARTIAL) | desc->flags;
 870
 871     /* Set partial bit if new opt-trans attribute is attached to non-local route */
 872     if ((s->src != NULL) && (a->type & EAF_ORIGINATED) &&
 873         (a->flags & BAF_OPTIONAL) && (a->flags & BAF_TRANSITIVE))
 874       a->flags |= BAF_PARTIAL;
 875
 876     /* Call specific hook */
 877     CALL(desc->export, s, a);
 878
 879     /* Attribute might become undefined in hook */
 880     if ((a->type & EAF_TYPE_MASK) == EAF_TYPE_UNDEF)
 881       return;
 882   }
 883   else
 884   {
 885     /* Don't re-export unknown non-transitive attributes */
 886     if (!(a->flags & BAF_TRANSITIVE))
 887       return;
 888
 889     a->flags |= BAF_PARTIAL;
 890   }
 891
 892   /* Append updated attribute */
 893   to->attrs[to->count++] = *a;
 894 }
 895
 896 /**
 897  * bgp_export_attrs - export BGP attributes
 898  * @s: BGP export state
 899  * @attrs: a list of extended attributes
 900  *
 901  * The bgp_export_attrs() function takes a list of attributes and merges it to
 902  * one newly allocated and sorted segment. Attributes are validated and
 903  * normalized by type-specific export hooks and attribute flags are updated.
 904  * Some attributes may be eliminated (e.g. unknown non-tranitive attributes, or
 905  * empty community sets).
 906  *
 907  * Result: one sorted attribute list segment, or NULL if attributes are unsuitable.
 908  */
 909 static inline ea_list *
 910 bgp_export_attrs(struct bgp_export_state *s, ea_list *attrs)
 911 {
 912   /* Merge the attribute list */
 913   ea_list *new = lp_alloc(s->pool, ea_scan(attrs));
 914   ea_merge(attrs, new);
 915   ea_sort(new);
 916
 917   uint i, count;
 918   count = new->count;
 919   new->count = 0;
 920
 921   /* Export each attribute */
 922   for (i = 0; i < count; i++)
 923     bgp_export_attr(s, &new->attrs[i], new);
 924
 925   if (s->err_withdraw)
 926     return NULL;
 927
 928   return new;
 929 }
 930
 931
 932 /*
 933  *      Attribute encoding
 934  */
 935
 936 static inline int
 937 bgp_encode_attr(struct bgp_write_state *s, eattr *a, byte *buf, uint size)
 938 {
 939   ASSERT(EA_PROTO(a->id) == EAP_BGP);
 940
 941   uint code = EA_ID(a->id);
 942
 943   if (bgp_attr_known(code))
 944     return bgp_attr_table[code].encode(s, a, buf, size);
 945   else
 946     return bgp_encode_raw(s, a, buf, size);
 947 }
 948
 949 /**
 950  * bgp_encode_attrs - encode BGP attributes
 951  * @s: BGP write state
 952  * @attrs: a list of extended attributes
 953  * @buf: buffer
 954  * @end: buffer end
 955  *
 956  * The bgp_encode_attrs() function takes a list of extended attributes
 957  * and converts it to its BGP representation (a part of an Update message).
 958  *
 959  * Result: Length of the attribute block generated or -1 if not enough space.
 960  */
 961 int
 962 bgp_encode_attrs(struct bgp_write_state *s, ea_list *attrs, byte *buf, byte *end)
 963 {
 964   byte *pos = buf;
 965   int i, len;
 966
 967   for (i = 0; i < attrs->count; i++)
 968   {
 969     len = bgp_encode_attr(s, &attrs->attrs[i], pos, end - pos);
 970
 971     if (len < 0)
 972       return -1;
 973
 974     pos += len;
 975   }
 976
 977   return pos - buf;
 978 }
 979
 980
 981 /*
 982  *      Attribute decoding
 983  */
 984
 985 static void bgp_process_as4_attrs(ea_list **attrs, struct linpool *pool);
 986
 987 static inline int
 988 bgp_as_path_loopy(struct bgp_proto *p, ea_list *attrs, u32 asn)
 989 {
 990   eattr *e = bgp_find_attr(attrs, BA_AS_PATH);
 991   int num = p->cf->allow_local_as + 1;
 992   return (e && (num > 0) && as_path_contains(e->u.ptr, asn, num));
 993 }
 994
 995 static inline int
 996 bgp_originator_id_loopy(struct bgp_proto *p, ea_list *attrs)
 997 {
 998   eattr *e = bgp_find_attr(attrs, BA_ORIGINATOR_ID);
 999   return (e && (e->u.data == p->local_id));
1000 }
1001
1002 static inline int
1003 bgp_cluster_list_loopy(struct bgp_proto *p, ea_list *attrs)
1004 {
1005   eattr *e = bgp_find_attr(attrs, BA_CLUSTER_LIST);
1006   return (e && int_set_contains(e->u.ptr, p->rr_cluster_id));
1007 }
1008
1009 static inline void
1010 bgp_decode_attr(struct bgp_parse_state *s, uint code, uint flags, byte *data, uint len, ea_list **to)
1011 {
1012   /* Handle duplicate attributes; RFC 7606 3 (g) */
1013   if (BIT32_TEST(s->attrs_seen, code))
1014   {
1015     if ((code == BA_MP_REACH_NLRI) || (code == BA_MP_UNREACH_NLRI))
1016       bgp_parse_error(s, 1);
1017     else
1018       DISCARD("Discarding duplicate attribute (code %u)", code);
1019   }
1020   BIT32_SET(s->attrs_seen, code);
1021
1022   if (bgp_attr_known(code))
1023   {
1024     const struct bgp_attr_desc *desc = &bgp_attr_table[code];
1025
1026     /* Handle conflicting flags; RFC 7606 3 (c) */
1027     if ((flags ^ desc->flags) & (BAF_OPTIONAL | BAF_TRANSITIVE))
1028       WITHDRAW("Malformed %s attribute - conflicting flags (%02x)", desc->name, flags);
1029
1030     desc->decode(s, code, flags, data, len, to);
1031   }
1032   else /* Unknown attribute */
1033   {
1034     if (!(flags & BAF_OPTIONAL))
1035       WITHDRAW("Unknown attribute (code %u) - conflicting flags (%02x)", code, flags);
1036
1037     bgp_decode_unknown(s, code, flags, data, len, to);
1038   }
1039 }
1040
1041 /**
1042  * bgp_decode_attrs - check and decode BGP attributes
1043  * @s: BGP parse state
1044  * @data: start of attribute block
1045  * @len: length of attribute block
1046  *
1047  * This function takes a BGP attribute block (a part of an Update message), checks
1048  * its consistency and converts it to a list of BIRD route attributes represented
1049  * by an (uncached) &rta.
1050  */
1051 ea_list *
1052 bgp_decode_attrs(struct bgp_parse_state *s, byte *data, uint len)
1053 {
1054   struct bgp_proto *p = s->proto;
1055   ea_list *attrs = NULL;
1056   uint code, flags, alen;
1057   byte *pos = data;
1058
1059   /* Parse the attributes */
1060   while (len)
1061   {
1062     alen = 0;
1063
1064     /* Read attribute type */
1065     if (len < 2)
1066       goto framing_error;
1067     flags = pos[0];
1068     code = pos[1];
1069     ADVANCE(pos, len, 2);
1070
1071     /* Read attribute length */
1072     if (flags & BAF_EXT_LEN)
1073     {
1074       if (len < 2)
1075         goto framing_error;
1076       alen = get_u16(pos);
1077       ADVANCE(pos, len, 2);
1078     }
1079     else
1080     {
1081       if (len < 1)
1082         goto framing_error;
1083       alen = *pos;
1084       ADVANCE(pos, len, 1);
1085     }
1086
1087     if (alen > len)
1088       goto framing_error;
1089
1090     DBG("Attr %02x %02x %u\n", code, flags, alen);
1091
1092     bgp_decode_attr(s, code, flags, pos, alen, &attrs);
1093     ADVANCE(pos, len, alen);
1094   }
1095
1096   if (s->err_withdraw)
1097     goto withdraw;
1098
1099   /* If there is no reachability NLRI, we are finished */
1100   if (!s->ip_reach_len && !s->mp_reach_len)
1101     return NULL;
1102
1103
1104   /* Handle missing mandatory attributes; RFC 7606 3 (d) */
1105   if (!BIT32_TEST(s->attrs_seen, BA_ORIGIN))
1106   { REPORT(NO_MANDATORY, "ORIGIN"); goto withdraw; }
1107
1108   if (!BIT32_TEST(s->attrs_seen, BA_AS_PATH))
1109   { REPORT(NO_MANDATORY, "AS_PATH"); goto withdraw; }
1110
1111   /* When receiving attributes from non-AS4-aware BGP speaker, we have to
1112      reconstruct AS_PATH and AGGREGATOR attributes; RFC 6793 4.2.3 */
1113   if (!p->as4_session)
1114     bgp_process_as4_attrs(&attrs, s->pool);
1115
1116   /* Reject routes with our ASN in AS_PATH attribute */
1117   if (bgp_as_path_loopy(p, attrs, p->local_as))
1118     goto withdraw;
1119
1120   /* Reject routes with our Confederation ID in AS_PATH attribute; RFC 5065 4.0 */
1121   if ((p->public_as != p->local_as) && bgp_as_path_loopy(p, attrs, p->public_as))
1122     goto withdraw;
1123
1124   /* Reject routes with our Router ID in ORIGINATOR_ID attribute; RFC 4456 8 */
1125   if (p->is_internal && bgp_originator_id_loopy(p, attrs))
1126     goto withdraw;
1127
1128   /* Reject routes with our Cluster ID in CLUSTER_LIST attribute; RFC 4456 8 */
1129   if (p->rr_client && bgp_cluster_list_loopy(p, attrs))
1130     goto withdraw;
1131
1132   /* If there is no local preference, define one */
1133   if (!BIT32_TEST(s->attrs_seen, BA_LOCAL_PREF))
1134     bgp_set_attr_u32(&attrs, s->pool, BA_LOCAL_PREF, 0, p->cf->default_local_pref);
1135
1136   return attrs;
1137
1138
1139 framing_error:
1140   /* RFC 7606 4 - handle attribute framing errors */
1141   REPORT("Malformed attribute list - framing error (%u/%u) at %d",
1142          alen, len, (int) (pos - s->attrs));
1143
1144 withdraw:
1145   /* RFC 7606 5.2 - handle missing NLRI during errors */
1146   if (!s->ip_reach_len && !s->mp_reach_len)
1147     bgp_parse_error(s, 1);
1148
1149   s->err_withdraw = 1;
1150   return NULL;
1151 }
1152
1153
1154 /*
1155  *      Route bucket hash table
1156  */
1157
1158 #define RBH_KEY(b)              b->eattrs, b->hash
1159 #define RBH_NEXT(b)             b->next
1160 #define RBH_EQ(a1,h1,a2,h2)     h1 == h2 && ea_same(a1, a2)
1161 #define RBH_FN(a,h)             h
1162
1163 #define RBH_REHASH              bgp_rbh_rehash
1164 #define RBH_PARAMS              /8, *2, 2, 2, 8, 20
1165
1166
1167 HASH_DEFINE_REHASH_FN(RBH, struct bgp_bucket)
1168
1169 void
1170 bgp_init_bucket_table(struct bgp_channel *c)
1171 {
1172   HASH_INIT(c->bucket_hash, c->pool, 8);
1173
1174   init_list(&c->bucket_queue);
1175   c->withdraw_bucket = NULL;
1176 }
1177
1178 void
1179 bgp_free_bucket_table(struct bgp_channel *c)
1180 {
1181   HASH_FREE(c->bucket_hash);
1182
1183   struct bgp_bucket *b;
1184   WALK_LIST_FIRST(b, c->bucket_queue)
1185   {
1186     rem_node(&b->send_node);
1187     mb_free(b);
1188   }
1189
1190   mb_free(c->withdraw_bucket);
1191   c->withdraw_bucket = NULL;
1192 }
1193
1194 static struct bgp_bucket *
1195 bgp_get_bucket(struct bgp_channel *c, ea_list *new)
1196 {
1197   /* Hash and lookup */
1198   u32 hash = ea_hash(new);
1199   struct bgp_bucket *b = HASH_FIND(c->bucket_hash, RBH, new, hash);
1200
1201   if (b)
1202     return b;
1203
1204   uint ea_size = sizeof(ea_list) + new->count * sizeof(eattr);
1205   uint ea_size_aligned = BIRD_ALIGN(ea_size, CPU_STRUCT_ALIGN);
1206   uint size = sizeof(struct bgp_bucket) + ea_size_aligned;
1207   uint i;
1208   byte *dest;
1209
1210   /* Gather total size of non-inline attributes */
1211   for (i = 0; i < new->count; i++)
1212   {
1213     eattr *a = &new->attrs[i];
1214
1215     if (!(a->type & EAF_EMBEDDED))
1216       size += BIRD_ALIGN(sizeof(struct adata) + a->u.ptr->length, CPU_STRUCT_ALIGN);
1217   }
1218
1219   /* Create the bucket */
1220   b = mb_alloc(c->pool, size);
1221   init_list(&b->prefixes);
1222   b->hash = hash;
1223
1224   /* Copy list of extended attributes */
1225   memcpy(b->eattrs, new, ea_size);
1226   dest = ((byte *) b->eattrs) + ea_size_aligned;
1227
1228   /* Copy values of non-inline attributes */
1229   for (i = 0; i < new->count; i++)
1230   {
1231     eattr *a = &b->eattrs->attrs[i];
1232
1233     if (!(a->type & EAF_EMBEDDED))
1234     {
1235       struct adata *oa = a->u.ptr;
1236       struct adata *na = (struct adata *) dest;
1237       memcpy(na, oa, sizeof(struct adata) + oa->length);
1238       a->u.ptr = na;
1239       dest += BIRD_ALIGN(sizeof(struct adata) + na->length, CPU_STRUCT_ALIGN);
1240     }
1241   }
1242
1243   /* Insert the bucket to send queue and bucket hash */
1244   add_tail(&c->bucket_queue, &b->send_node);
1245   HASH_INSERT2(c->bucket_hash, RBH, c->pool, b);
1246
1247   return b;
1248 }
1249
1250 static struct bgp_bucket *
1251 bgp_get_withdraw_bucket(struct bgp_channel *c)
1252 {
1253   if (!c->withdraw_bucket)
1254   {
1255     c->withdraw_bucket = mb_allocz(c->pool, sizeof(struct bgp_bucket));
1256     init_list(&c->withdraw_bucket->prefixes);
1257   }
1258
1259   return c->withdraw_bucket;
1260 }
1261
1262 void
1263 bgp_free_bucket(struct bgp_channel *c, struct bgp_bucket *b)
1264 {
1265   rem_node(&b->send_node);
1266   HASH_REMOVE2(c->bucket_hash, RBH, c->pool, b);
1267   mb_free(b);
1268 }
1269
1270 void
1271 bgp_defer_bucket(struct bgp_channel *c, struct bgp_bucket *b)
1272 {
1273   rem_node(&b->send_node);
1274   add_tail(&c->bucket_queue, &b->send_node);
1275 }
1276
1277 void
1278 bgp_withdraw_bucket(struct bgp_channel *c, struct bgp_bucket *b)
1279 {
1280   struct bgp_proto *p = (void *) c->c.proto;
1281   struct bgp_bucket *wb = bgp_get_withdraw_bucket(c);
1282
1283   log(L_ERR "%s: Attribute list too long", p->p.name);
1284   while (!EMPTY_LIST(b->prefixes))
1285   {
1286     struct bgp_prefix *px = HEAD(b->prefixes);
1287
1288     log(L_ERR "%s: - withdrawing %N", p->p.name, &px->net);
1289     rem_node(&px->buck_node);
1290     add_tail(&wb->prefixes, &px->buck_node);
1291   }
1292 }
1293
1294
1295 /*
1296  *      Prefix hash table
1297  */
1298
1299 #define PXH_KEY(px)             px->net, px->path_id, px->hash
1300 #define PXH_NEXT(px)            px->next
1301 #define PXH_EQ(n1,i1,h1,n2,i2,h2) h1 == h2 && i1 == i2 && net_equal(n1, n2)
1302 #define PXH_FN(n,i,h)           h
1303
1304 #define PXH_REHASH              bgp_pxh_rehash
1305 #define PXH_PARAMS              /8, *2, 2, 2, 8, 20
1306
1307
1308 HASH_DEFINE_REHASH_FN(PXH, struct bgp_prefix)
1309
1310 void
1311 bgp_init_prefix_table(struct bgp_channel *c)
1312 {
1313   HASH_INIT(c->prefix_hash, c->pool, 8);
1314
1315   uint alen = net_addr_length[c->c.net_type];
1316   c->prefix_slab = alen ? sl_new(c->pool, sizeof(struct bgp_prefix) + alen) : NULL;
1317 }
1318
1319 void
1320 bgp_free_prefix_table(struct bgp_channel *c)
1321 {
1322   HASH_FREE(c->prefix_hash);
1323
1324   rfree(c->prefix_slab);
1325   c->prefix_slab = NULL;
1326 }
1327
1328 static struct bgp_prefix *
1329 bgp_get_prefix(struct bgp_channel *c, net_addr *net, u32 path_id)
1330 {
1331   u32 hash = net_hash(net) ^ u32_hash(path_id);
1332   struct bgp_prefix *px = HASH_FIND(c->prefix_hash, PXH, net, path_id, hash);
1333
1334   if (px)
1335   {
1336     rem_node(&px->buck_node);
1337     return px;
1338   }
1339
1340   if (c->prefix_slab)
1341     px = sl_alloc(c->prefix_slab);
1342   else
1343     px = mb_alloc(c->pool, sizeof(struct bgp_prefix) + net->length);
1344
1345   px->buck_node.next = NULL;
1346   px->buck_node.prev = NULL;
1347   px->hash = hash;
1348   px->path_id = path_id;
1349   net_copy(px->net, net);
1350
1351   HASH_INSERT2(c->prefix_hash, PXH, c->pool, px);
1352
1353   return px;
1354 }
1355
1356 void
1357 bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *px)
1358 {
1359   rem_node(&px->buck_node);
1360   HASH_REMOVE2(c->prefix_hash, PXH, c->pool, px);
1361
1362   if (c->prefix_slab)
1363     sl_free(c->prefix_slab, px);
1364   else
1365     mb_free(px);
1366 }
1367
1368
1369 /*
1370  *      BGP protocol glue
1371  */
1372
1373 int
1374 bgp_import_control(struct proto *P, rte **new, ea_list **attrs UNUSED, struct linpool *pool UNUSED)
1375 {
1376   rte *e = *new;
1377   struct proto *SRC = e->attrs->src->proto;
1378   struct bgp_proto *p = (struct bgp_proto *) P;
1379   struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (struct bgp_proto *) SRC : NULL;
1380
1381   /* Reject our routes */
1382   if (src == p)
1383     return -1;
1384
1385   /* Accept non-BGP routes */
1386   if (src == NULL)
1387     return 0;
1388
1389   // XXXX: Check next hop AF
1390
1391   /* IBGP route reflection, RFC 4456 */
1392   if (p->is_internal && src->is_internal && (p->local_as == src->local_as))
1393   {
1394     /* Rejected unless configured as route reflector */
1395     if (!p->rr_client && !src->rr_client)
1396       return -1;
1397
1398     /* Generally, this should be handled when path is received, but we check it
1399        also here as rr_cluster_id may be undefined or different in src. */
1400     if (p->rr_cluster_id && bgp_cluster_list_loopy(p, e->attrs->eattrs))
1401       return -1;
1402   }
1403
1404   /* Handle well-known communities, RFC 1997 */
1405   struct eattr *c;
1406   if (p->cf->interpret_communities &&
1407       (c = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_COMMUNITY))))
1408   {
1409     struct adata *d = c->u.ptr;
1410
1411     /* Do not export anywhere */
1412     if (int_set_contains(d, BGP_COMM_NO_ADVERTISE))
1413       return -1;
1414
1415     /* Do not export outside of AS (or member-AS) */
1416     if (!p->is_internal && int_set_contains(d, BGP_COMM_NO_EXPORT_SUBCONFED))
1417       return -1;
1418
1419     /* Do not export outside of AS (or confederation) */
1420     if (!p->is_interior && int_set_contains(d, BGP_COMM_NO_EXPORT))
1421       return -1;
1422   }
1423
1424   return 0;
1425 }
1426
1427
1428 static adata null_adata;        /* adata of length 0 */
1429
1430 static ea_list *
1431 bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *attrs0, struct linpool *pool)
1432 {
1433   struct proto *SRC = e->attrs->src->proto;
1434   struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (void *) SRC : NULL;
1435   struct bgp_export_state s = { .proto = p, .channel = c, .pool = pool, .src = src, .route = e, .mpls = c->desc->mpls };
1436   ea_list *attrs = attrs0;
1437   eattr *a;
1438   adata *ad;
1439
1440   /* ORIGIN attribute - mandatory, attach if missing */
1441   if (! bgp_find_attr(attrs0, BA_ORIGIN))
1442     bgp_set_attr_u32(&attrs, pool, BA_ORIGIN, 0, src ? ORIGIN_INCOMPLETE : ORIGIN_IGP);
1443
1444   /* AS_PATH attribute - mandatory */
1445   a = bgp_find_attr(attrs0, BA_AS_PATH);
1446   ad = a ? a->u.ptr : &null_adata;
1447
1448   /* AS_PATH attribute - strip AS_CONFED* segments outside confederation */
1449   if ((!p->cf->confederation || !p->is_interior) && as_path_contains_confed(ad))
1450     ad = as_path_strip_confed(pool, ad);
1451
1452   /* AS_PATH attribute - keep or prepend ASN */
1453   if (p->is_internal ||
1454       (p->rs_client && src && src->rs_client))
1455   {
1456     /* IBGP or route server -> just ensure there is one */
1457     if (!a)
1458       bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, &null_adata);
1459   }
1460   else if (p->is_interior)
1461   {
1462     /* Confederation -> prepend ASN as AS_CONFED_SEQUENCE */
1463     ad = as_path_prepend2(pool, ad, AS_PATH_CONFED_SEQUENCE, p->public_as);
1464     bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, ad);
1465   }
1466   else /* Regular EBGP (no RS, no confederation) */
1467   {
1468     /* Regular EBGP -> prepend ASN as regular sequence */
1469     ad = as_path_prepend2(pool, ad, AS_PATH_SEQUENCE, p->public_as);
1470     bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, ad);
1471
1472     /* MULTI_EXIT_DESC attribute - accept only if set in export filter */
1473     a = bgp_find_attr(attrs0, BA_MULTI_EXIT_DISC);
1474     if (a && !(a->type & EAF_FRESH))
1475       bgp_unset_attr(&attrs, pool, BA_MULTI_EXIT_DISC);
1476   }
1477
1478   /* NEXT_HOP attribute - delegated to AF-specific hook */
1479   a = bgp_find_attr(attrs0, BA_NEXT_HOP);
1480   bgp_update_next_hop(&s, a, &attrs);
1481
1482   /* LOCAL_PREF attribute - required for IBGP, attach if missing */
1483   if (p->is_interior && ! bgp_find_attr(attrs0, BA_LOCAL_PREF))
1484     bgp_set_attr_u32(&attrs, pool, BA_LOCAL_PREF, 0, p->cf->default_local_pref);
1485
1486   /* IBGP route reflection, RFC 4456 */
1487   if (src && src->is_internal && p->is_internal && (src->local_as == p->local_as))
1488   {
1489     /* ORIGINATOR_ID attribute - attach if not already set */
1490     if (! bgp_find_attr(attrs0, BA_ORIGINATOR_ID))
1491       bgp_set_attr_u32(&attrs, pool, BA_ORIGINATOR_ID, 0, src->remote_id);
1492
1493     /* CLUSTER_LIST attribute - prepend cluster ID */
1494     a = bgp_find_attr(attrs0, BA_CLUSTER_LIST);
1495     ad = a ? a->u.ptr : NULL;
1496
1497     /* Prepend src cluster ID */
1498     if (src->rr_cluster_id)
1499       ad = int_set_prepend(pool, ad, src->rr_cluster_id);
1500
1501     /* Prepend dst cluster ID if src and dst clusters are different */
1502     if (p->rr_cluster_id && (src->rr_cluster_id != p->rr_cluster_id))
1503       ad = int_set_prepend(pool, ad, p->rr_cluster_id);
1504
1505     /* Should be at least one prepended cluster ID */
1506     bgp_set_attr_ptr(&attrs, pool, BA_CLUSTER_LIST, 0, ad);
1507   }
1508
1509   /* AS4_* transition attributes, RFC 6793 4.2.2 */
1510   if (! p->as4_session)
1511   {
1512     a = bgp_find_attr(attrs, BA_AS_PATH);
1513     if (a && as_path_contains_as4(a->u.ptr))
1514     {
1515       bgp_set_attr_ptr(&attrs, pool, BA_AS_PATH, 0, as_path_to_old(pool, a->u.ptr));
1516       bgp_set_attr_ptr(&attrs, pool, BA_AS4_PATH, 0, as_path_strip_confed(pool, a->u.ptr));
1517     }
1518
1519     a = bgp_find_attr(attrs, BA_AGGREGATOR);
1520     if (a && aggregator_contains_as4(a->u.ptr))
1521     {
1522       bgp_set_attr_ptr(&attrs, pool, BA_AGGREGATOR, 0, aggregator_to_old(pool, a->u.ptr));
1523       bgp_set_attr_ptr(&attrs, pool, BA_AS4_AGGREGATOR, 0, a->u.ptr);
1524     }
1525   }
1526
1527   /*
1528    * Presence of mandatory attributes ORIGIN and AS_PATH is ensured by above
1529    * conditions. Presence and validity of quasi-mandatory NEXT_HOP attribute
1530    * should be checked in AF-specific hooks.
1531    */
1532
1533   /* Apply per-attribute export hooks for validatation and normalization */
1534   return bgp_export_attrs(&s, attrs);
1535 }
1536
1537 void
1538 bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old, ea_list *attrs)
1539 {
1540   struct bgp_proto *p = (void *) P;
1541   struct bgp_channel *c = (void *) C;
1542   struct bgp_bucket *buck;
1543   struct bgp_prefix *px;
1544   u32 path;
1545
1546   if (new)
1547   {
1548     attrs = bgp_update_attrs(p, c, new, attrs, bgp_linpool2);
1549
1550     /* If attributes are invalid, we fail back to withdraw */
1551     buck = attrs ? bgp_get_bucket(c, attrs) : bgp_get_withdraw_bucket(c);
1552     path = new->attrs->src->global_id;
1553
1554     lp_flush(bgp_linpool2);
1555   }
1556   else
1557   {
1558     buck = bgp_get_withdraw_bucket(c);
1559     path = old->attrs->src->global_id;
1560   }
1561
1562   px = bgp_get_prefix(c, n->n.addr, c->add_path_tx ? path : 0);
1563   add_tail(&buck->prefixes, &px->buck_node);
1564
1565   bgp_schedule_packet(p->conn, c, PKT_UPDATE);
1566 }
1567
1568
1569 static inline u32
1570 bgp_get_neighbor(rte *r)
1571 {
1572   eattr *e = ea_find(r->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1573   u32 as;
1574
1575   if (e && as_path_get_first_regular(e->u.ptr, &as))
1576     return as;
1577
1578   /* If AS_PATH is not defined, we treat rte as locally originated */
1579   struct bgp_proto *p = (void *) r->attrs->src->proto;
1580   return p->cf->confederation ?: p->local_as;
1581 }
1582
1583 static inline int
1584 rte_resolvable(rte *rt)
1585 {
1586   return rt->attrs->dest == RTD_UNICAST;
1587 }
1588
1589 int
1590 bgp_rte_better(rte *new, rte *old)
1591 {
1592   struct bgp_proto *new_bgp = (struct bgp_proto *) new->attrs->src->proto;
1593   struct bgp_proto *old_bgp = (struct bgp_proto *) old->attrs->src->proto;
1594   eattr *x, *y;
1595   u32 n, o;
1596
1597   /* Skip suppressed routes (see bgp_rte_recalculate()) */
1598   n = new->u.bgp.suppressed;
1599   o = old->u.bgp.suppressed;
1600   if (n > o)
1601     return 0;
1602   if (n < o)
1603     return 1;
1604
1605   /* RFC 4271 9.1.2.1. Route resolvability test */
1606   n = rte_resolvable(new);
1607   o = rte_resolvable(old);
1608   if (n > o)
1609     return 1;
1610   if (n < o)
1611     return 0;
1612
1613   /* Start with local preferences */
1614   x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1615   y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1616   n = x ? x->u.data : new_bgp->cf->default_local_pref;
1617   o = y ? y->u.data : old_bgp->cf->default_local_pref;
1618   if (n > o)
1619     return 1;
1620   if (n < o)
1621     return 0;
1622
1623   /* RFC 4271 9.1.2.2. a)  Use AS path lengths */
1624   if (new_bgp->cf->compare_path_lengths || old_bgp->cf->compare_path_lengths)
1625   {
1626     x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1627     y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1628     n = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN;
1629     o = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN;
1630     if (n < o)
1631       return 1;
1632     if (n > o)
1633       return 0;
1634   }
1635
1636   /* RFC 4271 9.1.2.2. b) Use origins */
1637   x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1638   y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1639   n = x ? x->u.data : ORIGIN_INCOMPLETE;
1640   o = y ? y->u.data : ORIGIN_INCOMPLETE;
1641   if (n < o)
1642     return 1;
1643   if (n > o)
1644     return 0;
1645
1646   /* RFC 4271 9.1.2.2. c) Compare MED's */
1647   /* Proper RFC 4271 path selection cannot be interpreted as finding
1648    * the best path in some ordering. It is implemented partially in
1649    * bgp_rte_recalculate() when deterministic_med option is
1650    * active. Without that option, the behavior is just an
1651    * approximation, which in specific situations may lead to
1652    * persistent routing loops, because it is nondeterministic - it
1653    * depends on the order in which routes appeared. But it is also the
1654    * same behavior as used by default in Cisco routers, so it is
1655    * probably not a big issue.
1656    */
1657   if (new_bgp->cf->med_metric || old_bgp->cf->med_metric ||
1658       (bgp_get_neighbor(new) == bgp_get_neighbor(old)))
1659   {
1660     x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1661     y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1662     n = x ? x->u.data : new_bgp->cf->default_med;
1663     o = y ? y->u.data : old_bgp->cf->default_med;
1664     if (n < o)
1665       return 1;
1666     if (n > o)
1667       return 0;
1668   }
1669
1670   /* RFC 4271 9.1.2.2. d) Prefer external peers */
1671   if (new_bgp->is_interior > old_bgp->is_interior)
1672     return 0;
1673   if (new_bgp->is_interior < old_bgp->is_interior)
1674     return 1;
1675
1676   /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1677   n = new_bgp->cf->igp_metric ? new->attrs->igp_metric : 0;
1678   o = old_bgp->cf->igp_metric ? old->attrs->igp_metric : 0;
1679   if (n < o)
1680     return 1;
1681   if (n > o)
1682     return 0;
1683
1684   /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */
1685   /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighbor ID */
1686   x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
1687   y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
1688   n = x ? x->u.data : new_bgp->remote_id;
1689   o = y ? y->u.data : old_bgp->remote_id;
1690
1691   /* RFC 5004 - prefer older routes */
1692   /* (if both are external and from different peer) */
1693   if ((new_bgp->cf->prefer_older || old_bgp->cf->prefer_older) &&
1694       !new_bgp->is_internal && n != o)
1695     return 0;
1696
1697   /* rest of RFC 4271 9.1.2.2. f) */
1698   if (n < o)
1699     return 1;
1700   if (n > o)
1701     return 0;
1702
1703   /* RFC 4456 9. b) Compare cluster list lengths */
1704   x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
1705   y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
1706   n = x ? int_set_get_size(x->u.ptr) : 0;
1707   o = y ? int_set_get_size(y->u.ptr) : 0;
1708   if (n < o)
1709     return 1;
1710   if (n > o)
1711     return 0;
1712
1713   /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
1714   return (ipa_compare(new_bgp->cf->remote_ip, old_bgp->cf->remote_ip) < 0);
1715 }
1716
1717
1718 int
1719 bgp_rte_mergable(rte *pri, rte *sec)
1720 {
1721   struct bgp_proto *pri_bgp = (struct bgp_proto *) pri->attrs->src->proto;
1722   struct bgp_proto *sec_bgp = (struct bgp_proto *) sec->attrs->src->proto;
1723   eattr *x, *y;
1724   u32 p, s;
1725
1726   /* Skip suppressed routes (see bgp_rte_recalculate()) */
1727   if (pri->u.bgp.suppressed != sec->u.bgp.suppressed)
1728     return 0;
1729
1730   /* RFC 4271 9.1.2.1. Route resolvability test */
1731   if (!rte_resolvable(sec))
1732     return 0;
1733
1734   /* Start with local preferences */
1735   x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1736   y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1737   p = x ? x->u.data : pri_bgp->cf->default_local_pref;
1738   s = y ? y->u.data : sec_bgp->cf->default_local_pref;
1739   if (p != s)
1740     return 0;
1741
1742   /* RFC 4271 9.1.2.2. a)  Use AS path lengths */
1743   if (pri_bgp->cf->compare_path_lengths || sec_bgp->cf->compare_path_lengths)
1744   {
1745     x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1746     y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1747     p = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN;
1748     s = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN;
1749
1750     if (p != s)
1751       return 0;
1752
1753 //    if (DELTA(p, s) > pri_bgp->cf->relax_multipath)
1754 //      return 0;
1755   }
1756
1757   /* RFC 4271 9.1.2.2. b) Use origins */
1758   x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1759   y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1760   p = x ? x->u.data : ORIGIN_INCOMPLETE;
1761   s = y ? y->u.data : ORIGIN_INCOMPLETE;
1762   if (p != s)
1763     return 0;
1764
1765   /* RFC 4271 9.1.2.2. c) Compare MED's */
1766   if (pri_bgp->cf->med_metric || sec_bgp->cf->med_metric ||
1767       (bgp_get_neighbor(pri) == bgp_get_neighbor(sec)))
1768   {
1769     x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1770     y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1771     p = x ? x->u.data : pri_bgp->cf->default_med;
1772     s = y ? y->u.data : sec_bgp->cf->default_med;
1773     if (p != s)
1774       return 0;
1775   }
1776
1777   /* RFC 4271 9.1.2.2. d) Prefer external peers */
1778   if (pri_bgp->is_interior != sec_bgp->is_interior)
1779     return 0;
1780
1781   /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1782   p = pri_bgp->cf->igp_metric ? pri->attrs->igp_metric : 0;
1783   s = sec_bgp->cf->igp_metric ? sec->attrs->igp_metric : 0;
1784   if (p != s)
1785     return 0;
1786
1787   /* Remaining criteria are ignored */
1788
1789   return 1;
1790 }
1791
1792
1793 static inline int
1794 same_group(rte *r, u32 lpref, u32 lasn)
1795 {
1796   return (r->pref == lpref) && (bgp_get_neighbor(r) == lasn);
1797 }
1798
1799 static inline int
1800 use_deterministic_med(rte *r)
1801 {
1802   struct proto *P = r->attrs->src->proto;
1803   return (P->proto == &proto_bgp) && ((struct bgp_proto *) P)->cf->deterministic_med;
1804 }
1805
1806 int
1807 bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best)
1808 {
1809   rte *r, *s;
1810   rte *key = new ? new : old;
1811   u32 lpref = key->pref;
1812   u32 lasn = bgp_get_neighbor(key);
1813   int old_is_group_best = 0;
1814
1815   /*
1816    * Proper RFC 4271 path selection is a bit complicated, it cannot be
1817    * implemented just by rte_better(), because it is not a linear
1818    * ordering. But it can be splitted to two levels, where the lower
1819    * level chooses the best routes in each group of routes from the
1820    * same neighboring AS and higher level chooses the best route (with
1821    * a slightly different ordering) between the best-in-group routes.
1822    *
1823    * When deterministic_med is disabled, we just ignore this issue and
1824    * choose the best route by bgp_rte_better() alone. If enabled, the
1825    * lower level of the route selection is done here (for the group
1826    * to which the changed route belongs), all routes in group are
1827    * marked as suppressed, just chosen best-in-group is not.
1828    *
1829    * Global best route selection then implements higher level by
1830    * choosing between non-suppressed routes (as they are always
1831    * preferred over suppressed routes). Routes from BGP protocols
1832    * that do not set deterministic_med are just never suppressed. As
1833    * they do not participate in the lower level selection, it is OK
1834    * that this fn is not called for them.
1835    *
1836    * The idea is simple, the implementation is more problematic,
1837    * mostly because of optimizations in rte_recalculate() that
1838    * avoids full recalculation in most cases.
1839    *
1840    * We can assume that at least one of new, old is non-NULL and both
1841    * are from the same protocol with enabled deterministic_med. We
1842    * group routes by both neighbor AS (lasn) and preference (lpref),
1843    * because bgp_rte_better() does not handle preference itself.
1844    */
1845
1846   /* If new and old are from different groups, we just process that
1847      as two independent events */
1848   if (new && old && !same_group(old, lpref, lasn))
1849   {
1850     int i1, i2;
1851     i1 = bgp_rte_recalculate(table, net, NULL, old, old_best);
1852     i2 = bgp_rte_recalculate(table, net, new, NULL, old_best);
1853     return i1 || i2;
1854   }
1855
1856   /*
1857    * We could find the best-in-group and then make some shortcuts like
1858    * in rte_recalculate, but as we would have to walk through all
1859    * net->routes just to find it, it is probably not worth. So we
1860    * just have two simpler fast cases that use just the old route.
1861    * We also set suppressed flag to avoid using it in bgp_rte_better().
1862    */
1863
1864   if (new)
1865     new->u.bgp.suppressed = 1;
1866
1867   if (old)
1868   {
1869     old_is_group_best = !old->u.bgp.suppressed;
1870     old->u.bgp.suppressed = 1;
1871     int new_is_better = new && bgp_rte_better(new, old);
1872
1873     /* The first case - replace not best with worse (or remove not best) */
1874     if (!old_is_group_best && !new_is_better)
1875       return 0;
1876
1877     /* The second case - replace the best with better */
1878     if (old_is_group_best && new_is_better)
1879     {
1880       /* new is best-in-group, the see discussion below - this is
1881          a special variant of NBG && OBG. From OBG we can deduce
1882          that same_group(old_best) iff (old == old_best)  */
1883       new->u.bgp.suppressed = 0;
1884       return (old == old_best);
1885     }
1886   }
1887
1888   /* The default case - find a new best-in-group route */
1889   r = new; /* new may not be in the list */
1890   for (s=net->routes; rte_is_valid(s); s=s->next)
1891     if (use_deterministic_med(s) && same_group(s, lpref, lasn))
1892     {
1893       s->u.bgp.suppressed = 1;
1894       if (!r || bgp_rte_better(s, r))
1895         r = s;
1896     }
1897
1898   /* Simple case - the last route in group disappears */
1899   if (!r)
1900     return 0;
1901
1902   /* Found best-in-group */
1903   r->u.bgp.suppressed = 0;
1904
1905   /*
1906    * There are generally two reasons why we have to force
1907    * recalculation (return 1): First, the new route may be wrongfully
1908    * chosen to be the best in the first case check in
1909    * rte_recalculate(), this may happen only if old_best is from the
1910    * same group. Second, another (different than new route)
1911    * best-in-group is chosen and that may be the proper best (although
1912    * rte_recalculate() without ignore that possibility).
1913    *
1914    * There are three possible cases according to whether the old route
1915    * was the best in group (OBG, stored in old_is_group_best) and
1916    * whether the new route is the best in group (NBG, tested by r == new).
1917    * These cases work even if old or new is NULL.
1918    *
1919    * NBG -> new is a possible candidate for the best route, so we just
1920    *        check for the first reason using same_group().
1921    *
1922    * !NBG && OBG -> Second reason applies, return 1
1923    *
1924    * !NBG && !OBG -> Best in group does not change, old != old_best,
1925    *                 rte_better(new, old_best) is false and therefore
1926    *                 the first reason does not apply, return 0
1927    */
1928
1929   if (r == new)
1930     return old_best && same_group(old_best, lpref, lasn);
1931   else
1932     return old_is_group_best;
1933 }
1934
1935
1936 /*
1937  * Reconstruct AS_PATH and AGGREGATOR according to RFC 6793 4.2.3
1938  */
1939 static void
1940 bgp_process_as4_attrs(ea_list **attrs, struct linpool *pool)
1941 {
1942   eattr *p2 = bgp_find_attr(*attrs, BA_AS_PATH);
1943   eattr *p4 = bgp_find_attr(*attrs, BA_AS4_PATH);
1944   eattr *a2 = bgp_find_attr(*attrs, BA_AGGREGATOR);
1945   eattr *a4 = bgp_find_attr(*attrs, BA_AS4_AGGREGATOR);
1946
1947   /* First, unset AS4_* attributes */
1948   if (p4) bgp_unset_attr(attrs, pool, BA_AS4_PATH);
1949   if (a4) bgp_unset_attr(attrs, pool, BA_AS4_AGGREGATOR);
1950
1951   /* Handle AGGREGATOR attribute */
1952   if (a2 && a4)
1953   {
1954     u32 a2_asn = get_u32(a2->u.ptr->data);
1955
1956     /* If routes were aggregated by an old router, then AS4_PATH and
1957        AS4_AGGREGATOR are invalid. In that case we give up. */
1958     if (a2_asn != AS_TRANS)
1959       return;
1960
1961     /* Use AS4_AGGREGATOR instead of AGGREGATOR */
1962     a2->u.ptr = a4->u.ptr;
1963   }
1964
1965   /* Handle AS_PATH attribute */
1966   if (p2 && p4)
1967   {
1968     /* Both as_path_getlen() and as_path_cut() take AS_CONFED* as zero length */
1969     int p2_len = as_path_getlen(p2->u.ptr);
1970     int p4_len = as_path_getlen(p4->u.ptr);
1971
1972     /* AS_PATH is too short, give up */
1973     if (p2_len < p4_len)
1974       return;
1975
1976     /* Merge AS_PATH and AS4_PATH */
1977     as_path_cut(p2->u.ptr, p2_len - p4_len);
1978     p2->u.ptr = as_path_merge(pool, p2->u.ptr, p4->u.ptr);
1979   }
1980 }
1981
1982 int
1983 bgp_get_attr(eattr *a, byte *buf, int buflen)
1984 {
1985   uint i = EA_ID(a->id);
1986   const struct bgp_attr_desc *d;
1987   int len;
1988
1989   if (bgp_attr_known(i))
1990   {
1991     d = &bgp_attr_table[i];
1992     len = bsprintf(buf, "%s", d->name);
1993     buf += len;
1994     if (d->format)
1995     {
1996       *buf++ = ':';
1997       *buf++ = ' ';
1998       d->format(a, buf, buflen - len - 2);
1999       return GA_FULL;
2000     }
2001     return GA_NAME;
2002   }
2003
2004   bsprintf(buf, "%02x%s", i, (a->flags & BAF_TRANSITIVE) ? " [t]" : "");
2005   return GA_NAME;
2006 }
2007
2008 void
2009 bgp_get_route_info(rte *e, byte *buf, ea_list *attrs)
2010 {
2011   eattr *p = ea_find(attrs, EA_CODE(EAP_BGP, BA_AS_PATH));
2012   eattr *o = ea_find(attrs, EA_CODE(EAP_BGP, BA_ORIGIN));
2013   u32 origas;
2014
2015   buf += bsprintf(buf, " (%d", e->pref);
2016
2017   if (e->u.bgp.suppressed)
2018     buf += bsprintf(buf, "-");
2019
2020   if (e->attrs->hostentry)
2021   {
2022     if (!rte_resolvable(e))
2023       buf += bsprintf(buf, "/-");
2024     else if (e->attrs->igp_metric >= IGP_METRIC_UNKNOWN)
2025       buf += bsprintf(buf, "/?");
2026     else
2027       buf += bsprintf(buf, "/%d", e->attrs->igp_metric);
2028   }
2029   buf += bsprintf(buf, ") [");
2030
2031   if (p && as_path_get_last(p->u.ptr, &origas))
2032     buf += bsprintf(buf, "AS%u", origas);
2033   if (o)
2034     buf += bsprintf(buf, "%c", "ie?"[o->u.data]);
2035   strcpy(buf, "]");
2036 }