attrs.c 47.7 KB
Newer Older
1 2 3 4 5 6 7 8
/*
 *	BIRD -- BGP Attributes
 *
 *	(c) 2000 Martin Mares <mj@ucw.cz>
 *
 *	Can be freely distributed and used under the terms of the GNU GPL.
 */

Martin Mareš's avatar
Martin Mareš committed
9
#undef LOCAL_DEBUG
10

11 12
#include <stdlib.h>

13 14 15 16
#include "nest/bird.h"
#include "nest/iface.h"
#include "nest/protocol.h"
#include "nest/route.h"
17
#include "nest/attrs.h"
18
#include "conf/conf.h"
19 20 21
#include "lib/resource.h"
#include "lib/string.h"
#include "lib/unaligned.h"
22 23

#include "bgp.h"
24

25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
/*
 *   UPDATE message error handling
 *
 * All checks from RFC 4271 6.3 are done as specified with these exceptions:
 *  - The semantic check of an IP address from NEXT_HOP attribute is missing.
 *  - Checks of some optional attribute values are missing.
 *  - Syntactic and semantic checks of NLRIs (done in DECODE_PREFIX())
 *    are probably inadequate.
 *
 * Loop detection based on AS_PATH causes updates to be withdrawn. RFC
 * 4271 does not explicitly specifiy the behavior in that case.
 *
 * Loop detection related to route reflection (based on ORIGINATOR_ID
 * and CLUSTER_LIST) causes updates to be withdrawn. RFC 4456 8
 * specifies that such updates should be ignored, but that is generally
 * a bad idea.
 *
 * Error checking of optional transitive attributes is done according to
 * draft-ietf-idr-optional-transitive-03, but errors are handled always
 * as withdraws.
 *
 * Unexpected AS_CONFED_* segments in AS_PATH are logged and removed,
 * but unknown segments cause a session drop with Malformed AS_PATH
 * error (see validate_path()). The behavior in such case is not
 * explicitly specified by RFC 4271. RFC 5065 specifies that
 * inconsistent AS_CONFED_* segments should cause a session drop, but
 * implementations that pass invalid AS_CONFED_* segments are
 * widespread.
 *
 * Error handling of AS4_* attributes is done as specified by
 * draft-ietf-idr-rfc4893bis-03. There are several possible
 * inconsistencies between AGGREGATOR and AS4_AGGREGATOR that are not
 * handled by that draft, these are logged and ignored (see
 * bgp_reconstruct_4b_attrs()).
 */

Martin Mareš's avatar
Martin Mareš committed
61 62 63 64 65
static byte bgp_mandatory_attrs[] = { BA_ORIGIN, BA_AS_PATH
#ifndef IPV6
,BA_NEXT_HOP
#endif
};
66 67

struct attr_desc {
68
  char *name;
69 70 71
  int expected_length;
  int expected_flags;
  int type;
72
  int allow_in_ebgp;
73
  int (*validate)(struct bgp_proto *p, byte *attr, int len);
74
  void (*format)(eattr *ea, byte *buf, int buflen);
75 76
};

77 78 79
#define IGNORE -1
#define WITHDRAW -2

80
static int
81
bgp_check_origin(struct bgp_proto *p UNUSED, byte *a, int len UNUSED)
82
{
83
  if (*a > 2)
84 85 86 87 88
    return 6;
  return 0;
}

static void
89
bgp_format_origin(eattr *a, byte *buf, int buflen UNUSED)
90 91 92 93 94 95 96
{
  static char *bgp_origin_names[] = { "IGP", "EGP", "Incomplete" };

  bsprintf(buf, bgp_origin_names[a->u.data]);
}

static int
97
path_segment_contains(byte *p, int bs, u32 asn)
98
{
99 100 101 102 103
  int i;
  int len = p[1];
  p += 2;

  for(i=0; i<len; i++)
104
    {
105 106 107 108
      u32 asn2 = (bs == 4) ? get_u32(p) : get_u16(p);
      if (asn2 == asn)
	return 1;
      p += bs;
109
    }
110

111 112 113
  return 0;
}

114
/* Validates path attribute, removes AS_CONFED_* segments, and also returns path length */
115
static int
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
validate_path(struct bgp_proto *p, int as_path, int bs, byte *idata, unsigned int *ilength)
{
  int res = 0;
  u8 *a, *dst;
  int len, plen, copy;

  dst = a = idata;
  len = *ilength;

  while (len)
    {
      if (len < 2)
	return -1;

      plen = 2 + bs * a[1];
      if (len < plen)
	return -1;

      switch (a[0])
	{
	case AS_PATH_SET:
	  copy = 1;
	  res++;
	  break;

	case AS_PATH_SEQUENCE:
	  copy = 1;
	  res += a[1];
	  break;

	case AS_PATH_CONFED_SEQUENCE:
	case AS_PATH_CONFED_SET:
	  if (as_path && path_segment_contains(a, bs, p->remote_as))
	    {
	      log(L_WARN "%s: AS_CONFED_* segment with peer ASN found, misconfigured confederation?", p->p.name);
	      return -1;
	    }

	  log(L_WARN "%s: %s_PATH attribute contains AS_CONFED_* segment, skipping segment",
	      p->p.name, as_path ? "AS" : "AS4");
	  copy = 0;
	  break;

	default:
	  return -1;
	}

      if (copy)
	{
	  if (dst != a)
	    memmove(dst, a, plen);
	  dst += plen;
	}

      len -= plen;
      a += plen;
    }

  *ilength = dst - idata;
  return res;
}

static inline int
validate_as_path(struct bgp_proto *p, byte *a, int *len)
180
{
181 182 183 184 185 186 187
  return validate_path(p, 1, p->as4_session ? 4 : 2, a, len);
}

static inline int
validate_as4_path(struct bgp_proto *p, struct adata *path)
{
  return validate_path(p, 0, 4, path->data, &path->length);
188 189
}

190
static int
Martin Mareš's avatar
Martin Mareš committed
191
bgp_check_next_hop(struct bgp_proto *p UNUSED, byte *a, int len)
192
{
Martin Mareš's avatar
Martin Mareš committed
193
#ifdef IPV6
194
  return IGNORE;
Martin Mareš's avatar
Martin Mareš committed
195
#else
196 197 198 199 200 201 202 203
  ip_addr addr;

  memcpy(&addr, a, len);
  ipa_ntoh(addr);
  if (ipa_classify(addr) & IADDR_HOST)
    return 0;
  else
    return 8;
Martin Mareš's avatar
Martin Mareš committed
204 205 206
#endif
}

207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
static void
bgp_format_next_hop(eattr *a, byte *buf, int buflen UNUSED)
{
  ip_addr *ipp = (ip_addr *) a->u.ptr->data;
#ifdef IPV6
  /* in IPv6, we might have two addresses in NEXT HOP */
  if ((a->u.ptr->length == NEXT_HOP_LENGTH) && ipa_nonzero(ipp[1]))
    {
      bsprintf(buf, "%I %I", ipp[0], ipp[1]);
      return;
    }
#endif

  bsprintf(buf, "%I", ipp[0]);
}

223
static int
224
bgp_check_aggregator(struct bgp_proto *p, byte *a UNUSED, int len)
225
{
226
  int exp_len = p->as4_session ? 8 : 6;
227
  
228
  return (len == exp_len) ? 0 : WITHDRAW;
229 230
}

231 232 233 234 235 236 237
static void
bgp_format_aggregator(eattr *a, byte *buf, int buflen UNUSED)
{
  struct adata *ad =  a->u.ptr;
  byte *data = ad->data;
  u32 as;

238 239
  as = get_u32(data);
  data += 4;
240 241 242 243

  bsprintf(buf, "%d.%d.%d.%d AS%d", data[0], data[1], data[2], data[3], as);
}

244 245 246 247 248 249
static int
bgp_check_community(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
{
  return ((len % 4) == 0) ? 0 : WITHDRAW;
}

250
static int
251
bgp_check_cluster_list(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
252 253 254 255
{
  return ((len % 4) == 0) ? 0 : 5;
}

256
static void
257
bgp_format_cluster_list(eattr *a, byte *buf, int buflen)
258
{
259 260
  /* Truncates cluster lists larger than buflen, probably not a problem */
  int_set_format(a->u.ptr, 0, -1, buf, buflen);
261 262
}

Martin Mareš's avatar
Martin Mareš committed
263
static int
Martin Mareš's avatar
Martin Mareš committed
264
bgp_check_reach_nlri(struct bgp_proto *p UNUSED, byte *a UNUSED, int len UNUSED)
Martin Mareš's avatar
Martin Mareš committed
265 266 267 268 269
{
#ifdef IPV6
  p->mp_reach_start = a;
  p->mp_reach_len = len;
#endif
270
  return IGNORE;
Martin Mareš's avatar
Martin Mareš committed
271 272 273
}

static int
Martin Mareš's avatar
Martin Mareš committed
274
bgp_check_unreach_nlri(struct bgp_proto *p UNUSED, byte *a UNUSED, int len UNUSED)
Martin Mareš's avatar
Martin Mareš committed
275 276 277 278 279
{
#ifdef IPV6
  p->mp_unreach_start = a;
  p->mp_unreach_len = len;
#endif
280
  return IGNORE;
281 282
}

283 284 285 286 287 288 289
static int
bgp_check_ext_community(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
{
  return ((len % 8) == 0) ? 0 : WITHDRAW;
}


290
static struct attr_desc bgp_attr_table[] = {
Martin Mareš's avatar
Martin Mareš committed
291
  { NULL, -1, 0, 0, 0,								/* Undefined */
292
    NULL, NULL },
Martin Mareš's avatar
Martin Mareš committed
293
  { "origin", 1, BAF_TRANSITIVE, EAF_TYPE_INT, 1,				/* BA_ORIGIN */
294
    bgp_check_origin, bgp_format_origin },
Martin Mareš's avatar
Martin Mareš committed
295
  { "as_path", -1, BAF_TRANSITIVE, EAF_TYPE_AS_PATH, 1,				/* BA_AS_PATH */
296
    NULL, NULL }, /* is checked by validate_as_path() as a special case */
Martin Mareš's avatar
Martin Mareš committed
297
  { "next_hop", 4, BAF_TRANSITIVE, EAF_TYPE_IP_ADDRESS, 1,			/* BA_NEXT_HOP */
298
    bgp_check_next_hop, bgp_format_next_hop },
299
  { "med", 4, BAF_OPTIONAL, EAF_TYPE_INT, 1,					/* BA_MULTI_EXIT_DISC */
300
    NULL, NULL },
301
  { "local_pref", 4, BAF_TRANSITIVE, EAF_TYPE_INT, 0,				/* BA_LOCAL_PREF */
302
    NULL, NULL },
Martin Mareš's avatar
Martin Mareš committed
303
  { "atomic_aggr", 0, BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1,			/* BA_ATOMIC_AGGR */
304
    NULL, NULL },
305
  { "aggregator", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1,	/* BA_AGGREGATOR */
306
    bgp_check_aggregator, bgp_format_aggregator },
Martin Mareš's avatar
Martin Mareš committed
307
  { "community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_INT_SET, 1,	/* BA_COMMUNITY */
308
    bgp_check_community, NULL },
309
  { "originator_id", 4, BAF_OPTIONAL, EAF_TYPE_ROUTER_ID, 0,			/* BA_ORIGINATOR_ID */
310 311
    NULL, NULL },
  { "cluster_list", -1, BAF_OPTIONAL, EAF_TYPE_INT_SET, 0,			/* BA_CLUSTER_LIST */
312
    bgp_check_cluster_list, bgp_format_cluster_list }, 
313
  { .name = NULL },								/* BA_DPA */
314 315
  { .name = NULL },								/* BA_ADVERTISER */
  { .name = NULL },								/* BA_RCID_PATH */
Martin Mareš's avatar
Martin Mareš committed
316 317 318 319
  { "mp_reach_nlri", -1, BAF_OPTIONAL, EAF_TYPE_OPAQUE, 1,			/* BA_MP_REACH_NLRI */
    bgp_check_reach_nlri, NULL },
  { "mp_unreach_nlri", -1, BAF_OPTIONAL, EAF_TYPE_OPAQUE, 1,			/* BA_MP_UNREACH_NLRI */
    bgp_check_unreach_nlri, NULL },
320 321
  { "ext_community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_EC_SET, 1,	/* BA_EXT_COMMUNITY */
    bgp_check_ext_community, NULL },
322
  { "as4_path", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1,		/* BA_AS4_PATH */
323 324
    NULL, NULL },
  { "as4_aggregator", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1,	/* BA_AS4_PATH */
325
    NULL, NULL }
326 327
};

328 329
/* BA_AS4_PATH is type EAF_TYPE_OPAQUE and not type EAF_TYPE_AS_PATH.
 * It does not matter as this attribute does not appear on routes in the routing table.
330 331
 */

332 333
#define ATTR_KNOWN(code) ((code) < ARRAY_SIZE(bgp_attr_table) && bgp_attr_table[code].name)

334 335 336 337 338 339 340 341 342 343
static inline struct adata *
bgp_alloc_adata(struct linpool *pool, unsigned len)
{
  struct adata *ad = lp_alloc(pool, sizeof(struct adata) + len);
  ad->length = len;
  return ad;
}

static void
bgp_set_attr(eattr *e, unsigned attr, uintptr_t val)
344 345 346 347 348 349
{
  ASSERT(ATTR_KNOWN(attr));
  e->id = EA_CODE(EAP_BGP, attr);
  e->type = bgp_attr_table[attr].type;
  e->flags = bgp_attr_table[attr].expected_flags;
  if (e->type & EAF_EMBEDDED)
350
    e->u.data = val;
351
  else
352
    e->u.ptr = (struct adata *) val;
353 354
}

355 356 357 358 359 360 361 362 363 364
static byte *
bgp_set_attr_wa(eattr *e, struct linpool *pool, unsigned attr, unsigned len)
{
  struct adata *ad = bgp_alloc_adata(pool, len);
  bgp_set_attr(e, attr, (uintptr_t) ad);
  return ad->data;
}

void
bgp_attach_attr(ea_list **to, struct linpool *pool, unsigned attr, uintptr_t val)
365 366 367 368 369 370
{
  ea_list *a = lp_alloc(pool, sizeof(ea_list) + sizeof(eattr));
  a->next = *to;
  *to = a;
  a->flags = EALF_SORTED;
  a->count = 1;
371 372 373 374 375 376 377 378 379
  bgp_set_attr(a->attrs, attr, val);
}

byte *
bgp_attach_attr_wa(ea_list **to, struct linpool *pool, unsigned attr, unsigned len)
{
  struct adata *ad = bgp_alloc_adata(pool, len);
  bgp_attach_attr(to, pool, attr, (uintptr_t) ad);
  return ad->data;
380 381
}

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462
static int
bgp_encode_attr_hdr(byte *dst, unsigned int flags, unsigned code, int len)
{
  int wlen;

  DBG("\tAttribute %02x (%d bytes, flags %02x)\n", code, len, flags);

  if (len < 256)
    {
      *dst++ = flags;
      *dst++ = code;
      *dst++ = len;
      wlen = 3;
    }
  else
    {
      *dst++ = flags | BAF_EXT_LEN;
      *dst++ = code;
      put_u16(dst, len);
      wlen = 4;
    }

  return wlen;
}

static void
aggregator_convert_to_old(struct adata *aggr, byte *dst, int *new_used)
{
  byte *src = aggr->data;
  *new_used = 0;

  u32 as = get_u32(src);
  if (as > 0xFFFF) 
    {
      as = AS_TRANS;
      *new_used = 1;
    }
  put_u16(dst, as);

  /* Copy IPv4 address */
  memcpy(dst + 2, src + 4, 4);
}

static void
aggregator_convert_to_new(struct adata *aggr, byte *dst)
{
  byte *src = aggr->data;

  u32 as   = get_u16(src);
  put_u32(dst, as);

  /* Copy IPv4 address */
  memcpy(dst + 4, src + 2, 4);
}

static int
bgp_get_attr_len(eattr *a)
{
  int len;
  if (ATTR_KNOWN(EA_ID(a->id)))
    {
      int code = EA_ID(a->id);
      struct attr_desc *desc = &bgp_attr_table[code];
      len = desc->expected_length;
      if (len < 0)
	{
	  ASSERT(!(a->type & EAF_EMBEDDED));
	  len = a->u.ptr->length;
	}
    }
  else
    {
      ASSERT((a->type & EAF_TYPE_MASK) == EAF_TYPE_OPAQUE);
      len = a->u.ptr->length;
    }
  
  return len;
}

#define ADVANCE(w, r, l) do { r -= l; w += l; } while (0)

Martin Mareš's avatar
Martin Mareš committed
463 464
/**
 * bgp_encode_attrs - encode BGP attributes
465
 * @p: BGP instance
Martin Mareš's avatar
Martin Mareš committed
466 467 468 469 470 471 472
 * @w: buffer
 * @attrs: a list of extended attributes
 * @remains: remaining space in the buffer
 *
 * The bgp_encode_attrs() function takes a list of extended attributes
 * and converts it to its BGP representation (a part of an Update message).
 *
473
 * Result: Length of the attribute block generated or -1 if not enough space.
Martin Mareš's avatar
Martin Mareš committed
474
 */
475
unsigned int
476
bgp_encode_attrs(struct bgp_proto *p, byte *w, ea_list *attrs, int remains)
477
{
478
  unsigned int i, code, type, flags;
479
  byte *start = w;
480
  int len, rv;
481

482
  for(i=0; i<attrs->count; i++)
483
    {
484
      eattr *a = &attrs->attrs[i];
485 486
      ASSERT(EA_PROTO(a->id) == EAP_BGP);
      code = EA_ID(a->id);
487

488 489 490 491 492
#ifdef IPV6
      /* When talking multiprotocol BGP, the NEXT_HOP attributes are used only temporarily. */
      if (code == BA_NEXT_HOP)
	continue;
#endif
493 494 495 496 497

      /* When AS4-aware BGP speaker is talking to non-AS4-aware BGP speaker,
       * we have to convert our 4B AS_PATH to 2B AS_PATH and send our AS_PATH 
       * as optional AS4_PATH attribute.
       */
498
      if ((code == BA_AS_PATH) && (! p->as4_session))
499 500
	{
	  len = a->u.ptr->length;
501 502 503 504 505 506 507 508 509 510 511 512

	  if (remains < (len + 4))
	    goto err_no_buffer;

	  /* Using temporary buffer because don't know a length of created attr
	   * and therefore a length of a header. Perhaps i should better always
	   * use BAF_EXT_LEN. */
	  
	  byte buf[len];
	  int new_used;
	  int nl = as_path_convert_to_old(a->u.ptr, buf, &new_used);

Ondřej Zajíček's avatar
Ondřej Zajíček committed
513
	  DBG("BGP: Encoding old AS_PATH\n");
514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
	  rv = bgp_encode_attr_hdr(w, BAF_TRANSITIVE, BA_AS_PATH, nl);
	  ADVANCE(w, remains, rv);
	  memcpy(w, buf, nl);
	  ADVANCE(w, remains, nl);

	  if (! new_used)
	    continue;

	  if (remains < (len + 4))
	    goto err_no_buffer;

	  /* We should discard AS_CONFED_SEQUENCE or AS_CONFED_SET path segments 
	   * here but we don't support confederations and such paths we already
	   * discarded in bgp_check_as_path().
	   */

Ondřej Zajíček's avatar
Ondřej Zajíček committed
530
	  DBG("BGP: Encoding AS4_PATH\n");
531 532 533 534 535 536
	  rv = bgp_encode_attr_hdr(w, BAF_OPTIONAL | BAF_TRANSITIVE, BA_AS4_PATH, len);
	  ADVANCE(w, remains, rv);
	  memcpy(w, a->u.ptr->data, len);
	  ADVANCE(w, remains, len);

	  continue;
537
	}
538 539

      /* The same issue with AGGREGATOR attribute */
540
      if ((code == BA_AGGREGATOR) && (! p->as4_session))
541
	{
542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565
	  int new_used;

	  len = 6;
	  if (remains < (len + 3))
	    goto err_no_buffer;

	  rv = bgp_encode_attr_hdr(w, BAF_OPTIONAL | BAF_TRANSITIVE, BA_AGGREGATOR, len);
	  ADVANCE(w, remains, rv);
	  aggregator_convert_to_old(a->u.ptr, w, &new_used);
	  ADVANCE(w, remains, len);

	  if (! new_used)
	    continue;

	  len = 8;
	  if (remains < (len + 3))
	    goto err_no_buffer;

	  rv = bgp_encode_attr_hdr(w, BAF_OPTIONAL | BAF_TRANSITIVE, BA_AS4_AGGREGATOR, len);
	  ADVANCE(w, remains, rv);
	  memcpy(w, a->u.ptr->data, len);
	  ADVANCE(w, remains, len);

	  continue;
566
	}
567 568 569

      /* Standard path continues here ... */

570
      type = a->type & EAF_TYPE_MASK;
571 572 573
      flags = a->flags & (BAF_OPTIONAL | BAF_TRANSITIVE | BAF_PARTIAL);
      len = bgp_get_attr_len(a);

574 575
      /* Skip empty sets */ 
      if (((type == EAF_TYPE_INT_SET) || (type == EAF_TYPE_EC_SET)) && (len == 0))
576 577
	continue; 

578 579 580 581 582 583
      if (remains < len + 4)
	goto err_no_buffer;

      rv = bgp_encode_attr_hdr(w, flags, code, len);
      ADVANCE(w, remains, rv);

584
      switch (type)
585 586 587 588 589 590 591 592 593 594 595 596 597 598 599
	{
	case EAF_TYPE_INT:
	case EAF_TYPE_ROUTER_ID:
	  if (len == 4)
	    put_u32(w, a->u.data);
	  else
	    *w = a->u.data;
	  break;
	case EAF_TYPE_IP_ADDRESS:
	  {
	    ip_addr ip = *(ip_addr *)a->u.ptr->data;
	    ipa_hton(ip);
	    memcpy(w, &ip, len);
	    break;
	  }
600
	case EAF_TYPE_INT_SET:
601
	case EAF_TYPE_EC_SET:
602
	  {
603
	    u32 *z = int_set_get_data(a->u.ptr);
604 605 606 607 608
	    int i;
	    for(i=0; i<len; i+=4)
	      put_u32(w+i, *z++);
	    break;
	  }
609 610 611 612 613 614 615
	case EAF_TYPE_OPAQUE:
	case EAF_TYPE_AS_PATH:
	  memcpy(w, a->u.ptr->data, len);
	  break;
	default:
	  bug("bgp_encode_attrs: unknown attribute type %02x", a->type);
	}
616
      ADVANCE(w, remains, len);
617
    }
618
  return w - start;
619 620

 err_no_buffer:
621
  return -1;
622
}
623

Martin Mareš's avatar
Martin Mareš committed
624 625 626 627
static void
bgp_init_prefix(struct fib_node *N)
{
  struct bgp_prefix *p = (struct bgp_prefix *) N;
628
  p->bucket_node.next = NULL;
Martin Mareš's avatar
Martin Mareš committed
629 630
}

631 632 633 634 635 636
static int
bgp_compare_u32(const u32 *x, const u32 *y)
{
  return (*x < *y) ? -1 : (*x > *y) ? 1 : 0;
}

637 638
static inline void
bgp_normalize_int_set(u32 *dest, u32 *src, unsigned cnt)
639 640
{
  memcpy(dest, src, sizeof(u32) * cnt);
641
  qsort(dest, cnt, sizeof(u32), (int(*)(const void *, const void *)) bgp_compare_u32);
642 643
}

644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680
static int
bgp_compare_ec(const u32 *xp, const u32 *yp)
{
  u64 x = ec_get(xp, 0);
  u64 y = ec_get(yp, 0);
  return (x < y) ? -1 : (x > y) ? 1 : 0;
}

static inline void
bgp_normalize_ec_set(struct adata *ad, u32 *src, int internal)
{
  u32 *dst = int_set_get_data(ad);

  /* Remove non-transitive communities (EC_TBIT active) on external sessions */
  if (! internal)
    {
      int len = int_set_get_size(ad);
      u32 *t = dst;
      int i;

      for (i=0; i < len; i += 2)
	{
	  if (src[i] & EC_TBIT)
	    continue;
	  
	  *t++ = src[i];
	  *t++ = src[i+1];
	}

      ad->length = (t - dst) * 4;
    }
  else
    memcpy(dst, src, ad->length);

  qsort(dst, ad->length / 8, 8, (int(*)(const void *, const void *)) bgp_compare_ec);
}

681 682 683
static void
bgp_rehash_buckets(struct bgp_proto *p)
{
Martin Mareš's avatar
Martin Mareš committed
684
  struct bgp_bucket **old = p->bucket_hash;
685 686 687 688 689 690 691 692 693 694
  struct bgp_bucket **new;
  unsigned oldn = p->hash_size;
  unsigned i, e, mask;
  struct bgp_bucket *b;

  p->hash_size = p->hash_limit;
  DBG("BGP: Rehashing bucket table from %d to %d\n", oldn, p->hash_size);
  p->hash_limit *= 4;
  if (p->hash_limit >= 65536)
    p->hash_limit = ~0;
Martin Mareš's avatar
Martin Mareš committed
695
  new = p->bucket_hash = mb_allocz(p->p.pool, p->hash_size * sizeof(struct bgp_bucket *));
696 697 698 699
  mask = p->hash_size - 1;
  for (i=0; i<oldn; i++)
    while (b = old[i])
      {
Martin Mareš's avatar
Martin Mareš committed
700
	old[i] = b->hash_next;
701
	e = b->hash & mask;
Martin Mareš's avatar
Martin Mareš committed
702 703 704 705
	b->hash_next = new[e];
	if (b->hash_next)
	  b->hash_next->hash_prev = b;
	b->hash_prev = NULL;
706 707 708 709 710 711 712 713 714 715
	new[e] = b;
      }
  mb_free(old);
}

static struct bgp_bucket *
bgp_new_bucket(struct bgp_proto *p, ea_list *new, unsigned hash)
{
  struct bgp_bucket *b;
  unsigned ea_size = sizeof(ea_list) + new->count * sizeof(eattr);
Ondřej Filip's avatar
Ondřej Filip committed
716
  unsigned ea_size_aligned = BIRD_ALIGN(ea_size, CPU_STRUCT_ALIGN);
717 718 719 720 721 722 723 724 725 726
  unsigned size = sizeof(struct bgp_bucket) + ea_size;
  unsigned i;
  byte *dest;
  unsigned index = hash & (p->hash_size - 1);

  /* Gather total size of non-inline attributes */
  for (i=0; i<new->count; i++)
    {
      eattr *a = &new->attrs[i];
      if (!(a->type & EAF_EMBEDDED))
Ondřej Filip's avatar
Ondřej Filip committed
727
	size += BIRD_ALIGN(sizeof(struct adata) + a->u.ptr->length, CPU_STRUCT_ALIGN);
728 729 730 731
    }

  /* Create the bucket and hash it */
  b = mb_alloc(p->p.pool, size);
Martin Mareš's avatar
Martin Mareš committed
732 733 734 735 736
  b->hash_next = p->bucket_hash[index];
  if (b->hash_next)
    b->hash_next->hash_prev = b;
  p->bucket_hash[index] = b;
  b->hash_prev = NULL;
737
  b->hash = hash;
738 739
  add_tail(&p->bucket_queue, &b->send_node);
  init_list(&b->prefixes);
740 741 742 743 744 745
  memcpy(b->eattrs, new, ea_size);
  dest = ((byte *)b->eattrs) + ea_size_aligned;

  /* Copy values of non-inline attributes */
  for (i=0; i<new->count; i++)
    {
Martin Mareš's avatar
Martin Mareš committed
746
      eattr *a = &b->eattrs->attrs[i];
747 748 749 750 751 752
      if (!(a->type & EAF_EMBEDDED))
	{
	  struct adata *oa = a->u.ptr;
	  struct adata *na = (struct adata *) dest;
	  memcpy(na, oa, sizeof(struct adata) + oa->length);
	  a->u.ptr = na;
Ondřej Filip's avatar
Ondřej Filip committed
753
	  dest += BIRD_ALIGN(sizeof(struct adata) + na->length, CPU_STRUCT_ALIGN);
754 755 756 757 758 759 760 761 762 763 764 765
	}
    }

  /* If needed, rehash */
  p->hash_count++;
  if (p->hash_count > p->hash_limit)
    bgp_rehash_buckets(p);

  return b;
}

static struct bgp_bucket *
766
bgp_get_bucket(struct bgp_proto *p, net *n, ea_list *attrs, int originate)
767
{
Martin Mareš's avatar
Martin Mareš committed
768
  ea_list *new;
769
  unsigned i, cnt, hash, code;
770 771 772 773
  eattr *a, *d;
  u32 seen = 0;
  struct bgp_bucket *b;

774 775 776
  /* Merge the attribute list */
  new = alloca(ea_scan(attrs));
  ea_merge(attrs, new);
777
  ea_sort(new);
778 779 780 781 782 783 784 785 786 787

  /* Normalize attributes */
  d = new->attrs;
  cnt = new->count;
  new->count = 0;
  for(i=0; i<cnt; i++)
    {
      a = &new->attrs[i];
      if (EA_PROTO(a->id) != EAP_BGP)
	continue;
788
      code = EA_ID(a->id);
789
      if (ATTR_KNOWN(code))
790 791 792
	{
	  if (!bgp_attr_table[code].allow_in_ebgp && !p->is_internal)
	    continue;
793 794
	  /* The flags might have been zero if the attr was added by filters */
	  a->flags = (a->flags & BAF_PARTIAL) | bgp_attr_table[code].expected_flags;
795 796 797 798 799 800 801 802
	  if (code < 32)
	    seen |= 1 << code;
	}
      else
	{
	  /* Don't re-export unknown non-transitive attributes */
	  if (!(a->flags & BAF_TRANSITIVE))
	    continue;
803
	}
804
      *d = *a;
805 806
      if ((d->type & EAF_ORIGINATED) && !originate && (d->flags & BAF_TRANSITIVE) && (d->flags & BAF_OPTIONAL))
	d->flags |= BAF_PARTIAL;
807 808
      switch (d->type & EAF_TYPE_MASK)
	{
809
	case EAF_TYPE_INT_SET:
810 811 812
	  {
	    struct adata *z = alloca(sizeof(struct adata) + d->u.ptr->length);
	    z->length = d->u.ptr->length;
813 814 815 816 817 818 819 820 821
	    bgp_normalize_int_set((u32 *) z->data, (u32 *) d->u.ptr->data, z->length / 4);
	    d->u.ptr = z;
	    break;
	  }
	case EAF_TYPE_EC_SET:
	  {
	    struct adata *z = alloca(sizeof(struct adata) + d->u.ptr->length);
	    z->length = d->u.ptr->length;
	    bgp_normalize_ec_set(z, (u32 *) d->u.ptr->data, p->is_internal);
822 823 824
	    d->u.ptr = z;
	    break;
	  }
825
	default: ;
826 827 828 829 830 831 832
	}
      d++;
      new->count++;
    }

  /* Hash */
  hash = ea_hash(new);
Martin Mareš's avatar
Martin Mareš committed
833
  for(b=p->bucket_hash[hash & (p->hash_size - 1)]; b; b=b->hash_next)
834 835 836 837 838 839 840
    if (b->hash == hash && ea_same(b->eattrs, new))
      {
	DBG("Found bucket.\n");
	return b;
      }

  /* Ensure that there are all mandatory attributes */
841
  for(i=0; i<ARRAY_SIZE(bgp_mandatory_attrs); i++)
842 843
    if (!(seen & (1 << bgp_mandatory_attrs[i])))
      {
844
	log(L_ERR "%s: Mandatory attribute %s missing in route %I/%d", p->p.name, bgp_attr_table[bgp_mandatory_attrs[i]].name, n->n.prefix, n->n.pxlen);
845 846 847
	return NULL;
      }

848 849
  /* Check if next hop is valid */
  a = ea_find(new, EA_CODE(EAP_BGP, BA_NEXT_HOP));
850
  if (!a || ipa_equal(p->cf->remote_ip, *(ip_addr *)a->u.ptr->data))
851 852 853 854
    {
      log(L_ERR "%s: Invalid NEXT_HOP attribute in route %I/%d", p->p.name, n->n.prefix, n->n.pxlen);
      return NULL;
    }
855

856 857 858 859 860
  /* Create new bucket */
  DBG("Creating bucket.\n");
  return bgp_new_bucket(p, new, hash);
}

861 862 863 864 865 866 867 868 869 870 871 872
void
bgp_free_bucket(struct bgp_proto *p, struct bgp_bucket *buck)
{
  if (buck->hash_next)
    buck->hash_next->hash_prev = buck->hash_prev;
  if (buck->hash_prev)
    buck->hash_prev->hash_next = buck->hash_next;
  else
    p->bucket_hash[buck->hash & (p->hash_size-1)] = buck->hash_next;
  mb_free(buck);
}

Martin Mareš's avatar
Martin Mareš committed
873
void
874
bgp_rt_notify(struct proto *P, rtable *tbl UNUSED, net *n, rte *new, rte *old UNUSED, ea_list *attrs)
Martin Mareš's avatar
Martin Mareš committed
875
{
876
  struct bgp_proto *p = (struct bgp_proto *) P;
877 878
  struct bgp_bucket *buck;
  struct bgp_prefix *px;
879

880
  DBG("BGP: Got route %I/%d %s\n", n->n.prefix, n->n.pxlen, new ? "up" : "down");
881 882 883

  if (new)
    {
884
      buck = bgp_get_bucket(p, n, attrs, new->attrs->source != RTS_BGP);
885 886 887
      if (!buck)			/* Inconsistent attribute list */
	return;
    }
888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903
  else
    {
      if (!(buck = p->withdraw_bucket))
	{
	  buck = p->withdraw_bucket = mb_alloc(P->pool, sizeof(struct bgp_bucket));
	  init_list(&buck->prefixes);
	}
    }
  px = fib_get(&p->prefix_fib, &n->n.prefix, n->n.pxlen);
  if (px->bucket_node.next)
    {
      DBG("\tRemoving old entry.\n");
      rem_node(&px->bucket_node);
    }
  add_tail(&buck->prefixes, &px->bucket_node);
  bgp_schedule_packet(p->conn, PKT_UPDATE);
Martin Mareš's avatar
Martin Mareš committed
904 905
}

906 907
static int
bgp_create_attrs(struct bgp_proto *p, rte *e, ea_list **attrs, struct linpool *pool)
Martin Mareš's avatar
Martin Mareš committed
908
{
909
  ea_list *ea = lp_alloc(pool, sizeof(ea_list) + 4*sizeof(eattr));
Martin Mareš's avatar
Martin Mareš committed
910
  rta *rta = e->attrs;
911
  byte *z;
Martin Mareš's avatar
Martin Mareš committed
912

913 914
  ea->next = *attrs;
  *attrs = ea;
Martin Mareš's avatar
Martin Mareš committed
915
  ea->flags = EALF_SORTED;
916
  ea->count = 4;
Martin Mareš's avatar
Martin Mareš committed
917

918
  bgp_set_attr(ea->attrs, BA_ORIGIN,
Ondřej Filip's avatar
Ondřej Filip committed
919
       ((rta->source == RTS_OSPF_EXT1) || (rta->source == RTS_OSPF_EXT2)) ? ORIGIN_INCOMPLETE : ORIGIN_IGP);
Martin Mareš's avatar
Martin Mareš committed
920 921

  if (p->is_internal)
922
    bgp_set_attr_wa(ea->attrs+1, pool, BA_AS_PATH, 0);
Martin Mareš's avatar
Martin Mareš committed
923 924
  else
    {
925
      z = bgp_set_attr_wa(ea->attrs+1, pool, BA_AS_PATH, 6);
926
      z[0] = AS_PATH_SEQUENCE;
Martin Mareš's avatar
Martin Mareš committed
927
      z[1] = 1;				/* 1 AS */
928
      put_u32(z+2, p->local_as);
Martin Mareš's avatar
Martin Mareš committed
929 930
    }

931 932
  /* iBGP -> use gw, eBGP multi-hop -> use source_addr,
     eBGP single-hop -> use gw if on the same iface */
933
  z = bgp_set_attr_wa(ea->attrs+2, pool, BA_NEXT_HOP, NEXT_HOP_LENGTH);
934
  if (p->cf->next_hop_self ||
935
      rta->dest != RTD_ROUTER ||
936
      ipa_equal(rta->gw, IPA_NONE) ||
937
      ipa_has_link_scope(rta->gw) ||
938
      (!p->is_internal && (!p->neigh || (rta->iface != p->neigh->iface))))
939
    set_next_hop(z, p->source_addr);
Martin Mareš's avatar
Martin Mareš committed
940
  else
941
    set_next_hop(z, rta->gw);
942

943
  bgp_set_attr(ea->attrs+3, BA_LOCAL_PREF, p->cf->default_local_pref);
Martin Mareš's avatar
Martin Mareš committed
944

945
  return 0;				/* Leave decision to the filters */
Martin Mareš's avatar
Martin Mareš committed
946 947
}

948 949 950 951 952 953 954 955 956 957

static inline int
bgp_as_path_loopy(struct bgp_proto *p, rta *a)
{
  eattr *e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
  return (e && as_path_is_member(e->u.ptr, p->local_as));
}

static inline int
bgp_originator_id_loopy(struct bgp_proto *p, rta *a)
958
{
959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982
  eattr *e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
  return (e && (e->u.data == p->local_id));
}

static inline int
bgp_cluster_list_loopy(struct bgp_proto *p, rta *a)
{
  eattr *e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
  return (e && p->rr_client && int_set_contains(e->u.ptr, p->rr_cluster_id));
}


static inline void
bgp_path_prepend(rte *e, ea_list **attrs, struct linpool *pool, u32 as)
{
  eattr *a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
  bgp_attach_attr(attrs, pool, BA_AS_PATH, (uintptr_t) as_path_prepend(pool, a->u.ptr, as));
}

static inline void
bgp_cluster_list_prepend(rte *e, ea_list **attrs, struct linpool *pool, u32 cid)
{
  eattr *a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
  bgp_attach_attr(attrs, pool, BA_CLUSTER_LIST, (uintptr_t) int_set_add(pool, a ? a->u.ptr : NULL, cid));
Martin Mareš's avatar
Martin Mareš committed
983 984
}

985
static int
986
bgp_update_attrs(struct bgp_proto *p, rte *e, ea_list **attrs, struct linpool *pool, int rr)
Martin Mareš's avatar
Martin Mareš committed
987
{
988 989
  eattr *a;

990
  if (!p->is_internal && !p->rs_client)
991 992 993 994 995 996 997 998 999 1000 1001
    {
      bgp_path_prepend(e, attrs, pool, p->local_as);

      /* The MULTI_EXIT_DISC attribute received from a neighboring AS MUST NOT be
       * propagated to other neighboring ASes.
       * Perhaps it would be better to undefine it.
       */
      a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
      if (a)
	bgp_attach_attr(attrs, pool, BA_MULTI_EXIT_DISC, 0);
    }
Martin Mareš's avatar
Martin Mareš committed
1002

1003
  /* iBGP -> keep next_hop, eBGP multi-hop -> use source_addr,
1004 1005 1006
   * eBGP single-hop -> keep next_hop if on the same iface.
   * If the next_hop is zero (i.e. link-local), keep only if on the same iface.
   */
1007
  a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_NEXT_HOP));
1008
  if (a && !p->cf->next_hop_self && 
1009 1010
      ((p->is_internal && ipa_nonzero(*((ip_addr *) a->u.ptr->data))) ||
       (p->neigh && (e->attrs->iface == p->neigh->iface))))
1011 1012 1013 1014 1015 1016
    {
      /* Leave the original next hop attribute, will check later where does it point */
    }
  else
    {
      /* Need to create new one */
1017 1018
      byte *b = bgp_attach_attr_wa(attrs, pool, BA_NEXT_HOP, NEXT_HOP_LENGTH);
      set_next_hop(b, p->source_addr);
1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
    }

  if (rr)
    {
      /* Handling route reflection, RFC 4456 */
      struct bgp_proto *src = (struct bgp_proto *) e->attrs->proto;

      a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
      if (!a)
	bgp_attach_attr(attrs, pool, BA_ORIGINATOR_ID, src->remote_id);

      /* We attach proper cluster ID according to whether the route is entering or leaving the cluster */
      bgp_cluster_list_prepend(e, attrs, pool, src->rr_client ? src->rr_cluster_id : p->rr_cluster_id);

      /* Two RR clients with different cluster ID, hmmm */
      if (src->rr_client && p->rr_client && (src->rr_cluster_id != p->rr_cluster_id))
	bgp_cluster_list_prepend(e, attrs, pool, p->rr_cluster_id);
1036
    }
Martin Mareš's avatar
Martin Mareš committed
1037

1038
  return 0;				/* Leave decision to the filters */
Martin Mareš's avatar
Martin Mareš committed
1039 1040
}

1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068
static int
bgp_community_filter(struct bgp_proto *p, rte *e)
{
  eattr *a;
  struct adata *d;

  /* Check if we aren't forbidden to export the route by communities */
  a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_COMMUNITY));
  if (a)
    {
      d = a->u.ptr;
      if (int_set_contains(d, BGP_COMM_NO_ADVERTISE))
	{
	  DBG("\tNO_ADVERTISE\n");
	  return 1;
	}
      if (!p->is_internal &&
	  (int_set_contains(d, BGP_COMM_NO_EXPORT) ||
	   int_set_contains(d, BGP_COMM_NO_EXPORT_SUBCONFED)))
	{
	  DBG("\tNO_EXPORT\n");
	  return 1;
	}
    }

  return 0;
}

Martin Mareš's avatar
Martin Mareš committed
1069 1070 1071 1072 1073 1074 1075
int
bgp_import_control(struct proto *P, rte **new, ea_list **attrs, struct linpool *pool)
{
  rte *e = *new;
  struct bgp_proto *p = (struct bgp_proto *) P;
  struct bgp_proto *new_bgp = (e->attrs->proto->proto == &proto_bgp) ? (struct bgp_proto *) e->attrs->proto : NULL;

1076
  if (p == new_bgp)			/* Poison reverse updates */
Martin Mareš's avatar
Martin Mareš committed
1077 1078 1079
    return -1;
  if (new_bgp)
    {
1080 1081 1082 1083 1084
      /* We should check here for cluster list loop, because the receiving BGP instance
	 might have different cluster ID  */
      if (bgp_cluster_list_loopy(p, e->attrs))
	return -1;

1085
      if (p->cf->interpret_communities && bgp_community_filter(p, e))
1086 1087
	return -1;

Martin Mareš's avatar
Martin Mareš committed
1088
      if (p->local_as == new_bgp->local_as && p->is_internal && new_bgp->is_internal)
1089 1090 1091 1092 1093 1094 1095 1096 1097 1098
	{
	  /* Redistribution of internal routes with IBGP */
	  if (p->rr_client || new_bgp->rr_client)
	    /* Route reflection, RFC 4456 */
	    return bgp_update_attrs(p, e, attrs, pool, 1);
	  else
	    return -1;
	}
      else
	return bgp_update_attrs(p, e, attrs, pool, 0);
Martin Mareš's avatar
Martin Mareš committed
1099 1100
    }
  else
1101
    return bgp_create_attrs(p, e, attrs, pool);
Martin Mareš's avatar
Martin Mareš committed
1102 1103
}

1104 1105 1106 1107 1108 1109
static inline u32
bgp_get_neighbor(rte *r)
{
  eattr *e = ea_find(r->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
  u32 as;

1110
  if (e && as_path_get_first(e->u.ptr, &as))
1111 1112 1113 1114 1115
    return as;
  else
    return ((struct bgp_proto *) r->attrs->proto)->remote_as;
}

1116 1117 1118 1119 1120 1121 1122
static inline int
rte_resolvable(rte *rt)
{
  int rd = rt->attrs->dest;  
  return (rd == RTD_ROUTER) || (rd == RTD_DEVICE) || (rd == RTD_MULTIPATH);
}

Martin Mareš's avatar
Martin Mareš committed
1123 1124 1125 1126 1127
int
bgp_rte_better(rte *new, rte *old)
{
  struct bgp_proto *new_bgp = (struct bgp_proto *) new->attrs->proto;
  struct bgp_proto *old_bgp = (struct bgp_proto *) old->attrs->proto;
1128 1129
  eattr *x, *y;
  u32 n, o;
Martin Mareš's avatar
Martin Mareš committed
1130

1131 1132 1133 1134 1135 1136 1137 1138
  /* Skip suppressed routes (see bgp_rte_recalculate()) */
  n = new->u.bgp.suppressed;
  o = old->u.bgp.suppressed;
  if (n > o)
    return 0;
  if (n < o)
    return 1;

1139
  /* RFC 4271 9.1.2.1. Route resolvability test */
1140 1141
  n = rte_resolvable(new);
  o = rte_resolvable(old);
1142 1143 1144 1145 1146
  if (n > o)
    return 1;
  if (n < o)
    return 0;

Martin Mareš's avatar
Martin Mareš committed
1147
  /* Start with local preferences */
1148 1149 1150 1151 1152 1153 1154 1155 1156
  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
  n = x ? x->u.data : new_bgp->cf->default_local_pref;
  o = y ? y->u.data : old_bgp->cf->default_local_pref;
  if (n > o)
    return 1;
  if (n < o)
    return 0;

1157
  /* RFC 4271 9.1.2.2. a)  Use AS path lengths */
1158
  if (new_bgp->cf->compare_path_lengths || old_bgp->cf->compare_path_lengths)
Martin Mareš's avatar
Martin Mareš committed
1159
    {
1160 1161
      x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
      y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1162 1163
      n = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN;
      o = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN;
1164
      if (n < o)
Martin Mareš's avatar
Martin Mareš committed
1165
	return 1;
1166
      if (n > o)
Martin Mareš's avatar
Martin Mareš committed
1167 1168 1169
	return 0;
    }

1170
  /* RFC 4271 9.1.2.2. b) Use origins */
1171 1172
  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1173 1174
  n = x ? x->u.data : ORIGIN_INCOMPLETE;
  o = y ? y->u.data : ORIGIN_INCOMPLETE;
1175 1176 1177 1178 1179
  if (n < o)
    return 1;
  if (n > o)
    return 0;

1180
  /* RFC 4271 9.1.2.2. c) Compare MED's */
1181 1182 1183 1184 1185 1186 1187 1188 1189
  /* Proper RFC 4271 path selection cannot be interpreted as finding
   * the best path in some ordering. It is implemented partially in
   * bgp_rte_recalculate() when deterministic_med option is
   * active. Without that option, the behavior is just an
   * approximation, which in specific situations may lead to
   * persistent routing loops, because it is nondeterministic - it
   * depends on the order in which routes appeared. But it is also the
   * same behavior as used by default in Cisco routers, so it is
   * probably not a big issue.
1190 1191 1192
   */
  if (new_bgp->cf->med_metric || old_bgp->cf->med_metric ||
      (bgp_get_neighbor(new) == bgp_get_neighbor(old)))
1193 1194 1195 1196 1197 1198 1199 1200 1201 1202
    {
      x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
      y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
      n = x ? x->u.data : new_bgp->cf->default_med;
      o = y ? y->u.data : old_bgp->cf->default_med;
      if (n < o)
	return 1;
      if (n > o)
	return 0;
    }
1203

1204
  /* RFC 4271 9.1.2.2. d) Prefer external peers */
Martin Mareš's avatar
Martin Mareš committed
1205 1206 1207 1208 1209
  if (new_bgp->is_internal > old_bgp->is_internal)
    return 0;
  if (new_bgp->is_internal < old_bgp->is_internal)
    return 1;

1210 1211 1212 1213 1214 1215 1216
  /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
  n = new_bgp->cf->igp_metric ? new->attrs->igp_metric : 0;
  o = old_bgp->cf->igp_metric ? old->attrs->igp_metric : 0;
  if (n < o)
    return 1;
  if (n > o)
    return 0;
1217 1218 1219 1220 1221 1222 1223

  /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */
  /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighor ID */
  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
  n = x ? x->u.data : new_bgp->remote_id;
  o = y ? y->u.data : old_bgp->remote_id;
1224 1225 1226 1227 1228 1229 1230 1231

  /* RFC 5004 - prefer older routes */
  /* (if both are external and from different peer) */
  if ((new_bgp->cf->prefer_older || old_bgp->cf->prefer_older) &&
      !new_bgp->is_internal && n != o)
    return 0;

  /* rest of RFC 4271 9.1.2.2. f) */
1232 1233 1234 1235
  if (n < o)
    return 1;
  if (n > o)
    return 0;
1236

1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
  /* RFC 4456 9. b) Compare cluster list lengths */
  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
  n = x ? int_set_get_size(x->u.ptr) : 0;
  o = y ? int_set_get_size(y->u.ptr) : 0;
  if (n < o)
    return 1;
  if (n > o)
    return 0;

1247 1248 1249 1250
  /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
  return (ipa_compare(new_bgp->cf->remote_ip, old_bgp->cf->remote_ip) < 0);
}

1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392

static inline int
same_group(rte *r, u32 lpref, u32 lasn)
{
  return (r->pref == lpref) && (bgp_get_neighbor(r) == lasn);
}

static inline int
use_deterministic_med(rte *r)
{
  return ((struct bgp_proto *) r->attrs->proto)->cf->deterministic_med;
}

int
bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best)
{
  rte *r, *s;
  rte *key = new ? new : old;
  u32 lpref = key->pref;
  u32 lasn = bgp_get_neighbor(key);
  int old_is_group_best = 0;

  /*
   * Proper RFC 4271 path selection is a bit complicated, it cannot be
   * implemented just by rte_better(), because it is not a linear
   * ordering. But it can be splitted to two levels, where the lower
   * level chooses the best routes in each group of routes from the
   * same neighboring AS and higher level chooses the best route (with
   * a slightly different ordering) between the best-in-group routes.
   *
   * When deterministic_med is disabled, we just ignore this issue and
   * choose the best route by bgp_rte_better() alone. If enabled, the
   * lower level of the route selection is done here (for the group
   * to which the changed route belongs), all routes in group are
   * marked as suppressed, just chosen best-in-group is not.
   *
   * Global best route selection then implements higher level by
   * choosing between non-suppressed routes (as they are always
   * preferred over suppressed routes). Routes from BGP protocols
   * that do not set deterministic_med are just never suppressed. As
   * they do not participate in the lower level selection, it is OK
   * that this fn is not called for them.
   *
   * The idea is simple, the implementation is more problematic,
   * mostly because of optimizations in rte_recalculate() that 
   * avoids full recalculation in most cases.
   *
   * We can assume that at least one of new, old is non-NULL and both
   * are from the same protocol with enabled deterministic_med. We
   * group routes by both neighbor AS (lasn) and preference (lpref),
   * because bgp_rte_better() does not handle preference itself.
   */

  /* If new and old are from different groups, we just process that
     as two independent events */
  if (new && old && !same_group(old, lpref, lasn))
    {
      int i1, i2;
      i1 = bgp_rte_recalculate(table, net, NULL, old, old_best);
      i2 = bgp_rte_recalculate(table, net, new, NULL, old_best);
      return i1 || i2;
    }

  /* 
   * We could find the best-in-group and then make some shortcuts like
   * in rte_recalculate, but as we would have to walk through all
   * net->routes just to find it, it is probably not worth. So we
   * just have two simpler fast cases that use just the old route.
   * We also set suppressed flag to avoid using it in bgp_rte_better().
   */

  if (new)
    new->u.bgp.suppressed = 1;

  if (old)
    {
      old_is_group_best = !old->u.bgp.suppressed;
      old->u.bgp.suppressed = 1;
      int new_is_better = new && bgp_rte_better(new, old);

      /* The first case - replace not best with worse (or remove not best) */
      if (!old_is_group_best && !new_is_better)
	return 0;

      /* The second case - replace the best with better */
      if (old_is_group_best && new_is_better)
	{
	  /* new is best-in-group, the see discussion below - this is
	     a special variant of NBG && OBG. From OBG we can deduce
	     that same_group(old_best) iff (old == old_best)  */
	  new->u.bgp.suppressed = 0;
	  return (old == old_best);
	}
    }

  /* The default case - find a new best-in-group route */
  r = new; /* new may not be in the list */
  for (s=net->routes; s; s=s->next)
    if (use_deterministic_med(s) && same_group(s, lpref, lasn))
      {
	s->u.bgp.suppressed = 1;
	if (!r || bgp_rte_better(s, r))
	  r = s;
      }

  /* Simple case - the last route in group disappears */
  if (!r)
    return 0;

  /* Found best-in-group */
  r->u.bgp.suppressed = 0;

  /*
   * There are generally two reasons why we have to force
   * recalculation (return 1): First, the new route may be wrongfully
   * chosen to be the best in the first case check in
   * rte_recalculate(), this may happen only if old_best is from the
   * same group. Second, another (different than new route)
   * best-in-group is chosen and that may be the proper best (although
   * rte_recalculate() without ignore that possibility).
   *
   * There are three possible cases according to whether the old route
   * was the best in group (OBG, stored in old_is_group_best) and
   * whether the new route is the best in group (NBG, tested by r == new).
   * These cases work even if old or new is NULL.
   *
   * NBG -> new is a possible candidate for the best route, so we just
   *        check for the first reason using same_group().
   *
   * !NBG && OBG -> Second reason applies, return 1
   *
   * !NBG && !OBG -> Best in group does not change, old != old_best,
   *                 rte_better(new, old_best) is false and therefore
   *                 the first reason does not apply, return 0
   */

  if (r == new)
    return old_best && same_group(old_best, lpref, lasn);
  else
    return old_is_group_best;
}

1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412
static struct adata *
bgp_aggregator_convert_to_new(struct adata *old, struct linpool *pool)
{
  struct adata *newa = lp_alloc(pool, sizeof(struct adata) + 8);
  newa->length = 8;
  aggregator_convert_to_new(old, newa->data);
  return newa;
}


/* Take last req_as ASNs from path old2 (in 2B format), convert to 4B format
 * and append path old4 (in 4B format).
 */
static struct adata *
bgp_merge_as_paths(struct adata *old2, struct adata *old4, int req_as, struct linpool *pool)
{
  byte buf[old2->length * 2];

  int ol = as_path_convert_to_new(old2, buf, req_as);
  int nl = ol + (old4 ? old4->length : 0);
Martin Mareš's avatar
Martin Mareš committed
1413

1414 1415 1416 1417 1418 1419 1420 1421
  struct adata *newa = lp_alloc(pool, sizeof(struct adata) + nl);
  newa->length = nl;
  memcpy(newa->data, buf, ol);
  if (old4) memcpy(newa->data + ol, old4->data, old4->length);

  return newa;
}

1422 1423 1424
static int
as4_aggregator_valid(struct adata *aggr)
{
1425
  return aggr->length == 8;
1426 1427
}

1428

1429
/* Reconstruct 4B AS_PATH and AGGREGATOR according to RFC 4893 4.2.3 */
1430 1431 1432 1433 1434 1435 1436
static void
bgp_reconstruct_4b_atts(struct bgp_proto *p, rta *a, struct linpool *pool)
{
  eattr *p2 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
  eattr *p4 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS4_PATH));
  eattr *a2 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AGGREGATOR));
  eattr *a4 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS4_AGGREGATOR));
1437 1438 1439 1440
  int a4_removed = 0;

  if (a4 && !as4_aggregator_valid(a4->u.ptr))
    {
1441
      log(L_WARN "%s: AS4_AGGREGATOR attribute is invalid, skipping attribute", p->p.name);
1442 1443 1444
      a4 = NULL;
      a4_removed = 1;
    }
1445 1446

  if (a2)
Martin Mareš's avatar
Martin Mareš committed
1447
    {
1448 1449 1450
      u32 a2_as = get_u16(a2->u.ptr->data);

      if (a4)
Martin Mareš's avatar
Martin Mareš committed
1451
	{
1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475
	  if (a2_as != AS_TRANS)
	    {
	      /* Routes were aggregated by old router and therefore AS4_PATH
	       * and AS4_AGGREGATOR is invalid
	       *
	       * Convert AS_PATH and AGGREGATOR to 4B format and finish.
	       */

	      a2->u.ptr = bgp_aggregator_convert_to_new(a2->u.ptr, pool);
	      p2->u.ptr = bgp_merge_as_paths(p2->u.ptr, NULL, AS_PATH_MAXLEN, pool);

	      return;
	    }
	  else
	    {
	      /* Common case, use AS4_AGGREGATOR attribute */
	      a2->u.ptr = a4->u.ptr;
	    }
	}
      else
	{
	  /* Common case, use old AGGREGATOR attribute */
	  a2->u.ptr = bgp_aggregator_convert_to_new(a2->u.ptr, pool);

1476
	  if ((a2_as == AS_TRANS) && !a4_removed)
1477
	    log(L_WARN "%s: AGGREGATOR attribute contain AS_TRANS, but AS4_AGGREGATOR is missing", p->p.name);
Martin Mareš's avatar
Martin Mareš committed
1478 1479
	}
    }
1480 1481
  else
    if (a4)
1482
      log(L_WARN "%s: AS4_AGGREGATOR attribute received, but AGGREGATOR attribute is missing", p->p.name);
1483

1484
  int p2_len = as_path_getlen_int(p2->u.ptr, 2);
1485 1486 1487 1488
  int p4_len = p4 ? validate_as4_path(p, p4->u.ptr) : -1;

  if (p4 && (p4_len < 0))
    log(L_WARN "%s: AS4_PATH attribute is malformed, skipping attribute", p->p.name);
1489

1490
  if ((p4_len <= 0) || (p2_len < p4_len))
1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502
    p2->u.ptr = bgp_merge_as_paths(p2->u.ptr, NULL, AS_PATH_MAXLEN, pool);
  else
    p2->u.ptr = bgp_merge_as_paths(p2->u.ptr, p4->u.ptr, p2_len - p4_len, pool);
}

static void
bgp_remove_as4_attrs(struct bgp_proto *p, rta *a)
{
  unsigned id1 = EA_CODE(EAP_BGP, BA_AS4_PATH);
  unsigned id2 = EA_CODE(EAP_BGP, BA_AS4_AGGREGATOR);
  ea_list **el = &(a->eattrs);

1503
  /* We know that ea_lists constructed in bgp_decode attrs have one attribute per ea_list struct */
1504 1505 1506 1507 1508 1509 1510
  while (*el != NULL)
    {
      unsigned fid = (*el)->attrs[0].id;

      if ((fid == id1) || (fid == id2))
	{
	  *el = (*el)->next;
1511
	  if (p->as4_session)
1512
	    log(L_WARN "%s: Unexpected AS4_* attributes received", p->p.name);
1513 1514 1515 1516
	}
      else
	el = &((*el)->next);
    }
Martin Mareš's avatar
Martin Mareš committed
1517 1518
}

Martin Mareš's avatar
Martin Mareš committed
1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530
/**
 * bgp_decode_attrs - check and decode BGP attributes
 * @conn: connection
 * @attr: start of attribute block
 * @len: length of attribute block
 * @pool: linear pool to make all the allocations in
 * @mandatory: 1 iff presence of mandatory attributes has to be checked
 *
 * This function takes a BGP attribute block (a part of an Update message), checks
 * its consistency and converts it to a list of BIRD route attributes represented
 * by a &rta.
 */
1531
struct rta *
1532
bgp_decode_attrs(struct bgp_conn *conn, byte *attr, unsigned int len, struct linpool *pool, int mandatory)
1533 1534 1535
{
  struct bgp_proto *bgp = conn->bgp;
  rta *a = lp_alloc(pool, sizeof(struct rta));
Martin Mareš's avatar
Martin Mareš committed
1536 1537
  unsigned int flags, code, l, i, type;
  int errcode;
1538 1539 1540 1541
  byte *z, *attr_start;
  byte seen[256/8];
  ea_list *ea;
  struct adata *ad;
1542
  int withdraw = 0;
1543

1544
  bzero(a, sizeof(rta));
1545 1546 1547 1548
  a->proto = &bgp->p;
  a->source = RTS_BGP;
  a->scope = SCOPE_UNIVERSE;
  a->cast = RTC_UNICAST;
1549
  /* a->dest = RTD_ROUTER;  -- set in bgp_set_next_hop() */
1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585
  a->from = bgp->cf->remote_ip;

  /* Parse the attributes */
  bzero(seen, sizeof(seen));
  DBG("BGP: Parsing attributes\n");
  while (len)
    {
      if (len < 2)
	goto malformed;
      attr_start = attr;
      flags = *attr++;
      code = *attr++;
      len -= 2;
      if (flags & BAF_EXT_LEN)
	{
	  if (len < 2)
	    goto malformed;
	  l = get_u16(attr);
	  attr += 2;
	  len -= 2;
	}
      else
	{
	  if (len < 1)
	    goto malformed;
	  l = *attr++;
	  len--;
	}
      if (l > len)
	goto malformed;
      len -= l;
      z = attr;
      attr += l;
      DBG("Attr %02x %02x %d\n", code, flags, l);
      if (seen[code/8] & (1 << (code%8)))
	goto malformed;
1586
      if (ATTR_KNOWN(code))
1587 1588 1589 1590 1591 1592
	{
	  struct attr_desc *desc = &bgp_attr_table[code];
	  if (desc->expected_length >= 0 && desc->expected_length != (int) l)
	    { errcode = 5; goto err; }
	  if ((desc->expected_flags ^ flags) & (BAF_OPTIONAL | BAF_TRANSITIVE))
	    { errcode = 4; goto err; }
1593 1594
	  if (!desc->allow_in_ebgp && !bgp->is_internal)
	    continue;
Martin Mareš's avatar
Martin Mareš committed
1595 1596 1597 1598 1599
	  if (desc->validate)
	    {
	      errcode = desc->validate(bgp, z, l);
	      if (errcode > 0)
		goto err;
1600
	      if (errcode == IGNORE)
Martin Mareš's avatar
Martin Mareš committed
1601
		continue;
1602 1603 1604 1605 1606 1607
	      if (errcode <= WITHDRAW)
		{
		  log(L_WARN "%s: Attribute %s is malformed, withdrawing update",
		      bgp->p.name, desc->name);
		  withdraw = 1;
		}
Martin Mareš's avatar
Martin Mareš committed
1608
	    }
1609 1610 1611 1612 1613 1614
	  else if (code == BA_AS_PATH)
	    {
	      /* Special case as it might also trim the attribute */
	      if (validate_as_path(bgp, z, &l) < 0)
		{ errcode = 11; goto err; }
	    }
1615 1616 1617
	  type = desc->type;
	}
      else				/* Unknown attribute */
1618
	{
1619 1620 1621 1622
	  if (!(flags & BAF_OPTIONAL))
	    { errcode = 2; goto err; }
	  type = EAF_TYPE_OPAQUE;
	}