bgpd/bgp_mpath.c

   1 /*
   2  * BGP Multipath
   3  * Copyright (C) 2010 Google Inc.
   4  *
   5  * This file is part of Quagga
   6  *
   7  * Quagga is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License as published by the
   9  * Free Software Foundation; either version 2, or (at your option) any
  10  * later version.
  11  *
  12  * Quagga is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along
  18  * with this program; see the file COPYING; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include <zebra.h>
  23
  24 #include "command.h"
  25 #include "prefix.h"
  26 #include "linklist.h"
  27 #include "sockunion.h"
  28 #include "memory.h"
  29 #include "queue.h"
  30 #include "filter.h"
  31
  32 #include "bgpd/bgpd.h"
  33 #include "bgpd/bgp_table.h"
  34 #include "bgpd/bgp_route.h"
  35 #include "bgpd/bgp_attr.h"
  36 #include "bgpd/bgp_debug.h"
  37 #include "bgpd/bgp_aspath.h"
  38 #include "bgpd/bgp_community.h"
  39 #include "bgpd/bgp_ecommunity.h"
  40 #include "bgpd/bgp_lcommunity.h"
  41 #include "bgpd/bgp_mpath.h"
  42
  43 /*
  44  * bgp_maximum_paths_set
  45  *
  46  * Record maximum-paths configuration for BGP instance
  47  */
  48 int bgp_maximum_paths_set(struct bgp *bgp, afi_t afi, safi_t safi, int peertype,
  49                           u_int16_t maxpaths, u_int16_t options)
  50 {
  51         if (!bgp || (afi >= AFI_MAX) || (safi >= SAFI_MAX))
  52                 return -1;
  53
  54         switch (peertype) {
  55         case BGP_PEER_IBGP:
  56                 bgp->maxpaths[afi][safi].maxpaths_ibgp = maxpaths;
  57                 bgp->maxpaths[afi][safi].ibgp_flags |= options;
  58                 break;
  59         case BGP_PEER_EBGP:
  60                 bgp->maxpaths[afi][safi].maxpaths_ebgp = maxpaths;
  61                 break;
  62         default:
  63                 return -1;
  64         }
  65
  66         return 0;
  67 }
  68
  69 /*
  70  * bgp_maximum_paths_unset
  71  *
  72  * Remove maximum-paths configuration from BGP instance
  73  */
  74 int bgp_maximum_paths_unset(struct bgp *bgp, afi_t afi, safi_t safi,
  75                             int peertype)
  76 {
  77         if (!bgp || (afi >= AFI_MAX) || (safi >= SAFI_MAX))
  78                 return -1;
  79
  80         switch (peertype) {
  81         case BGP_PEER_IBGP:
  82                 bgp->maxpaths[afi][safi].maxpaths_ibgp = multipath_num;
  83                 bgp->maxpaths[afi][safi].ibgp_flags = 0;
  84                 break;
  85         case BGP_PEER_EBGP:
  86                 bgp->maxpaths[afi][safi].maxpaths_ebgp = multipath_num;
  87                 break;
  88         default:
  89                 return -1;
  90         }
  91
  92         return 0;
  93 }
  94
  95 /*
  96  * bgp_interface_same
  97  *
  98  * Return true if ifindex for ifp1 and ifp2 are the same, else return false.
  99  */
 100 static int bgp_interface_same(struct interface *ifp1, struct interface *ifp2)
 101 {
 102         if (!ifp1 && !ifp2)
 103                 return 1;
 104
 105         if (!ifp1 && ifp2)
 106                 return 0;
 107
 108         if (ifp1 && !ifp2)
 109                 return 0;
 110
 111         return (ifp1->ifindex == ifp2->ifindex);
 112 }
 113
 114
 115 /*
 116  * bgp_info_nexthop_cmp
 117  *
 118  * Compare the nexthops of two paths. Return value is less than, equal to,
 119  * or greater than zero if bi1 is respectively less than, equal to,
 120  * or greater than bi2.
 121  */
 122 int bgp_info_nexthop_cmp(struct bgp_info *bi1, struct bgp_info *bi2)
 123 {
 124         int compare;
 125         struct in6_addr addr1, addr2;
 126
 127         compare = IPV4_ADDR_CMP(&bi1->attr->nexthop, &bi2->attr->nexthop);
 128         if (!compare) {
 129                 if (bi1->attr->mp_nexthop_len == bi2->attr->mp_nexthop_len) {
 130                         switch (bi1->attr->mp_nexthop_len) {
 131                         case BGP_ATTR_NHLEN_IPV4:
 132                         case BGP_ATTR_NHLEN_VPNV4:
 133                                 compare = IPV4_ADDR_CMP(
 134                                         &bi1->attr->mp_nexthop_global_in,
 135                                         &bi2->attr->mp_nexthop_global_in);
 136                                 break;
 137                         case BGP_ATTR_NHLEN_IPV6_GLOBAL:
 138                         case BGP_ATTR_NHLEN_VPNV6_GLOBAL:
 139                                 compare = IPV6_ADDR_CMP(
 140                                         &bi1->attr->mp_nexthop_global,
 141                                         &bi2->attr->mp_nexthop_global);
 142                                 break;
 143                         case BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL:
 144                                 addr1 = (bi1->attr->mp_nexthop_prefer_global) ?
 145                                         bi1->attr->mp_nexthop_global
 146                                         : bi1->attr->mp_nexthop_local;
 147                                 addr2 = (bi2->attr->mp_nexthop_prefer_global) ?
 148                                         bi2->attr->mp_nexthop_global
 149                                         : bi2->attr->mp_nexthop_local;
 150
 151                                 if (!bi1->attr->mp_nexthop_prefer_global &&
 152                                     !bi2->attr->mp_nexthop_prefer_global)
 153                                         compare = !bgp_interface_same(bi1->peer->ifp, bi2->peer->ifp);
 154
 155                                 if (!compare)
 156                                         compare = IPV6_ADDR_CMP(&addr1, &addr2);
 157                                 break;
 158                         }
 159                 }
 160
 161                 /* This can happen if one IPv6 peer sends you global and
 162                  * link-local
 163                  * nexthops but another IPv6 peer only sends you global
 164                  */
 165                 else if (bi1->attr->mp_nexthop_len == BGP_ATTR_NHLEN_IPV6_GLOBAL
 166                          || bi1->attr->mp_nexthop_len
 167                                     == BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL) {
 168                         compare = IPV6_ADDR_CMP(&bi1->attr->mp_nexthop_global,
 169                                                 &bi2->attr->mp_nexthop_global);
 170                         if (!compare) {
 171                                 if (bi1->attr->mp_nexthop_len
 172                                     < bi2->attr->mp_nexthop_len)
 173                                         compare = -1;
 174                                 else
 175                                         compare = 1;
 176                         }
 177                 }
 178         }
 179
 180         return compare;
 181 }
 182
 183 /*
 184  * bgp_info_mpath_cmp
 185  *
 186  * This function determines our multipath list ordering. By ordering
 187  * the list we can deterministically select which paths are included
 188  * in the multipath set. The ordering also helps in detecting changes
 189  * in the multipath selection so we can detect whether to send an
 190  * update to zebra.
 191  *
 192  * The order of paths is determined first by received nexthop, and then
 193  * by peer address if the nexthops are the same.
 194  */
 195 static int bgp_info_mpath_cmp(void *val1, void *val2)
 196 {
 197         struct bgp_info *bi1, *bi2;
 198         int compare;
 199
 200         bi1 = val1;
 201         bi2 = val2;
 202
 203         compare = bgp_info_nexthop_cmp(bi1, bi2);
 204
 205         if (!compare) {
 206                 if (!bi1->peer->su_remote && !bi2->peer->su_remote)
 207                         compare = 0;
 208                 else if (!bi1->peer->su_remote)
 209                         compare = 1;
 210                 else if (!bi2->peer->su_remote)
 211                         compare = -1;
 212                 else
 213                         compare = sockunion_cmp(bi1->peer->su_remote,
 214                                                 bi2->peer->su_remote);
 215         }
 216
 217         return compare;
 218 }
 219
 220 /*
 221  * bgp_mp_list_init
 222  *
 223  * Initialize the mp_list, which holds the list of multipaths
 224  * selected by bgp_best_selection
 225  */
 226 void bgp_mp_list_init(struct list *mp_list)
 227 {
 228         assert(mp_list);
 229         memset(mp_list, 0, sizeof(struct list));
 230         mp_list->cmp = bgp_info_mpath_cmp;
 231 }
 232
 233 /*
 234  * bgp_mp_list_clear
 235  *
 236  * Clears all entries out of the mp_list
 237  */
 238 void bgp_mp_list_clear(struct list *mp_list)
 239 {
 240         assert(mp_list);
 241         list_delete_all_node(mp_list);
 242 }
 243
 244 /*
 245  * bgp_mp_list_add
 246  *
 247  * Adds a multipath entry to the mp_list
 248  */
 249 void bgp_mp_list_add(struct list *mp_list, struct bgp_info *mpinfo)
 250 {
 251         assert(mp_list && mpinfo);
 252         listnode_add_sort(mp_list, mpinfo);
 253 }
 254
 255 /*
 256  * bgp_info_mpath_new
 257  *
 258  * Allocate and zero memory for a new bgp_info_mpath element
 259  */
 260 static struct bgp_info_mpath *bgp_info_mpath_new(void)
 261 {
 262         struct bgp_info_mpath *new_mpath;
 263         new_mpath =
 264                 XCALLOC(MTYPE_BGP_MPATH_INFO, sizeof(struct bgp_info_mpath));
 265         return new_mpath;
 266 }
 267
 268 /*
 269  * bgp_info_mpath_free
 270  *
 271  * Release resources for a bgp_info_mpath element and zero out pointer
 272  */
 273 void bgp_info_mpath_free(struct bgp_info_mpath **mpath)
 274 {
 275         if (mpath && *mpath) {
 276                 if ((*mpath)->mp_attr)
 277                         bgp_attr_unintern(&(*mpath)->mp_attr);
 278                 XFREE(MTYPE_BGP_MPATH_INFO, *mpath);
 279                 *mpath = NULL;
 280         }
 281 }
 282
 283 /*
 284  * bgp_info_mpath_get
 285  *
 286  * Fetch the mpath element for the given bgp_info. Used for
 287  * doing lazy allocation.
 288  */
 289 static struct bgp_info_mpath *bgp_info_mpath_get(struct bgp_info *binfo)
 290 {
 291         struct bgp_info_mpath *mpath;
 292         if (!binfo->mpath) {
 293                 mpath = bgp_info_mpath_new();
 294                 if (!mpath)
 295                         return NULL;
 296                 binfo->mpath = mpath;
 297                 mpath->mp_info = binfo;
 298         }
 299         return binfo->mpath;
 300 }
 301
 302 /*
 303  * bgp_info_mpath_enqueue
 304  *
 305  * Enqueue a path onto the multipath list given the previous multipath
 306  * list entry
 307  */
 308 static void bgp_info_mpath_enqueue(struct bgp_info *prev_info,
 309                                    struct bgp_info *binfo)
 310 {
 311         struct bgp_info_mpath *prev, *mpath;
 312
 313         prev = bgp_info_mpath_get(prev_info);
 314         mpath = bgp_info_mpath_get(binfo);
 315         if (!prev || !mpath)
 316                 return;
 317
 318         mpath->mp_next = prev->mp_next;
 319         mpath->mp_prev = prev;
 320         if (prev->mp_next)
 321                 prev->mp_next->mp_prev = mpath;
 322         prev->mp_next = mpath;
 323
 324         SET_FLAG(binfo->flags, BGP_INFO_MULTIPATH);
 325 }
 326
 327 /*
 328  * bgp_info_mpath_dequeue
 329  *
 330  * Remove a path from the multipath list
 331  */
 332 void bgp_info_mpath_dequeue(struct bgp_info *binfo)
 333 {
 334         struct bgp_info_mpath *mpath = binfo->mpath;
 335         if (!mpath)
 336                 return;
 337         if (mpath->mp_prev)
 338                 mpath->mp_prev->mp_next = mpath->mp_next;
 339         if (mpath->mp_next)
 340                 mpath->mp_next->mp_prev = mpath->mp_prev;
 341         mpath->mp_next = mpath->mp_prev = NULL;
 342         UNSET_FLAG(binfo->flags, BGP_INFO_MULTIPATH);
 343 }
 344
 345 /*
 346  * bgp_info_mpath_next
 347  *
 348  * Given a bgp_info, return the next multipath entry
 349  */
 350 struct bgp_info *bgp_info_mpath_next(struct bgp_info *binfo)
 351 {
 352         if (!binfo->mpath || !binfo->mpath->mp_next)
 353                 return NULL;
 354         return binfo->mpath->mp_next->mp_info;
 355 }
 356
 357 /*
 358  * bgp_info_mpath_first
 359  *
 360  * Given bestpath bgp_info, return the first multipath entry.
 361  */
 362 struct bgp_info *bgp_info_mpath_first(struct bgp_info *binfo)
 363 {
 364         return bgp_info_mpath_next(binfo);
 365 }
 366
 367 /*
 368  * bgp_info_mpath_count
 369  *
 370  * Given the bestpath bgp_info, return the number of multipath entries
 371  */
 372 u_int32_t bgp_info_mpath_count(struct bgp_info *binfo)
 373 {
 374         if (!binfo->mpath)
 375                 return 0;
 376         return binfo->mpath->mp_count;
 377 }
 378
 379 /*
 380  * bgp_info_mpath_count_set
 381  *
 382  * Sets the count of multipaths into bestpath's mpath element
 383  */
 384 static void bgp_info_mpath_count_set(struct bgp_info *binfo, u_int32_t count)
 385 {
 386         struct bgp_info_mpath *mpath;
 387         if (!count && !binfo->mpath)
 388                 return;
 389         mpath = bgp_info_mpath_get(binfo);
 390         if (!mpath)
 391                 return;
 392         mpath->mp_count = count;
 393 }
 394
 395 /*
 396  * bgp_info_mpath_attr
 397  *
 398  * Given bestpath bgp_info, return aggregated attribute set used
 399  * for advertising the multipath route
 400  */
 401 struct attr *bgp_info_mpath_attr(struct bgp_info *binfo)
 402 {
 403         if (!binfo->mpath)
 404                 return NULL;
 405         return binfo->mpath->mp_attr;
 406 }
 407
 408 /*
 409  * bgp_info_mpath_attr_set
 410  *
 411  * Sets the aggregated attribute into bestpath's mpath element
 412  */
 413 static void bgp_info_mpath_attr_set(struct bgp_info *binfo, struct attr *attr)
 414 {
 415         struct bgp_info_mpath *mpath;
 416         if (!attr && !binfo->mpath)
 417                 return;
 418         mpath = bgp_info_mpath_get(binfo);
 419         if (!mpath)
 420                 return;
 421         mpath->mp_attr = attr;
 422 }
 423
 424 /*
 425  * bgp_info_mpath_update
 426  *
 427  * Compare and sync up the multipath list with the mp_list generated by
 428  * bgp_best_selection
 429  */
 430 void bgp_info_mpath_update(struct bgp_node *rn, struct bgp_info *new_best,
 431                            struct bgp_info *old_best, struct list *mp_list,
 432                            struct bgp_maxpaths_cfg *mpath_cfg)
 433 {
 434         u_int16_t maxpaths, mpath_count, old_mpath_count;
 435         struct listnode *mp_node, *mp_next_node;
 436         struct bgp_info *cur_mpath, *new_mpath, *next_mpath, *prev_mpath;
 437         int mpath_changed, debug;
 438         char pfx_buf[PREFIX2STR_BUFFER], nh_buf[2][INET6_ADDRSTRLEN];
 439         char path_buf[PATH_ADDPATH_STR_BUFFER];
 440
 441         mpath_changed = 0;
 442         maxpaths = multipath_num;
 443         mpath_count = 0;
 444         cur_mpath = NULL;
 445         old_mpath_count = 0;
 446         prev_mpath = new_best;
 447         mp_node = listhead(mp_list);
 448         debug = bgp_debug_bestpath(&rn->p);
 449
 450         if (debug)
 451                 prefix2str(&rn->p, pfx_buf, sizeof(pfx_buf));
 452
 453         if (new_best) {
 454                 mpath_count++;
 455                 if (new_best != old_best)
 456                         bgp_info_mpath_dequeue(new_best);
 457                 maxpaths = (new_best->peer->sort == BGP_PEER_IBGP)
 458                                    ? mpath_cfg->maxpaths_ibgp
 459                                    : mpath_cfg->maxpaths_ebgp;
 460         }
 461
 462         if (old_best) {
 463                 cur_mpath = bgp_info_mpath_first(old_best);
 464                 old_mpath_count = bgp_info_mpath_count(old_best);
 465                 bgp_info_mpath_count_set(old_best, 0);
 466                 bgp_info_mpath_dequeue(old_best);
 467         }
 468
 469         if (debug)
 470                 zlog_debug(
 471                         "%s: starting mpath update, newbest %s num candidates %d old-mpath-count %d",
 472                         pfx_buf, new_best ? new_best->peer->host : "NONE",
 473                         listcount(mp_list), old_mpath_count);
 474
 475         /*
 476          * We perform an ordered walk through both lists in parallel.
 477          * The reason for the ordered walk is that if there are paths
 478          * that were previously multipaths and are still multipaths, the walk
 479          * should encounter them in both lists at the same time. Otherwise
 480          * there will be paths that are in one list or another, and we
 481          * will deal with these separately.
 482          *
 483          * Note that new_best might be somewhere in the mp_list, so we need
 484          * to skip over it
 485          */
 486         while (mp_node || cur_mpath) {
 487                 struct bgp_info *tmp_info;
 488
 489                 /*
 490                  * We can bail out of this loop if all existing paths on the
 491                  * multipath list have been visited (for cleanup purposes) and
 492                  * the maxpath requirement is fulfulled
 493                  */
 494                 if (!cur_mpath && (mpath_count >= maxpaths))
 495                         break;
 496
 497                 mp_next_node = mp_node ? listnextnode(mp_node) : NULL;
 498                 next_mpath = cur_mpath ? bgp_info_mpath_next(cur_mpath) : NULL;
 499                 tmp_info = mp_node ? listgetdata(mp_node) : NULL;
 500
 501                 if (debug)
 502                         zlog_debug(
 503                                 "%s: comparing candidate %s with existing mpath %s",
 504                                 pfx_buf,
 505                                 tmp_info ? tmp_info->peer->host : "NONE",
 506                                 cur_mpath ? cur_mpath->peer->host : "NONE");
 507
 508                 /*
 509                  * If equal, the path was a multipath and is still a multipath.
 510                  * Insert onto new multipath list if maxpaths allows.
 511                  */
 512                 if (mp_node && (listgetdata(mp_node) == cur_mpath)) {
 513                         list_delete_node(mp_list, mp_node);
 514                         bgp_info_mpath_dequeue(cur_mpath);
 515                         if ((mpath_count < maxpaths)
 516                             && bgp_info_nexthop_cmp(prev_mpath, cur_mpath)) {
 517                                 bgp_info_mpath_enqueue(prev_mpath, cur_mpath);
 518                                 prev_mpath = cur_mpath;
 519                                 mpath_count++;
 520                                 if (debug) {
 521                                         bgp_info_path_with_addpath_rx_str(
 522                                                 cur_mpath, path_buf);
 523                                         zlog_debug(
 524                                                 "%s: %s is still multipath, cur count %d",
 525                                                 pfx_buf, path_buf, mpath_count);
 526                                 }
 527                         } else {
 528                                 mpath_changed = 1;
 529                                 if (debug) {
 530                                         bgp_info_path_with_addpath_rx_str(
 531                                                 cur_mpath, path_buf);
 532                                         zlog_debug(
 533                                                 "%s: remove mpath %s nexthop %s, cur count %d",
 534                                                 pfx_buf, path_buf,
 535                                                 inet_ntop(AF_INET,
 536                                                           &cur_mpath->attr
 537                                                                    ->nexthop,
 538                                                           nh_buf[0],
 539                                                           sizeof(nh_buf[0])),
 540                                                 mpath_count);
 541                                 }
 542                         }
 543                         mp_node = mp_next_node;
 544                         cur_mpath = next_mpath;
 545                         continue;
 546                 }
 547
 548                 if (cur_mpath
 549                     && (!mp_node
 550                         || (bgp_info_mpath_cmp(cur_mpath, listgetdata(mp_node))
 551                             < 0))) {
 552                         /*
 553                          * If here, we have an old multipath and either the
 554                          * mp_list
 555                          * is finished or the next mp_node points to a later
 556                          * multipath, so we need to purge this path from the
 557                          * multipath list
 558                          */
 559                         bgp_info_mpath_dequeue(cur_mpath);
 560                         mpath_changed = 1;
 561                         if (debug) {
 562                                 bgp_info_path_with_addpath_rx_str(cur_mpath,
 563                                                                   path_buf);
 564                                 zlog_debug(
 565                                         "%s: remove mpath %s nexthop %s, cur count %d",
 566                                         pfx_buf, path_buf,
 567                                         inet_ntop(AF_INET,
 568                                                   &cur_mpath->attr->nexthop,
 569                                                   nh_buf[0], sizeof(nh_buf[0])),
 570                                         mpath_count);
 571                         }
 572                         cur_mpath = next_mpath;
 573                 } else {
 574                         /*
 575                          * If here, we have a path on the mp_list that was not
 576                          * previously
 577                          * a multipath (due to non-equivalance or maxpaths
 578                          * exceeded),
 579                          * or the matching multipath is sorted later in the
 580                          * multipath
 581                          * list. Before we enqueue the path on the new multipath
 582                          * list,
 583                          * make sure its not on the old_best multipath list or
 584                          * referenced
 585                          * via next_mpath:
 586                          * - If next_mpath points to this new path, update
 587                          * next_mpath to
 588                          *   point to the multipath after this one
 589                          * - Dequeue the path from the multipath list just to
 590                          * make sure
 591                          */
 592                         new_mpath = listgetdata(mp_node);
 593                         list_delete_node(mp_list, mp_node);
 594                         if ((mpath_count < maxpaths) && (new_mpath != new_best)
 595                             && bgp_info_nexthop_cmp(prev_mpath, new_mpath)) {
 596                                 if (new_mpath == next_mpath)
 597                                         bgp_info_mpath_next(new_mpath);
 598                                 bgp_info_mpath_dequeue(new_mpath);
 599
 600                                 bgp_info_mpath_enqueue(prev_mpath, new_mpath);
 601                                 prev_mpath = new_mpath;
 602                                 mpath_changed = 1;
 603                                 mpath_count++;
 604                                 if (debug) {
 605                                         bgp_info_path_with_addpath_rx_str(
 606                                                 new_mpath, path_buf);
 607                                         zlog_debug(
 608                                                 "%s: add mpath %s nexthop %s, cur count %d",
 609                                                 pfx_buf, path_buf,
 610                                                 inet_ntop(AF_INET,
 611                                                           &new_mpath->attr
 612                                                                    ->nexthop,
 613                                                           nh_buf[0],
 614                                                           sizeof(nh_buf[0])),
 615                                                 mpath_count);
 616                                 }
 617                         }
 618                         mp_node = mp_next_node;
 619                 }
 620         }
 621
 622         if (new_best) {
 623                 if (debug)
 624                         zlog_debug(
 625                                 "%s: New mpath count (incl newbest) %d mpath-change %s",
 626                                 pfx_buf, mpath_count,
 627                                 mpath_changed ? "YES" : "NO");
 628
 629                 bgp_info_mpath_count_set(new_best, mpath_count - 1);
 630                 if (mpath_changed
 631                     || (bgp_info_mpath_count(new_best) != old_mpath_count))
 632                         SET_FLAG(new_best->flags, BGP_INFO_MULTIPATH_CHG);
 633         }
 634 }
 635
 636 /*
 637  * bgp_mp_dmed_deselect
 638  *
 639  * Clean up multipath information for BGP_INFO_DMED_SELECTED path that
 640  * is not selected as best path
 641  */
 642 void bgp_mp_dmed_deselect(struct bgp_info *dmed_best)
 643 {
 644         struct bgp_info *mpinfo, *mpnext;
 645
 646         if (!dmed_best)
 647                 return;
 648
 649         for (mpinfo = bgp_info_mpath_first(dmed_best); mpinfo;
 650              mpinfo = mpnext) {
 651                 mpnext = bgp_info_mpath_next(mpinfo);
 652                 bgp_info_mpath_dequeue(mpinfo);
 653         }
 654
 655         bgp_info_mpath_count_set(dmed_best, 0);
 656         UNSET_FLAG(dmed_best->flags, BGP_INFO_MULTIPATH_CHG);
 657         assert(bgp_info_mpath_first(dmed_best) == 0);
 658 }
 659
 660 /*
 661  * bgp_info_mpath_aggregate_update
 662  *
 663  * Set the multipath aggregate attribute. We need to see if the
 664  * aggregate has changed and then set the ATTR_CHANGED flag on the
 665  * bestpath info so that a peer update will be generated. The
 666  * change is detected by generating the current attribute,
 667  * interning it, and then comparing the interned pointer with the
 668  * current value. We can skip this generate/compare step if there
 669  * is no change in multipath selection and no attribute change in
 670  * any multipath.
 671  */
 672 void bgp_info_mpath_aggregate_update(struct bgp_info *new_best,
 673                                      struct bgp_info *old_best)
 674 {
 675         struct bgp_info *mpinfo;
 676         struct aspath *aspath;
 677         struct aspath *asmerge;
 678         struct attr *new_attr, *old_attr;
 679         u_char origin;
 680         struct community *community, *commerge;
 681         struct ecommunity *ecomm, *ecommerge;
 682         struct lcommunity *lcomm, *lcommerge;
 683         struct attr attr = {0};
 684
 685         if (old_best && (old_best != new_best)
 686             && (old_attr = bgp_info_mpath_attr(old_best))) {
 687                 bgp_attr_unintern(&old_attr);
 688                 bgp_info_mpath_attr_set(old_best, NULL);
 689         }
 690
 691         if (!new_best)
 692                 return;
 693
 694         if (!bgp_info_mpath_count(new_best)) {
 695                 if ((new_attr = bgp_info_mpath_attr(new_best))) {
 696                         bgp_attr_unintern(&new_attr);
 697                         bgp_info_mpath_attr_set(new_best, NULL);
 698                         SET_FLAG(new_best->flags, BGP_INFO_ATTR_CHANGED);
 699                 }
 700                 return;
 701         }
 702
 703         bgp_attr_dup(&attr, new_best->attr);
 704
 705         if (new_best->peer && bgp_flag_check(new_best->peer->bgp,
 706                                              BGP_FLAG_MULTIPATH_RELAX_AS_SET)) {
 707
 708                 /* aggregate attribute from multipath constituents */
 709                 aspath = aspath_dup(attr.aspath);
 710                 origin = attr.origin;
 711                 community =
 712                         attr.community ? community_dup(attr.community) : NULL;
 713                 ecomm = (attr.ecommunity) ? ecommunity_dup(attr.ecommunity)
 714                                           : NULL;
 715                 lcomm = (attr.lcommunity) ? lcommunity_dup(attr.lcommunity)
 716                                           : NULL;
 717
 718                 for (mpinfo = bgp_info_mpath_first(new_best); mpinfo;
 719                      mpinfo = bgp_info_mpath_next(mpinfo)) {
 720                         asmerge =
 721                                 aspath_aggregate(aspath, mpinfo->attr->aspath);
 722                         aspath_free(aspath);
 723                         aspath = asmerge;
 724
 725                         if (origin < mpinfo->attr->origin)
 726                                 origin = mpinfo->attr->origin;
 727
 728                         if (mpinfo->attr->community) {
 729                                 if (community) {
 730                                         commerge = community_merge(
 731                                                 community,
 732                                                 mpinfo->attr->community);
 733                                         community =
 734                                                 community_uniq_sort(commerge);
 735                                         community_free(commerge);
 736                                 } else
 737                                         community = community_dup(
 738                                                 mpinfo->attr->community);
 739                         }
 740
 741                         if (mpinfo->attr->ecommunity) {
 742                                 if (ecomm) {
 743                                         ecommerge = ecommunity_merge(
 744                                                 ecomm,
 745                                                 mpinfo->attr->ecommunity);
 746                                         ecomm = ecommunity_uniq_sort(ecommerge);
 747                                         ecommunity_free(&ecommerge);
 748                                 } else
 749                                         ecomm = ecommunity_dup(
 750                                                 mpinfo->attr->ecommunity);
 751                         }
 752                         if (mpinfo->attr->lcommunity) {
 753                                 if (lcomm) {
 754                                         lcommerge = lcommunity_merge(
 755                                                 lcomm,
 756                                                 mpinfo->attr->lcommunity);
 757                                         lcomm = lcommunity_uniq_sort(lcommerge);
 758                                         lcommunity_free(&lcommerge);
 759                                 } else
 760                                         lcomm = lcommunity_dup(
 761                                                 mpinfo->attr->lcommunity);
 762                         }
 763                 }
 764
 765                 attr.aspath = aspath;
 766                 attr.origin = origin;
 767                 if (community) {
 768                         attr.community = community;
 769                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_COMMUNITIES);
 770                 }
 771                 if (ecomm) {
 772                         attr.ecommunity = ecomm;
 773                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_EXT_COMMUNITIES);
 774                 }
 775                 if (lcomm) {
 776                         attr.lcommunity = lcomm;
 777                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_LARGE_COMMUNITIES);
 778                 }
 779
 780                 /* Zap multipath attr nexthop so we set nexthop to self */
 781                 attr.nexthop.s_addr = 0;
 782                 memset(&attr.mp_nexthop_global, 0, sizeof(struct in6_addr));
 783
 784                 /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */
 785         }
 786
 787         new_attr = bgp_attr_intern(&attr);
 788
 789         if (new_attr != bgp_info_mpath_attr(new_best)) {
 790                 if ((old_attr = bgp_info_mpath_attr(new_best)))
 791                         bgp_attr_unintern(&old_attr);
 792                 bgp_info_mpath_attr_set(new_best, new_attr);
 793                 SET_FLAG(new_best->flags, BGP_INFO_ATTR_CHANGED);
 794         } else
 795                 bgp_attr_unintern(&new_attr);
 796 }