]>
Commit | Line | Data |
---|---|---|
8ef0791c | 1 | /* |
165b5fff JB |
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 | * | |
896014f4 DL |
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 | |
165b5fff JB |
20 | */ |
21 | ||
22 | #include <zebra.h> | |
23 | ||
24 | #include "command.h" | |
96450faf JB |
25 | #include "prefix.h" |
26 | #include "linklist.h" | |
27 | #include "sockunion.h" | |
de8d5dff | 28 | #include "memory.h" |
3f9c7369 | 29 | #include "queue.h" |
039f3a34 | 30 | #include "filter.h" |
165b5fff JB |
31 | |
32 | #include "bgpd/bgpd.h" | |
96450faf JB |
33 | #include "bgpd/bgp_table.h" |
34 | #include "bgpd/bgp_route.h" | |
35 | #include "bgpd/bgp_attr.h" | |
de8d5dff | 36 | #include "bgpd/bgp_debug.h" |
0b597ef0 JB |
37 | #include "bgpd/bgp_aspath.h" |
38 | #include "bgpd/bgp_community.h" | |
39 | #include "bgpd/bgp_ecommunity.h" | |
57d187bc | 40 | #include "bgpd/bgp_lcommunity.h" |
165b5fff JB |
41 | #include "bgpd/bgp_mpath.h" |
42 | ||
43 | /* | |
44 | * bgp_maximum_paths_set | |
45 | * | |
46 | * Record maximum-paths configuration for BGP instance | |
47 | */ | |
d62a17ae | 48 | int bgp_maximum_paths_set(struct bgp *bgp, afi_t afi, safi_t safi, int peertype, |
d7c0a89a | 49 | uint16_t maxpaths, uint16_t options) |
165b5fff | 50 | { |
d62a17ae | 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; | |
165b5fff JB |
67 | } |
68 | ||
69 | /* | |
70 | * bgp_maximum_paths_unset | |
71 | * | |
72 | * Remove maximum-paths configuration from BGP instance | |
73 | */ | |
d62a17ae | 74 | int bgp_maximum_paths_unset(struct bgp *bgp, afi_t afi, safi_t safi, |
75 | int peertype) | |
165b5fff | 76 | { |
d62a17ae | 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; | |
165b5fff | 93 | } |
96450faf | 94 | |
194a4f2c DW |
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 | ||
96450faf | 115 | /* |
18ee8310 | 116 | * bgp_path_info_nexthop_cmp |
96450faf JB |
117 | * |
118 | * Compare the nexthops of two paths. Return value is less than, equal to, | |
40381db7 DS |
119 | * or greater than zero if bpi1 is respectively less than, equal to, |
120 | * or greater than bpi2. | |
96450faf | 121 | */ |
40381db7 DS |
122 | int bgp_path_info_nexthop_cmp(struct bgp_path_info *bpi1, |
123 | struct bgp_path_info *bpi2) | |
96450faf | 124 | { |
d62a17ae | 125 | int compare; |
05a38c0c | 126 | struct in6_addr addr1, addr2; |
d62a17ae | 127 | |
40381db7 | 128 | compare = IPV4_ADDR_CMP(&bpi1->attr->nexthop, &bpi2->attr->nexthop); |
d62a17ae | 129 | if (!compare) { |
40381db7 DS |
130 | if (bpi1->attr->mp_nexthop_len == bpi2->attr->mp_nexthop_len) { |
131 | switch (bpi1->attr->mp_nexthop_len) { | |
d62a17ae | 132 | case BGP_ATTR_NHLEN_IPV4: |
133 | case BGP_ATTR_NHLEN_VPNV4: | |
134 | compare = IPV4_ADDR_CMP( | |
40381db7 DS |
135 | &bpi1->attr->mp_nexthop_global_in, |
136 | &bpi2->attr->mp_nexthop_global_in); | |
d62a17ae | 137 | break; |
138 | case BGP_ATTR_NHLEN_IPV6_GLOBAL: | |
139 | case BGP_ATTR_NHLEN_VPNV6_GLOBAL: | |
140 | compare = IPV6_ADDR_CMP( | |
40381db7 DS |
141 | &bpi1->attr->mp_nexthop_global, |
142 | &bpi2->attr->mp_nexthop_global); | |
d62a17ae | 143 | break; |
144 | case BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL: | |
40381db7 DS |
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) | |
996c9314 | 154 | compare = !bgp_interface_same( |
40381db7 DS |
155 | bpi1->peer->ifp, |
156 | bpi2->peer->ifp); | |
194a4f2c | 157 | |
d62a17ae | 158 | if (!compare) |
05a38c0c | 159 | compare = IPV6_ADDR_CMP(&addr1, &addr2); |
d62a17ae | 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 | */ | |
40381db7 DS |
168 | else if (bpi1->attr->mp_nexthop_len |
169 | == BGP_ATTR_NHLEN_IPV6_GLOBAL | |
170 | || bpi1->attr->mp_nexthop_len | |
d62a17ae | 171 | == BGP_ATTR_NHLEN_IPV6_GLOBAL_AND_LL) { |
40381db7 DS |
172 | compare = IPV6_ADDR_CMP(&bpi1->attr->mp_nexthop_global, |
173 | &bpi2->attr->mp_nexthop_global); | |
d62a17ae | 174 | if (!compare) { |
40381db7 DS |
175 | if (bpi1->attr->mp_nexthop_len |
176 | < bpi2->attr->mp_nexthop_len) | |
d62a17ae | 177 | compare = -1; |
178 | else | |
179 | compare = 1; | |
180 | } | |
181 | } | |
182 | } | |
183 | ||
184 | return compare; | |
96450faf JB |
185 | } |
186 | ||
187 | /* | |
18ee8310 | 188 | * bgp_path_info_mpath_cmp |
96450faf JB |
189 | * |
190 | * This function determines our multipath list ordering. By ordering | |
191 | * the list we can deterministically select which paths are included | |
192 | * in the multipath set. The ordering also helps in detecting changes | |
193 | * in the multipath selection so we can detect whether to send an | |
194 | * update to zebra. | |
195 | * | |
196 | * The order of paths is determined first by received nexthop, and then | |
197 | * by peer address if the nexthops are the same. | |
198 | */ | |
18ee8310 | 199 | static int bgp_path_info_mpath_cmp(void *val1, void *val2) |
96450faf | 200 | { |
40381db7 | 201 | struct bgp_path_info *bpi1, *bpi2; |
d62a17ae | 202 | int compare; |
203 | ||
40381db7 DS |
204 | bpi1 = val1; |
205 | bpi2 = val2; | |
d62a17ae | 206 | |
40381db7 | 207 | compare = bgp_path_info_nexthop_cmp(bpi1, bpi2); |
d62a17ae | 208 | |
209 | if (!compare) { | |
40381db7 | 210 | if (!bpi1->peer->su_remote && !bpi2->peer->su_remote) |
d62a17ae | 211 | compare = 0; |
40381db7 | 212 | else if (!bpi1->peer->su_remote) |
d62a17ae | 213 | compare = 1; |
40381db7 | 214 | else if (!bpi2->peer->su_remote) |
d62a17ae | 215 | compare = -1; |
216 | else | |
40381db7 DS |
217 | compare = sockunion_cmp(bpi1->peer->su_remote, |
218 | bpi2->peer->su_remote); | |
d62a17ae | 219 | } |
220 | ||
221 | return compare; | |
96450faf JB |
222 | } |
223 | ||
224 | /* | |
225 | * bgp_mp_list_init | |
226 | * | |
227 | * Initialize the mp_list, which holds the list of multipaths | |
228 | * selected by bgp_best_selection | |
229 | */ | |
d62a17ae | 230 | void bgp_mp_list_init(struct list *mp_list) |
96450faf | 231 | { |
d62a17ae | 232 | assert(mp_list); |
233 | memset(mp_list, 0, sizeof(struct list)); | |
18ee8310 | 234 | mp_list->cmp = bgp_path_info_mpath_cmp; |
96450faf JB |
235 | } |
236 | ||
237 | /* | |
238 | * bgp_mp_list_clear | |
239 | * | |
240 | * Clears all entries out of the mp_list | |
241 | */ | |
d62a17ae | 242 | void bgp_mp_list_clear(struct list *mp_list) |
96450faf | 243 | { |
d62a17ae | 244 | assert(mp_list); |
245 | list_delete_all_node(mp_list); | |
96450faf JB |
246 | } |
247 | ||
248 | /* | |
249 | * bgp_mp_list_add | |
250 | * | |
251 | * Adds a multipath entry to the mp_list | |
252 | */ | |
4b7e6066 | 253 | void bgp_mp_list_add(struct list *mp_list, struct bgp_path_info *mpinfo) |
96450faf | 254 | { |
d62a17ae | 255 | assert(mp_list && mpinfo); |
256 | listnode_add_sort(mp_list, mpinfo); | |
96450faf | 257 | } |
de8d5dff JB |
258 | |
259 | /* | |
18ee8310 | 260 | * bgp_path_info_mpath_new |
de8d5dff | 261 | * |
18ee8310 | 262 | * Allocate and zero memory for a new bgp_path_info_mpath element |
de8d5dff | 263 | */ |
18ee8310 | 264 | static struct bgp_path_info_mpath *bgp_path_info_mpath_new(void) |
de8d5dff | 265 | { |
4b7e6066 DS |
266 | struct bgp_path_info_mpath *new_mpath; |
267 | new_mpath = XCALLOC(MTYPE_BGP_MPATH_INFO, | |
268 | sizeof(struct bgp_path_info_mpath)); | |
d62a17ae | 269 | return new_mpath; |
de8d5dff JB |
270 | } |
271 | ||
272 | /* | |
18ee8310 | 273 | * bgp_path_info_mpath_free |
de8d5dff | 274 | * |
18ee8310 | 275 | * Release resources for a bgp_path_info_mpath element and zero out pointer |
de8d5dff | 276 | */ |
18ee8310 | 277 | void bgp_path_info_mpath_free(struct bgp_path_info_mpath **mpath) |
de8d5dff | 278 | { |
d62a17ae | 279 | if (mpath && *mpath) { |
280 | if ((*mpath)->mp_attr) | |
281 | bgp_attr_unintern(&(*mpath)->mp_attr); | |
282 | XFREE(MTYPE_BGP_MPATH_INFO, *mpath); | |
283 | *mpath = NULL; | |
284 | } | |
de8d5dff JB |
285 | } |
286 | ||
287 | /* | |
18ee8310 | 288 | * bgp_path_info_mpath_get |
de8d5dff | 289 | * |
18ee8310 | 290 | * Fetch the mpath element for the given bgp_path_info. Used for |
de8d5dff JB |
291 | * doing lazy allocation. |
292 | */ | |
4b7e6066 | 293 | static struct bgp_path_info_mpath * |
9b6d8fcf | 294 | bgp_path_info_mpath_get(struct bgp_path_info *path) |
de8d5dff | 295 | { |
4b7e6066 | 296 | struct bgp_path_info_mpath *mpath; |
fa3bf3a2 MS |
297 | |
298 | if (!path) | |
299 | return NULL; | |
300 | ||
9b6d8fcf | 301 | if (!path->mpath) { |
18ee8310 | 302 | mpath = bgp_path_info_mpath_new(); |
d62a17ae | 303 | if (!mpath) |
304 | return NULL; | |
9b6d8fcf DS |
305 | path->mpath = mpath; |
306 | mpath->mp_info = path; | |
d62a17ae | 307 | } |
9b6d8fcf | 308 | return path->mpath; |
de8d5dff JB |
309 | } |
310 | ||
311 | /* | |
18ee8310 | 312 | * bgp_path_info_mpath_enqueue |
de8d5dff JB |
313 | * |
314 | * Enqueue a path onto the multipath list given the previous multipath | |
315 | * list entry | |
316 | */ | |
18ee8310 | 317 | static void bgp_path_info_mpath_enqueue(struct bgp_path_info *prev_info, |
9b6d8fcf | 318 | struct bgp_path_info *path) |
de8d5dff | 319 | { |
4b7e6066 | 320 | struct bgp_path_info_mpath *prev, *mpath; |
de8d5dff | 321 | |
18ee8310 | 322 | prev = bgp_path_info_mpath_get(prev_info); |
9b6d8fcf | 323 | mpath = bgp_path_info_mpath_get(path); |
d62a17ae | 324 | if (!prev || !mpath) |
325 | return; | |
de8d5dff | 326 | |
d62a17ae | 327 | mpath->mp_next = prev->mp_next; |
328 | mpath->mp_prev = prev; | |
329 | if (prev->mp_next) | |
330 | prev->mp_next->mp_prev = mpath; | |
331 | prev->mp_next = mpath; | |
de8d5dff | 332 | |
9b6d8fcf | 333 | SET_FLAG(path->flags, BGP_PATH_MULTIPATH); |
de8d5dff JB |
334 | } |
335 | ||
336 | /* | |
18ee8310 | 337 | * bgp_path_info_mpath_dequeue |
de8d5dff JB |
338 | * |
339 | * Remove a path from the multipath list | |
340 | */ | |
9b6d8fcf | 341 | void bgp_path_info_mpath_dequeue(struct bgp_path_info *path) |
de8d5dff | 342 | { |
9b6d8fcf | 343 | struct bgp_path_info_mpath *mpath = path->mpath; |
d62a17ae | 344 | if (!mpath) |
345 | return; | |
346 | if (mpath->mp_prev) | |
347 | mpath->mp_prev->mp_next = mpath->mp_next; | |
348 | if (mpath->mp_next) | |
349 | mpath->mp_next->mp_prev = mpath->mp_prev; | |
350 | mpath->mp_next = mpath->mp_prev = NULL; | |
9b6d8fcf | 351 | UNSET_FLAG(path->flags, BGP_PATH_MULTIPATH); |
de8d5dff JB |
352 | } |
353 | ||
354 | /* | |
18ee8310 | 355 | * bgp_path_info_mpath_next |
de8d5dff | 356 | * |
18ee8310 | 357 | * Given a bgp_path_info, return the next multipath entry |
de8d5dff | 358 | */ |
9b6d8fcf | 359 | struct bgp_path_info *bgp_path_info_mpath_next(struct bgp_path_info *path) |
de8d5dff | 360 | { |
9b6d8fcf | 361 | if (!path->mpath || !path->mpath->mp_next) |
d62a17ae | 362 | return NULL; |
9b6d8fcf | 363 | return path->mpath->mp_next->mp_info; |
de8d5dff JB |
364 | } |
365 | ||
366 | /* | |
18ee8310 | 367 | * bgp_path_info_mpath_first |
de8d5dff | 368 | * |
18ee8310 | 369 | * Given bestpath bgp_path_info, return the first multipath entry. |
de8d5dff | 370 | */ |
9b6d8fcf | 371 | struct bgp_path_info *bgp_path_info_mpath_first(struct bgp_path_info *path) |
de8d5dff | 372 | { |
9b6d8fcf | 373 | return bgp_path_info_mpath_next(path); |
de8d5dff JB |
374 | } |
375 | ||
376 | /* | |
18ee8310 | 377 | * bgp_path_info_mpath_count |
de8d5dff | 378 | * |
18ee8310 | 379 | * Given the bestpath bgp_path_info, return the number of multipath entries |
de8d5dff | 380 | */ |
9b6d8fcf | 381 | uint32_t bgp_path_info_mpath_count(struct bgp_path_info *path) |
de8d5dff | 382 | { |
9b6d8fcf | 383 | if (!path->mpath) |
d62a17ae | 384 | return 0; |
9b6d8fcf | 385 | return path->mpath->mp_count; |
de8d5dff JB |
386 | } |
387 | ||
388 | /* | |
18ee8310 | 389 | * bgp_path_info_mpath_count_set |
de8d5dff JB |
390 | * |
391 | * Sets the count of multipaths into bestpath's mpath element | |
392 | */ | |
9b6d8fcf | 393 | static void bgp_path_info_mpath_count_set(struct bgp_path_info *path, |
18ee8310 | 394 | uint32_t count) |
de8d5dff | 395 | { |
4b7e6066 | 396 | struct bgp_path_info_mpath *mpath; |
9b6d8fcf | 397 | if (!count && !path->mpath) |
d62a17ae | 398 | return; |
9b6d8fcf | 399 | mpath = bgp_path_info_mpath_get(path); |
d62a17ae | 400 | if (!mpath) |
401 | return; | |
402 | mpath->mp_count = count; | |
de8d5dff JB |
403 | } |
404 | ||
0b597ef0 | 405 | /* |
18ee8310 | 406 | * bgp_path_info_mpath_attr |
0b597ef0 | 407 | * |
18ee8310 | 408 | * Given bestpath bgp_path_info, return aggregated attribute set used |
0b597ef0 JB |
409 | * for advertising the multipath route |
410 | */ | |
9b6d8fcf | 411 | struct attr *bgp_path_info_mpath_attr(struct bgp_path_info *path) |
0b597ef0 | 412 | { |
9b6d8fcf | 413 | if (!path->mpath) |
d62a17ae | 414 | return NULL; |
9b6d8fcf | 415 | return path->mpath->mp_attr; |
0b597ef0 JB |
416 | } |
417 | ||
418 | /* | |
18ee8310 | 419 | * bgp_path_info_mpath_attr_set |
0b597ef0 JB |
420 | * |
421 | * Sets the aggregated attribute into bestpath's mpath element | |
422 | */ | |
9b6d8fcf | 423 | static void bgp_path_info_mpath_attr_set(struct bgp_path_info *path, |
18ee8310 | 424 | struct attr *attr) |
0b597ef0 | 425 | { |
4b7e6066 | 426 | struct bgp_path_info_mpath *mpath; |
9b6d8fcf | 427 | if (!attr && !path->mpath) |
d62a17ae | 428 | return; |
9b6d8fcf | 429 | mpath = bgp_path_info_mpath_get(path); |
d62a17ae | 430 | if (!mpath) |
431 | return; | |
432 | mpath->mp_attr = attr; | |
0b597ef0 JB |
433 | } |
434 | ||
de8d5dff | 435 | /* |
18ee8310 | 436 | * bgp_path_info_mpath_update |
de8d5dff JB |
437 | * |
438 | * Compare and sync up the multipath list with the mp_list generated by | |
439 | * bgp_best_selection | |
440 | */ | |
18ee8310 DS |
441 | void bgp_path_info_mpath_update(struct bgp_node *rn, |
442 | struct bgp_path_info *new_best, | |
443 | struct bgp_path_info *old_best, | |
444 | struct list *mp_list, | |
445 | struct bgp_maxpaths_cfg *mpath_cfg) | |
de8d5dff | 446 | { |
d7c0a89a | 447 | uint16_t maxpaths, mpath_count, old_mpath_count; |
d62a17ae | 448 | struct listnode *mp_node, *mp_next_node; |
4b7e6066 | 449 | struct bgp_path_info *cur_mpath, *new_mpath, *next_mpath, *prev_mpath; |
d62a17ae | 450 | int mpath_changed, debug; |
451 | char pfx_buf[PREFIX2STR_BUFFER], nh_buf[2][INET6_ADDRSTRLEN]; | |
452 | char path_buf[PATH_ADDPATH_STR_BUFFER]; | |
453 | ||
454 | mpath_changed = 0; | |
455 | maxpaths = multipath_num; | |
456 | mpath_count = 0; | |
457 | cur_mpath = NULL; | |
458 | old_mpath_count = 0; | |
459 | prev_mpath = new_best; | |
460 | mp_node = listhead(mp_list); | |
461 | debug = bgp_debug_bestpath(&rn->p); | |
462 | ||
463 | if (debug) | |
464 | prefix2str(&rn->p, pfx_buf, sizeof(pfx_buf)); | |
465 | ||
466 | if (new_best) { | |
467 | mpath_count++; | |
468 | if (new_best != old_best) | |
18ee8310 | 469 | bgp_path_info_mpath_dequeue(new_best); |
d62a17ae | 470 | maxpaths = (new_best->peer->sort == BGP_PEER_IBGP) |
471 | ? mpath_cfg->maxpaths_ibgp | |
472 | : mpath_cfg->maxpaths_ebgp; | |
473 | } | |
474 | ||
475 | if (old_best) { | |
18ee8310 DS |
476 | cur_mpath = bgp_path_info_mpath_first(old_best); |
477 | old_mpath_count = bgp_path_info_mpath_count(old_best); | |
478 | bgp_path_info_mpath_count_set(old_best, 0); | |
479 | bgp_path_info_mpath_dequeue(old_best); | |
d62a17ae | 480 | } |
481 | ||
482 | if (debug) | |
483 | zlog_debug( | |
484 | "%s: starting mpath update, newbest %s num candidates %d old-mpath-count %d", | |
485 | pfx_buf, new_best ? new_best->peer->host : "NONE", | |
cc35bc7c | 486 | mp_list ? listcount(mp_list) : 0, old_mpath_count); |
d62a17ae | 487 | |
488 | /* | |
489 | * We perform an ordered walk through both lists in parallel. | |
490 | * The reason for the ordered walk is that if there are paths | |
491 | * that were previously multipaths and are still multipaths, the walk | |
492 | * should encounter them in both lists at the same time. Otherwise | |
493 | * there will be paths that are in one list or another, and we | |
494 | * will deal with these separately. | |
495 | * | |
496 | * Note that new_best might be somewhere in the mp_list, so we need | |
497 | * to skip over it | |
498 | */ | |
499 | while (mp_node || cur_mpath) { | |
4b7e6066 | 500 | struct bgp_path_info *tmp_info; |
d62a17ae | 501 | |
502 | /* | |
503 | * We can bail out of this loop if all existing paths on the | |
504 | * multipath list have been visited (for cleanup purposes) and | |
505 | * the maxpath requirement is fulfulled | |
506 | */ | |
507 | if (!cur_mpath && (mpath_count >= maxpaths)) | |
508 | break; | |
509 | ||
510 | mp_next_node = mp_node ? listnextnode(mp_node) : NULL; | |
18ee8310 DS |
511 | next_mpath = |
512 | cur_mpath ? bgp_path_info_mpath_next(cur_mpath) : NULL; | |
d62a17ae | 513 | tmp_info = mp_node ? listgetdata(mp_node) : NULL; |
514 | ||
515 | if (debug) | |
516 | zlog_debug( | |
517 | "%s: comparing candidate %s with existing mpath %s", | |
518 | pfx_buf, | |
519 | tmp_info ? tmp_info->peer->host : "NONE", | |
520 | cur_mpath ? cur_mpath->peer->host : "NONE"); | |
521 | ||
522 | /* | |
523 | * If equal, the path was a multipath and is still a multipath. | |
524 | * Insert onto new multipath list if maxpaths allows. | |
525 | */ | |
526 | if (mp_node && (listgetdata(mp_node) == cur_mpath)) { | |
527 | list_delete_node(mp_list, mp_node); | |
18ee8310 | 528 | bgp_path_info_mpath_dequeue(cur_mpath); |
d62a17ae | 529 | if ((mpath_count < maxpaths) |
fa3bf3a2 | 530 | && prev_mpath |
18ee8310 DS |
531 | && bgp_path_info_nexthop_cmp(prev_mpath, |
532 | cur_mpath)) { | |
533 | bgp_path_info_mpath_enqueue(prev_mpath, | |
534 | cur_mpath); | |
d62a17ae | 535 | prev_mpath = cur_mpath; |
536 | mpath_count++; | |
537 | if (debug) { | |
18ee8310 | 538 | bgp_path_info_path_with_addpath_rx_str( |
d62a17ae | 539 | cur_mpath, path_buf); |
540 | zlog_debug( | |
541 | "%s: %s is still multipath, cur count %d", | |
542 | pfx_buf, path_buf, mpath_count); | |
543 | } | |
544 | } else { | |
545 | mpath_changed = 1; | |
546 | if (debug) { | |
18ee8310 | 547 | bgp_path_info_path_with_addpath_rx_str( |
d62a17ae | 548 | cur_mpath, path_buf); |
549 | zlog_debug( | |
550 | "%s: remove mpath %s nexthop %s, cur count %d", | |
551 | pfx_buf, path_buf, | |
552 | inet_ntop(AF_INET, | |
553 | &cur_mpath->attr | |
554 | ->nexthop, | |
555 | nh_buf[0], | |
556 | sizeof(nh_buf[0])), | |
557 | mpath_count); | |
558 | } | |
559 | } | |
560 | mp_node = mp_next_node; | |
561 | cur_mpath = next_mpath; | |
562 | continue; | |
563 | } | |
564 | ||
565 | if (cur_mpath | |
566 | && (!mp_node | |
18ee8310 DS |
567 | || (bgp_path_info_mpath_cmp(cur_mpath, |
568 | listgetdata(mp_node)) | |
d62a17ae | 569 | < 0))) { |
570 | /* | |
571 | * If here, we have an old multipath and either the | |
572 | * mp_list | |
573 | * is finished or the next mp_node points to a later | |
574 | * multipath, so we need to purge this path from the | |
575 | * multipath list | |
576 | */ | |
18ee8310 | 577 | bgp_path_info_mpath_dequeue(cur_mpath); |
d62a17ae | 578 | mpath_changed = 1; |
579 | if (debug) { | |
18ee8310 DS |
580 | bgp_path_info_path_with_addpath_rx_str( |
581 | cur_mpath, path_buf); | |
d62a17ae | 582 | zlog_debug( |
583 | "%s: remove mpath %s nexthop %s, cur count %d", | |
584 | pfx_buf, path_buf, | |
585 | inet_ntop(AF_INET, | |
586 | &cur_mpath->attr->nexthop, | |
587 | nh_buf[0], sizeof(nh_buf[0])), | |
588 | mpath_count); | |
589 | } | |
590 | cur_mpath = next_mpath; | |
591 | } else { | |
592 | /* | |
593 | * If here, we have a path on the mp_list that was not | |
594 | * previously | |
595 | * a multipath (due to non-equivalance or maxpaths | |
596 | * exceeded), | |
597 | * or the matching multipath is sorted later in the | |
598 | * multipath | |
599 | * list. Before we enqueue the path on the new multipath | |
600 | * list, | |
601 | * make sure its not on the old_best multipath list or | |
602 | * referenced | |
603 | * via next_mpath: | |
604 | * - If next_mpath points to this new path, update | |
605 | * next_mpath to | |
606 | * point to the multipath after this one | |
607 | * - Dequeue the path from the multipath list just to | |
608 | * make sure | |
609 | */ | |
610 | new_mpath = listgetdata(mp_node); | |
611 | list_delete_node(mp_list, mp_node); | |
2165e2ea QY |
612 | assert(new_mpath); |
613 | assert(prev_mpath); | |
d62a17ae | 614 | if ((mpath_count < maxpaths) && (new_mpath != new_best) |
18ee8310 DS |
615 | && bgp_path_info_nexthop_cmp(prev_mpath, |
616 | new_mpath)) { | |
d62a17ae | 617 | if (new_mpath == next_mpath) |
18ee8310 DS |
618 | bgp_path_info_mpath_next(new_mpath); |
619 | bgp_path_info_mpath_dequeue(new_mpath); | |
d62a17ae | 620 | |
18ee8310 DS |
621 | bgp_path_info_mpath_enqueue(prev_mpath, |
622 | new_mpath); | |
d62a17ae | 623 | prev_mpath = new_mpath; |
624 | mpath_changed = 1; | |
625 | mpath_count++; | |
626 | if (debug) { | |
18ee8310 | 627 | bgp_path_info_path_with_addpath_rx_str( |
d62a17ae | 628 | new_mpath, path_buf); |
629 | zlog_debug( | |
630 | "%s: add mpath %s nexthop %s, cur count %d", | |
631 | pfx_buf, path_buf, | |
632 | inet_ntop(AF_INET, | |
633 | &new_mpath->attr | |
634 | ->nexthop, | |
635 | nh_buf[0], | |
636 | sizeof(nh_buf[0])), | |
637 | mpath_count); | |
638 | } | |
639 | } | |
640 | mp_node = mp_next_node; | |
641 | } | |
642 | } | |
643 | ||
644 | if (new_best) { | |
645 | if (debug) | |
646 | zlog_debug( | |
647 | "%s: New mpath count (incl newbest) %d mpath-change %s", | |
648 | pfx_buf, mpath_count, | |
649 | mpath_changed ? "YES" : "NO"); | |
650 | ||
18ee8310 | 651 | bgp_path_info_mpath_count_set(new_best, mpath_count - 1); |
d62a17ae | 652 | if (mpath_changed |
18ee8310 | 653 | || (bgp_path_info_mpath_count(new_best) != old_mpath_count)) |
1defdda8 | 654 | SET_FLAG(new_best->flags, BGP_PATH_MULTIPATH_CHG); |
d62a17ae | 655 | } |
de8d5dff | 656 | } |
6918e74b JB |
657 | |
658 | /* | |
659 | * bgp_mp_dmed_deselect | |
660 | * | |
1defdda8 | 661 | * Clean up multipath information for BGP_PATH_DMED_SELECTED path that |
6918e74b JB |
662 | * is not selected as best path |
663 | */ | |
4b7e6066 | 664 | void bgp_mp_dmed_deselect(struct bgp_path_info *dmed_best) |
6918e74b | 665 | { |
4b7e6066 | 666 | struct bgp_path_info *mpinfo, *mpnext; |
6918e74b | 667 | |
d62a17ae | 668 | if (!dmed_best) |
669 | return; | |
6918e74b | 670 | |
18ee8310 | 671 | for (mpinfo = bgp_path_info_mpath_first(dmed_best); mpinfo; |
d62a17ae | 672 | mpinfo = mpnext) { |
18ee8310 DS |
673 | mpnext = bgp_path_info_mpath_next(mpinfo); |
674 | bgp_path_info_mpath_dequeue(mpinfo); | |
d62a17ae | 675 | } |
6918e74b | 676 | |
18ee8310 | 677 | bgp_path_info_mpath_count_set(dmed_best, 0); |
1defdda8 | 678 | UNSET_FLAG(dmed_best->flags, BGP_PATH_MULTIPATH_CHG); |
831600c3 | 679 | assert(bgp_path_info_mpath_first(dmed_best) == NULL); |
6918e74b | 680 | } |
0b597ef0 JB |
681 | |
682 | /* | |
18ee8310 | 683 | * bgp_path_info_mpath_aggregate_update |
0b597ef0 JB |
684 | * |
685 | * Set the multipath aggregate attribute. We need to see if the | |
686 | * aggregate has changed and then set the ATTR_CHANGED flag on the | |
687 | * bestpath info so that a peer update will be generated. The | |
688 | * change is detected by generating the current attribute, | |
689 | * interning it, and then comparing the interned pointer with the | |
690 | * current value. We can skip this generate/compare step if there | |
691 | * is no change in multipath selection and no attribute change in | |
692 | * any multipath. | |
693 | */ | |
18ee8310 DS |
694 | void bgp_path_info_mpath_aggregate_update(struct bgp_path_info *new_best, |
695 | struct bgp_path_info *old_best) | |
0b597ef0 | 696 | { |
4b7e6066 | 697 | struct bgp_path_info *mpinfo; |
d62a17ae | 698 | struct aspath *aspath; |
699 | struct aspath *asmerge; | |
700 | struct attr *new_attr, *old_attr; | |
d7c0a89a | 701 | uint8_t origin; |
d62a17ae | 702 | struct community *community, *commerge; |
703 | struct ecommunity *ecomm, *ecommerge; | |
704 | struct lcommunity *lcomm, *lcommerge; | |
705 | struct attr attr = {0}; | |
706 | ||
707 | if (old_best && (old_best != new_best) | |
18ee8310 | 708 | && (old_attr = bgp_path_info_mpath_attr(old_best))) { |
d62a17ae | 709 | bgp_attr_unintern(&old_attr); |
18ee8310 | 710 | bgp_path_info_mpath_attr_set(old_best, NULL); |
d62a17ae | 711 | } |
712 | ||
713 | if (!new_best) | |
714 | return; | |
715 | ||
18ee8310 DS |
716 | if (!bgp_path_info_mpath_count(new_best)) { |
717 | if ((new_attr = bgp_path_info_mpath_attr(new_best))) { | |
d62a17ae | 718 | bgp_attr_unintern(&new_attr); |
18ee8310 | 719 | bgp_path_info_mpath_attr_set(new_best, NULL); |
1defdda8 | 720 | SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED); |
d62a17ae | 721 | } |
722 | return; | |
723 | } | |
724 | ||
725 | bgp_attr_dup(&attr, new_best->attr); | |
726 | ||
9d303b37 DL |
727 | if (new_best->peer && bgp_flag_check(new_best->peer->bgp, |
728 | BGP_FLAG_MULTIPATH_RELAX_AS_SET)) { | |
d62a17ae | 729 | |
730 | /* aggregate attribute from multipath constituents */ | |
731 | aspath = aspath_dup(attr.aspath); | |
732 | origin = attr.origin; | |
733 | community = | |
734 | attr.community ? community_dup(attr.community) : NULL; | |
735 | ecomm = (attr.ecommunity) ? ecommunity_dup(attr.ecommunity) | |
736 | : NULL; | |
737 | lcomm = (attr.lcommunity) ? lcommunity_dup(attr.lcommunity) | |
738 | : NULL; | |
739 | ||
18ee8310 DS |
740 | for (mpinfo = bgp_path_info_mpath_first(new_best); mpinfo; |
741 | mpinfo = bgp_path_info_mpath_next(mpinfo)) { | |
d62a17ae | 742 | asmerge = |
743 | aspath_aggregate(aspath, mpinfo->attr->aspath); | |
744 | aspath_free(aspath); | |
745 | aspath = asmerge; | |
746 | ||
747 | if (origin < mpinfo->attr->origin) | |
748 | origin = mpinfo->attr->origin; | |
749 | ||
750 | if (mpinfo->attr->community) { | |
751 | if (community) { | |
752 | commerge = community_merge( | |
753 | community, | |
754 | mpinfo->attr->community); | |
755 | community = | |
756 | community_uniq_sort(commerge); | |
3c1f53de | 757 | community_free(&commerge); |
d62a17ae | 758 | } else |
759 | community = community_dup( | |
760 | mpinfo->attr->community); | |
761 | } | |
762 | ||
763 | if (mpinfo->attr->ecommunity) { | |
764 | if (ecomm) { | |
765 | ecommerge = ecommunity_merge( | |
766 | ecomm, | |
767 | mpinfo->attr->ecommunity); | |
768 | ecomm = ecommunity_uniq_sort(ecommerge); | |
769 | ecommunity_free(&ecommerge); | |
770 | } else | |
771 | ecomm = ecommunity_dup( | |
772 | mpinfo->attr->ecommunity); | |
773 | } | |
774 | if (mpinfo->attr->lcommunity) { | |
775 | if (lcomm) { | |
776 | lcommerge = lcommunity_merge( | |
777 | lcomm, | |
778 | mpinfo->attr->lcommunity); | |
779 | lcomm = lcommunity_uniq_sort(lcommerge); | |
780 | lcommunity_free(&lcommerge); | |
781 | } else | |
782 | lcomm = lcommunity_dup( | |
783 | mpinfo->attr->lcommunity); | |
784 | } | |
785 | } | |
786 | ||
787 | attr.aspath = aspath; | |
788 | attr.origin = origin; | |
789 | if (community) { | |
790 | attr.community = community; | |
791 | attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_COMMUNITIES); | |
792 | } | |
793 | if (ecomm) { | |
794 | attr.ecommunity = ecomm; | |
795 | attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_EXT_COMMUNITIES); | |
796 | } | |
44f12f20 RW |
797 | if (lcomm) { |
798 | attr.lcommunity = lcomm; | |
799 | attr.flag |= ATTR_FLAG_BIT(BGP_ATTR_LARGE_COMMUNITIES); | |
800 | } | |
d62a17ae | 801 | |
802 | /* Zap multipath attr nexthop so we set nexthop to self */ | |
803 | attr.nexthop.s_addr = 0; | |
804 | memset(&attr.mp_nexthop_global, 0, sizeof(struct in6_addr)); | |
805 | ||
806 | /* TODO: should we set ATOMIC_AGGREGATE and AGGREGATOR? */ | |
807 | } | |
808 | ||
809 | new_attr = bgp_attr_intern(&attr); | |
810 | ||
18ee8310 DS |
811 | if (new_attr != bgp_path_info_mpath_attr(new_best)) { |
812 | if ((old_attr = bgp_path_info_mpath_attr(new_best))) | |
d62a17ae | 813 | bgp_attr_unintern(&old_attr); |
18ee8310 | 814 | bgp_path_info_mpath_attr_set(new_best, new_attr); |
1defdda8 | 815 | SET_FLAG(new_best->flags, BGP_PATH_ATTR_CHANGED); |
d62a17ae | 816 | } else |
817 | bgp_attr_unintern(&new_attr); | |
0b597ef0 | 818 | } |