Annotation of embedaddon/bird/proto/bgp/attrs.c, revision 1.1.1.2
1.1 misho 1: /*
2: * BIRD -- BGP Attributes
3: *
4: * (c) 2000 Martin Mares <mj@ucw.cz>
5: *
6: * Can be freely distributed and used under the terms of the GNU GPL.
7: */
8:
9: #undef LOCAL_DEBUG
10:
11: #include <stdlib.h>
12:
13: #include "nest/bird.h"
14: #include "nest/iface.h"
15: #include "nest/protocol.h"
16: #include "nest/route.h"
17: #include "nest/attrs.h"
18: #include "conf/conf.h"
19: #include "lib/resource.h"
20: #include "lib/string.h"
21: #include "lib/unaligned.h"
22:
23: #include "bgp.h"
24:
25: /*
26: * UPDATE message error handling
27: *
28: * All checks from RFC 4271 6.3 are done as specified with these exceptions:
29: * - The semantic check of an IP address from NEXT_HOP attribute is missing.
30: * - Checks of some optional attribute values are missing.
31: * - Syntactic and semantic checks of NLRIs (done in DECODE_PREFIX())
32: * are probably inadequate.
33: *
34: * Loop detection based on AS_PATH causes updates to be withdrawn. RFC
35: * 4271 does not explicitly specifiy the behavior in that case.
36: *
37: * Loop detection related to route reflection (based on ORIGINATOR_ID
38: * and CLUSTER_LIST) causes updates to be withdrawn. RFC 4456 8
39: * specifies that such updates should be ignored, but that is generally
40: * a bad idea.
41: *
42: * Error checking of optional transitive attributes is done according to
43: * draft-ietf-idr-optional-transitive-03, but errors are handled always
44: * as withdraws.
45: *
46: * Unexpected AS_CONFED_* segments in AS_PATH are logged and removed,
47: * but unknown segments cause a session drop with Malformed AS_PATH
48: * error (see validate_path()). The behavior in such case is not
49: * explicitly specified by RFC 4271. RFC 5065 specifies that
50: * inconsistent AS_CONFED_* segments should cause a session drop, but
51: * implementations that pass invalid AS_CONFED_* segments are
52: * widespread.
53: *
1.1.1.2 ! misho 54: * Error handling of AS4_* attributes is done as specified by RFC 6793. There
! 55: * are several possible inconsistencies between AGGREGATOR and AS4_AGGREGATOR
! 56: * that are not handled by that RFC, these are logged and ignored (see
1.1 misho 57: * bgp_reconstruct_4b_attrs()).
58: */
59:
60:
61: static byte bgp_mandatory_attrs[] = { BA_ORIGIN, BA_AS_PATH
62: #ifndef IPV6
63: ,BA_NEXT_HOP
64: #endif
65: };
66:
67: struct attr_desc {
68: char *name;
69: int expected_length;
70: int expected_flags;
71: 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: #define IGNORE -1
78: #define WITHDRAW -2
79:
80: static int
81: bgp_check_origin(struct bgp_proto *p UNUSED, byte *a, int len UNUSED)
82: {
83: if (*a > 2)
84: return 6;
85: return 0;
86: }
87:
88: static void
89: bgp_format_origin(eattr *a, byte *buf, int buflen UNUSED)
90: {
91: static char *bgp_origin_names[] = { "IGP", "EGP", "Incomplete" };
92:
93: bsprintf(buf, bgp_origin_names[a->u.data]);
94: }
95:
96: static int
97: path_segment_contains(byte *p, int bs, u32 asn)
98: {
99: int i;
100: int len = p[1];
101: p += 2;
102:
103: for(i=0; i<len; i++)
104: {
105: u32 asn2 = (bs == 4) ? get_u32(p) : get_u16(p);
106: if (asn2 == asn)
107: return 1;
108: p += bs;
109: }
110:
111: return 0;
112: }
113:
114: /* Validates path attribute, removes AS_CONFED_* segments, and also returns path length */
115: static int
116: validate_path(struct bgp_proto *p, int as_path, int bs, byte *idata, uint *ilength)
117: {
118: int res = 0;
119: u8 *a, *dst;
120: int len, plen;
121:
122: dst = a = idata;
123: len = *ilength;
124:
125: while (len)
126: {
127: if (len < 2)
128: return -1;
129:
130: plen = 2 + bs * a[1];
131: if (len < plen)
132: return -1;
133:
134: if (a[1] == 0)
135: {
136: log(L_WARN "%s: %s_PATH attribute contains empty segment, skipping it",
137: p->p.name, as_path ? "AS" : "AS4");
138: goto skip;
139: }
140:
141: switch (a[0])
142: {
143: case AS_PATH_SET:
144: res++;
145: break;
146:
147: case AS_PATH_SEQUENCE:
148: res += a[1];
149: break;
150:
151: case AS_PATH_CONFED_SEQUENCE:
152: case AS_PATH_CONFED_SET:
153: if (as_path && path_segment_contains(a, bs, p->remote_as))
154: {
155: log(L_WARN "%s: AS_CONFED_* segment with peer ASN found, misconfigured confederation?", p->p.name);
156: return -1;
157: }
158:
159: log(L_WARN "%s: %s_PATH attribute contains AS_CONFED_* segment, skipping segment",
160: p->p.name, as_path ? "AS" : "AS4");
161: goto skip;
162:
163: default:
164: return -1;
165: }
166:
167: if (dst != a)
168: memmove(dst, a, plen);
169: dst += plen;
170:
171: skip:
172: len -= plen;
173: a += plen;
174: }
175:
176: *ilength = dst - idata;
177: return res;
178: }
179:
180: static inline int
181: validate_as_path(struct bgp_proto *p, byte *a, int *len)
182: {
183: return validate_path(p, 1, p->as4_session ? 4 : 2, a, len);
184: }
185:
186: static inline int
187: validate_as4_path(struct bgp_proto *p, struct adata *path)
188: {
189: return validate_path(p, 0, 4, path->data, &path->length);
190: }
191:
192: static int
193: bgp_check_next_hop(struct bgp_proto *p UNUSED, byte *a UNUSED6, int len UNUSED6)
194: {
195: #ifdef IPV6
196: return IGNORE;
197: #else
198: ip_addr addr;
199:
200: memcpy(&addr, a, len);
201: ipa_ntoh(addr);
202: if (ipa_classify(addr) & IADDR_HOST)
203: return 0;
204: else
205: return 8;
206: #endif
207: }
208:
209: static void
210: bgp_format_next_hop(eattr *a, byte *buf, int buflen UNUSED)
211: {
212: ip_addr *ipp = (ip_addr *) a->u.ptr->data;
213: #ifdef IPV6
214: /* in IPv6, we might have two addresses in NEXT HOP */
215: if ((a->u.ptr->length == NEXT_HOP_LENGTH) && ipa_nonzero(ipp[1]))
216: {
217: bsprintf(buf, "%I %I", ipp[0], ipp[1]);
218: return;
219: }
220: #endif
221:
222: bsprintf(buf, "%I", ipp[0]);
223: }
224:
225: static int
226: bgp_check_aggregator(struct bgp_proto *p, byte *a UNUSED, int len)
227: {
228: int exp_len = p->as4_session ? 8 : 6;
229:
230: return (len == exp_len) ? 0 : WITHDRAW;
231: }
232:
233: static void
234: bgp_format_aggregator(eattr *a, byte *buf, int buflen UNUSED)
235: {
236: struct adata *ad = a->u.ptr;
237: byte *data = ad->data;
238: u32 as;
239:
240: as = get_u32(data);
241: data += 4;
242:
243: bsprintf(buf, "%d.%d.%d.%d AS%u", data[0], data[1], data[2], data[3], as);
244: }
245:
246: static int
247: bgp_check_community(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
248: {
249: return ((len % 4) == 0) ? 0 : WITHDRAW;
250: }
251:
252: static int
253: bgp_check_cluster_list(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
254: {
255: return ((len % 4) == 0) ? 0 : 5;
256: }
257:
258: static void
259: bgp_format_cluster_list(eattr *a, byte *buf, int buflen)
260: {
261: /* Truncates cluster lists larger than buflen, probably not a problem */
262: int_set_format(a->u.ptr, 0, -1, buf, buflen);
263: }
264:
265: static int
266: bgp_check_reach_nlri(struct bgp_proto *p UNUSED, byte *a UNUSED, int len UNUSED)
267: {
268: #ifdef IPV6
269: p->mp_reach_start = a;
270: p->mp_reach_len = len;
271: #endif
272: return IGNORE;
273: }
274:
275: static int
276: bgp_check_unreach_nlri(struct bgp_proto *p UNUSED, byte *a UNUSED, int len UNUSED)
277: {
278: #ifdef IPV6
279: p->mp_unreach_start = a;
280: p->mp_unreach_len = len;
281: #endif
282: return IGNORE;
283: }
284:
285: static int
286: bgp_check_ext_community(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
287: {
288: return ((len % 8) == 0) ? 0 : WITHDRAW;
289: }
290:
291: static int
292: bgp_check_large_community(struct bgp_proto *p UNUSED, byte *a UNUSED, int len)
293: {
294: return ((len % 12) == 0) ? 0 : WITHDRAW;
295: }
296:
297:
298: static struct attr_desc bgp_attr_table[] = {
299: { NULL, -1, 0, 0, 0, /* Undefined */
300: NULL, NULL },
301: { "origin", 1, BAF_TRANSITIVE, EAF_TYPE_INT, 1, /* BA_ORIGIN */
302: bgp_check_origin, bgp_format_origin },
303: { "as_path", -1, BAF_TRANSITIVE, EAF_TYPE_AS_PATH, 1, /* BA_AS_PATH */
304: NULL, NULL }, /* is checked by validate_as_path() as a special case */
305: { "next_hop", 4, BAF_TRANSITIVE, EAF_TYPE_IP_ADDRESS, 1, /* BA_NEXT_HOP */
306: bgp_check_next_hop, bgp_format_next_hop },
307: { "med", 4, BAF_OPTIONAL, EAF_TYPE_INT, 1, /* BA_MULTI_EXIT_DISC */
308: NULL, NULL },
1.1.1.2 ! misho 309: { "local_pref", 4, BAF_TRANSITIVE, EAF_TYPE_INT, 1, /* BA_LOCAL_PREF */
1.1 misho 310: NULL, NULL },
311: { "atomic_aggr", 0, BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1, /* BA_ATOMIC_AGGR */
312: NULL, NULL },
313: { "aggregator", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1, /* BA_AGGREGATOR */
314: bgp_check_aggregator, bgp_format_aggregator },
315: { "community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_INT_SET, 1, /* BA_COMMUNITY */
316: bgp_check_community, NULL },
317: { "originator_id", 4, BAF_OPTIONAL, EAF_TYPE_ROUTER_ID, 0, /* BA_ORIGINATOR_ID */
318: NULL, NULL },
319: { "cluster_list", -1, BAF_OPTIONAL, EAF_TYPE_INT_SET, 0, /* BA_CLUSTER_LIST */
320: bgp_check_cluster_list, bgp_format_cluster_list },
321: { .name = NULL }, /* BA_DPA */
322: { .name = NULL }, /* BA_ADVERTISER */
323: { .name = NULL }, /* BA_RCID_PATH */
324: { "mp_reach_nlri", -1, BAF_OPTIONAL, EAF_TYPE_OPAQUE, 1, /* BA_MP_REACH_NLRI */
325: bgp_check_reach_nlri, NULL },
326: { "mp_unreach_nlri", -1, BAF_OPTIONAL, EAF_TYPE_OPAQUE, 1, /* BA_MP_UNREACH_NLRI */
327: bgp_check_unreach_nlri, NULL },
328: { "ext_community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_EC_SET, 1, /* BA_EXT_COMMUNITY */
329: bgp_check_ext_community, NULL },
330: { "as4_path", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1, /* BA_AS4_PATH */
331: NULL, NULL },
332: { "as4_aggregator", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_OPAQUE, 1, /* BA_AS4_PATH */
333: NULL, NULL },
334: [BA_LARGE_COMMUNITY] =
335: { "large_community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_LC_SET, 1,
336: bgp_check_large_community, NULL }
337: };
338:
339: /* BA_AS4_PATH is type EAF_TYPE_OPAQUE and not type EAF_TYPE_AS_PATH.
340: * It does not matter as this attribute does not appear on routes in the routing table.
341: */
342:
343: #define ATTR_KNOWN(code) ((code) < ARRAY_SIZE(bgp_attr_table) && bgp_attr_table[code].name)
344:
345: static inline struct adata *
346: bgp_alloc_adata(struct linpool *pool, unsigned len)
347: {
348: struct adata *ad = lp_alloc(pool, sizeof(struct adata) + len);
349: ad->length = len;
350: return ad;
351: }
352:
353: static void
354: bgp_set_attr(eattr *e, unsigned attr, uintptr_t val)
355: {
356: ASSERT(ATTR_KNOWN(attr));
357: e->id = EA_CODE(EAP_BGP, attr);
358: e->type = bgp_attr_table[attr].type;
359: e->flags = bgp_attr_table[attr].expected_flags;
360: if (e->type & EAF_EMBEDDED)
361: e->u.data = val;
362: else
363: e->u.ptr = (struct adata *) val;
364: }
365:
366: static byte *
367: bgp_set_attr_wa(eattr *e, struct linpool *pool, unsigned attr, unsigned len)
368: {
369: struct adata *ad = bgp_alloc_adata(pool, len);
370: bgp_set_attr(e, attr, (uintptr_t) ad);
371: return ad->data;
372: }
373:
374: void
375: bgp_attach_attr(ea_list **to, struct linpool *pool, unsigned attr, uintptr_t val)
376: {
377: ea_list *a = lp_alloc(pool, sizeof(ea_list) + sizeof(eattr));
378: a->next = *to;
379: *to = a;
380: a->flags = EALF_SORTED;
381: a->count = 1;
382: bgp_set_attr(a->attrs, attr, val);
383: }
384:
385: byte *
386: bgp_attach_attr_wa(ea_list **to, struct linpool *pool, unsigned attr, unsigned len)
387: {
388: struct adata *ad = bgp_alloc_adata(pool, len);
389: bgp_attach_attr(to, pool, attr, (uintptr_t) ad);
390: return ad->data;
391: }
392:
393: static int
394: bgp_encode_attr_hdr(byte *dst, uint flags, unsigned code, int len)
395: {
396: int wlen;
397:
398: DBG("\tAttribute %02x (%d bytes, flags %02x)\n", code, len, flags);
399:
400: if (len < 256)
401: {
402: *dst++ = flags;
403: *dst++ = code;
404: *dst++ = len;
405: wlen = 3;
406: }
407: else
408: {
409: *dst++ = flags | BAF_EXT_LEN;
410: *dst++ = code;
411: put_u16(dst, len);
412: wlen = 4;
413: }
414:
415: return wlen;
416: }
417:
418: static void
419: aggregator_convert_to_old(struct adata *aggr, byte *dst, int *new_used)
420: {
421: byte *src = aggr->data;
422: *new_used = 0;
423:
424: u32 as = get_u32(src);
425: if (as > 0xFFFF)
426: {
427: as = AS_TRANS;
428: *new_used = 1;
429: }
430: put_u16(dst, as);
431:
432: /* Copy IPv4 address */
433: memcpy(dst + 2, src + 4, 4);
434: }
435:
436: static void
437: aggregator_convert_to_new(struct adata *aggr, byte *dst)
438: {
439: byte *src = aggr->data;
440:
441: u32 as = get_u16(src);
442: put_u32(dst, as);
443:
444: /* Copy IPv4 address */
445: memcpy(dst + 4, src + 2, 4);
446: }
447:
448: static int
449: bgp_get_attr_len(eattr *a)
450: {
451: int len;
452: if (ATTR_KNOWN(EA_ID(a->id)))
453: {
454: int code = EA_ID(a->id);
455: struct attr_desc *desc = &bgp_attr_table[code];
456: len = desc->expected_length;
457: if (len < 0)
458: {
459: ASSERT(!(a->type & EAF_EMBEDDED));
460: len = a->u.ptr->length;
461: }
462: }
463: else
464: {
465: ASSERT((a->type & EAF_TYPE_MASK) == EAF_TYPE_OPAQUE);
466: len = a->u.ptr->length;
467: }
468:
469: return len;
470: }
471:
472: /**
473: * bgp_encode_attrs - encode BGP attributes
1.1.1.2 ! misho 474: * @p: BGP instance (or NULL)
1.1 misho 475: * @w: buffer
476: * @attrs: a list of extended attributes
477: * @remains: remaining space in the buffer
478: *
479: * The bgp_encode_attrs() function takes a list of extended attributes
480: * and converts it to its BGP representation (a part of an Update message).
481: *
482: * Result: Length of the attribute block generated or -1 if not enough space.
483: */
484: uint
485: bgp_encode_attrs(struct bgp_proto *p, byte *w, ea_list *attrs, int remains)
486: {
487: uint i, code, type, flags;
1.1.1.2 ! misho 488: int as4_session = p ? p->as4_session : 1;
1.1 misho 489: byte *start = w;
490: int len, rv;
491:
492: for(i=0; i<attrs->count; i++)
493: {
494: eattr *a = &attrs->attrs[i];
495: ASSERT(EA_PROTO(a->id) == EAP_BGP);
496: code = EA_ID(a->id);
497:
498: #ifdef IPV6
499: /* When talking multiprotocol BGP, the NEXT_HOP attributes are used only temporarily. */
500: if (code == BA_NEXT_HOP)
501: continue;
502: #endif
503:
504: /* When AS4-aware BGP speaker is talking to non-AS4-aware BGP speaker,
505: * we have to convert our 4B AS_PATH to 2B AS_PATH and send our AS_PATH
506: * as optional AS4_PATH attribute.
507: */
1.1.1.2 ! misho 508: if ((code == BA_AS_PATH) && !as4_session)
1.1 misho 509: {
510: len = a->u.ptr->length;
511:
512: if (remains < (len + 4))
513: goto err_no_buffer;
514:
515: /* Using temporary buffer because don't know a length of created attr
516: * and therefore a length of a header. Perhaps i should better always
517: * use BAF_EXT_LEN. */
518:
519: byte buf[len];
520: int new_used;
521: int nl = as_path_convert_to_old(a->u.ptr, buf, &new_used);
522:
523: DBG("BGP: Encoding old AS_PATH\n");
524: rv = bgp_encode_attr_hdr(w, BAF_TRANSITIVE, BA_AS_PATH, nl);
525: ADVANCE(w, remains, rv);
526: memcpy(w, buf, nl);
527: ADVANCE(w, remains, nl);
528:
529: if (! new_used)
530: continue;
531:
532: if (remains < (len + 4))
533: goto err_no_buffer;
534:
535: /* We should discard AS_CONFED_SEQUENCE or AS_CONFED_SET path segments
536: * here but we don't support confederations and such paths we already
537: * discarded in bgp_check_as_path().
538: */
539:
540: DBG("BGP: Encoding AS4_PATH\n");
541: rv = bgp_encode_attr_hdr(w, BAF_OPTIONAL | BAF_TRANSITIVE, BA_AS4_PATH, len);
542: ADVANCE(w, remains, rv);
543: memcpy(w, a->u.ptr->data, len);
544: ADVANCE(w, remains, len);
545:
546: continue;
547: }
548:
549: /* The same issue with AGGREGATOR attribute */
1.1.1.2 ! misho 550: if ((code == BA_AGGREGATOR) && !as4_session)
1.1 misho 551: {
552: int new_used;
553:
554: len = 6;
555: if (remains < (len + 3))
556: goto err_no_buffer;
557:
558: rv = bgp_encode_attr_hdr(w, BAF_OPTIONAL | BAF_TRANSITIVE, BA_AGGREGATOR, len);
559: ADVANCE(w, remains, rv);
560: aggregator_convert_to_old(a->u.ptr, w, &new_used);
561: ADVANCE(w, remains, len);
562:
563: if (! new_used)
564: continue;
565:
566: len = 8;
567: if (remains < (len + 3))
568: goto err_no_buffer;
569:
570: rv = bgp_encode_attr_hdr(w, BAF_OPTIONAL | BAF_TRANSITIVE, BA_AS4_AGGREGATOR, len);
571: ADVANCE(w, remains, rv);
572: memcpy(w, a->u.ptr->data, len);
573: ADVANCE(w, remains, len);
574:
575: continue;
576: }
577:
578: /* Standard path continues here ... */
579:
580: type = a->type & EAF_TYPE_MASK;
581: flags = a->flags & (BAF_OPTIONAL | BAF_TRANSITIVE | BAF_PARTIAL);
582: len = bgp_get_attr_len(a);
583:
584: /* Skip empty sets */
585: if (((type == EAF_TYPE_INT_SET) || (type == EAF_TYPE_EC_SET) || (type == EAF_TYPE_LC_SET)) && (len == 0))
586: continue;
587:
588: if (remains < len + 4)
589: goto err_no_buffer;
590:
591: rv = bgp_encode_attr_hdr(w, flags, code, len);
592: ADVANCE(w, remains, rv);
593:
594: switch (type)
595: {
596: case EAF_TYPE_INT:
597: case EAF_TYPE_ROUTER_ID:
598: if (len == 4)
599: put_u32(w, a->u.data);
600: else
601: *w = a->u.data;
602: break;
603: case EAF_TYPE_IP_ADDRESS:
604: {
605: ip_addr ip = *(ip_addr *)a->u.ptr->data;
606: ipa_hton(ip);
607: memcpy(w, &ip, len);
608: break;
609: }
610: case EAF_TYPE_INT_SET:
611: case EAF_TYPE_LC_SET:
612: case EAF_TYPE_EC_SET:
613: {
614: u32 *z = int_set_get_data(a->u.ptr);
615: int i;
616: for(i=0; i<len; i+=4)
617: put_u32(w+i, *z++);
618: break;
619: }
620: case EAF_TYPE_OPAQUE:
621: case EAF_TYPE_AS_PATH:
622: memcpy(w, a->u.ptr->data, len);
623: break;
624: default:
625: bug("bgp_encode_attrs: unknown attribute type %02x", a->type);
626: }
627: ADVANCE(w, remains, len);
628: }
629: return w - start;
630:
631: err_no_buffer:
632: return -1;
633: }
634:
635: /*
636: static void
637: bgp_init_prefix(struct fib_node *N)
638: {
639: struct bgp_prefix *p = (struct bgp_prefix *) N;
640: p->bucket_node.next = NULL;
641: }
642: */
643:
644: static int
645: bgp_compare_u32(const u32 *x, const u32 *y)
646: {
647: return (*x < *y) ? -1 : (*x > *y) ? 1 : 0;
648: }
649:
650: static inline void
651: bgp_normalize_int_set(u32 *dest, u32 *src, unsigned cnt)
652: {
653: memcpy(dest, src, sizeof(u32) * cnt);
654: qsort(dest, cnt, sizeof(u32), (int(*)(const void *, const void *)) bgp_compare_u32);
655: }
656:
657: static int
658: bgp_compare_ec(const u32 *xp, const u32 *yp)
659: {
660: u64 x = ec_get(xp, 0);
661: u64 y = ec_get(yp, 0);
662: return (x < y) ? -1 : (x > y) ? 1 : 0;
663: }
664:
665: static inline void
666: bgp_normalize_ec_set(struct adata *ad, u32 *src, int internal)
667: {
668: u32 *dst = int_set_get_data(ad);
669:
670: /* Remove non-transitive communities (EC_TBIT active) on external sessions */
671: if (! internal)
672: {
673: int len = int_set_get_size(ad);
674: u32 *t = dst;
675: int i;
676:
677: for (i=0; i < len; i += 2)
678: {
679: if (src[i] & EC_TBIT)
680: continue;
681:
682: *t++ = src[i];
683: *t++ = src[i+1];
684: }
685:
686: ad->length = (t - dst) * 4;
687: }
688: else
689: memcpy(dst, src, ad->length);
690:
691: qsort(dst, ad->length / 8, 8, (int(*)(const void *, const void *)) bgp_compare_ec);
692: }
693:
694: static int
695: bgp_compare_lc(const u32 *x, const u32 *y)
696: {
697: if (x[0] != y[0])
698: return (x[0] > y[0]) ? 1 : -1;
699: if (x[1] != y[1])
700: return (x[1] > y[1]) ? 1 : -1;
701: if (x[2] != y[2])
702: return (x[2] > y[2]) ? 1 : -1;
703: return 0;
704: }
705:
706: static inline void
707: bgp_normalize_lc_set(u32 *dest, u32 *src, unsigned cnt)
708: {
709: memcpy(dest, src, LCOMM_LENGTH * cnt);
710: qsort(dest, cnt, LCOMM_LENGTH, (int(*)(const void *, const void *)) bgp_compare_lc);
711: }
712:
713: static void
714: bgp_rehash_buckets(struct bgp_proto *p)
715: {
716: struct bgp_bucket **old = p->bucket_hash;
717: struct bgp_bucket **new;
718: unsigned oldn = p->hash_size;
719: unsigned i, e, mask;
720: struct bgp_bucket *b;
721:
722: p->hash_size = p->hash_limit;
723: DBG("BGP: Rehashing bucket table from %d to %d\n", oldn, p->hash_size);
724: p->hash_limit *= 4;
725: if (p->hash_limit >= 65536)
726: p->hash_limit = ~0;
727: new = p->bucket_hash = mb_allocz(p->p.pool, p->hash_size * sizeof(struct bgp_bucket *));
728: mask = p->hash_size - 1;
729: for (i=0; i<oldn; i++)
730: while (b = old[i])
731: {
732: old[i] = b->hash_next;
733: e = b->hash & mask;
734: b->hash_next = new[e];
735: if (b->hash_next)
736: b->hash_next->hash_prev = b;
737: b->hash_prev = NULL;
738: new[e] = b;
739: }
740: mb_free(old);
741: }
742:
743: static struct bgp_bucket *
744: bgp_new_bucket(struct bgp_proto *p, ea_list *new, unsigned hash)
745: {
746: struct bgp_bucket *b;
747: unsigned ea_size = sizeof(ea_list) + new->count * sizeof(eattr);
748: unsigned ea_size_aligned = BIRD_ALIGN(ea_size, CPU_STRUCT_ALIGN);
749: unsigned size = sizeof(struct bgp_bucket) + ea_size_aligned;
750: unsigned i;
751: byte *dest;
752: unsigned index = hash & (p->hash_size - 1);
753:
754: /* Gather total size of non-inline attributes */
755: for (i=0; i<new->count; i++)
756: {
757: eattr *a = &new->attrs[i];
758: if (!(a->type & EAF_EMBEDDED))
759: size += BIRD_ALIGN(sizeof(struct adata) + a->u.ptr->length, CPU_STRUCT_ALIGN);
760: }
761:
762: /* Create the bucket and hash it */
763: b = mb_alloc(p->p.pool, size);
764: b->hash_next = p->bucket_hash[index];
765: if (b->hash_next)
766: b->hash_next->hash_prev = b;
767: p->bucket_hash[index] = b;
768: b->hash_prev = NULL;
769: b->hash = hash;
770: add_tail(&p->bucket_queue, &b->send_node);
771: init_list(&b->prefixes);
772: memcpy(b->eattrs, new, ea_size);
773: dest = ((byte *)b->eattrs) + ea_size_aligned;
774:
775: /* Copy values of non-inline attributes */
776: for (i=0; i<new->count; i++)
777: {
778: eattr *a = &b->eattrs->attrs[i];
779: if (!(a->type & EAF_EMBEDDED))
780: {
781: struct adata *oa = a->u.ptr;
782: struct adata *na = (struct adata *) dest;
783: memcpy(na, oa, sizeof(struct adata) + oa->length);
784: a->u.ptr = na;
785: dest += BIRD_ALIGN(sizeof(struct adata) + na->length, CPU_STRUCT_ALIGN);
786: }
787: }
788:
789: /* If needed, rehash */
790: p->hash_count++;
791: if (p->hash_count > p->hash_limit)
792: bgp_rehash_buckets(p);
793:
794: return b;
795: }
796:
797: static struct bgp_bucket *
798: bgp_get_bucket(struct bgp_proto *p, net *n, ea_list *attrs, int originate)
799: {
800: ea_list *new;
801: unsigned i, cnt, hash, code;
802: eattr *a, *d;
803: u32 seen = 0;
804: struct bgp_bucket *b;
805:
806: /* Merge the attribute list */
807: new = alloca(ea_scan(attrs));
808: ea_merge(attrs, new);
809: ea_sort(new);
810:
811: /* Normalize attributes */
812: d = new->attrs;
813: cnt = new->count;
814: new->count = 0;
815: for(i=0; i<cnt; i++)
816: {
817: a = &new->attrs[i];
818: if (EA_PROTO(a->id) != EAP_BGP)
819: continue;
820: code = EA_ID(a->id);
821: if (ATTR_KNOWN(code))
822: {
1.1.1.2 ! misho 823: if (!p->is_internal)
! 824: {
! 825: if (!bgp_attr_table[code].allow_in_ebgp)
! 826: continue;
! 827: if ((code == BA_LOCAL_PREF) && !p->cf->allow_local_pref)
! 828: continue;
! 829: }
1.1 misho 830: /* The flags might have been zero if the attr was added by filters */
831: a->flags = (a->flags & BAF_PARTIAL) | bgp_attr_table[code].expected_flags;
832: if (code < 32)
833: seen |= 1 << code;
834: }
835: else
836: {
837: /* Don't re-export unknown non-transitive attributes */
838: if (!(a->flags & BAF_TRANSITIVE))
839: continue;
840: }
841: *d = *a;
842: if ((d->type & EAF_ORIGINATED) && !originate && (d->flags & BAF_TRANSITIVE) && (d->flags & BAF_OPTIONAL))
843: d->flags |= BAF_PARTIAL;
844: switch (d->type & EAF_TYPE_MASK)
845: {
846: case EAF_TYPE_INT_SET:
847: {
848: struct adata *z = alloca(sizeof(struct adata) + d->u.ptr->length);
849: z->length = d->u.ptr->length;
850: bgp_normalize_int_set((u32 *) z->data, (u32 *) d->u.ptr->data, z->length / 4);
851: d->u.ptr = z;
852: break;
853: }
854: case EAF_TYPE_EC_SET:
855: {
856: struct adata *z = alloca(sizeof(struct adata) + d->u.ptr->length);
857: z->length = d->u.ptr->length;
858: bgp_normalize_ec_set(z, (u32 *) d->u.ptr->data, p->is_internal);
859: d->u.ptr = z;
860: break;
861: }
862: case EAF_TYPE_LC_SET:
863: {
864: struct adata *z = alloca(sizeof(struct adata) + d->u.ptr->length);
865: z->length = d->u.ptr->length;
866: bgp_normalize_lc_set((u32 *) z->data, (u32 *) d->u.ptr->data, z->length / LCOMM_LENGTH);
867: d->u.ptr = z;
868: break;
869: }
870: default: ;
871: }
872: d++;
873: new->count++;
874: }
875:
876: /* Hash */
877: hash = ea_hash(new);
878: for(b=p->bucket_hash[hash & (p->hash_size - 1)]; b; b=b->hash_next)
879: if (b->hash == hash && ea_same(b->eattrs, new))
880: {
881: DBG("Found bucket.\n");
882: return b;
883: }
884:
885: /* Ensure that there are all mandatory attributes */
886: for(i=0; i<ARRAY_SIZE(bgp_mandatory_attrs); i++)
887: if (!(seen & (1 << bgp_mandatory_attrs[i])))
888: {
889: 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);
890: return NULL;
891: }
892:
893: /* Check if next hop is valid */
894: a = ea_find(new, EA_CODE(EAP_BGP, BA_NEXT_HOP));
895: if (!a || ipa_equal(p->cf->remote_ip, *(ip_addr *)a->u.ptr->data))
896: {
897: log(L_ERR "%s: Invalid NEXT_HOP attribute in route %I/%d", p->p.name, n->n.prefix, n->n.pxlen);
898: return NULL;
899: }
900:
901: /* Create new bucket */
902: DBG("Creating bucket.\n");
903: return bgp_new_bucket(p, new, hash);
904: }
905:
906: void
907: bgp_free_bucket(struct bgp_proto *p, struct bgp_bucket *buck)
908: {
909: if (buck->hash_next)
910: buck->hash_next->hash_prev = buck->hash_prev;
911: if (buck->hash_prev)
912: buck->hash_prev->hash_next = buck->hash_next;
913: else
914: p->bucket_hash[buck->hash & (p->hash_size-1)] = buck->hash_next;
915: mb_free(buck);
916: }
917:
918:
919: /* Prefix hash table */
920:
921: #define PXH_KEY(n1) n1->n.prefix, n1->n.pxlen, n1->path_id
922: #define PXH_NEXT(n) n->next
923: #define PXH_EQ(p1,l1,i1,p2,l2,i2) ipa_equal(p1, p2) && l1 == l2 && i1 == i2
924: #define PXH_FN(p,l,i) ipa_hash32(p) ^ u32_hash((l << 16) ^ i)
925:
926: #define PXH_REHASH bgp_pxh_rehash
1.1.1.2 ! misho 927: #define PXH_PARAMS /8, *2, 2, 2, 8, 24
1.1 misho 928:
929:
930: HASH_DEFINE_REHASH_FN(PXH, struct bgp_prefix)
931:
932: void
933: bgp_init_prefix_table(struct bgp_proto *p, u32 order)
934: {
935: HASH_INIT(p->prefix_hash, p->p.pool, order);
936:
937: p->prefix_slab = sl_new(p->p.pool, sizeof(struct bgp_prefix));
938: }
939:
940: void
941: bgp_free_prefix_table(struct bgp_proto *p)
942: {
943: HASH_FREE(p->prefix_hash);
944:
945: rfree(p->prefix_slab);
946: p->prefix_slab = NULL;
947: }
948:
949: static struct bgp_prefix *
950: bgp_get_prefix(struct bgp_proto *p, ip_addr prefix, int pxlen, u32 path_id)
951: {
952: struct bgp_prefix *bp = HASH_FIND(p->prefix_hash, PXH, prefix, pxlen, path_id);
953:
954: if (bp)
955: return bp;
956:
957: bp = sl_alloc(p->prefix_slab);
958: bp->n.prefix = prefix;
959: bp->n.pxlen = pxlen;
960: bp->path_id = path_id;
961: bp->bucket_node.next = NULL;
962:
963: HASH_INSERT2(p->prefix_hash, PXH, p->p.pool, bp);
964:
965: return bp;
966: }
967:
968: void
969: bgp_free_prefix(struct bgp_proto *p, struct bgp_prefix *bp)
970: {
971: HASH_REMOVE2(p->prefix_hash, PXH, p->p.pool, bp);
972: sl_free(p->prefix_slab, bp);
973: }
974:
975:
976: void
977: bgp_rt_notify(struct proto *P, rtable *tbl UNUSED, net *n, rte *new, rte *old UNUSED, ea_list *attrs)
978: {
979: struct bgp_proto *p = (struct bgp_proto *) P;
980: struct bgp_bucket *buck;
981: struct bgp_prefix *px;
982: rte *key;
983: u32 path_id;
984:
985: DBG("BGP: Got route %I/%d %s\n", n->n.prefix, n->n.pxlen, new ? "up" : "down");
986:
987: if (new)
988: {
989: key = new;
990: buck = bgp_get_bucket(p, n, attrs, new->attrs->source != RTS_BGP);
991: if (!buck) /* Inconsistent attribute list */
992: return;
993: }
994: else
995: {
996: key = old;
997: if (!(buck = p->withdraw_bucket))
998: {
999: buck = p->withdraw_bucket = mb_alloc(P->pool, sizeof(struct bgp_bucket));
1000: init_list(&buck->prefixes);
1001: }
1002: }
1003: path_id = p->add_path_tx ? key->attrs->src->global_id : 0;
1004: px = bgp_get_prefix(p, n->n.prefix, n->n.pxlen, path_id);
1005: if (px->bucket_node.next)
1006: {
1007: DBG("\tRemoving old entry.\n");
1008: rem_node(&px->bucket_node);
1009: }
1010: add_tail(&buck->prefixes, &px->bucket_node);
1011: bgp_schedule_packet(p->conn, PKT_UPDATE);
1012: }
1013:
1014: static int
1015: bgp_create_attrs(struct bgp_proto *p, rte *e, ea_list **attrs, struct linpool *pool)
1016: {
1017: ea_list *ea = lp_alloc(pool, sizeof(ea_list) + 4*sizeof(eattr));
1018: rta *rta = e->attrs;
1019: byte *z;
1020:
1021: ea->next = *attrs;
1022: *attrs = ea;
1023: ea->flags = EALF_SORTED;
1024: ea->count = 4;
1025:
1026: bgp_set_attr(ea->attrs, BA_ORIGIN,
1027: ((rta->source == RTS_OSPF_EXT1) || (rta->source == RTS_OSPF_EXT2)) ? ORIGIN_INCOMPLETE : ORIGIN_IGP);
1028:
1029: if (p->is_internal)
1030: bgp_set_attr_wa(ea->attrs+1, pool, BA_AS_PATH, 0);
1031: else
1032: {
1033: z = bgp_set_attr_wa(ea->attrs+1, pool, BA_AS_PATH, 6);
1034: z[0] = AS_PATH_SEQUENCE;
1035: z[1] = 1; /* 1 AS */
1036: put_u32(z+2, p->local_as);
1037: }
1038:
1039: /* iBGP -> use gw, eBGP multi-hop -> use source_addr,
1040: eBGP single-hop -> use gw if on the same iface */
1041: z = bgp_set_attr_wa(ea->attrs+2, pool, BA_NEXT_HOP, NEXT_HOP_LENGTH);
1042: if (p->cf->next_hop_self ||
1043: rta->dest != RTD_ROUTER ||
1044: ipa_equal(rta->gw, IPA_NONE) ||
1045: ipa_is_link_local(rta->gw) ||
1046: (!p->is_internal && !p->cf->next_hop_keep &&
1047: (!p->neigh || (rta->iface != p->neigh->iface))))
1048: set_next_hop(z, p->source_addr);
1049: else
1050: set_next_hop(z, rta->gw);
1051:
1052: bgp_set_attr(ea->attrs+3, BA_LOCAL_PREF, p->cf->default_local_pref);
1053:
1054: return 0; /* Leave decision to the filters */
1055: }
1056:
1057:
1058: static inline int
1059: bgp_as_path_loopy(struct bgp_proto *p, rta *a)
1060: {
1061: int num = p->cf->allow_local_as + 1;
1062: eattr *e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1063: return (e && (num > 0) && as_path_contains(e->u.ptr, p->local_as, num));
1064: }
1065:
1066: static inline int
1067: bgp_originator_id_loopy(struct bgp_proto *p, rta *a)
1068: {
1069: eattr *e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
1070: return (e && (e->u.data == p->local_id));
1071: }
1072:
1073: static inline int
1074: bgp_cluster_list_loopy(struct bgp_proto *p, rta *a)
1075: {
1076: eattr *e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
1077: return (e && p->rr_client && int_set_contains(e->u.ptr, p->rr_cluster_id));
1078: }
1079:
1080:
1081: static inline void
1082: bgp_path_prepend(rte *e, ea_list **attrs, struct linpool *pool, u32 as)
1083: {
1084: eattr *a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1085: bgp_attach_attr(attrs, pool, BA_AS_PATH, (uintptr_t) as_path_prepend(pool, a->u.ptr, as));
1086: }
1087:
1088: static inline void
1089: bgp_cluster_list_prepend(rte *e, ea_list **attrs, struct linpool *pool, u32 cid)
1090: {
1091: eattr *a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
1092: bgp_attach_attr(attrs, pool, BA_CLUSTER_LIST, (uintptr_t) int_set_prepend(pool, a ? a->u.ptr : NULL, cid));
1093: }
1094:
1095: static int
1096: bgp_update_attrs(struct bgp_proto *p, rte *e, ea_list **attrs, struct linpool *pool, int rr)
1097: {
1098: eattr *a;
1099:
1100: if (!p->is_internal && !p->rs_client)
1101: {
1102: bgp_path_prepend(e, attrs, pool, p->local_as);
1103:
1104: /* The MULTI_EXIT_DISC attribute received from a neighboring AS MUST NOT be
1105: * propagated to other neighboring ASes.
1106: * Perhaps it would be better to undefine it.
1107: */
1108: a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1109: if (a)
1110: bgp_attach_attr(attrs, pool, BA_MULTI_EXIT_DISC, 0);
1111: }
1112:
1113: /* iBGP -> keep next_hop, eBGP multi-hop -> use source_addr,
1114: * eBGP single-hop -> keep next_hop if on the same iface.
1115: * If the next_hop is zero (i.e. link-local), keep only if on the same iface.
1116: *
1117: * Note that same-iface-check uses iface from route, which is based on gw.
1118: */
1119: a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_NEXT_HOP));
1120: if (a && !p->cf->next_hop_self &&
1121: (p->cf->next_hop_keep ||
1122: (p->is_internal && ipa_nonzero(*((ip_addr *) a->u.ptr->data))) ||
1123: (p->neigh && (e->attrs->iface == p->neigh->iface))))
1124: {
1125: /* Leave the original next hop attribute, will check later where does it point */
1126: }
1127: else
1128: {
1129: /* Need to create new one */
1130: byte *b = bgp_attach_attr_wa(attrs, pool, BA_NEXT_HOP, NEXT_HOP_LENGTH);
1131: set_next_hop(b, p->source_addr);
1132: }
1133:
1134: if (rr)
1135: {
1136: /* Handling route reflection, RFC 4456 */
1137: struct bgp_proto *src = (struct bgp_proto *) e->attrs->src->proto;
1138:
1139: a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
1140: if (!a)
1141: bgp_attach_attr(attrs, pool, BA_ORIGINATOR_ID, src->remote_id);
1142:
1143: /* We attach proper cluster ID according to whether the route is entering or leaving the cluster */
1144: bgp_cluster_list_prepend(e, attrs, pool, src->rr_client ? src->rr_cluster_id : p->rr_cluster_id);
1145:
1146: /* Two RR clients with different cluster ID, hmmm */
1147: if (src->rr_client && p->rr_client && (src->rr_cluster_id != p->rr_cluster_id))
1148: bgp_cluster_list_prepend(e, attrs, pool, p->rr_cluster_id);
1149: }
1150:
1151: return 0; /* Leave decision to the filters */
1152: }
1153:
1154: static int
1155: bgp_community_filter(struct bgp_proto *p, rte *e)
1156: {
1157: eattr *a;
1158: struct adata *d;
1159:
1160: /* Check if we aren't forbidden to export the route by communities */
1161: a = ea_find(e->attrs->eattrs, EA_CODE(EAP_BGP, BA_COMMUNITY));
1162: if (a)
1163: {
1164: d = a->u.ptr;
1165: if (int_set_contains(d, BGP_COMM_NO_ADVERTISE))
1166: {
1167: DBG("\tNO_ADVERTISE\n");
1168: return 1;
1169: }
1170: if (!p->is_internal &&
1171: (int_set_contains(d, BGP_COMM_NO_EXPORT) ||
1172: int_set_contains(d, BGP_COMM_NO_EXPORT_SUBCONFED)))
1173: {
1174: DBG("\tNO_EXPORT\n");
1175: return 1;
1176: }
1.1.1.2 ! misho 1177:
! 1178: if (!p->conn->peer_llgr_aware && int_set_contains(d, BGP_COMM_LLGR_STALE))
! 1179: return 1;
1.1 misho 1180: }
1181:
1182: return 0;
1183: }
1184:
1185: int
1186: bgp_import_control(struct proto *P, rte **new, ea_list **attrs, struct linpool *pool)
1187: {
1188: rte *e = *new;
1189: struct bgp_proto *p = (struct bgp_proto *) P;
1190: struct bgp_proto *new_bgp = (e->attrs->src->proto->proto == &proto_bgp) ?
1191: (struct bgp_proto *) e->attrs->src->proto : NULL;
1192:
1193: if (p == new_bgp) /* Poison reverse updates */
1194: return -1;
1195: if (new_bgp)
1196: {
1197: /* We should check here for cluster list loop, because the receiving BGP instance
1198: might have different cluster ID */
1199: if (bgp_cluster_list_loopy(p, e->attrs))
1200: return -1;
1201:
1202: if (p->cf->interpret_communities && bgp_community_filter(p, e))
1203: return -1;
1204:
1205: if (p->local_as == new_bgp->local_as && p->is_internal && new_bgp->is_internal)
1206: {
1207: /* Redistribution of internal routes with IBGP */
1208: if (p->rr_client || new_bgp->rr_client)
1209: /* Route reflection, RFC 4456 */
1210: return bgp_update_attrs(p, e, attrs, pool, 1);
1211: else
1212: return -1;
1213: }
1214: else
1215: return bgp_update_attrs(p, e, attrs, pool, 0);
1216: }
1217: else
1218: return bgp_create_attrs(p, e, attrs, pool);
1219: }
1220:
1221: static inline u32
1222: bgp_get_neighbor(rte *r)
1223: {
1224: eattr *e = ea_find(r->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1225: u32 as;
1226:
1227: if (e && as_path_get_first(e->u.ptr, &as))
1228: return as;
1229: else
1230: return ((struct bgp_proto *) r->attrs->src->proto)->remote_as;
1231: }
1232:
1233: static inline int
1234: rte_resolvable(rte *rt)
1235: {
1236: int rd = rt->attrs->dest;
1237: return (rd == RTD_ROUTER) || (rd == RTD_DEVICE) || (rd == RTD_MULTIPATH);
1238: }
1239:
1.1.1.2 ! misho 1240: static inline int
! 1241: rte_stale(rte *r)
! 1242: {
! 1243: if (r->u.bgp.stale < 0)
! 1244: {
! 1245: /* If staleness is unknown, compute and cache it */
! 1246: eattr *a = ea_find(r->attrs->eattrs, EA_CODE(EAP_BGP, BA_COMMUNITY));
! 1247: r->u.bgp.stale = a && int_set_contains(a->u.ptr, BGP_COMM_LLGR_STALE);
! 1248: }
! 1249:
! 1250: return r->u.bgp.stale;
! 1251: }
! 1252:
1.1 misho 1253: int
1254: bgp_rte_better(rte *new, rte *old)
1255: {
1256: struct bgp_proto *new_bgp = (struct bgp_proto *) new->attrs->src->proto;
1257: struct bgp_proto *old_bgp = (struct bgp_proto *) old->attrs->src->proto;
1258: eattr *x, *y;
1259: u32 n, o;
1260:
1261: /* Skip suppressed routes (see bgp_rte_recalculate()) */
1262: n = new->u.bgp.suppressed;
1263: o = old->u.bgp.suppressed;
1264: if (n > o)
1265: return 0;
1266: if (n < o)
1267: return 1;
1268:
1269: /* RFC 4271 9.1.2.1. Route resolvability test */
1270: n = rte_resolvable(new);
1271: o = rte_resolvable(old);
1272: if (n > o)
1273: return 1;
1274: if (n < o)
1275: return 0;
1276:
1.1.1.2 ! misho 1277: /* LLGR draft - depreference stale routes */
! 1278: n = rte_stale(new);
! 1279: o = rte_stale(old);
! 1280: if (n > o)
! 1281: return 0;
! 1282: if (n < o)
! 1283: return 1;
! 1284:
1.1 misho 1285: /* Start with local preferences */
1286: x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1287: y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1288: n = x ? x->u.data : new_bgp->cf->default_local_pref;
1289: o = y ? y->u.data : old_bgp->cf->default_local_pref;
1290: if (n > o)
1291: return 1;
1292: if (n < o)
1293: return 0;
1294:
1295: /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1296: if (new_bgp->cf->compare_path_lengths || old_bgp->cf->compare_path_lengths)
1297: {
1298: x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1299: y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1300: n = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN;
1301: o = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN;
1302: if (n < o)
1303: return 1;
1304: if (n > o)
1305: return 0;
1306: }
1307:
1308: /* RFC 4271 9.1.2.2. b) Use origins */
1309: x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1310: y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1311: n = x ? x->u.data : ORIGIN_INCOMPLETE;
1312: o = y ? y->u.data : ORIGIN_INCOMPLETE;
1313: if (n < o)
1314: return 1;
1315: if (n > o)
1316: return 0;
1317:
1318: /* RFC 4271 9.1.2.2. c) Compare MED's */
1319: /* Proper RFC 4271 path selection cannot be interpreted as finding
1320: * the best path in some ordering. It is implemented partially in
1321: * bgp_rte_recalculate() when deterministic_med option is
1322: * active. Without that option, the behavior is just an
1323: * approximation, which in specific situations may lead to
1324: * persistent routing loops, because it is nondeterministic - it
1325: * depends on the order in which routes appeared. But it is also the
1326: * same behavior as used by default in Cisco routers, so it is
1327: * probably not a big issue.
1328: */
1329: if (new_bgp->cf->med_metric || old_bgp->cf->med_metric ||
1330: (bgp_get_neighbor(new) == bgp_get_neighbor(old)))
1331: {
1332: x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1333: y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1334: n = x ? x->u.data : new_bgp->cf->default_med;
1335: o = y ? y->u.data : old_bgp->cf->default_med;
1336: if (n < o)
1337: return 1;
1338: if (n > o)
1339: return 0;
1340: }
1341:
1342: /* RFC 4271 9.1.2.2. d) Prefer external peers */
1343: if (new_bgp->is_internal > old_bgp->is_internal)
1344: return 0;
1345: if (new_bgp->is_internal < old_bgp->is_internal)
1346: return 1;
1347:
1348: /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1349: n = new_bgp->cf->igp_metric ? new->attrs->igp_metric : 0;
1350: o = old_bgp->cf->igp_metric ? old->attrs->igp_metric : 0;
1351: if (n < o)
1352: return 1;
1353: if (n > o)
1354: return 0;
1355:
1356: /* RFC 4271 9.1.2.2. f) Compare BGP identifiers */
1357: /* RFC 4456 9. a) Use ORIGINATOR_ID instead of local neighor ID */
1358: x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
1359: y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGINATOR_ID));
1360: n = x ? x->u.data : new_bgp->remote_id;
1361: o = y ? y->u.data : old_bgp->remote_id;
1362:
1363: /* RFC 5004 - prefer older routes */
1364: /* (if both are external and from different peer) */
1365: if ((new_bgp->cf->prefer_older || old_bgp->cf->prefer_older) &&
1366: !new_bgp->is_internal && n != o)
1367: return 0;
1368:
1369: /* rest of RFC 4271 9.1.2.2. f) */
1370: if (n < o)
1371: return 1;
1372: if (n > o)
1373: return 0;
1374:
1375: /* RFC 4456 9. b) Compare cluster list lengths */
1376: x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
1377: y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_CLUSTER_LIST));
1378: n = x ? int_set_get_size(x->u.ptr) : 0;
1379: o = y ? int_set_get_size(y->u.ptr) : 0;
1380: if (n < o)
1381: return 1;
1382: if (n > o)
1383: return 0;
1384:
1385: /* RFC 4271 9.1.2.2. g) Compare peer IP adresses */
1386: return (ipa_compare(new_bgp->cf->remote_ip, old_bgp->cf->remote_ip) < 0);
1387: }
1388:
1389:
1390: int
1391: bgp_rte_mergable(rte *pri, rte *sec)
1392: {
1393: struct bgp_proto *pri_bgp = (struct bgp_proto *) pri->attrs->src->proto;
1394: struct bgp_proto *sec_bgp = (struct bgp_proto *) sec->attrs->src->proto;
1395: eattr *x, *y;
1396: u32 p, s;
1397:
1398: /* Skip suppressed routes (see bgp_rte_recalculate()) */
1399: if (pri->u.bgp.suppressed != sec->u.bgp.suppressed)
1400: return 0;
1401:
1402: /* RFC 4271 9.1.2.1. Route resolvability test */
1403: if (!rte_resolvable(sec))
1404: return 0;
1405:
1.1.1.2 ! misho 1406: /* LLGR draft - depreference stale routes */
! 1407: if (rte_stale(pri) != rte_stale(sec))
! 1408: return 0;
! 1409:
1.1 misho 1410: /* Start with local preferences */
1411: x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1412: y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
1413: p = x ? x->u.data : pri_bgp->cf->default_local_pref;
1414: s = y ? y->u.data : sec_bgp->cf->default_local_pref;
1415: if (p != s)
1416: return 0;
1417:
1418: /* RFC 4271 9.1.2.2. a) Use AS path lengths */
1419: if (pri_bgp->cf->compare_path_lengths || sec_bgp->cf->compare_path_lengths)
1420: {
1421: x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1422: y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1423: p = x ? as_path_getlen(x->u.ptr) : AS_PATH_MAXLEN;
1424: s = y ? as_path_getlen(y->u.ptr) : AS_PATH_MAXLEN;
1425:
1426: if (p != s)
1427: return 0;
1428:
1429: // if (DELTA(p, s) > pri_bgp->cf->relax_multipath)
1430: // return 0;
1431: }
1432:
1433: /* RFC 4271 9.1.2.2. b) Use origins */
1434: x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1435: y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
1436: p = x ? x->u.data : ORIGIN_INCOMPLETE;
1437: s = y ? y->u.data : ORIGIN_INCOMPLETE;
1438: if (p != s)
1439: return 0;
1440:
1441: /* RFC 4271 9.1.2.2. c) Compare MED's */
1442: if (pri_bgp->cf->med_metric || sec_bgp->cf->med_metric ||
1443: (bgp_get_neighbor(pri) == bgp_get_neighbor(sec)))
1444: {
1445: x = ea_find(pri->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1446: y = ea_find(sec->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
1447: p = x ? x->u.data : pri_bgp->cf->default_med;
1448: s = y ? y->u.data : sec_bgp->cf->default_med;
1449: if (p != s)
1450: return 0;
1451: }
1452:
1453: /* RFC 4271 9.1.2.2. d) Prefer external peers */
1454: if (pri_bgp->is_internal != sec_bgp->is_internal)
1455: return 0;
1456:
1457: /* RFC 4271 9.1.2.2. e) Compare IGP metrics */
1458: p = pri_bgp->cf->igp_metric ? pri->attrs->igp_metric : 0;
1459: s = sec_bgp->cf->igp_metric ? sec->attrs->igp_metric : 0;
1460: if (p != s)
1461: return 0;
1462:
1463: /* Remaining criteria are ignored */
1464:
1465: return 1;
1466: }
1467:
1468:
1469:
1470: static inline int
1471: same_group(rte *r, u32 lpref, u32 lasn)
1472: {
1473: return (r->pref == lpref) && (bgp_get_neighbor(r) == lasn);
1474: }
1475:
1476: static inline int
1477: use_deterministic_med(rte *r)
1478: {
1479: struct proto *P = r->attrs->src->proto;
1480: return (P->proto == &proto_bgp) && ((struct bgp_proto *) P)->cf->deterministic_med;
1481: }
1482:
1483: int
1484: bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best)
1485: {
1486: rte *r, *s;
1487: rte *key = new ? new : old;
1488: u32 lpref = key->pref;
1489: u32 lasn = bgp_get_neighbor(key);
1490: int old_is_group_best = 0;
1491:
1492: /*
1493: * Proper RFC 4271 path selection is a bit complicated, it cannot be
1494: * implemented just by rte_better(), because it is not a linear
1495: * ordering. But it can be splitted to two levels, where the lower
1496: * level chooses the best routes in each group of routes from the
1497: * same neighboring AS and higher level chooses the best route (with
1498: * a slightly different ordering) between the best-in-group routes.
1499: *
1500: * When deterministic_med is disabled, we just ignore this issue and
1501: * choose the best route by bgp_rte_better() alone. If enabled, the
1502: * lower level of the route selection is done here (for the group
1503: * to which the changed route belongs), all routes in group are
1504: * marked as suppressed, just chosen best-in-group is not.
1505: *
1506: * Global best route selection then implements higher level by
1507: * choosing between non-suppressed routes (as they are always
1508: * preferred over suppressed routes). Routes from BGP protocols
1509: * that do not set deterministic_med are just never suppressed. As
1510: * they do not participate in the lower level selection, it is OK
1511: * that this fn is not called for them.
1512: *
1513: * The idea is simple, the implementation is more problematic,
1514: * mostly because of optimizations in rte_recalculate() that
1515: * avoids full recalculation in most cases.
1516: *
1517: * We can assume that at least one of new, old is non-NULL and both
1518: * are from the same protocol with enabled deterministic_med. We
1519: * group routes by both neighbor AS (lasn) and preference (lpref),
1520: * because bgp_rte_better() does not handle preference itself.
1521: */
1522:
1523: /* If new and old are from different groups, we just process that
1524: as two independent events */
1525: if (new && old && !same_group(old, lpref, lasn))
1526: {
1527: int i1, i2;
1528: i1 = bgp_rte_recalculate(table, net, NULL, old, old_best);
1529: i2 = bgp_rte_recalculate(table, net, new, NULL, old_best);
1530: return i1 || i2;
1531: }
1532:
1533: /*
1534: * We could find the best-in-group and then make some shortcuts like
1535: * in rte_recalculate, but as we would have to walk through all
1536: * net->routes just to find it, it is probably not worth. So we
1537: * just have two simpler fast cases that use just the old route.
1538: * We also set suppressed flag to avoid using it in bgp_rte_better().
1539: */
1540:
1541: if (new)
1542: new->u.bgp.suppressed = 1;
1543:
1544: if (old)
1545: {
1546: old_is_group_best = !old->u.bgp.suppressed;
1547: old->u.bgp.suppressed = 1;
1548: int new_is_better = new && bgp_rte_better(new, old);
1549:
1550: /* The first case - replace not best with worse (or remove not best) */
1551: if (!old_is_group_best && !new_is_better)
1552: return 0;
1553:
1554: /* The second case - replace the best with better */
1555: if (old_is_group_best && new_is_better)
1556: {
1557: /* new is best-in-group, the see discussion below - this is
1558: a special variant of NBG && OBG. From OBG we can deduce
1559: that same_group(old_best) iff (old == old_best) */
1560: new->u.bgp.suppressed = 0;
1561: return (old == old_best);
1562: }
1563: }
1564:
1565: /* The default case - find a new best-in-group route */
1566: r = new; /* new may not be in the list */
1567: for (s=net->routes; rte_is_valid(s); s=s->next)
1568: if (use_deterministic_med(s) && same_group(s, lpref, lasn))
1569: {
1570: s->u.bgp.suppressed = 1;
1571: if (!r || bgp_rte_better(s, r))
1572: r = s;
1573: }
1574:
1575: /* Simple case - the last route in group disappears */
1576: if (!r)
1577: return 0;
1578:
1579: /* Found best-in-group */
1580: r->u.bgp.suppressed = 0;
1581:
1582: /*
1583: * There are generally two reasons why we have to force
1584: * recalculation (return 1): First, the new route may be wrongfully
1585: * chosen to be the best in the first case check in
1586: * rte_recalculate(), this may happen only if old_best is from the
1587: * same group. Second, another (different than new route)
1588: * best-in-group is chosen and that may be the proper best (although
1589: * rte_recalculate() without ignore that possibility).
1590: *
1591: * There are three possible cases according to whether the old route
1592: * was the best in group (OBG, stored in old_is_group_best) and
1593: * whether the new route is the best in group (NBG, tested by r == new).
1594: * These cases work even if old or new is NULL.
1595: *
1596: * NBG -> new is a possible candidate for the best route, so we just
1597: * check for the first reason using same_group().
1598: *
1599: * !NBG && OBG -> Second reason applies, return 1
1600: *
1601: * !NBG && !OBG -> Best in group does not change, old != old_best,
1602: * rte_better(new, old_best) is false and therefore
1603: * the first reason does not apply, return 0
1604: */
1605:
1606: if (r == new)
1607: return old_best && same_group(old_best, lpref, lasn);
1608: else
1609: return old_is_group_best;
1610: }
1611:
1.1.1.2 ! misho 1612: struct rte *
! 1613: bgp_rte_modify_stale(struct rte *r, struct linpool *pool)
! 1614: {
! 1615: eattr *a = ea_find(r->attrs->eattrs, EA_CODE(EAP_BGP, BA_COMMUNITY));
! 1616: struct adata *ad = a ? a->u.ptr : NULL;
! 1617:
! 1618: if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR))
! 1619: return NULL;
! 1620:
! 1621: if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE))
! 1622: return r;
! 1623:
! 1624: r = rte_cow_rta(r, pool);
! 1625: bgp_attach_attr(&(r->attrs->eattrs), pool, BA_COMMUNITY,
! 1626: (uintptr_t) int_set_add(pool, ad, BGP_COMM_LLGR_STALE));
! 1627: r->u.bgp.stale = 1;
! 1628:
! 1629: return r;
! 1630: }
! 1631:
! 1632:
1.1 misho 1633: static struct adata *
1634: bgp_aggregator_convert_to_new(struct adata *old, struct linpool *pool)
1635: {
1636: struct adata *newa = lp_alloc(pool, sizeof(struct adata) + 8);
1637: newa->length = 8;
1638: aggregator_convert_to_new(old, newa->data);
1639: return newa;
1640: }
1641:
1642: /* Take last req_as ASNs from path old2 (in 2B format), convert to 4B format
1643: * and append path old4 (in 4B format).
1644: */
1645: static struct adata *
1646: bgp_merge_as_paths(struct adata *old2, struct adata *old4, int req_as, struct linpool *pool)
1647: {
1648: byte buf[old2->length * 2];
1649:
1650: int ol = as_path_convert_to_new(old2, buf, req_as);
1651: int nl = ol + (old4 ? old4->length : 0);
1652:
1653: struct adata *newa = lp_alloc(pool, sizeof(struct adata) + nl);
1654: newa->length = nl;
1655: memcpy(newa->data, buf, ol);
1656: if (old4) memcpy(newa->data + ol, old4->data, old4->length);
1657:
1658: return newa;
1659: }
1660:
1661: static int
1662: as4_aggregator_valid(struct adata *aggr)
1663: {
1664: return aggr->length == 8;
1665: }
1666:
1667:
1668: /* Reconstruct 4B AS_PATH and AGGREGATOR according to RFC 4893 4.2.3 */
1669: static void
1670: bgp_reconstruct_4b_atts(struct bgp_proto *p, rta *a, struct linpool *pool)
1671: {
1672: eattr *p2 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
1673: eattr *p4 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS4_PATH));
1674: eattr *a2 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AGGREGATOR));
1675: eattr *a4 =ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS4_AGGREGATOR));
1676: int a4_removed = 0;
1677:
1678: if (a4 && !as4_aggregator_valid(a4->u.ptr))
1679: {
1680: log(L_WARN "%s: AS4_AGGREGATOR attribute is invalid, skipping attribute", p->p.name);
1681: a4 = NULL;
1682: a4_removed = 1;
1683: }
1684:
1685: if (a2)
1686: {
1687: u32 a2_as = get_u16(a2->u.ptr->data);
1688:
1689: if (a4)
1690: {
1691: if (a2_as != AS_TRANS)
1692: {
1693: /* Routes were aggregated by old router and therefore AS4_PATH
1694: * and AS4_AGGREGATOR is invalid
1695: *
1696: * Convert AS_PATH and AGGREGATOR to 4B format and finish.
1697: */
1698:
1699: a2->u.ptr = bgp_aggregator_convert_to_new(a2->u.ptr, pool);
1700: p2->u.ptr = bgp_merge_as_paths(p2->u.ptr, NULL, AS_PATH_MAXLEN, pool);
1701:
1702: return;
1703: }
1704: else
1705: {
1706: /* Common case, use AS4_AGGREGATOR attribute */
1707: a2->u.ptr = a4->u.ptr;
1708: }
1709: }
1710: else
1711: {
1712: /* Common case, use old AGGREGATOR attribute */
1713: a2->u.ptr = bgp_aggregator_convert_to_new(a2->u.ptr, pool);
1714:
1715: if ((a2_as == AS_TRANS) && !a4_removed)
1716: log(L_WARN "%s: AGGREGATOR attribute contain AS_TRANS, but AS4_AGGREGATOR is missing", p->p.name);
1717: }
1718: }
1719: else
1720: if (a4)
1721: log(L_WARN "%s: AS4_AGGREGATOR attribute received, but AGGREGATOR attribute is missing", p->p.name);
1722:
1723: int p2_len = as_path_getlen_int(p2->u.ptr, 2);
1724: int p4_len = p4 ? validate_as4_path(p, p4->u.ptr) : -1;
1725:
1726: if (p4 && (p4_len < 0))
1727: log(L_WARN "%s: AS4_PATH attribute is malformed, skipping attribute", p->p.name);
1728:
1729: if ((p4_len <= 0) || (p2_len < p4_len))
1730: p2->u.ptr = bgp_merge_as_paths(p2->u.ptr, NULL, AS_PATH_MAXLEN, pool);
1731: else
1732: p2->u.ptr = bgp_merge_as_paths(p2->u.ptr, p4->u.ptr, p2_len - p4_len, pool);
1733: }
1734:
1735: static void
1736: bgp_remove_as4_attrs(struct bgp_proto *p, rta *a)
1737: {
1738: unsigned id1 = EA_CODE(EAP_BGP, BA_AS4_PATH);
1739: unsigned id2 = EA_CODE(EAP_BGP, BA_AS4_AGGREGATOR);
1740: ea_list **el = &(a->eattrs);
1741:
1742: /* We know that ea_lists constructed in bgp_decode attrs have one attribute per ea_list struct */
1743: while (*el != NULL)
1744: {
1745: unsigned fid = (*el)->attrs[0].id;
1746:
1747: if ((fid == id1) || (fid == id2))
1748: {
1749: *el = (*el)->next;
1750: if (p->as4_session)
1751: log(L_WARN "%s: Unexpected AS4_* attributes received", p->p.name);
1752: }
1753: else
1754: el = &((*el)->next);
1755: }
1756: }
1757:
1758: /**
1759: * bgp_decode_attrs - check and decode BGP attributes
1760: * @conn: connection
1761: * @attr: start of attribute block
1762: * @len: length of attribute block
1763: * @pool: linear pool to make all the allocations in
1764: * @mandatory: 1 iff presence of mandatory attributes has to be checked
1765: *
1766: * This function takes a BGP attribute block (a part of an Update message), checks
1767: * its consistency and converts it to a list of BIRD route attributes represented
1768: * by a &rta.
1769: */
1770: struct rta *
1771: bgp_decode_attrs(struct bgp_conn *conn, byte *attr, uint len, struct linpool *pool, int mandatory)
1772: {
1773: struct bgp_proto *bgp = conn->bgp;
1774: rta *a = lp_alloc(pool, sizeof(struct rta));
1775: uint flags, code, l, i, type;
1776: int errcode;
1777: byte *z, *attr_start;
1778: byte seen[256/8];
1779: ea_list *ea;
1780: struct adata *ad;
1781: int withdraw = 0;
1782:
1783: bzero(a, sizeof(rta));
1784: a->source = RTS_BGP;
1785: a->scope = SCOPE_UNIVERSE;
1786: a->cast = RTC_UNICAST;
1787: /* a->dest = RTD_ROUTER; -- set in bgp_set_next_hop() */
1788: a->from = bgp->cf->remote_ip;
1789:
1790: /* Parse the attributes */
1791: bzero(seen, sizeof(seen));
1792: DBG("BGP: Parsing attributes\n");
1793: while (len)
1794: {
1795: if (len < 2)
1796: goto malformed;
1797: attr_start = attr;
1798: flags = *attr++;
1799: code = *attr++;
1800: len -= 2;
1801: if (flags & BAF_EXT_LEN)
1802: {
1803: if (len < 2)
1804: goto malformed;
1805: l = get_u16(attr);
1806: attr += 2;
1807: len -= 2;
1808: }
1809: else
1810: {
1811: if (len < 1)
1812: goto malformed;
1813: l = *attr++;
1814: len--;
1815: }
1816: if (l > len)
1817: goto malformed;
1818: len -= l;
1819: z = attr;
1820: attr += l;
1821: DBG("Attr %02x %02x %d\n", code, flags, l);
1822: if (seen[code/8] & (1 << (code%8)))
1823: goto malformed;
1824: if (ATTR_KNOWN(code))
1825: {
1826: struct attr_desc *desc = &bgp_attr_table[code];
1827: if (desc->expected_length >= 0 && desc->expected_length != (int) l)
1828: { errcode = 5; goto err; }
1829: if ((desc->expected_flags ^ flags) & (BAF_OPTIONAL | BAF_TRANSITIVE))
1830: { errcode = 4; goto err; }
1.1.1.2 ! misho 1831: if (!bgp->is_internal)
! 1832: {
! 1833: if (!desc->allow_in_ebgp)
! 1834: continue;
! 1835: if ((code == BA_LOCAL_PREF) && !bgp->cf->allow_local_pref)
! 1836: continue;
! 1837: }
1.1 misho 1838: if (desc->validate)
1839: {
1840: errcode = desc->validate(bgp, z, l);
1841: if (errcode > 0)
1842: goto err;
1843: if (errcode == IGNORE)
1844: continue;
1845: if (errcode <= WITHDRAW)
1846: {
1847: log(L_WARN "%s: Attribute %s is malformed, withdrawing update",
1848: bgp->p.name, desc->name);
1849: withdraw = 1;
1850: }
1851: }
1852: else if (code == BA_AS_PATH)
1853: {
1854: /* Special case as it might also trim the attribute */
1855: if (validate_as_path(bgp, z, &l) < 0)
1856: { errcode = 11; goto err; }
1857: }
1858: type = desc->type;
1859: }
1860: else /* Unknown attribute */
1861: {
1862: if (!(flags & BAF_OPTIONAL))
1863: { errcode = 2; goto err; }
1864: type = EAF_TYPE_OPAQUE;
1865: }
1866:
1867: // Only OPTIONAL and TRANSITIVE attributes may have non-zero PARTIAL flag
1868: // if (!((flags & BAF_OPTIONAL) && (flags & BAF_TRANSITIVE)) && (flags & BAF_PARTIAL))
1869: // { errcode = 4; goto err; }
1870:
1871: seen[code/8] |= (1 << (code%8));
1872: ea = lp_alloc(pool, sizeof(ea_list) + sizeof(eattr));
1873: ea->next = a->eattrs;
1874: a->eattrs = ea;
1875: ea->flags = 0;
1876: ea->count = 1;
1877: ea->attrs[0].id = EA_CODE(EAP_BGP, code);
1878: ea->attrs[0].flags = flags;
1879: ea->attrs[0].type = type;
1880: if (type & EAF_EMBEDDED)
1881: ad = NULL;
1882: else
1883: {
1884: ad = lp_alloc(pool, sizeof(struct adata) + l);
1885: ea->attrs[0].u.ptr = ad;
1886: ad->length = l;
1887: memcpy(ad->data, z, l);
1888: }
1889: switch (type)
1890: {
1891: case EAF_TYPE_ROUTER_ID:
1892: case EAF_TYPE_INT:
1893: if (l == 1)
1894: ea->attrs[0].u.data = *z;
1895: else
1896: ea->attrs[0].u.data = get_u32(z);
1897: break;
1898: case EAF_TYPE_IP_ADDRESS:
1899: ipa_ntoh(*(ip_addr *)ad->data);
1900: break;
1901: case EAF_TYPE_INT_SET:
1902: case EAF_TYPE_LC_SET:
1903: case EAF_TYPE_EC_SET:
1904: {
1905: u32 *z = (u32 *) ad->data;
1906: for(i=0; i<ad->length/4; i++)
1907: z[i] = ntohl(z[i]);
1908: break;
1909: }
1910: }
1911: }
1912:
1913: if (withdraw)
1914: goto withdraw;
1915:
1916: #ifdef IPV6
1917: /* If we received MP_REACH_NLRI we should check mandatory attributes */
1918: if (bgp->mp_reach_len != 0)
1919: mandatory = 1;
1920: #endif
1921:
1922: /* If there is no (reachability) NLRI, we should exit now */
1923: if (! mandatory)
1924: return a;
1925:
1926: /* Check if all mandatory attributes are present */
1927: for(i=0; i < ARRAY_SIZE(bgp_mandatory_attrs); i++)
1928: {
1929: code = bgp_mandatory_attrs[i];
1930: if (!(seen[code/8] & (1 << (code%8))))
1931: {
1932: bgp_error(conn, 3, 3, &bgp_mandatory_attrs[i], 1);
1933: return NULL;
1934: }
1935: }
1936:
1937: /* When receiving attributes from non-AS4-aware BGP speaker,
1938: * we have to reconstruct 4B AS_PATH and AGGREGATOR attributes
1939: */
1940: if (! bgp->as4_session)
1941: bgp_reconstruct_4b_atts(bgp, a, pool);
1942:
1943: bgp_remove_as4_attrs(bgp, a);
1944:
1945: /* If the AS path attribute contains our AS, reject the routes */
1946: if (bgp_as_path_loopy(bgp, a))
1947: goto withdraw;
1948:
1949: /* Two checks for IBGP loops caused by route reflection, RFC 4456 */
1950: if (bgp_originator_id_loopy(bgp, a) ||
1951: bgp_cluster_list_loopy(bgp, a))
1952: goto withdraw;
1953:
1954: /* If there's no local preference, define one */
1955: if (!(seen[0] & (1 << BA_LOCAL_PREF)))
1956: bgp_attach_attr(&a->eattrs, pool, BA_LOCAL_PREF, bgp->cf->default_local_pref);
1957:
1958: return a;
1959:
1960: withdraw:
1961: return NULL;
1962:
1963: malformed:
1964: bgp_error(conn, 3, 1, NULL, 0);
1965: return NULL;
1966:
1967: err:
1968: bgp_error(conn, 3, errcode, attr_start, z+l-attr_start);
1969: return NULL;
1970: }
1971:
1972: int
1973: bgp_get_attr(eattr *a, byte *buf, int buflen)
1974: {
1975: uint i = EA_ID(a->id);
1976: struct attr_desc *d;
1977: int len;
1978:
1979: if (ATTR_KNOWN(i))
1980: {
1981: d = &bgp_attr_table[i];
1982: len = bsprintf(buf, "%s", d->name);
1983: buf += len;
1984: if (d->format)
1985: {
1986: *buf++ = ':';
1987: *buf++ = ' ';
1988: d->format(a, buf, buflen - len - 2);
1989: return GA_FULL;
1990: }
1991: return GA_NAME;
1992: }
1993: bsprintf(buf, "%02x%s", i, (a->flags & BAF_TRANSITIVE) ? " [t]" : "");
1994: return GA_NAME;
1995: }
1996:
1997: void
1998: bgp_init_bucket_table(struct bgp_proto *p)
1999: {
2000: p->hash_size = 256;
2001: p->hash_limit = p->hash_size * 4;
2002: p->bucket_hash = mb_allocz(p->p.pool, p->hash_size * sizeof(struct bgp_bucket *));
2003: init_list(&p->bucket_queue);
2004: p->withdraw_bucket = NULL;
2005: // fib_init(&p->prefix_fib, p->p.pool, sizeof(struct bgp_prefix), 0, bgp_init_prefix);
2006: }
2007:
2008: void
2009: bgp_free_bucket_table(struct bgp_proto *p)
2010: {
2011: mb_free(p->bucket_hash);
2012: p->bucket_hash = NULL;
2013:
2014: struct bgp_bucket *b;
2015: WALK_LIST_FIRST(b, p->bucket_queue)
2016: {
2017: rem_node(&b->send_node);
2018: mb_free(b);
2019: }
2020:
2021: mb_free(p->withdraw_bucket);
2022: p->withdraw_bucket = NULL;
2023: }
2024:
2025: void
2026: bgp_get_route_info(rte *e, byte *buf, ea_list *attrs)
2027: {
2028: eattr *p = ea_find(attrs, EA_CODE(EAP_BGP, BA_AS_PATH));
2029: eattr *o = ea_find(attrs, EA_CODE(EAP_BGP, BA_ORIGIN));
2030: u32 origas;
2031:
2032: buf += bsprintf(buf, " (%d", e->pref);
2033:
2034: if (e->u.bgp.suppressed)
2035: buf += bsprintf(buf, "-");
2036:
1.1.1.2 ! misho 2037: if (rte_stale(e))
! 2038: buf += bsprintf(buf, "s");
! 2039:
1.1 misho 2040: if (e->attrs->hostentry)
2041: {
2042: if (!rte_resolvable(e))
2043: buf += bsprintf(buf, "/-");
2044: else if (e->attrs->igp_metric >= IGP_METRIC_UNKNOWN)
2045: buf += bsprintf(buf, "/?");
2046: else
2047: buf += bsprintf(buf, "/%d", e->attrs->igp_metric);
2048: }
2049: buf += bsprintf(buf, ") [");
2050:
2051: if (p && as_path_get_last(p->u.ptr, &origas))
2052: buf += bsprintf(buf, "AS%u", origas);
2053: if (o)
2054: buf += bsprintf(buf, "%c", "ie?"[o->u.data]);
2055: strcpy(buf, "]");
2056: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to attrs.c CVS log ![[TXT]](/icons/cvsweb/text.gif){kind=icon}
![[DIR]](/icons/cvsweb/dir.gif){kind=icon}
Up to [ELWIX - Embedded LightWeight unIX -] / embedaddon / bird / proto / bgp