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, bool same_clusterlen)
  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].same_clusterlen = same_clusterlen;
  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].same_clusterlen = false;
  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         /*
 185          * If both nexthops are same then check
 186          * if they belong to same VRF
 187          */
 188         if (!compare && bpi1->attr->nh_type != NEXTHOP_TYPE_BLACKHOLE) {
 189                 if (bpi1->extra && bpi1->extra->bgp_orig && bpi2->extra
 190                     && bpi2->extra->bgp_orig) {
 191                         if (bpi1->extra->bgp_orig->vrf_id
 192                             != bpi2->extra->bgp_orig->vrf_id) {
 193                                 compare = 1;
 194                         }
 195                 }
 196         }
 197
 198         return compare;
 199 }
 200
 201 /*
 202  * bgp_path_info_mpath_cmp
 203  *
 204  * This function determines our multipath list ordering. By ordering
 205  * the list we can deterministically select which paths are included
 206  * in the multipath set. The ordering also helps in detecting changes
 207  * in the multipath selection so we can detect whether to send an
 208  * update to zebra.
 209  *
 210  * The order of paths is determined first by received nexthop, and then
 211  * by peer address if the nexthops are the same.
 212  */
 213 static int bgp_path_info_mpath_cmp(void *val1, void *val2)
 214 {
 215         struct bgp_path_info *bpi1, *bpi2;
 216         int compare;
 217
 218         bpi1 = val1;
 219         bpi2 = val2;
 220
 221         compare = bgp_path_info_nexthop_cmp(bpi1, bpi2);
 222
 223         if (!compare) {
 224                 if (!bpi1->peer->su_remote && !bpi2->peer->su_remote)
 225                         compare = 0;
 226                 else if (!bpi1->peer->su_remote)
 227                         compare = 1;
 228                 else if (!bpi2->peer->su_remote)
 229                         compare = -1;
 230                 else
 231                         compare = sockunion_cmp(bpi1->peer->su_remote,
 232                                                 bpi2->peer->su_remote);
 233         }
 234
 235         return compare;
 236 }
 237
 238 /*
 239  * bgp_mp_list_init
 240  *
 241  * Initialize the mp_list, which holds the list of multipaths
 242  * selected by bgp_best_selection
 243  */
 244 void bgp_mp_list_init(struct list *mp_list)
 245 {
 246         assert(mp_list);
 247         memset(mp_list, 0, sizeof(struct list));
 248         mp_list->cmp = bgp_path_info_mpath_cmp;
 249 }
 250
 251 /*
 252  * bgp_mp_list_clear
 253  *
 254  * Clears all entries out of the mp_list
 255  */
 256 void bgp_mp_list_clear(struct list *mp_list)
 257 {
 258         assert(mp_list);
 259         list_delete_all_node(mp_list);
 260 }
 261
 262 /*
 263  * bgp_mp_list_add
 264  *
 265  * Adds a multipath entry to the mp_list
 266  */
 267 void bgp_mp_list_add(struct list *mp_list, struct bgp_path_info *mpinfo)
 268 {
 269         assert(mp_list && mpinfo);
 270         listnode_add_sort(mp_list, mpinfo);
 271 }
 272
 273 /*
 274  * bgp_path_info_mpath_new
 275  *
 276  * Allocate and zero memory for a new bgp_path_info_mpath element
 277  */
 278 static struct bgp_path_info_mpath *bgp_path_info_mpath_new(void)
 279 {
 280         struct bgp_path_info_mpath *new_mpath;
 281         new_mpath = XCALLOC(MTYPE_BGP_MPATH_INFO,
 282                             sizeof(struct bgp_path_info_mpath));
 283         return new_mpath;
 284 }
 285
 286 /*
 287  * bgp_path_info_mpath_free
 288  *
 289  * Release resources for a bgp_path_info_mpath element and zero out pointer
 290  */
 291 void bgp_path_info_mpath_free(struct bgp_path_info_mpath **mpath)
 292 {
 293         if (mpath && *mpath) {
 294                 if ((*mpath)->mp_attr)
 295                         bgp_attr_unintern(&(*mpath)->mp_attr);
 296                 XFREE(MTYPE_BGP_MPATH_INFO, *mpath);
 297         }
 298 }
 299
 300 /*
 301  * bgp_path_info_mpath_get
 302  *
 303  * Fetch the mpath element for the given bgp_path_info. Used for
 304  * doing lazy allocation.
 305  */
 306 static struct bgp_path_info_mpath *
 307 bgp_path_info_mpath_get(struct bgp_path_info *path)
 308 {
 309         struct bgp_path_info_mpath *mpath;
 310
 311         if (!path)
 312                 return NULL;
 313
 314         if (!path->mpath) {
 315                 mpath = bgp_path_info_mpath_new();
 316                 if (!mpath)
 317                         return NULL;
 318                 path->mpath = mpath;
 319                 mpath->mp_info = path;
 320         }
 321         return path->mpath;
 322 }
 323
 324 /*
 325  * bgp_path_info_mpath_enqueue
 326  *
 327  * Enqueue a path onto the multipath list given the previous multipath
 328  * list entry
 329  */
 330 static void bgp_path_info_mpath_enqueue(struct bgp_path_info *prev_info,
 331                                         struct bgp_path_info *path)
 332 {
 333         struct bgp_path_info_mpath *prev, *mpath;
 334
 335         prev = bgp_path_info_mpath_get(prev_info);
 336         mpath = bgp_path_info_mpath_get(path);
 337         if (!prev || !mpath)
 338                 return;
 339
 340         mpath->mp_next = prev->mp_next;
 341         mpath->mp_prev = prev;
 342         if (prev->mp_next)
 343                 prev->mp_next->mp_prev = mpath;
 344         prev->mp_next = mpath;
 345
 346         SET_FLAG(path->flags, BGP_PATH_MULTIPATH);
 347 }
 348
 349 /*
 350  * bgp_path_info_mpath_dequeue
 351  *
 352  * Remove a path from the multipath list
 353  */
 354 void bgp_path_info_mpath_dequeue(struct bgp_path_info *path)
 355 {
 356         struct bgp_path_info_mpath *mpath = path->mpath;
 357         if (!mpath)
 358                 return;
 359         if (mpath->mp_prev)
 360                 mpath->mp_prev->mp_next = mpath->mp_next;
 361         if (mpath->mp_next)
 362                 mpath->mp_next->mp_prev = mpath->mp_prev;
 363         mpath->mp_next = mpath->mp_prev = NULL;
 364         UNSET_FLAG(path->flags, BGP_PATH_MULTIPATH);
 365 }
 366
 367 /*
 368  * bgp_path_info_mpath_next
 369  *
 370  * Given a bgp_path_info, return the next multipath entry
 371  */
 372 struct bgp_path_info *bgp_path_info_mpath_next(struct bgp_path_info *path)
 373 {
 374         if (!path->mpath || !path->mpath->mp_next)
 375                 return NULL;
 376         return path->mpath->mp_next->mp_info;
 377 }
 378
 379 /*
 380  * bgp_path_info_mpath_first
 381  *
 382  * Given bestpath bgp_path_info, return the first multipath entry.
 383  */
 384 struct bgp_path_info *bgp_path_info_mpath_first(struct bgp_path_info *path)
 385 {
 386         return bgp_path_info_mpath_next(path);
 387 }
 388
 389 /*
 390  * bgp_path_info_mpath_count
 391  *
 392  * Given the bestpath bgp_path_info, return the number of multipath entries
 393  */
 394 uint32_t bgp_path_info_mpath_count(struct bgp_path_info *path)
 395 {
 396         if (!path->mpath)
 397                 return 0;
 398         return path->mpath->mp_count;
 399 }
 400
 401 /*
 402  * bgp_path_info_mpath_count_set
 403  *
 404  * Sets the count of multipaths into bestpath's mpath element
 405  */
 406 static void bgp_path_info_mpath_count_set(struct bgp_path_info *path,
 407                                           uint16_t count)
 408 {
 409         struct bgp_path_info_mpath *mpath;
 410         if (!count && !path->mpath)
 411                 return;
 412         mpath = bgp_path_info_mpath_get(path);
 413         if (!mpath)
 414                 return;
 415         mpath->mp_count = count;
 416 }
 417
 418 /*
 419  * bgp_path_info_mpath_lb_update
 420  *
 421  * Update cumulative info related to link-bandwidth
 422  */
 423 static void bgp_path_info_mpath_lb_update(struct bgp_path_info *path, bool set,
 424                                           bool all_paths_lb, uint64_t cum_bw)
 425 {
 426         struct bgp_path_info_mpath *mpath;
 427
 428         mpath = path->mpath;
 429         if (mpath == NULL) {
 430                 if (!set || (cum_bw == 0 && !all_paths_lb))
 431                         return;
 432
 433                 mpath = bgp_path_info_mpath_get(path);
 434                 if (!mpath)
 435                         return;
 436         }
 437         if (set) {
 438                 if (cum_bw)
 439                         SET_FLAG(mpath->mp_flags, BGP_MP_LB_PRESENT);
 440                 else
 441                         UNSET_FLAG(mpath->mp_flags, BGP_MP_LB_PRESENT);
 442                 if (all_paths_lb)
 443                         SET_FLAG(mpath->mp_flags, BGP_MP_LB_ALL);
 444                 else
 445                         UNSET_FLAG(mpath->mp_flags, BGP_MP_LB_ALL);
 446                 mpath->cum_bw = cum_bw;
 447         } else {
 448                 mpath->mp_flags = 0;
 449                 mpath->cum_bw = 0;
 450         }
 451 }
 452
 453 /*
 454  * bgp_path_info_mpath_attr
 455  *
 456  * Given bestpath bgp_path_info, return aggregated attribute set used
 457  * for advertising the multipath route
 458  */
 459 struct attr *bgp_path_info_mpath_attr(struct bgp_path_info *path)
 460 {
 461         if (!path->mpath)
 462                 return NULL;
 463         return path->mpath->mp_attr;
 464 }
 465
 466 /*
 467  * bgp_path_info_chkwtd
 468  *
 469  * Return if we should attempt to do weighted ECMP or not
 470  * The path passed in is the bestpath.
 471  */
 472 bool bgp_path_info_mpath_chkwtd(struct bgp *bgp, struct bgp_path_info *path)
 473 {
 474         /* Check if told to ignore weights or not multipath */
 475         if (bgp->lb_handling == BGP_LINK_BW_IGNORE_BW || !path->mpath)
 476                 return false;
 477
 478         /* All paths in multipath should have associated weight (bandwidth)
 479          * unless told explicitly otherwise.
 480          */
 481         if (bgp->lb_handling != BGP_LINK_BW_SKIP_MISSING &&
 482             bgp->lb_handling != BGP_LINK_BW_DEFWT_4_MISSING)
 483                 return (path->mpath->mp_flags & BGP_MP_LB_ALL);
 484
 485         /* At least one path should have bandwidth. */
 486         return (path->mpath->mp_flags & BGP_MP_LB_PRESENT);
 487 }
 488
 489 /*
 490  * bgp_path_info_mpath_attr
 491  *
 492  * Given bestpath bgp_path_info, return cumulative bandwidth
 493  * computed for all multipaths with bandwidth info
 494  */
 495 uint64_t bgp_path_info_mpath_cumbw(struct bgp_path_info *path)
 496 {
 497         if (!path->mpath)
 498                 return 0;
 499         return path->mpath->cum_bw;
 500 }
 501
 502 /*
 503  * bgp_path_info_mpath_attr_set
 504  *
 505  * Sets the aggregated attribute into bestpath's mpath element
 506  */
 507 static void bgp_path_info_mpath_attr_set(struct bgp_path_info *path,
 508                                          struct attr *attr)
 509 {
 510         struct bgp_path_info_mpath *mpath;
 511         if (!attr && !path->mpath)
 512                 return;
 513         mpath = bgp_path_info_mpath_get(path);
 514         if (!mpath)
 515                 return;
 516         mpath->mp_attr = attr;
 517 }
 518
 519 /*
 520  * bgp_path_info_mpath_update
 521  *
 522  * Compare and sync up the multipath list with the mp_list generated by
 523  * bgp_best_selection
 524  */
 525 void bgp_path_info_mpath_update(struct bgp *bgp, struct bgp_dest *dest,
 526                                 struct bgp_path_info *new_best,
 527                                 struct bgp_path_info *old_best,
 528                                 struct list *mp_list,
 529                                 struct bgp_maxpaths_cfg *mpath_cfg)
 530 {
 531         uint16_t maxpaths, mpath_count, old_mpath_count;
 532         uint32_t bwval;
 533         uint64_t cum_bw, old_cum_bw;
 534         struct listnode *mp_node, *mp_next_node;
 535         struct bgp_path_info *cur_mpath, *new_mpath, *next_mpath, *prev_mpath;
 536         int mpath_changed, debug;
 537         bool all_paths_lb;
 538         char path_buf[PATH_ADDPATH_STR_BUFFER];
 539
 540         mpath_changed = 0;
 541         maxpaths = multipath_num;
 542         mpath_count = 0;
 543         cur_mpath = NULL;
 544         old_mpath_count = 0;
 545         old_cum_bw = cum_bw = 0;
 546         prev_mpath = new_best;
 547         mp_node = listhead(mp_list);
 548         debug = bgp_debug_bestpath(dest);
 549
 550         if (new_best) {
 551                 mpath_count++;
 552                 if (new_best != old_best)
 553                         bgp_path_info_mpath_dequeue(new_best);
 554                 maxpaths = (new_best->peer->sort == BGP_PEER_IBGP)
 555                                    ? mpath_cfg->maxpaths_ibgp
 556                                    : mpath_cfg->maxpaths_ebgp;
 557         }
 558
 559         if (old_best) {
 560                 cur_mpath = bgp_path_info_mpath_first(old_best);
 561                 old_mpath_count = bgp_path_info_mpath_count(old_best);
 562                 old_cum_bw = bgp_path_info_mpath_cumbw(old_best);
 563                 bgp_path_info_mpath_count_set(old_best, 0);
 564                 bgp_path_info_mpath_lb_update(old_best, false, false, 0);
 565                 bgp_path_info_mpath_dequeue(old_best);
 566         }
 567
 568         if (debug)
 569                 zlog_debug(
 570                         "%pRN(%s): starting mpath update, newbest %s num candidates %d old-mpath-count %d old-cum-bw u%" PRIu64,
 571                         bgp_dest_to_rnode(dest), bgp->name_pretty,
 572                         new_best ? new_best->peer->host : "NONE",
 573                         mp_list ? listcount(mp_list) : 0, old_mpath_count,
 574                         old_cum_bw);
 575
 576         /*
 577          * We perform an ordered walk through both lists in parallel.
 578          * The reason for the ordered walk is that if there are paths
 579          * that were previously multipaths and are still multipaths, the walk
 580          * should encounter them in both lists at the same time. Otherwise
 581          * there will be paths that are in one list or another, and we
 582          * will deal with these separately.
 583          *
 584          * Note that new_best might be somewhere in the mp_list, so we need
 585          * to skip over it
 586          */
 587         all_paths_lb = true; /* We'll reset if any path doesn't have LB. */
 588         while (mp_node || cur_mpath) {
 589                 struct bgp_path_info *tmp_info;
 590
 591                 /*
 592                  * We can bail out of this loop if all existing paths on the
 593                  * multipath list have been visited (for cleanup purposes) and
 594                  * the maxpath requirement is fulfulled
 595                  */
 596                 if (!cur_mpath && (mpath_count >= maxpaths))
 597                         break;
 598
 599                 mp_next_node = mp_node ? listnextnode(mp_node) : NULL;
 600                 next_mpath =
 601                         cur_mpath ? bgp_path_info_mpath_next(cur_mpath) : NULL;
 602                 tmp_info = mp_node ? listgetdata(mp_node) : NULL;
 603
 604                 if (debug)
 605                         zlog_debug(
 606                                 "%pRN(%s): comparing candidate %s with existing mpath %s",
 607                                 bgp_dest_to_rnode(dest), bgp->name_pretty,
 608                                 tmp_info ? tmp_info->peer->host : "NONE",
 609                                 cur_mpath ? cur_mpath->peer->host : "NONE");
 610
 611                 /*
 612                  * If equal, the path was a multipath and is still a multipath.
 613                  * Insert onto new multipath list if maxpaths allows.
 614                  */
 615                 if (mp_node && (listgetdata(mp_node) == cur_mpath)) {
 616                         list_delete_node(mp_list, mp_node);
 617                         bgp_path_info_mpath_dequeue(cur_mpath);
 618                         if ((mpath_count < maxpaths)
 619                             && prev_mpath
 620                             && bgp_path_info_nexthop_cmp(prev_mpath,
 621                                                          cur_mpath)) {
 622                                 bgp_path_info_mpath_enqueue(prev_mpath,
 623                                                             cur_mpath);
 624                                 prev_mpath = cur_mpath;
 625                                 mpath_count++;
 626                                 if (ecommunity_linkbw_present(
 627                                             bgp_attr_get_ecommunity(
 628                                                     cur_mpath->attr),
 629                                             &bwval))
 630                                         cum_bw += bwval;
 631                                 else
 632                                         all_paths_lb = false;
 633                                 if (debug) {
 634                                         bgp_path_info_path_with_addpath_rx_str(
 635                                                 cur_mpath, path_buf,
 636                                                 sizeof(path_buf));
 637                                         zlog_debug(
 638                                                 "%pRN: %s is still multipath, cur count %d",
 639                                                 bgp_dest_to_rnode(dest),
 640                                                 path_buf, mpath_count);
 641                                 }
 642                         } else {
 643                                 mpath_changed = 1;
 644                                 if (debug) {
 645                                         bgp_path_info_path_with_addpath_rx_str(
 646                                                 cur_mpath, path_buf,
 647                                                 sizeof(path_buf));
 648                                         zlog_debug(
 649                                                 "%pRN: remove mpath %s nexthop %pI4, cur count %d",
 650                                                 bgp_dest_to_rnode(dest),
 651                                                 path_buf,
 652                                                 &cur_mpath->attr->nexthop,
 653                                                 mpath_count);
 654                                 }
 655                         }
 656                         mp_node = mp_next_node;
 657                         cur_mpath = next_mpath;
 658                         continue;
 659                 }
 660
 661                 if (cur_mpath
 662                     && (!mp_node
 663                         || (bgp_path_info_mpath_cmp(cur_mpath,
 664                                                     listgetdata(mp_node))
 665                             < 0))) {
 666                         /*
 667                          * If here, we have an old multipath and either the
 668                          * mp_list
 669                          * is finished or the next mp_node points to a later
 670                          * multipath, so we need to purge this path from the
 671                          * multipath list
 672                          */
 673                         bgp_path_info_mpath_dequeue(cur_mpath);
 674                         mpath_changed = 1;
 675                         if (debug) {
 676                                 bgp_path_info_path_with_addpath_rx_str(
 677                                         cur_mpath, path_buf, sizeof(path_buf));
 678                                 zlog_debug(
 679                                         "%pRN: remove mpath %s nexthop %pI4, cur count %d",
 680                                         bgp_dest_to_rnode(dest), path_buf,
 681                                         &cur_mpath->attr->nexthop, mpath_count);
 682                         }
 683                         cur_mpath = next_mpath;
 684                 } else {
 685                         /*
 686                          * If here, we have a path on the mp_list that was not
 687                          * previously
 688                          * a multipath (due to non-equivalance or maxpaths
 689                          * exceeded),
 690                          * or the matching multipath is sorted later in the
 691                          * multipath
 692                          * list. Before we enqueue the path on the new multipath
 693                          * list,
 694                          * make sure its not on the old_best multipath list or
 695                          * referenced
 696                          * via next_mpath:
 697                          * - If next_mpath points to this new path, update
 698                          * next_mpath to
 699                          *   point to the multipath after this one
 700                          * - Dequeue the path from the multipath list just to
 701                          * make sure
 702                          */
 703                         new_mpath = listgetdata(mp_node);
 704                         list_delete_node(mp_list, mp_node);
 705                         assert(new_mpath);
 706                         assert(prev_mpath);
 707                         if ((mpath_count < maxpaths) && (new_mpath != new_best)
 708                             && bgp_path_info_nexthop_cmp(prev_mpath,
 709                                                          new_mpath)) {
 710                                 bgp_path_info_mpath_dequeue(new_mpath);
 711
 712                                 bgp_path_info_mpath_enqueue(prev_mpath,
 713                                                             new_mpath);
 714                                 prev_mpath = new_mpath;
 715                                 mpath_changed = 1;
 716                                 mpath_count++;
 717                                 if (ecommunity_linkbw_present(
 718                                             bgp_attr_get_ecommunity(
 719                                                     new_mpath->attr),
 720                                             &bwval))
 721                                         cum_bw += bwval;
 722                                 else
 723                                         all_paths_lb = false;
 724                                 if (debug) {
 725                                         bgp_path_info_path_with_addpath_rx_str(
 726                                                 new_mpath, path_buf,
 727                                                 sizeof(path_buf));
 728                                         zlog_debug(
 729                                                 "%pRN: add mpath %s nexthop %pI4, cur count %d",
 730                                                 bgp_dest_to_rnode(dest),
 731                                                 path_buf,
 732                                                 &new_mpath->attr->nexthop,
 733                                                 mpath_count);
 734                                 }
 735                         }
 736                         mp_node = mp_next_node;
 737                 }
 738         }
 739
 740         if (new_best) {
 741                 bgp_path_info_mpath_count_set(new_best, mpath_count - 1);
 742                 if (mpath_count <= 1 ||
 743                     !ecommunity_linkbw_present(
 744                             bgp_attr_get_ecommunity(new_best->attr), &bwval))
 745                         all_paths_lb = false;
 746                 else
 747                         cum_bw += bwval;
 748                 bgp_path_info_mpath_lb_update(new_best, true,
 749                                               all_paths_lb, cum_bw);
 750
 751                 if (debug)
 752                         zlog_debug(
 753                                 "%pRN(%s): New mpath count (incl newbest) %d mpath-change %s all_paths_lb %d cum_bw u%" PRIu64,
 754                                 bgp_dest_to_rnode(dest), bgp->name_pretty,
 755                                 mpath_count, mpath_changed ? "YES" : "NO",
 756                                 all_paths_lb, cum_bw);
 757
 758                 if (mpath_changed
 759                     || (bgp_path_info_mpath_count(new_best) != old_mpath_count))
 760                         SET_FLAG(new_best->flags, BGP_PATH_MULTIPATH_CHG);
 761                 if ((mpath_count - 1) != old_mpath_count ||
 762                     old_cum_bw != cum_bw)
 763                         SET_FLAG(new_best->flags, BGP_PATH_LINK_BW_CHG);
 764         }
 765 }
 766
 767 /*
 768  * bgp_mp_dmed_deselect
 769  *
 770  * Clean up multipath information for BGP_PATH_DMED_SELECTED path that
 771  * is not selected as best path
 772  */
 773 void bgp_mp_dmed_deselect(struct bgp_path_info *dmed_best)
 774 {
 775         struct bgp_path_info *mpinfo, *mpnext;
 776
 777         if (!dmed_best)
 778                 return;
 779
 780         for (mpinfo = bgp_path_info_mpath_first(dmed_best); mpinfo;
 781              mpinfo = mpnext) {
 782                 mpnext = bgp_path_info_mpath_next(mpinfo);
 783                 bgp_path_info_mpath_dequeue(mpinfo);
 784         }
 785
 786         bgp_path_info_mpath_count_set(dmed_best, 0);
 787         UNSET_FLAG(dmed_best->flags, BGP_PATH_MULTIPATH_CHG);
 788         UNSET_FLAG(dmed_best->flags, BGP_PATH_LINK_BW_CHG);
 789         assert(bgp_path_info_mpath_first(dmed_best) == NULL);
 790 }
 791
 792 /*
 793  * bgp_path_info_mpath_aggregate_update
 794  *
 795  * Set the multipath aggregate attribute. We need to see if the
 796  * aggregate has changed and then set the ATTR_CHANGED flag on the
 797  * bestpath info so that a peer update will be generated. The
 798  * change is detected by generating the current attribute,
 799  * interning it, and then comparing the interned pointer with the
 800  * current value. We can skip this generate/compare step if there
 801  * is no change in multipath selection and no attribute change in
 802  * any multipath.
 803  */
 804 void bgp_path_info_mpath_aggregate_update(struct bgp_path_info *new_best,
 805                                           struct bgp_path_info *old_best)
 806 {
 807         struct bgp_path_info *mpinfo;
 808         struct aspath *aspath;
 809         struct aspath *asmerge;
 810         struct attr *new_attr, *old_attr;
 811         uint8_t origin;
 812         struct community *community, *commerge;
 813         struct ecommunity *ecomm, *ecommerge;
 814         struct lcommunity *lcomm, *lcommerge;
 815         struct attr attr = {0};
 816
 817         if (old_best && (old_best != new_best)
 818             && (old_attr = bgp_path_info_mpath_attr(old_best))) {
 819                 bgp_attr_unintern(&old_attr);
 820                 bgp_path_info_mpath_attr_set(old_best, NULL);
 821         }
 822
 823         if (!new_best)
 824                 return;
 825
 826         if (!bgp_path_info_mpath_count(new_best)) {
 827                 if ((new_attr = bgp_path_info_mpath_attr(new_best))) {
 828                         bgp_attr_unintern(&new_attr);
 829                         bgp_path_info_mpath_attr_set(new_best, NULL);
 830                         SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 831                 }
 832                 return;
 833         }
 834
 835         attr = *new_best->attr;
 836
 837         if (new_best->peer
 838             && CHECK_FLAG(new_best->peer->bgp->flags,
 839                           BGP_FLAG_MULTIPATH_RELAX_AS_SET)) {
 840
 841                 /* aggregate attribute from multipath constituents */
 842                 aspath = aspath_dup(attr.aspath);
 843                 origin = attr.origin;
 844                 community =
 845                         bgp_attr_get_community(&attr)
 846                                 ? community_dup(bgp_attr_get_community(&attr))
 847                                 : NULL;
 848                 ecomm = (bgp_attr_get_ecommunity(&attr))
 849                                 ? ecommunity_dup(bgp_attr_get_ecommunity(&attr))
 850                                 : NULL;
 851                 lcomm = (bgp_attr_get_lcommunity(&attr))
 852                                 ? lcommunity_dup(bgp_attr_get_lcommunity(&attr))
 853                                 : NULL;
 854
 855                 for (mpinfo = bgp_path_info_mpath_first(new_best); mpinfo;
 856                      mpinfo = bgp_path_info_mpath_next(mpinfo)) {
 857                         asmerge =
 858                                 aspath_aggregate(aspath, mpinfo->attr->aspath);
 859                         aspath_free(aspath);
 860                         aspath = asmerge;
 861
 862                         if (origin < mpinfo->attr->origin)
 863                                 origin = mpinfo->attr->origin;
 864
 865                         if (bgp_attr_get_community(mpinfo->attr)) {
 866                                 if (community) {
 867                                         commerge = community_merge(
 868                                                 community,
 869                                                 bgp_attr_get_community(
 870                                                         mpinfo->attr));
 871                                         community =
 872                                                 community_uniq_sort(commerge);
 873                                         community_free(&commerge);
 874                                 } else
 875                                         community = community_dup(
 876                                                 bgp_attr_get_community(
 877                                                         mpinfo->attr));
 878                         }
 879
 880                         if (bgp_attr_get_ecommunity(mpinfo->attr)) {
 881                                 if (ecomm) {
 882                                         ecommerge = ecommunity_merge(
 883                                                 ecomm, bgp_attr_get_ecommunity(
 884                                                                mpinfo->attr));
 885                                         ecomm = ecommunity_uniq_sort(ecommerge);
 886                                         ecommunity_free(&ecommerge);
 887                                 } else
 888                                         ecomm = ecommunity_dup(
 889                                                 bgp_attr_get_ecommunity(
 890                                                         mpinfo->attr));
 891                         }
 892                         if (bgp_attr_get_lcommunity(mpinfo->attr)) {
 893                                 if (lcomm) {
 894                                         lcommerge = lcommunity_merge(
 895                                                 lcomm, bgp_attr_get_lcommunity(
 896                                                                mpinfo->attr));
 897                                         lcomm = lcommunity_uniq_sort(lcommerge);
 898                                         lcommunity_free(&lcommerge);
 899                                 } else
 900                                         lcomm = lcommunity_dup(
 901                                                 bgp_attr_get_lcommunity(
 902                                                         mpinfo->attr));
 903                         }
 904                 }
 905
 906                 attr.aspath = aspath;
 907                 attr.origin = origin;
 908                 if (community)
 909                         bgp_attr_set_community(&attr, community);
 910                 if (ecomm)
 911                         bgp_attr_set_ecommunity(&attr, ecomm);
 912                 if (lcomm)
 913                         bgp_attr_set_lcommunity(&attr, lcomm);
 914
 915                 /* Zap multipath attr nexthop so we set nexthop to self */
 916                 attr.nexthop.s_addr = INADDR_ANY;
 917                 memset(&attr.mp_nexthop_global, 0, sizeof(struct in6_addr));
 918
 919                 /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */
 920         }
 921
 922         new_attr = bgp_attr_intern(&attr);
 923
 924         if (new_attr != bgp_path_info_mpath_attr(new_best)) {
 925                 if ((old_attr = bgp_path_info_mpath_attr(new_best)))
 926                         bgp_attr_unintern(&old_attr);
 927                 bgp_path_info_mpath_attr_set(new_best, new_attr);
 928                 SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED);
 929         } else
 930                 bgp_attr_unintern(&new_attr);
 931 }