bgpd/bgp_addpath.c

   1 /*
   2  * Addpath TX ID selection, and related utilities
   3  * Copyright (C) 2018  Amazon.com, Inc. or its affiliates
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License as published by the Free
   7  * Software Foundation; either version 2 of the License, or (at your option)
   8  * any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; see the file COPYING; if not, write to the Free Software
  17  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  18  */
  19
  20 #ifdef HAVE_CONFIG_H
  21 #include "config.h"
  22 #endif
  23
  24 #include "bgp_addpath.h"
  25 #include "bgp_route.h"
  26
  27 static const struct bgp_addpath_strategy_names strat_names[BGP_ADDPATH_MAX] = {
  28         {
  29                 .config_name = "addpath-tx-all-paths",
  30                 .human_name = "All",
  31                 .human_description = "Advertise all paths via addpath",
  32                 .type_json_name = "addpathTxAllPaths",
  33                 .id_json_name = "addpathTxIdAll"
  34         },
  35         {
  36                 .config_name = "addpath-tx-bestpath-per-AS",
  37                 .human_name = "Best-Per-AS",
  38                 .human_description = "Advertise bestpath per AS via addpath",
  39                 .type_json_name = "addpathTxBestpathPerAS",
  40                 .id_json_name = "addpathTxIdBestPerAS"
  41         }
  42 };
  43
  44 static const struct bgp_addpath_strategy_names unknown_names = {
  45         .config_name = "addpath-tx-unknown",
  46         .human_name = "Unknown-Addpath-Strategy",
  47         .human_description = "Unknown Addpath Strategy",
  48         .type_json_name = "addpathTxUnknown",
  49         .id_json_name = "addpathTxIdUnknown"
  50 };
  51
  52 /*
  53  * Returns a structure full of strings associated with an addpath type. Will
  54  * never return null.
  55  */
  56 const struct bgp_addpath_strategy_names *
  57 bgp_addpath_names(enum bgp_addpath_strat strat)
  58 {
  59         if (strat < BGP_ADDPATH_MAX)
  60                 return &(strat_names[strat]);
  61         else
  62                 return &unknown_names;
  63 };
  64
  65 /*
  66  * Returns if any peer is transmitting addpaths for a given afi/safi.
  67  */
  68 bool bgp_addpath_is_addpath_used(struct bgp_addpath_bgp_data *d, afi_t afi,
  69                                  safi_t safi)
  70 {
  71         return d->total_peercount[afi][safi] > 0;
  72 }
  73
  74 /*
  75  * Initialize the BGP instance level data for addpath.
  76  */
  77 void bgp_addpath_init_bgp_data(struct bgp_addpath_bgp_data *d)
  78 {
  79         safi_t safi;
  80         afi_t afi;
  81         int i;
  82
  83         for (afi = AFI_IP; afi < AFI_MAX; afi++) {
  84                 for (safi = SAFI_UNICAST; safi < SAFI_MAX; safi++) {
  85                         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
  86                                 d->id_allocators[afi][safi][i] = NULL;
  87                                 d->peercount[afi][safi][i] = 0;
  88                         }
  89                         d->total_peercount[afi][safi] = 0;
  90                 }
  91         }
  92 }
  93
  94 /*
  95  * Free up resources associated with BGP route info structures.
  96  */
  97 void bgp_addpath_free_info_data(struct bgp_addpath_info_data *d,
  98                               struct bgp_addpath_node_data *nd)
  99 {
 100         int i;
 101
 102         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 103                 if (d->addpath_tx_id[i] != IDALLOC_INVALID)
 104                         idalloc_free_to_pool(&nd->free_ids[i],
 105                                              d->addpath_tx_id[i]);
 106         }
 107 }
 108
 109 /*
 110  * Return the addpath ID used to send a particular route, to a particular peer,
 111  * in a particular AFI/SAFI.
 112  */
 113 uint32_t bgp_addpath_id_for_peer(struct peer *peer, afi_t afi, safi_t safi,
 114                                 struct bgp_addpath_info_data *d)
 115 {
 116         if (peer->addpath_type[afi][safi] < BGP_ADDPATH_MAX)
 117                 return d->addpath_tx_id[peer->addpath_type[afi][safi]];
 118         else
 119                 return IDALLOC_INVALID;
 120 }
 121
 122 /*
 123  * Returns true if the path has an assigned addpath ID for any of the addpath
 124  * strategies.
 125  */
 126 bool bgp_addpath_info_has_ids(struct bgp_addpath_info_data *d)
 127 {
 128         int i;
 129
 130         for (i = 0; i < BGP_ADDPATH_MAX; i++)
 131                 if (d->addpath_tx_id[i] != 0)
 132                         return true;
 133
 134         return false;
 135 }
 136
 137 /*
 138  * Releases any ID's associated with the BGP prefix.
 139  */
 140 void bgp_addpath_free_node_data(struct bgp_addpath_bgp_data *bd,
 141                               struct bgp_addpath_node_data *nd, afi_t afi,
 142                               safi_t safi)
 143 {
 144         int i;
 145
 146         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 147                 idalloc_drain_pool(bd->id_allocators[afi][safi][i],
 148                                    &(nd->free_ids[i]));
 149         }
 150 }
 151
 152 /*
 153  * Check to see if the addpath strategy requires DMED to be configured to work.
 154  */
 155 bool bgp_addpath_dmed_required(int strategy)
 156 {
 157         return strategy == BGP_ADDPATH_BEST_PER_AS;
 158 }
 159
 160 /*
 161  * Return true if this is a path we should advertise due to a
 162  * configured addpath-tx knob
 163  */
 164 bool bgp_addpath_tx_path(enum bgp_addpath_strat strat, struct bgp_path_info *pi)
 165 {
 166         switch (strat) {
 167         case BGP_ADDPATH_NONE:
 168                 return false;
 169         case BGP_ADDPATH_ALL:
 170                 return true;
 171         case BGP_ADDPATH_BEST_PER_AS:
 172                 if (CHECK_FLAG(pi->flags, BGP_PATH_DMED_SELECTED))
 173                         return true;
 174                 else
 175                         return false;
 176         default:
 177                 return false;
 178         }
 179 }
 180
 181 static void bgp_addpath_flush_type_rn(struct bgp *bgp, afi_t afi, safi_t safi,
 182                                       enum bgp_addpath_strat addpath_type,
 183                                       struct bgp_dest *dest)
 184 {
 185         struct bgp_path_info *pi;
 186
 187         idalloc_drain_pool(
 188                 bgp->tx_addpath.id_allocators[afi][safi][addpath_type],
 189                 &(dest->tx_addpath.free_ids[addpath_type]));
 190         for (pi = bgp_dest_get_bgp_path_info(dest); pi; pi = pi->next) {
 191                 if (pi->tx_addpath.addpath_tx_id[addpath_type]
 192                     != IDALLOC_INVALID) {
 193                         idalloc_free(
 194                                 bgp->tx_addpath
 195                                         .id_allocators[afi][safi][addpath_type],
 196                                 pi->tx_addpath.addpath_tx_id[addpath_type]);
 197                         pi->tx_addpath.addpath_tx_id[addpath_type] =
 198                                 IDALLOC_INVALID;
 199                 }
 200         }
 201 }
 202
 203 /*
 204  * Purge all addpath ID's on a BGP instance associated with the addpath
 205  * strategy, and afi/safi combination. This lets us let go of all memory held to
 206  * track ID numbers associated with an addpath type not in use. Since
 207  * post-bestpath ID processing is skipped for types not used, this is the only
 208  * chance to free this data.
 209  */
 210 static void bgp_addpath_flush_type(struct bgp *bgp, afi_t afi, safi_t safi,
 211                                    enum bgp_addpath_strat addpath_type)
 212 {
 213         struct bgp_dest *dest, *ndest;
 214
 215         for (dest = bgp_table_top(bgp->rib[afi][safi]); dest;
 216              dest = bgp_route_next(dest)) {
 217                 if (safi == SAFI_MPLS_VPN) {
 218                         struct bgp_table *table;
 219
 220                         table = bgp_dest_get_bgp_table_info(dest);
 221                         if (!table)
 222                                 continue;
 223
 224                         for (ndest = bgp_table_top(table); ndest;
 225                              ndest = bgp_route_next(ndest))
 226                                 bgp_addpath_flush_type_rn(bgp, afi, safi,
 227                                                           addpath_type, ndest);
 228                 } else {
 229                         bgp_addpath_flush_type_rn(bgp, afi, safi, addpath_type,
 230                                                   dest);
 231                 }
 232         }
 233
 234         idalloc_destroy(bgp->tx_addpath.id_allocators[afi][safi][addpath_type]);
 235         bgp->tx_addpath.id_allocators[afi][safi][addpath_type] = NULL;
 236 }
 237
 238 /*
 239  * Allocate an Addpath ID for the given type on a path, if necessary.
 240  */
 241 static void bgp_addpath_populate_path(struct id_alloc *allocator,
 242                                       struct bgp_path_info *path,
 243                                       enum bgp_addpath_strat addpath_type)
 244 {
 245         if (bgp_addpath_tx_path(addpath_type, path)) {
 246                 path->tx_addpath.addpath_tx_id[addpath_type] =
 247                         idalloc_allocate(allocator);
 248         }
 249 }
 250
 251 /*
 252  * Compute addpath ID's on a BGP instance associated with the addpath strategy,
 253  * and afi/safi combination. Since we won't waste the time computing addpath IDs
 254  * for unused strategies, the first time a peer is configured to use a strategy,
 255  * we have to backfill the data.
 256  */
 257 static void bgp_addpath_populate_type(struct bgp *bgp, afi_t afi, safi_t safi,
 258                                     enum bgp_addpath_strat addpath_type)
 259 {
 260         struct bgp_dest *dest, *ndest;
 261         char buf[200];
 262         struct id_alloc *allocator;
 263
 264         snprintf(buf, sizeof(buf), "Addpath ID Allocator %s:%d/%d",
 265                  bgp_addpath_names(addpath_type)->config_name, (int)afi,
 266                  (int)safi);
 267         buf[sizeof(buf) - 1] = '\0';
 268         zlog_info("Computing addpath IDs for addpath type %s",
 269                 bgp_addpath_names(addpath_type)->human_name);
 270
 271         bgp->tx_addpath.id_allocators[afi][safi][addpath_type] =
 272                 idalloc_new(buf);
 273
 274         idalloc_reserve(bgp->tx_addpath.id_allocators[afi][safi][addpath_type],
 275                 BGP_ADDPATH_TX_ID_FOR_DEFAULT_ORIGINATE);
 276
 277         allocator = bgp->tx_addpath.id_allocators[afi][safi][addpath_type];
 278
 279         for (dest = bgp_table_top(bgp->rib[afi][safi]); dest;
 280              dest = bgp_route_next(dest)) {
 281                 struct bgp_path_info *bi;
 282
 283                 if (safi == SAFI_MPLS_VPN) {
 284                         struct bgp_table *table;
 285
 286                         table = bgp_dest_get_bgp_table_info(dest);
 287                         if (!table)
 288                                 continue;
 289
 290                         for (ndest = bgp_table_top(table); ndest;
 291                              ndest = bgp_route_next(ndest))
 292                                 for (bi = bgp_dest_get_bgp_path_info(ndest); bi;
 293                                      bi = bi->next)
 294                                         bgp_addpath_populate_path(allocator, bi,
 295                                                                   addpath_type);
 296                 } else {
 297                         for (bi = bgp_dest_get_bgp_path_info(dest); bi;
 298                              bi = bi->next)
 299                                 bgp_addpath_populate_path(allocator, bi,
 300                                                           addpath_type);
 301                 }
 302         }
 303 }
 304
 305 /*
 306  * Handle updates to a peer or group's addpath strategy. If after adjusting
 307  * counts a addpath strategy is in use for the first time, or no longer in use,
 308  * the IDs for that strategy will be populated or flushed.
 309  */
 310 void bgp_addpath_type_changed(struct bgp *bgp)
 311 {
 312         afi_t afi;
 313         safi_t safi;
 314         struct listnode *node, *nnode;
 315         struct peer *peer;
 316         int peer_count[AFI_MAX][SAFI_MAX][BGP_ADDPATH_MAX];
 317         enum bgp_addpath_strat type;
 318
 319         FOREACH_AFI_SAFI(afi, safi) {
 320                 for (type=0; type<BGP_ADDPATH_MAX; type++) {
 321                         peer_count[afi][safi][type] = 0;
 322                 }
 323                 bgp->tx_addpath.total_peercount[afi][safi] = 0;
 324         }
 325
 326         for (ALL_LIST_ELEMENTS(bgp->peer, node, nnode, peer)) {
 327                 FOREACH_AFI_SAFI(afi, safi) {
 328                         type = peer->addpath_type[afi][safi];
 329                         if (type != BGP_ADDPATH_NONE) {
 330                                 peer_count[afi][safi][type] += 1;
 331                                 bgp->tx_addpath.total_peercount[afi][safi] += 1;
 332                         }
 333                 }
 334         }
 335
 336         FOREACH_AFI_SAFI(afi, safi) {
 337                 for (type=0; type<BGP_ADDPATH_MAX; type++) {
 338                         int old = bgp->tx_addpath.peercount[afi][safi][type];
 339                         int new = peer_count[afi][safi][type];
 340
 341                         bgp->tx_addpath.peercount[afi][safi][type] = new;
 342
 343                         if (old == 0 && new != 0) {
 344                                 bgp_addpath_populate_type(bgp, afi, safi,
 345                                         type);
 346                         } else if (old != 0 && new == 0) {
 347                                 bgp_addpath_flush_type(bgp, afi, safi, type);
 348                         }
 349                 }
 350         }
 351 }
 352
 353 /*
 354  * Change the addpath type assigned to a peer, or peer group. In addition to
 355  * adjusting the counts, peer sessions will be reset as needed to make the
 356  * change take effect.
 357  */
 358 void bgp_addpath_set_peer_type(struct peer *peer, afi_t afi, safi_t safi,
 359                               enum bgp_addpath_strat addpath_type)
 360 {
 361         struct bgp *bgp = peer->bgp;
 362         enum bgp_addpath_strat old_type = peer->addpath_type[afi][safi];
 363         struct listnode *node, *nnode;
 364         struct peer *tmp_peer;
 365         struct peer_group *group;
 366
 367         if (addpath_type == old_type)
 368                 return;
 369
 370         if (addpath_type == BGP_ADDPATH_NONE && peer->group &&
 371             !CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP)) {
 372                 /* A "no" config on a group member inherits group */
 373                 addpath_type = peer->group->conf->addpath_type[afi][safi];
 374         }
 375
 376         peer->addpath_type[afi][safi] = addpath_type;
 377
 378         bgp_addpath_type_changed(bgp);
 379
 380         if (addpath_type != BGP_ADDPATH_NONE) {
 381                 if (bgp_addpath_dmed_required(addpath_type)) {
 382                         if (!CHECK_FLAG(bgp->flags,
 383                                         BGP_FLAG_DETERMINISTIC_MED)) {
 384                                 zlog_warn(
 385                                         "%s: enabling bgp deterministic-med, this is required for addpath-tx-bestpath-per-AS",
 386                                         peer->host);
 387                                 SET_FLAG(bgp->flags,
 388                                          BGP_FLAG_DETERMINISTIC_MED);
 389                                 bgp_recalculate_all_bestpaths(bgp);
 390                         }
 391                 }
 392         }
 393
 394         zlog_info("Resetting peer %s%s due to change in addpath config",
 395                   CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP) ? "group " : "",
 396                   peer->host);
 397
 398         if (CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP)) {
 399                 group = peer->group;
 400
 401                 /* group will be null as peer_group_delete calls peer_delete on
 402                  * group->conf. That peer_delete will eventuallly end up here
 403                  * if the group was configured to tx addpaths.
 404                  */
 405                 if (group != NULL) {
 406                         for (ALL_LIST_ELEMENTS(group->peer, node, nnode,
 407                              tmp_peer)) {
 408                                 if (tmp_peer->addpath_type[afi][safi] ==
 409                                     old_type) {
 410                                         bgp_addpath_set_peer_type(tmp_peer,
 411                                                                  afi,
 412                                                                  safi,
 413                                                                  addpath_type);
 414                                 }
 415                         }
 416                 }
 417         } else {
 418                 peer_change_action(peer, afi, safi, peer_change_reset);
 419         }
 420
 421 }
 422
 423 /*
 424  * Intended to run after bestpath. This function will take TX IDs from paths
 425  * that no longer need them, and give them to paths that do. This prevents
 426  * best-per-as updates from needing to do a separate withdraw and update just to
 427  * swap out which path is sent.
 428  */
 429 void bgp_addpath_update_ids(struct bgp *bgp, struct bgp_dest *bn, afi_t afi,
 430                             safi_t safi)
 431 {
 432         int i;
 433         struct bgp_path_info *pi;
 434         struct id_alloc_pool **pool_ptr;
 435
 436         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 437                 struct id_alloc *alloc =
 438                         bgp->tx_addpath.id_allocators[afi][safi][i];
 439                 pool_ptr = &(bn->tx_addpath.free_ids[i]);
 440
 441                 if (bgp->tx_addpath.peercount[afi][safi][i] == 0)
 442                         continue;
 443
 444                 /* Free Unused IDs back to the pool.*/
 445                 for (pi = bgp_dest_get_bgp_path_info(bn); pi; pi = pi->next) {
 446                         if (pi->tx_addpath.addpath_tx_id[i] != IDALLOC_INVALID
 447                             && !bgp_addpath_tx_path(i, pi)) {
 448                                 idalloc_free_to_pool(pool_ptr,
 449                                         pi->tx_addpath.addpath_tx_id[i]);
 450                                 pi->tx_addpath.addpath_tx_id[i] =
 451                                         IDALLOC_INVALID;
 452                         }
 453                 }
 454
 455                 /* Give IDs to paths that need them (pulling from the pool) */
 456                 for (pi = bgp_dest_get_bgp_path_info(bn); pi; pi = pi->next) {
 457                         if (pi->tx_addpath.addpath_tx_id[i] == IDALLOC_INVALID
 458                             && bgp_addpath_tx_path(i, pi)) {
 459                                 pi->tx_addpath.addpath_tx_id[i] =
 460                                         idalloc_allocate_prefer_pool(
 461                                                 alloc, pool_ptr);
 462                         }
 463                 }
 464
 465                 /* Free any IDs left in the pool to the main allocator */
 466                 idalloc_drain_pool(alloc, pool_ptr);
 467         }
 468 }