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_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_path_info *bi1, struct bgp_path_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(
 154                                                 bi1->peer->ifp, bi2->peer->ifp);
 155
 156                                 if (!compare)
 157                                         compare = IPV6_ADDR_CMP(&addr1, &addr2);
 158                                 break;
 159                         }
 160                 }
 161
 162                 /* This can happen if one IPv6 peer sends you global and
 163                  * link-local
 164                  * nexthops but another IPv6 peer only sends you global
 165                  */
 166                 else if (bi1->attr->mp_nexthop_len == BGP_ATTR_NHLEN_IPV6_GLOBAL
 167                          || bi1->attr->mp_nexthop_len
 168                                     == BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL) {
 169                         compare = IPV6_ADDR_CMP(&bi1->attr->mp_nexthop_global,
 170                                                 &bi2->attr->mp_nexthop_global);
 171                         if (!compare) {
 172                                 if (bi1->attr->mp_nexthop_len
 173                                     < bi2->attr->mp_nexthop_len)
 174                                         compare = -1;
 175                                 else
 176                                         compare = 1;
 177                         }
 178                 }
 179         }
 180
 181         return compare;
 182 }
 183
 184 /*
 185  * bgp_info_mpath_cmp
 186  *
 187  * This function determines our multipath list ordering. By ordering
 188  * the list we can deterministically select which paths are included
 189  * in the multipath set. The ordering also helps in detecting changes
 190  * in the multipath selection so we can detect whether to send an
 191  * update to zebra.
 192  *
 193  * The order of paths is determined first by received nexthop, and then
 194  * by peer address if the nexthops are the same.
 195  */
 196 static int bgp_info_mpath_cmp(void *val1, void *val2)
 197 {
 198         struct bgp_path_info *bi1, *bi2;
 199         int compare;
 200
 201         bi1 = val1;
 202         bi2 = val2;
 203
 204         compare = bgp_info_nexthop_cmp(bi1, bi2);
 205
 206         if (!compare) {
 207                 if (!bi1->peer->su_remote && !bi2->peer->su_remote)
 208                         compare = 0;
 209                 else if (!bi1->peer->su_remote)
 210                         compare = 1;
 211                 else if (!bi2->peer->su_remote)
 212                         compare = -1;
 213                 else
 214                         compare = sockunion_cmp(bi1->peer->su_remote,
 215                                                 bi2->peer->su_remote);
 216         }
 217
 218         return compare;
 219 }
 220
 221 /*
 222  * bgp_mp_list_init
 223  *
 224  * Initialize the mp_list, which holds the list of multipaths
 225  * selected by bgp_best_selection
 226  */
 227 void bgp_mp_list_init(struct list *mp_list)
 228 {
 229         assert(mp_list);
 230         memset(mp_list, 0, sizeof(struct list));
 231         mp_list->cmp = bgp_info_mpath_cmp;
 232 }
 233
 234 /*
 235  * bgp_mp_list_clear
 236  *
 237  * Clears all entries out of the mp_list
 238  */
 239 void bgp_mp_list_clear(struct list *mp_list)
 240 {
 241         assert(mp_list);
 242         list_delete_all_node(mp_list);
 243 }
 244
 245 /*
 246  * bgp_mp_list_add
 247  *
 248  * Adds a multipath entry to the mp_list
 249  */
 250 void bgp_mp_list_add(struct list *mp_list, struct bgp_path_info *mpinfo)
 251 {
 252         assert(mp_list && mpinfo);
 253         listnode_add_sort(mp_list, mpinfo);
 254 }
 255
 256 /*
 257  * bgp_info_mpath_new
 258  *
 259  * Allocate and zero memory for a new bgp_info_mpath element
 260  */
 261 static struct bgp_path_info_mpath *bgp_info_mpath_new(void)
 262 {
 263         struct bgp_path_info_mpath *new_mpath;
 264         new_mpath = XCALLOC(MTYPE_BGP_MPATH_INFO,
 265                             sizeof(struct bgp_path_info_mpath));
 266         return new_mpath;
 267 }
 268
 269 /*
 270  * bgp_info_mpath_free
 271  *
 272  * Release resources for a bgp_info_mpath element and zero out pointer
 273  */
 274 void bgp_info_mpath_free(struct bgp_path_info_mpath **mpath)
 275 {
 276         if (mpath && *mpath) {
 277                 if ((*mpath)->mp_attr)
 278                         bgp_attr_unintern(&(*mpath)->mp_attr);
 279                 XFREE(MTYPE_BGP_MPATH_INFO, *mpath);
 280                 *mpath = NULL;
 281         }
 282 }
 283
 284 /*
 285  * bgp_info_mpath_get
 286  *
 287  * Fetch the mpath element for the given bgp_info. Used for
 288  * doing lazy allocation.
 289  */
 290 static struct bgp_path_info_mpath *
 291 bgp_info_mpath_get(struct bgp_path_info *binfo)
 292 {
 293         struct bgp_path_info_mpath *mpath;
 294         if (!binfo->mpath) {
 295                 mpath = bgp_info_mpath_new();
 296                 if (!mpath)
 297                         return NULL;
 298                 binfo->mpath = mpath;
 299                 mpath->mp_info = binfo;
 300         }
 301         return binfo->mpath;
 302 }
 303
 304 /*
 305  * bgp_info_mpath_enqueue
 306  *
 307  * Enqueue a path onto the multipath list given the previous multipath
 308  * list entry
 309  */
 310 static void bgp_info_mpath_enqueue(struct bgp_path_info *prev_info,
 311                                    struct bgp_path_info *binfo)
 312 {
 313         struct bgp_path_info_mpath *prev, *mpath;
 314
 315         prev = bgp_info_mpath_get(prev_info);
 316         mpath = bgp_info_mpath_get(binfo);
 317         if (!prev || !mpath)
 318                 return;
 319
 320         mpath->mp_next = prev->mp_next;
 321         mpath->mp_prev = prev;
 322         if (prev->mp_next)
 323                 prev->mp_next->mp_prev = mpath;
 324         prev->mp_next = mpath;
 325
 326         SET_FLAG(binfo->flags, BGP_PATH_MULTIPATH);
 327 }
 328
 329 /*
 330  * bgp_info_mpath_dequeue
 331  *
 332  * Remove a path from the multipath list
 333  */
 334 void bgp_info_mpath_dequeue(struct bgp_path_info *binfo)
 335 {
 336         struct bgp_path_info_mpath *mpath = binfo->mpath;
 337         if (!mpath)
 338                 return;
 339         if (mpath->mp_prev)
 340                 mpath->mp_prev->mp_next = mpath->mp_next;
 341         if (mpath->mp_next)
 342                 mpath->mp_next->mp_prev = mpath->mp_prev;
 343         mpath->mp_next = mpath->mp_prev = NULL;
 344         UNSET_FLAG(binfo->flags, BGP_PATH_MULTIPATH);
 345 }
 346
 347 /*
 348  * bgp_info_mpath_next
 349  *
 350  * Given a bgp_info, return the next multipath entry
 351  */
 352 struct bgp_path_info *bgp_info_mpath_next(struct bgp_path_info *binfo)
 353 {
 354         if (!binfo->mpath || !binfo->mpath->mp_next)
 355                 return NULL;
 356         return binfo->mpath->mp_next->mp_info;
 357 }
 358
 359 /*
 360  * bgp_info_mpath_first
 361  *
 362  * Given bestpath bgp_info, return the first multipath entry.
 363  */
 364 struct bgp_path_info *bgp_info_mpath_first(struct bgp_path_info *binfo)
 365 {
 366         return bgp_info_mpath_next(binfo);
 367 }
 368
 369 /*
 370  * bgp_info_mpath_count
 371  *
 372  * Given the bestpath bgp_info, return the number of multipath entries
 373  */
 374 uint32_t bgp_info_mpath_count(struct bgp_path_info *binfo)
 375 {
 376         if (!binfo->mpath)
 377                 return 0;
 378         return binfo->mpath->mp_count;
 379 }
 380
 381 /*
 382  * bgp_info_mpath_count_set
 383  *
 384  * Sets the count of multipaths into bestpath's mpath element
 385  */
 386 static void bgp_info_mpath_count_set(struct bgp_path_info *binfo,
 387                                      uint32_t count)
 388 {
 389         struct bgp_path_info_mpath *mpath;
 390         if (!count && !binfo->mpath)
 391                 return;
 392         mpath = bgp_info_mpath_get(binfo);
 393         if (!mpath)
 394                 return;
 395         mpath->mp_count = count;
 396 }
 397
 398 /*
 399  * bgp_info_mpath_attr
 400  *
 401  * Given bestpath bgp_info, return aggregated attribute set used
 402  * for advertising the multipath route
 403  */
 404 struct attr *bgp_info_mpath_attr(struct bgp_path_info *binfo)
 405 {
 406         if (!binfo->mpath)
 407                 return NULL;
 408         return binfo->mpath->mp_attr;
 409 }
 410
 411 /*
 412  * bgp_info_mpath_attr_set
 413  *
 414  * Sets the aggregated attribute into bestpath's mpath element
 415  */
 416 static void bgp_info_mpath_attr_set(struct bgp_path_info *binfo,
 417                                     struct attr *attr)
 418 {
 419         struct bgp_path_info_mpath *mpath;
 420         if (!attr && !binfo->mpath)
 421                 return;
 422         mpath = bgp_info_mpath_get(binfo);
 423         if (!mpath)
 424                 return;
 425         mpath->mp_attr = attr;
 426 }
 427
 428 /*
 429  * bgp_info_mpath_update
 430  *
 431  * Compare and sync up the multipath list with the mp_list generated by
 432  * bgp_best_selection
 433  */
 434 void bgp_info_mpath_update(struct bgp_node *rn, struct bgp_path_info *new_best,
 435                            struct bgp_path_info *old_best, struct list *mp_list,
 436                            struct bgp_maxpaths_cfg *mpath_cfg)
 437 {
 438         uint16_t maxpaths, mpath_count, old_mpath_count;
 439         struct listnode *mp_node, *mp_next_node;
 440         struct bgp_path_info *cur_mpath, *new_mpath, *next_mpath, *prev_mpath;
 441         int mpath_changed, debug;
 442         char pfx_buf[PREFIX2STR_BUFFER], nh_buf[2][INET6_ADDRSTRLEN];
 443         char path_buf[PATH_ADDPATH_STR_BUFFER];
 444
 445         mpath_changed = 0;
 446         maxpaths = multipath_num;
 447         mpath_count = 0;
 448         cur_mpath = NULL;
 449         old_mpath_count = 0;
 450         prev_mpath = new_best;
 451         mp_node = listhead(mp_list);
 452         debug = bgp_debug_bestpath(&rn->p);
 453
 454         if (debug)
 455                 prefix2str(&rn->p, pfx_buf, sizeof(pfx_buf));
 456
 457         if (new_best) {
 458                 mpath_count++;
 459                 if (new_best != old_best)
 460                         bgp_info_mpath_dequeue(new_best);
 461                 maxpaths = (new_best->peer->sort == BGP_PEER_IBGP)
 462                                    ? mpath_cfg->maxpaths_ibgp
 463                                    : mpath_cfg->maxpaths_ebgp;
 464         }
 465
 466         if (old_best) {
 467                 cur_mpath = bgp_info_mpath_first(old_best);
 468                 old_mpath_count = bgp_info_mpath_count(old_best);
 469                 bgp_info_mpath_count_set(old_best, 0);
 470                 bgp_info_mpath_dequeue(old_best);
 471         }
 472
 473         if (debug)
 474                 zlog_debug(
 475                         "%s: starting mpath update, newbest %s num candidates %d old-mpath-count %d",
 476                         pfx_buf, new_best ? new_best->peer->host : "NONE",
 477                         mp_list ? listcount(mp_list) : 0, old_mpath_count);
 478
 479         /*
 480          * We perform an ordered walk through both lists in parallel.
 481          * The reason for the ordered walk is that if there are paths
 482          * that were previously multipaths and are still multipaths, the walk
 483          * should encounter them in both lists at the same time. Otherwise
 484          * there will be paths that are in one list or another, and we
 485          * will deal with these separately.
 486          *
 487          * Note that new_best might be somewhere in the mp_list, so we need
 488          * to skip over it
 489          */
 490         while (mp_node || cur_mpath) {
 491                 struct bgp_path_info *tmp_info;
 492
 493                 /*
 494                  * We can bail out of this loop if all existing paths on the
 495                  * multipath list have been visited (for cleanup purposes) and
 496                  * the maxpath requirement is fulfulled
 497                  */
 498                 if (!cur_mpath && (mpath_count >= maxpaths))
 499                         break;
 500
 501                 mp_next_node = mp_node ? listnextnode(mp_node) : NULL;
 502                 next_mpath = cur_mpath ? bgp_info_mpath_next(cur_mpath) : NULL;
 503                 tmp_info = mp_node ? listgetdata(mp_node) : NULL;
 504
 505                 if (debug)
 506                         zlog_debug(
 507                                 "%s: comparing candidate %s with existing mpath %s",
 508                                 pfx_buf,
 509                                 tmp_info ? tmp_info->peer->host : "NONE",
 510                                 cur_mpath ? cur_mpath->peer->host : "NONE");
 511
 512                 /*
 513                  * If equal, the path was a multipath and is still a multipath.
 514                  * Insert onto new multipath list if maxpaths allows.
 515                  */
 516                 if (mp_node && (listgetdata(mp_node) == cur_mpath)) {
 517                         list_delete_node(mp_list, mp_node);
 518                         bgp_info_mpath_dequeue(cur_mpath);
 519                         if ((mpath_count < maxpaths)
 520                             && bgp_info_nexthop_cmp(prev_mpath, cur_mpath)) {
 521                                 bgp_info_mpath_enqueue(prev_mpath, cur_mpath);
 522                                 prev_mpath = cur_mpath;
 523                                 mpath_count++;
 524                                 if (debug) {
 525                                         bgp_info_path_with_addpath_rx_str(
 526                                                 cur_mpath, path_buf);
 527                                         zlog_debug(
 528                                                 "%s: %s is still multipath, cur count %d",
 529                                                 pfx_buf, path_buf, mpath_count);
 530                                 }
 531                         } else {
 532                                 mpath_changed = 1;
 533                                 if (debug) {
 534                                         bgp_info_path_with_addpath_rx_str(
 535                                                 cur_mpath, path_buf);
 536                                         zlog_debug(
 537                                                 "%s: remove mpath %s nexthop %s, cur count %d",
 538                                                 pfx_buf, path_buf,
 539                                                 inet_ntop(AF_INET,
 540                                                           &cur_mpath->attr
 541                                                                    ->nexthop,
 542                                                           nh_buf[0],
 543                                                           sizeof(nh_buf[0])),
 544                                                 mpath_count);
 545                                 }
 546                         }
 547                         mp_node = mp_next_node;
 548                         cur_mpath = next_mpath;
 549                         continue;
 550                 }
 551
 552                 if (cur_mpath
 553                     && (!mp_node
 554                         || (bgp_info_mpath_cmp(cur_mpath, listgetdata(mp_node))
 555                             < 0))) {
 556                         /*
 557                          * If here, we have an old multipath and either the
 558                          * mp_list
 559                          * is finished or the next mp_node points to a later
 560                          * multipath, so we need to purge this path from the
 561                          * multipath list
 562                          */
 563                         bgp_info_mpath_dequeue(cur_mpath);
 564                         mpath_changed = 1;
 565                         if (debug) {
 566                                 bgp_info_path_with_addpath_rx_str(cur_mpath,
 567                                                                   path_buf);
 568                                 zlog_debug(
 569                                         "%s: remove mpath %s nexthop %s, cur count %d",
 570                                         pfx_buf, path_buf,
 571                                         inet_ntop(AF_INET,
 572                                                   &cur_mpath->attr->nexthop,
 573                                                   nh_buf[0], sizeof(nh_buf[0])),
 574                                         mpath_count);
 575                         }
 576                         cur_mpath = next_mpath;
 577                 } else {
 578                         /*
 579                          * If here, we have a path on the mp_list that was not
 580                          * previously
 581                          * a multipath (due to non-equivalance or maxpaths
 582                          * exceeded),
 583                          * or the matching multipath is sorted later in the
 584                          * multipath
 585                          * list. Before we enqueue the path on the new multipath
 586                          * list,
 587                          * make sure its not on the old_best multipath list or
 588                          * referenced
 589                          * via next_mpath:
 590                          * - If next_mpath points to this new path, update
 591                          * next_mpath to
 592                          *   point to the multipath after this one
 593                          * - Dequeue the path from the multipath list just to
 594                          * make sure
 595                          */
 596                         new_mpath = listgetdata(mp_node);
 597                         list_delete_node(mp_list, mp_node);
 598                         assert(new_mpath);
 599                         assert(prev_mpath);
 600                         if ((mpath_count < maxpaths) && (new_mpath != new_best)
 601                             && bgp_info_nexthop_cmp(prev_mpath, new_mpath)) {
 602                                 if (new_mpath == next_mpath)
 603                                         bgp_info_mpath_next(new_mpath);
 604                                 bgp_info_mpath_dequeue(new_mpath);
 605
 606                                 bgp_info_mpath_enqueue(prev_mpath, new_mpath);
 607                                 prev_mpath = new_mpath;
 608                                 mpath_changed = 1;
 609                                 mpath_count++;
 610                                 if (debug) {
 611                                         bgp_info_path_with_addpath_rx_str(
 612                                                 new_mpath, path_buf);
 613                                         zlog_debug(
 614                                                 "%s: add mpath %s nexthop %s, cur count %d",
 615                                                 pfx_buf, path_buf,
 616                                                 inet_ntop(AF_INET,
 617                                                           &new_mpath->attr
 618                                                                    ->nexthop,
 619                                                           nh_buf[0],
 620                                                           sizeof(nh_buf[0])),
 621                                                 mpath_count);
 622                                 }
 623                         }
 624                         mp_node = mp_next_node;
 625                 }
 626         }
 627
 628         if (new_best) {
 629                 if (debug)
 630                         zlog_debug(
 631                                 "%s: New mpath count (incl newbest) %d mpath-change %s",
 632                                 pfx_buf, mpath_count,
 633                                 mpath_changed ? "YES" : "NO");
 634
 635                 bgp_info_mpath_count_set(new_best, mpath_count - 1);
 636                 if (mpath_changed
 637                     || (bgp_info_mpath_count(new_best) != old_mpath_count))
 638                         SET_FLAG(new_best->flags, BGP_PATH_MULTIPATH_CHG);
 639         }
 640 }
 641
 642 /*
 643  * bgp_mp_dmed_deselect
 644  *
 645  * Clean up multipath information for BGP_PATH_DMED_SELECTED path that
 646  * is not selected as best path
 647  */
 648 void bgp_mp_dmed_deselect(struct bgp_path_info *dmed_best)
 649 {
 650         struct bgp_path_info *mpinfo, *mpnext;
 651
 652         if (!dmed_best)
 653                 return;
 654
 655         for (mpinfo = bgp_info_mpath_first(dmed_best); mpinfo;
 656              mpinfo = mpnext) {
 657                 mpnext = bgp_info_mpath_next(mpinfo);
 658                 bgp_info_mpath_dequeue(mpinfo);
 659         }
 660
 661         bgp_info_mpath_count_set(dmed_best, 0);
 662         UNSET_FLAG(dmed_best->flags, BGP_PATH_MULTIPATH_CHG);
 663         assert(bgp_info_mpath_first(dmed_best) == 0);
 664 }
 665
 666 /*
 667  * bgp_info_mpath_aggregate_update
 668  *
 669  * Set the multipath aggregate attribute. We need to see if the
 670  * aggregate has changed and then set the ATTR_CHANGED flag on the
 671  * bestpath info so that a peer update will be generated. The
 672  * change is detected by generating the current attribute,
 673  * interning it, and then comparing the interned pointer with the
 674  * current value. We can skip this generate/compare step if there
 675  * is no change in multipath selection and no attribute change in
 676  * any multipath.
 677  */
 678 void bgp_info_mpath_aggregate_update(struct bgp_path_info *new_best,
 679                                      struct bgp_path_info *old_best)
 680 {
 681         struct bgp_path_info *mpinfo;
 682         struct aspath *aspath;
 683         struct aspath *asmerge;
 684         struct attr *new_attr, *old_attr;
 685         uint8_t origin;
 686         struct community *community, *commerge;
 687         struct ecommunity *ecomm, *ecommerge;
 688         struct lcommunity *lcomm, *lcommerge;
 689         struct attr attr = {0};
 690
 691         if (old_best && (old_best != new_best)
 692             && (old_attr = bgp_info_mpath_attr(old_best))) {
 693                 bgp_attr_unintern(&old_attr);
 694                 bgp_info_mpath_attr_set(old_best, NULL);
 695         }
 696
 697         if (!new_best)
 698                 return;
 699
 700         if (!bgp_info_mpath_count(new_best)) {
 701                 if ((new_attr = bgp_info_mpath_attr(new_best))) {
 702                         bgp_attr_unintern(&new_attr);
 703                         bgp_info_mpath_attr_set(new_best, NULL);
 704                         SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 705                 }
 706                 return;
 707         }
 708
 709         bgp_attr_dup(&attr, new_best->attr);
 710
 711         if (new_best->peer && bgp_flag_check(new_best->peer->bgp,
 712                                              BGP_FLAG_MULTIPATH_RELAX_AS_SET)) {
 713
 714                 /* aggregate attribute from multipath constituents */
 715                 aspath = aspath_dup(attr.aspath);
 716                 origin = attr.origin;
 717                 community =
 718                         attr.community ? community_dup(attr.community) : NULL;
 719                 ecomm = (attr.ecommunity) ? ecommunity_dup(attr.ecommunity)
 720                                           : NULL;
 721                 lcomm = (attr.lcommunity) ? lcommunity_dup(attr.lcommunity)
 722                                           : NULL;
 723
 724                 for (mpinfo = bgp_info_mpath_first(new_best); mpinfo;
 725                      mpinfo = bgp_info_mpath_next(mpinfo)) {
 726                         asmerge =
 727                                 aspath_aggregate(aspath, mpinfo->attr->aspath);
 728                         aspath_free(aspath);
 729                         aspath = asmerge;
 730
 731                         if (origin < mpinfo->attr->origin)
 732                                 origin = mpinfo->attr->origin;
 733
 734                         if (mpinfo->attr->community) {
 735                                 if (community) {
 736                                         commerge = community_merge(
 737                                                 community,
 738                                                 mpinfo->attr->community);
 739                                         community =
 740                                                 community_uniq_sort(commerge);
 741                                         community_free(commerge);
 742                                 } else
 743                                         community = community_dup(
 744                                                 mpinfo->attr->community);
 745                         }
 746
 747                         if (mpinfo->attr->ecommunity) {
 748                                 if (ecomm) {
 749                                         ecommerge = ecommunity_merge(
 750                                                 ecomm,
 751                                                 mpinfo->attr->ecommunity);
 752                                         ecomm = ecommunity_uniq_sort(ecommerge);
 753                                         ecommunity_free(&ecommerge);
 754                                 } else
 755                                         ecomm = ecommunity_dup(
 756                                                 mpinfo->attr->ecommunity);
 757                         }
 758                         if (mpinfo->attr->lcommunity) {
 759                                 if (lcomm) {
 760                                         lcommerge = lcommunity_merge(
 761                                                 lcomm,
 762                                                 mpinfo->attr->lcommunity);
 763                                         lcomm = lcommunity_uniq_sort(lcommerge);
 764                                         lcommunity_free(&lcommerge);
 765                                 } else
 766                                         lcomm = lcommunity_dup(
 767                                                 mpinfo->attr->lcommunity);
 768                         }
 769                 }
 770
 771                 attr.aspath = aspath;
 772                 attr.origin = origin;
 773                 if (community) {
 774                         attr.community = community;
 775                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_COMMUNITIES);
 776                 }
 777                 if (ecomm) {
 778                         attr.ecommunity = ecomm;
 779                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_EXT_COMMUNITIES);
 780                 }
 781                 if (lcomm) {
 782                         attr.lcommunity = lcomm;
 783                         attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_LARGE_COMMUNITIES);
 784                 }
 785
 786                 /* Zap multipath attr nexthop so we set nexthop to self */
 787                 attr.nexthop.s_addr = 0;
 788                 memset(&attr.mp_nexthop_global, 0, sizeof(struct in6_addr));
 789
 790                 /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */
 791         }
 792
 793         new_attr = bgp_attr_intern(&attr);
 794
 795         if (new_attr != bgp_info_mpath_attr(new_best)) {
 796                 if ((old_attr = bgp_info_mpath_attr(new_best)))
 797                         bgp_attr_unintern(&old_attr);
 798                 bgp_info_mpath_attr_set(new_best, new_attr);
 799                 SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 800         } else
 801                 bgp_attr_unintern(&new_attr);
 802 }