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                           uint16_t maxpaths, uint16_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_path_info_nexthop_cmp
 117  *
 118  * Compare the nexthops of two paths. Return value is less than, equal to,
 119  * or greater than zero if bpi1 is respectively less than, equal to,
 120  * or greater than bpi2.
 121  */
 122 int bgp_path_info_nexthop_cmp(struct bgp_path_info *bpi1,
 123                               struct bgp_path_info *bpi2)
 124 {
 125         int compare;
 126         struct in6_addr addr1, addr2;
 127
 128         compare = IPV4_ADDR_CMP(&bpi1->attr->nexthop, &bpi2->attr->nexthop);
 129         if (!compare) {
 130                 if (bpi1->attr->mp_nexthop_len == bpi2->attr->mp_nexthop_len) {
 131                         switch (bpi1->attr->mp_nexthop_len) {
 132                         case BGP_ATTR_NHLEN_IPV4:
 133                         case BGP_ATTR_NHLEN_VPNV4:
 134                                 compare = IPV4_ADDR_CMP(
 135                                         &bpi1->attr->mp_nexthop_global_in,
 136                                         &bpi2->attr->mp_nexthop_global_in);
 137                                 break;
 138                         case BGP_ATTR_NHLEN_IPV6_GLOBAL:
 139                         case BGP_ATTR_NHLEN_VPNV6_GLOBAL:
 140                                 compare = IPV6_ADDR_CMP(
 141                                         &bpi1->attr->mp_nexthop_global,
 142                                         &bpi2->attr->mp_nexthop_global);
 143                                 break;
 144                         case BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL:
 145                                 addr1 = (bpi1->attr->mp_nexthop_prefer_global)
 146                                                 ? bpi1->attr->mp_nexthop_global
 147                                                 : bpi1->attr->mp_nexthop_local;
 148                                 addr2 = (bpi2->attr->mp_nexthop_prefer_global)
 149                                                 ? bpi2->attr->mp_nexthop_global
 150                                                 : bpi2->attr->mp_nexthop_local;
 151
 152                                 if (!bpi1->attr->mp_nexthop_prefer_global
 153                                     && !bpi2->attr->mp_nexthop_prefer_global)
 154                                         compare = !bgp_interface_same(
 155                                                 bpi1->peer->ifp,
 156                                                 bpi2->peer->ifp);
 157
 158                                 if (!compare)
 159                                         compare = IPV6_ADDR_CMP(&addr1, &addr2);
 160                                 break;
 161                         }
 162                 }
 163
 164                 /* This can happen if one IPv6 peer sends you global and
 165                  * link-local
 166                  * nexthops but another IPv6 peer only sends you global
 167                  */
 168                 else if (bpi1->attr->mp_nexthop_len
 169                                  == BGP_ATTR_NHLEN_IPV6_GLOBAL
 170                          || bpi1->attr->mp_nexthop_len
 171                                     == BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL) {
 172                         compare = IPV6_ADDR_CMP(&bpi1->attr->mp_nexthop_global,
 173                                                 &bpi2->attr->mp_nexthop_global);
 174                         if (!compare) {
 175                                 if (bpi1->attr->mp_nexthop_len
 176                                     < bpi2->attr->mp_nexthop_len)
 177                                         compare = -1;
 178                                 else
 179                                         compare = 1;
 180                         }
 181                 }
 182         }
 183
 184         return compare;
 185 }
 186
 187 /*
 188  * bgp_path_info_mpath_cmp
 189  *
 190  * This function determines our multipath list ordering. By ordering
 191  * the list we can deterministically select which paths are included
 192  * in the multipath set. The ordering also helps in detecting changes
 193  * in the multipath selection so we can detect whether to send an
 194  * update to zebra.
 195  *
 196  * The order of paths is determined first by received nexthop, and then
 197  * by peer address if the nexthops are the same.
 198  */
 199 static int bgp_path_info_mpath_cmp(void *val1, void *val2)
 200 {
 201         struct bgp_path_info *bpi1, *bpi2;
 202         int compare;
 203
 204         bpi1 = val1;
 205         bpi2 = val2;
 206
 207         compare = bgp_path_info_nexthop_cmp(bpi1, bpi2);
 208
 209         if (!compare) {
 210                 if (!bpi1->peer->su_remote && !bpi2->peer->su_remote)
 211                         compare = 0;
 212                 else if (!bpi1->peer->su_remote)
 213                         compare = 1;
 214                 else if (!bpi2->peer->su_remote)
 215                         compare = -1;
 216                 else
 217                         compare = sockunion_cmp(bpi1->peer->su_remote,
 218                                                 bpi2->peer->su_remote);
 219         }
 220
 221         return compare;
 222 }
 223
 224 /*
 225  * bgp_mp_list_init
 226  *
 227  * Initialize the mp_list, which holds the list of multipaths
 228  * selected by bgp_best_selection
 229  */
 230 void bgp_mp_list_init(struct list *mp_list)
 231 {
 232         assert(mp_list);
 233         memset(mp_list, 0, sizeof(struct list));
 234         mp_list->cmp = bgp_path_info_mpath_cmp;
 235 }
 236
 237 /*
 238  * bgp_mp_list_clear
 239  *
 240  * Clears all entries out of the mp_list
 241  */
 242 void bgp_mp_list_clear(struct list *mp_list)
 243 {
 244         assert(mp_list);
 245         list_delete_all_node(mp_list);
 246 }
 247
 248 /*
 249  * bgp_mp_list_add
 250  *
 251  * Adds a multipath entry to the mp_list
 252  */
 253 void bgp_mp_list_add(struct list *mp_list, struct bgp_path_info *mpinfo)
 254 {
 255         assert(mp_list && mpinfo);
 256         listnode_add_sort(mp_list, mpinfo);
 257 }
 258
 259 /*
 260  * bgp_path_info_mpath_new
 261  *
 262  * Allocate and zero memory for a new bgp_path_info_mpath element
 263  */
 264 static struct bgp_path_info_mpath *bgp_path_info_mpath_new(void)
 265 {
 266         struct bgp_path_info_mpath *new_mpath;
 267         new_mpath = XCALLOC(MTYPE_BGP_MPATH_INFO,
 268                             sizeof(struct bgp_path_info_mpath));
 269         return new_mpath;
 270 }
 271
 272 /*
 273  * bgp_path_info_mpath_free
 274  *
 275  * Release resources for a bgp_path_info_mpath element and zero out pointer
 276  */
 277 void bgp_path_info_mpath_free(struct bgp_path_info_mpath **mpath)
 278 {
 279         if (mpath && *mpath) {
 280                 if ((*mpath)->mp_attr)
 281                         bgp_attr_unintern(&(*mpath)->mp_attr);
 282                 XFREE(MTYPE_BGP_MPATH_INFO, *mpath);
 283                 *mpath = NULL;
 284         }
 285 }
 286
 287 /*
 288  * bgp_path_info_mpath_get
 289  *
 290  * Fetch the mpath element for the given bgp_path_info. Used for
 291  * doing lazy allocation.
 292  */
 293 static struct bgp_path_info_mpath *
 294 bgp_path_info_mpath_get(struct bgp_path_info *path)
 295 {
 296         struct bgp_path_info_mpath *mpath;
 297
 298         if (!path)
 299                 return NULL;
 300
 301         if (!path->mpath) {
 302                 mpath = bgp_path_info_mpath_new();
 303                 if (!mpath)
 304                         return NULL;
 305                 path->mpath = mpath;
 306                 mpath->mp_info = path;
 307         }
 308         return path->mpath;
 309 }
 310
 311 /*
 312  * bgp_path_info_mpath_enqueue
 313  *
 314  * Enqueue a path onto the multipath list given the previous multipath
 315  * list entry
 316  */
 317 static void bgp_path_info_mpath_enqueue(struct bgp_path_info *prev_info,
 318                                         struct bgp_path_info *path)
 319 {
 320         struct bgp_path_info_mpath *prev, *mpath;
 321
 322         prev = bgp_path_info_mpath_get(prev_info);
 323         mpath = bgp_path_info_mpath_get(path);
 324         if (!prev || !mpath)
 325                 return;
 326
 327         mpath->mp_next = prev->mp_next;
 328         mpath->mp_prev = prev;
 329         if (prev->mp_next)
 330                 prev->mp_next->mp_prev = mpath;
 331         prev->mp_next = mpath;
 332
 333         SET_FLAG(path->flags, BGP_PATH_MULTIPATH);
 334 }
 335
 336 /*
 337  * bgp_path_info_mpath_dequeue
 338  *
 339  * Remove a path from the multipath list
 340  */
 341 void bgp_path_info_mpath_dequeue(struct bgp_path_info *path)
 342 {
 343         struct bgp_path_info_mpath *mpath = path->mpath;
 344         if (!mpath)
 345                 return;
 346         if (mpath->mp_prev)
 347                 mpath->mp_prev->mp_next = mpath->mp_next;
 348         if (mpath->mp_next)
 349                 mpath->mp_next->mp_prev = mpath->mp_prev;
 350         mpath->mp_next = mpath->mp_prev = NULL;
 351         UNSET_FLAG(path->flags, BGP_PATH_MULTIPATH);
 352 }
 353
 354 /*
 355  * bgp_path_info_mpath_next
 356  *
 357  * Given a bgp_path_info, return the next multipath entry
 358  */
 359 struct bgp_path_info *bgp_path_info_mpath_next(struct bgp_path_info *path)
 360 {
 361         if (!path->mpath || !path->mpath->mp_next)
 362                 return NULL;
 363         return path->mpath->mp_next->mp_info;
 364 }
 365
 366 /*
 367  * bgp_path_info_mpath_first
 368  *
 369  * Given bestpath bgp_path_info, return the first multipath entry.
 370  */
 371 struct bgp_path_info *bgp_path_info_mpath_first(struct bgp_path_info *path)
 372 {
 373         return bgp_path_info_mpath_next(path);
 374 }
 375
 376 /*
 377  * bgp_path_info_mpath_count
 378  *
 379  * Given the bestpath bgp_path_info, return the number of multipath entries
 380  */
 381 uint32_t bgp_path_info_mpath_count(struct bgp_path_info *path)
 382 {
 383         if (!path->mpath)
 384                 return 0;
 385         return path->mpath->mp_count;
 386 }
 387
 388 /*
 389  * bgp_path_info_mpath_count_set
 390  *
 391  * Sets the count of multipaths into bestpath's mpath element
 392  */
 393 static void bgp_path_info_mpath_count_set(struct bgp_path_info *path,
 394                                           uint32_t count)
 395 {
 396         struct bgp_path_info_mpath *mpath;
 397         if (!count && !path->mpath)
 398                 return;
 399         mpath = bgp_path_info_mpath_get(path);
 400         if (!mpath)
 401                 return;
 402         mpath->mp_count = count;
 403 }
 404
 405 /*
 406  * bgp_path_info_mpath_attr
 407  *
 408  * Given bestpath bgp_path_info, return aggregated attribute set used
 409  * for advertising the multipath route
 410  */
 411 struct attr *bgp_path_info_mpath_attr(struct bgp_path_info *path)
 412 {
 413         if (!path->mpath)
 414                 return NULL;
 415         return path->mpath->mp_attr;
 416 }
 417
 418 /*
 419  * bgp_path_info_mpath_attr_set
 420  *
 421  * Sets the aggregated attribute into bestpath's mpath element
 422  */
 423 static void bgp_path_info_mpath_attr_set(struct bgp_path_info *path,
 424                                          struct attr *attr)
 425 {
 426         struct bgp_path_info_mpath *mpath;
 427         if (!attr && !path->mpath)
 428                 return;
 429         mpath = bgp_path_info_mpath_get(path);
 430         if (!mpath)
 431                 return;
 432         mpath->mp_attr = attr;
 433 }
 434
 435 /*
 436  * bgp_path_info_mpath_update
 437  *
 438  * Compare and sync up the multipath list with the mp_list generated by
 439  * bgp_best_selection
 440  */
 441 void bgp_path_info_mpath_update(struct bgp_node *rn,
 442                                 struct bgp_path_info *new_best,
 443                                 struct bgp_path_info *old_best,
 444                                 struct list *mp_list,
 445                                 struct bgp_maxpaths_cfg *mpath_cfg)
 446 {
 447         uint16_t maxpaths, mpath_count, old_mpath_count;
 448         struct listnode *mp_node, *mp_next_node;
 449         struct bgp_path_info *cur_mpath, *new_mpath, *next_mpath, *prev_mpath;
 450         int mpath_changed, debug;
 451         char pfx_buf[PREFIX2STR_BUFFER], nh_buf[2][INET6_ADDRSTRLEN];
 452         char path_buf[PATH_ADDPATH_STR_BUFFER];
 453
 454         mpath_changed = 0;
 455         maxpaths = multipath_num;
 456         mpath_count = 0;
 457         cur_mpath = NULL;
 458         old_mpath_count = 0;
 459         prev_mpath = new_best;
 460         mp_node = listhead(mp_list);
 461         debug = bgp_debug_bestpath(&rn->p);
 462
 463         if (debug)
 464                 prefix2str(&rn->p, pfx_buf, sizeof(pfx_buf));
 465
 466         if (new_best) {
 467                 mpath_count++;
 468                 if (new_best != old_best)
 469                         bgp_path_info_mpath_dequeue(new_best);
 470                 maxpaths = (new_best->peer->sort == BGP_PEER_IBGP)
 471                                    ? mpath_cfg->maxpaths_ibgp
 472                                    : mpath_cfg->maxpaths_ebgp;
 473         }
 474
 475         if (old_best) {
 476                 cur_mpath = bgp_path_info_mpath_first(old_best);
 477                 old_mpath_count = bgp_path_info_mpath_count(old_best);
 478                 bgp_path_info_mpath_count_set(old_best, 0);
 479                 bgp_path_info_mpath_dequeue(old_best);
 480         }
 481
 482         if (debug)
 483                 zlog_debug(
 484                         "%s: starting mpath update, newbest %s num candidates %d old-mpath-count %d",
 485                         pfx_buf, new_best ? new_best->peer->host : "NONE",
 486                         mp_list ? listcount(mp_list) : 0, old_mpath_count);
 487
 488         /*
 489          * We perform an ordered walk through both lists in parallel.
 490          * The reason for the ordered walk is that if there are paths
 491          * that were previously multipaths and are still multipaths, the walk
 492          * should encounter them in both lists at the same time. Otherwise
 493          * there will be paths that are in one list or another, and we
 494          * will deal with these separately.
 495          *
 496          * Note that new_best might be somewhere in the mp_list, so we need
 497          * to skip over it
 498          */
 499         while (mp_node || cur_mpath) {
 500                 struct bgp_path_info *tmp_info;
 501
 502                 /*
 503                  * We can bail out of this loop if all existing paths on the
 504                  * multipath list have been visited (for cleanup purposes) and
 505                  * the maxpath requirement is fulfulled
 506                  */
 507                 if (!cur_mpath && (mpath_count >= maxpaths))
 508                         break;
 509
 510                 mp_next_node = mp_node ? listnextnode(mp_node) : NULL;
 511                 next_mpath =
 512                         cur_mpath ? bgp_path_info_mpath_next(cur_mpath) : NULL;
 513                 tmp_info = mp_node ? listgetdata(mp_node) : NULL;
 514
 515                 if (debug)
 516                         zlog_debug(
 517                                 "%s: comparing candidate %s with existing mpath %s",
 518                                 pfx_buf,
 519                                 tmp_info ? tmp_info->peer->host : "NONE",
 520                                 cur_mpath ? cur_mpath->peer->host : "NONE");
 521
 522                 /*
 523                  * If equal, the path was a multipath and is still a multipath.
 524                  * Insert onto new multipath list if maxpaths allows.
 525                  */
 526                 if (mp_node && (listgetdata(mp_node) == cur_mpath)) {
 527                         list_delete_node(mp_list, mp_node);
 528                         bgp_path_info_mpath_dequeue(cur_mpath);
 529                         if ((mpath_count < maxpaths)
 530                             && prev_mpath
 531                             && bgp_path_info_nexthop_cmp(prev_mpath,
 532                                                          cur_mpath)) {
 533                                 bgp_path_info_mpath_enqueue(prev_mpath,
 534                                                             cur_mpath);
 535                                 prev_mpath = cur_mpath;
 536                                 mpath_count++;
 537                                 if (debug) {
 538                                         bgp_path_info_path_with_addpath_rx_str(
 539                                                 cur_mpath, path_buf);
 540                                         zlog_debug(
 541                                                 "%s: %s is still multipath, cur count %d",
 542                                                 pfx_buf, path_buf, mpath_count);
 543                                 }
 544                         } else {
 545                                 mpath_changed = 1;
 546                                 if (debug) {
 547                                         bgp_path_info_path_with_addpath_rx_str(
 548                                                 cur_mpath, path_buf);
 549                                         zlog_debug(
 550                                                 "%s: remove mpath %s nexthop %s, cur count %d",
 551                                                 pfx_buf, path_buf,
 552                                                 inet_ntop(AF_INET,
 553                                                           &cur_mpath->attr
 554                                                                    ->nexthop,
 555                                                           nh_buf[0],
 556                                                           sizeof(nh_buf[0])),
 557                                                 mpath_count);
 558                                 }
 559                         }
 560                         mp_node = mp_next_node;
 561                         cur_mpath = next_mpath;
 562                         continue;
 563                 }
 564
 565                 if (cur_mpath
 566                     && (!mp_node
 567                         || (bgp_path_info_mpath_cmp(cur_mpath,
 568                                                     listgetdata(mp_node))
 569                             < 0))) {
 570                         /*
 571                          * If here, we have an old multipath and either the
 572                          * mp_list
 573                          * is finished or the next mp_node points to a later
 574                          * multipath, so we need to purge this path from the
 575                          * multipath list
 576                          */
 577                         bgp_path_info_mpath_dequeue(cur_mpath);
 578                         mpath_changed = 1;
 579                         if (debug) {
 580                                 bgp_path_info_path_with_addpath_rx_str(
 581                                         cur_mpath, path_buf);
 582                                 zlog_debug(
 583                                         "%s: remove mpath %s nexthop %s, cur count %d",
 584                                         pfx_buf, path_buf,
 585                                         inet_ntop(AF_INET,
 586                                                   &cur_mpath->attr->nexthop,
 587                                                   nh_buf[0], sizeof(nh_buf[0])),
 588                                         mpath_count);
 589                         }
 590                         cur_mpath = next_mpath;
 591                 } else {
 592                         /*
 593                          * If here, we have a path on the mp_list that was not
 594                          * previously
 595                          * a multipath (due to non-equivalance or maxpaths
 596                          * exceeded),
 597                          * or the matching multipath is sorted later in the
 598                          * multipath
 599                          * list. Before we enqueue the path on the new multipath
 600                          * list,
 601                          * make sure its not on the old_best multipath list or
 602                          * referenced
 603                          * via next_mpath:
 604                          * - If next_mpath points to this new path, update
 605                          * next_mpath to
 606                          *   point to the multipath after this one
 607                          * - Dequeue the path from the multipath list just to
 608                          * make sure
 609                          */
 610                         new_mpath = listgetdata(mp_node);
 611                         list_delete_node(mp_list, mp_node);
 612                         assert(new_mpath);
 613                         assert(prev_mpath);
 614                         if ((mpath_count < maxpaths) && (new_mpath != new_best)
 615                             && bgp_path_info_nexthop_cmp(prev_mpath,
 616                                                          new_mpath)) {
 617                                 bgp_path_info_mpath_dequeue(new_mpath);
 618
 619                                 bgp_path_info_mpath_enqueue(prev_mpath,
 620                                                             new_mpath);
 621                                 prev_mpath = new_mpath;
 622                                 mpath_changed = 1;
 623                                 mpath_count++;
 624                                 if (debug) {
 625                                         bgp_path_info_path_with_addpath_rx_str(
 626                                                 new_mpath, path_buf);
 627                                         zlog_debug(
 628                                                 "%s: add mpath %s nexthop %s, cur count %d",
 629                                                 pfx_buf, path_buf,
 630                                                 inet_ntop(AF_INET,
 631                                                           &new_mpath->attr
 632                                                                    ->nexthop,
 633                                                           nh_buf[0],
 634                                                           sizeof(nh_buf[0])),
 635                                                 mpath_count);
 636                                 }
 637                         }
 638                         mp_node = mp_next_node;
 639                 }
 640         }
 641
 642         if (new_best) {
 643                 if (debug)
 644                         zlog_debug(
 645                                 "%s: New mpath count (incl newbest) %d mpath-change %s",
 646                                 pfx_buf, mpath_count,
 647                                 mpath_changed ? "YES" : "NO");
 648
 649                 bgp_path_info_mpath_count_set(new_best, mpath_count - 1);
 650                 if (mpath_changed
 651                     || (bgp_path_info_mpath_count(new_best) != old_mpath_count))
 652                         SET_FLAG(new_best->flags, BGP_PATH_MULTIPATH_CHG);
 653         }
 654 }
 655
 656 /*
 657  * bgp_mp_dmed_deselect
 658  *
 659  * Clean up multipath information for BGP_PATH_DMED_SELECTED path that
 660  * is not selected as best path
 661  */
 662 void bgp_mp_dmed_deselect(struct bgp_path_info *dmed_best)
 663 {
 664         struct bgp_path_info *mpinfo, *mpnext;
 665
 666         if (!dmed_best)
 667                 return;
 668
 669         for (mpinfo = bgp_path_info_mpath_first(dmed_best); mpinfo;
 670              mpinfo = mpnext) {
 671                 mpnext = bgp_path_info_mpath_next(mpinfo);
 672                 bgp_path_info_mpath_dequeue(mpinfo);
 673         }
 674
 675         bgp_path_info_mpath_count_set(dmed_best, 0);
 676         UNSET_FLAG(dmed_best->flags, BGP_PATH_MULTIPATH_CHG);
 677         assert(bgp_path_info_mpath_first(dmed_best) == NULL);
 678 }
 679
 680 /*
 681  * bgp_path_info_mpath_aggregate_update
 682  *
 683  * Set the multipath aggregate attribute. We need to see if the
 684  * aggregate has changed and then set the ATTR_CHANGED flag on the
 685  * bestpath info so that a peer update will be generated. The
 686  * change is detected by generating the current attribute,
 687  * interning it, and then comparing the interned pointer with the
 688  * current value. We can skip this generate/compare step if there
 689  * is no change in multipath selection and no attribute change in
 690  * any multipath.
 691  */
 692 void bgp_path_info_mpath_aggregate_update(struct bgp_path_info *new_best,
 693                                           struct bgp_path_info *old_best)
 694 {
 695         struct bgp_path_info *mpinfo;
 696         struct aspath *aspath;
 697         struct aspath *asmerge;
 698         struct attr *new_attr, *old_attr;
 699         uint8_t origin;
 700         struct community *community, *commerge;
 701         struct ecommunity *ecomm, *ecommerge;
 702         struct lcommunity *lcomm, *lcommerge;
 703         struct attr attr = {0};
 704
 705         if (old_best && (old_best != new_best)
 706             && (old_attr = bgp_path_info_mpath_attr(old_best))) {
 707                 bgp_attr_unintern(&old_attr);
 708                 bgp_path_info_mpath_attr_set(old_best, NULL);
 709         }
 710
 711         if (!new_best)
 712                 return;
 713
 714         if (!bgp_path_info_mpath_count(new_best)) {
 715                 if ((new_attr = bgp_path_info_mpath_attr(new_best))) {
 716                         bgp_attr_unintern(&new_attr);
 717                         bgp_path_info_mpath_attr_set(new_best, NULL);
 718                         SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 719                 }
 720                 return;
 721         }
 722
 723         attr = *new_best->attr;
 724
 725         if (new_best->peer && bgp_flag_check(new_best->peer->bgp,
 726                                              BGP_FLAG_MULTIPATH_RELAX_AS_SET)) {
 727
 728                 /* aggregate attribute from multipath constituents */
 729                 aspath = aspath_dup(attr.aspath);
 730                 origin = attr.origin;
 731                 community =
 732                         attr.community ? community_dup(attr.community) : NULL;
 733                 ecomm = (attr.ecommunity) ? ecommunity_dup(attr.ecommunity)
 734                                           : NULL;
 735                 lcomm = (attr.lcommunity) ? lcommunity_dup(attr.lcommunity)
 736                                           : NULL;
 737
 738                 for (mpinfo = bgp_path_info_mpath_first(new_best); mpinfo;
 739                      mpinfo = bgp_path_info_mpath_next(mpinfo)) {
 740                         asmerge =
 741                                 aspath_aggregate(aspath, mpinfo->attr->aspath);
 742                         aspath_free(aspath);
 743                         aspath = asmerge;
 744
 745                         if (origin < mpinfo->attr->origin)
 746                                 origin = mpinfo->attr->origin;
 747
 748                         if (mpinfo->attr->community) {
 749                                 if (community) {
 750                                         commerge = community_merge(
 751                                                 community,
 752                                                 mpinfo->attr->community);
 753                                         community =
 754                                                 community_uniq_sort(commerge);
 755                                         community_free(&commerge);
 756                                 } else
 757                                         community = community_dup(
 758                                                 mpinfo->attr->community);
 759                         }
 760
 761                         if (mpinfo->attr->ecommunity) {
 762                                 if (ecomm) {
 763                                         ecommerge = ecommunity_merge(
 764                                                 ecomm,
 765                                                 mpinfo->attr->ecommunity);
 766                                         ecomm = ecommunity_uniq_sort(ecommerge);
 767                                         ecommunity_free(&ecommerge);
 768                                 } else
 769                                         ecomm = ecommunity_dup(
 770                                                 mpinfo->attr->ecommunity);
 771                         }
 772                         if (mpinfo->attr->lcommunity) {
 773                                 if (lcomm) {
 774                                         lcommerge = lcommunity_merge(
 775                                                 lcomm,
 776                                                 mpinfo->attr->lcommunity);
 777                                         lcomm = lcommunity_uniq_sort(lcommerge);
 778                                         lcommunity_free(&lcommerge);
 779                                 } else
 780                                         lcomm = lcommunity_dup(
 781                                                 mpinfo->attr->lcommunity);
 782                         }
 783                 }
 784
 785                 attr.aspath = aspath;
 786                 attr.origin = origin;
 787                 if (community) {
 788                         attr.community = community;
 789                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_COMMUNITIES);
 790                 }
 791                 if (ecomm) {
 792                         attr.ecommunity = ecomm;
 793                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_EXT_COMMUNITIES);
 794                 }
 795                 if (lcomm) {
 796                         attr.lcommunity = lcomm;
 797                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_LARGE_COMMUNITIES);
 798                 }
 799
 800                 /* Zap multipath attr nexthop so we set nexthop to self */
 801                 attr.nexthop.s_addr = 0;
 802                 memset(&attr.mp_nexthop_global, 0, sizeof(struct in6_addr));
 803
 804                 /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */
 805         }
 806
 807         new_attr = bgp_attr_intern(&attr);
 808
 809         if (new_attr != bgp_path_info_mpath_attr(new_best)) {
 810                 if ((old_attr = bgp_path_info_mpath_attr(new_best)))
 811                         bgp_attr_unintern(&old_attr);
 812                 bgp_path_info_mpath_attr_set(new_best, new_attr);
 813                 SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 814         } else
 815                 bgp_attr_unintern(&new_attr);
 816 }