Skip to content

Commit 545aeef

Browse files
rwestphalGalaxyGorilla
rwestphal
authored and
GalaxyGorilla
committed
bgpd: extend the NHT code to understand SR-TE colors
Extend the NHT code so that only the affected BGP routes are affected whenever an SR-policy is updated on zebra. Signed-off-by: Renato Westphal <[email protected]>
1 parent ef3e0d0 commit 545aeef

File tree

3 files changed

+121
-70
lines changed

3 files changed

+121
-70
lines changed

‎bgpd/bgp_nexthop.c

Lines changed: 11 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -50,6 +50,11 @@ DEFINE_MTYPE_STATIC(BGPD, MARTIAN_STRING, "BGP Martian Address Intf String");
5050
int bgp_nexthop_cache_compare(const struct bgp_nexthop_cache *a,
5151
const struct bgp_nexthop_cache *b)
5252
{
53+
if (a->srte_color < b->srte_color)
54+
return -1;
55+
if (a->srte_color > b->srte_color)
56+
return 1;
57+
5358
return prefix_cmp(&a->prefix, &b->prefix);
5459
}
5560

@@ -64,13 +69,14 @@ void bnc_nexthop_free(struct bgp_nexthop_cache *bnc)
6469
}
6570

6671
struct bgp_nexthop_cache *bnc_new(struct bgp_nexthop_cache_head *tree,
67-
struct prefix *prefix)
72+
struct prefix *prefix, uint32_t srte_color)
6873
{
6974
struct bgp_nexthop_cache *bnc;
7075

7176
bnc = XCALLOC(MTYPE_BGP_NEXTHOP_CACHE,
7277
sizeof(struct bgp_nexthop_cache));
7378
bnc->prefix = *prefix;
79+
bnc->srte_color = srte_color;
7480
bnc->tree = tree;
7581
LIST_INIT(&(bnc->paths));
7682
bgp_nexthop_cache_add(tree, bnc);
@@ -86,14 +92,15 @@ void bnc_free(struct bgp_nexthop_cache *bnc)
8692
}
8793

8894
struct bgp_nexthop_cache *bnc_find(struct bgp_nexthop_cache_head *tree,
89-
struct prefix *prefix)
95+
struct prefix *prefix, uint32_t srte_color)
9096
{
9197
struct bgp_nexthop_cache bnc = {};
9298

9399
if (!tree)
94100
return NULL;
95101

96102
bnc.prefix = *prefix;
103+
bnc.srte_color = srte_color;
97104
return bgp_nexthop_cache_find(tree, &bnc);
98105
}
99106

@@ -799,6 +806,8 @@ static void bgp_show_nexthop(struct vty *vty, struct bgp *bgp,
799806

800807
peer = (struct peer *)bnc->nht_info;
801808

809+
if (bnc->srte_color)
810+
vty_out(vty, " SR-TE color %u -", bnc->srte_color);
802811
if (CHECK_FLAG(bnc->flags, BGP_NEXTHOP_VALID)) {
803812
vty_out(vty, " %s valid [IGP metric %d], #paths %d",
804813
inet_ntop(bnc->prefix.family, &bnc->prefix.u.prefix,

‎bgpd/bgp_nexthop.h

Lines changed: 5 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -70,6 +70,7 @@ struct bgp_nexthop_cache {
7070
/* Back pointer to the cache tree this entry belongs to. */
7171
struct bgp_nexthop_cache_head *tree;
7272

73+
uint32_t srte_color;
7374
struct prefix prefix;
7475
void *nht_info; /* In BGP, peer session */
7576
LIST_HEAD(path_list, bgp_path_info) paths;
@@ -115,10 +116,12 @@ extern bool bgp_nexthop_self(struct bgp *bgp, afi_t afi, uint8_t type,
115116
uint8_t sub_type, struct attr *attr,
116117
struct bgp_dest *dest);
117118
extern struct bgp_nexthop_cache *bnc_new(struct bgp_nexthop_cache_head *tree,
118-
struct prefix *prefix);
119+
struct prefix *prefix,
120+
uint32_t srte_color);
119121
extern void bnc_free(struct bgp_nexthop_cache *bnc);
120122
extern struct bgp_nexthop_cache *bnc_find(struct bgp_nexthop_cache_head *tree,
121-
struct prefix *prefix);
123+
struct prefix *prefix,
124+
uint32_t srte_color);
122125
extern void bnc_nexthop_free(struct bgp_nexthop_cache *bnc);
123126
extern const char *bnc_str(struct bgp_nexthop_cache *bnc, char *buf, int size);
124127
extern void bgp_scan_init(struct bgp *bgp);

‎bgpd/bgp_nht.c

Lines changed: 105 additions & 66 deletions
Original file line numberDiff line numberDiff line change
@@ -72,9 +72,9 @@ static void bgp_unlink_nexthop_check(struct bgp_nexthop_cache *bnc)
7272
if (LIST_EMPTY(&(bnc->paths)) && !bnc->nht_info) {
7373
if (BGP_DEBUG(nht, NHT)) {
7474
char buf[PREFIX2STR_BUFFER];
75-
zlog_debug("bgp_unlink_nexthop: freeing bnc %s(%s)",
75+
zlog_debug("bgp_unlink_nexthop: freeing bnc %s(%u)(%s)",
7676
bnc_str(bnc, buf, PREFIX2STR_BUFFER),
77-
bnc->bgp->name_pretty);
77+
bnc->srte_color, bnc->bgp->name_pretty);
7878
}
7979
unregister_zebra_rnh(bnc,
8080
CHECK_FLAG(bnc->flags, BGP_STATIC_ROUTE));
@@ -103,7 +103,7 @@ void bgp_unlink_nexthop_by_peer(struct peer *peer)
103103
if (!sockunion2hostprefix(&peer->su, &p))
104104
return;
105105

106-
bnc = bnc_find(&peer->bgp->nexthop_cache_table[afi], &p);
106+
bnc = bnc_find(&peer->bgp->nexthop_cache_table[afi], &p, 0);
107107
if (!bnc)
108108
return;
109109

@@ -124,6 +124,7 @@ int bgp_find_or_add_nexthop(struct bgp *bgp_route, struct bgp *bgp_nexthop,
124124
struct bgp_nexthop_cache_head *tree = NULL;
125125
struct bgp_nexthop_cache *bnc;
126126
struct prefix p;
127+
uint32_t srte_color = 0;
127128
int is_bgp_static_route = 0;
128129

129130
if (pi) {
@@ -148,6 +149,8 @@ int bgp_find_or_add_nexthop(struct bgp *bgp_route, struct bgp *bgp_nexthop,
148149
* addr */
149150
if (make_prefix(afi, pi, &p) < 0)
150151
return 1;
152+
153+
srte_color = pi->attr->srte_color;
151154
} else if (peer) {
152155
if (!sockunion2hostprefix(&peer->su, &p)) {
153156
if (BGP_DEBUG(nht, NHT)) {
@@ -165,16 +168,17 @@ int bgp_find_or_add_nexthop(struct bgp *bgp_route, struct bgp *bgp_nexthop,
165168
else
166169
tree = &bgp_nexthop->nexthop_cache_table[afi];
167170

168-
bnc = bnc_find(tree, &p);
171+
bnc = bnc_find(tree, &p, srte_color);
169172
if (!bnc) {
170-
bnc = bnc_new(tree, &p);
173+
bnc = bnc_new(tree, &p, srte_color);
171174
bnc->bgp = bgp_nexthop;
172175
if (BGP_DEBUG(nht, NHT)) {
173176
char buf[PREFIX2STR_BUFFER];
174177

175-
zlog_debug("Allocated bnc %s(%s) peer %p",
178+
zlog_debug("Allocated bnc %s(%u)(%s) peer %p",
176179
bnc_str(bnc, buf, PREFIX2STR_BUFFER),
177-
bnc->bgp->name_pretty, peer);
180+
bnc->srte_color, bnc->bgp->name_pretty,
181+
peer);
178182
}
179183
}
180184

@@ -266,7 +270,7 @@ void bgp_delete_connected_nexthop(afi_t afi, struct peer *peer)
266270
return;
267271

268272
bnc = bnc_find(&peer->bgp->nexthop_cache_table[family2afi(p.family)],
269-
&p);
273+
&p, 0);
270274
if (!bnc) {
271275
if (BGP_DEBUG(nht, NHT))
272276
zlog_debug(
@@ -296,89 +300,52 @@ void bgp_delete_connected_nexthop(afi_t afi, struct peer *peer)
296300
}
297301
}
298302

299-
void bgp_parse_nexthop_update(int command, vrf_id_t vrf_id)
303+
static void bgp_process_nexthop_update(struct bgp_nexthop_cache *bnc,
304+
struct zapi_route *nhr)
300305
{
301-
struct bgp_nexthop_cache_head *tree = NULL;
302-
struct bgp_nexthop_cache *bnc;
303306
struct nexthop *nexthop;
304307
struct nexthop *oldnh;
305308
struct nexthop *nhlist_head = NULL;
306309
struct nexthop *nhlist_tail = NULL;
307310
int i;
308-
struct bgp *bgp;
309-
struct zapi_route nhr;
310-
311-
bgp = bgp_lookup_by_vrf_id(vrf_id);
312-
if (!bgp) {
313-
flog_err(
314-
EC_BGP_NH_UPD,
315-
"parse nexthop update: instance not found for vrf_id %u",
316-
vrf_id);
317-
return;
318-
}
319-
320-
if (!zapi_nexthop_update_decode(zclient->ibuf, &nhr)) {
321-
if (BGP_DEBUG(nht, NHT))
322-
zlog_debug("%s[%s]: Failure to decode nexthop update",
323-
__func__, bgp->name_pretty);
324-
return;
325-
}
326-
327-
if (command == ZEBRA_NEXTHOP_UPDATE)
328-
tree = &bgp->nexthop_cache_table[family2afi(nhr.prefix.family)];
329-
else if (command == ZEBRA_IMPORT_CHECK_UPDATE)
330-
tree = &bgp->import_check_table[family2afi(nhr.prefix.family)];
331-
332-
bnc = bnc_find(tree, &nhr.prefix);
333-
if (!bnc) {
334-
if (BGP_DEBUG(nht, NHT)) {
335-
char buf[PREFIX2STR_BUFFER];
336-
337-
prefix2str(&nhr.prefix, buf, sizeof(buf));
338-
zlog_debug(
339-
"parse nexthop update(%s(%s)): bnc info not found",
340-
buf, bgp->name_pretty);
341-
}
342-
return;
343-
}
344311

345312
bnc->last_update = bgp_clock();
346313
bnc->change_flags = 0;
347314

348315
/* debug print the input */
349316
if (BGP_DEBUG(nht, NHT)) {
350317
char buf[PREFIX2STR_BUFFER];
351-
prefix2str(&nhr.prefix, buf, sizeof(buf));
318+
prefix2str(&nhr->prefix, buf, sizeof(buf));
352319
zlog_debug(
353-
"%s(%u): Rcvd NH update %s - metric %d/%d #nhops %d/%d flags 0x%x",
354-
bnc->bgp->name_pretty, vrf_id, buf, nhr.metric,
355-
bnc->metric, nhr.nexthop_num, bnc->nexthop_num,
356-
bnc->flags);
320+
"%s(%u): Rcvd NH update %s(%u) - metric %d/%d #nhops %d/%d flags 0x%x",
321+
bnc->bgp->name_pretty, bnc->bgp->vrf_id, buf,
322+
bnc->srte_color, nhr->metric, bnc->metric,
323+
nhr->nexthop_num, bnc->nexthop_num, bnc->flags);
357324
}
358325

359-
if (nhr.metric != bnc->metric)
326+
if (nhr->metric != bnc->metric)
360327
bnc->change_flags |= BGP_NEXTHOP_METRIC_CHANGED;
361328

362-
if (nhr.nexthop_num != bnc->nexthop_num)
329+
if (nhr->nexthop_num != bnc->nexthop_num)
363330
bnc->change_flags |= BGP_NEXTHOP_CHANGED;
364331

365-
if (nhr.nexthop_num) {
332+
if (nhr->nexthop_num) {
366333
struct peer *peer = bnc->nht_info;
367334

368335
/* notify bgp fsm if nbr ip goes from invalid->valid */
369336
if (!bnc->nexthop_num)
370337
UNSET_FLAG(bnc->flags, BGP_NEXTHOP_PEER_NOTIFIED);
371338

372339
bnc->flags |= BGP_NEXTHOP_VALID;
373-
bnc->metric = nhr.metric;
374-
bnc->nexthop_num = nhr.nexthop_num;
340+
bnc->metric = nhr->metric;
341+
bnc->nexthop_num = nhr->nexthop_num;
375342

376343
bnc->flags &= ~BGP_NEXTHOP_LABELED_VALID; /* check below */
377344

378-
for (i = 0; i < nhr.nexthop_num; i++) {
345+
for (i = 0; i < nhr->nexthop_num; i++) {
379346
int num_labels = 0;
380347

381-
nexthop = nexthop_from_zapi_nexthop(&nhr.nexthops[i]);
348+
nexthop = nexthop_from_zapi_nexthop(&nhr->nexthops[i]);
382349

383350
/*
384351
* Turn on RA for the v6 nexthops
@@ -388,7 +355,7 @@ void bgp_parse_nexthop_update(int command, vrf_id_t vrf_id)
388355
if (peer && !peer->ifp
389356
&& CHECK_FLAG(peer->flags,
390357
PEER_FLAG_CAPABILITY_ENHE)
391-
&& nhr.prefix.family == AF_INET6
358+
&& nhr->prefix.family == AF_INET6
392359
&& nexthop->type != NEXTHOP_TYPE_BLACKHOLE) {
393360
struct interface *ifp;
394361

@@ -442,7 +409,7 @@ void bgp_parse_nexthop_update(int command, vrf_id_t vrf_id)
442409
bnc->nexthop = nhlist_head;
443410
} else {
444411
bnc->flags &= ~BGP_NEXTHOP_VALID;
445-
bnc->nexthop_num = nhr.nexthop_num;
412+
bnc->nexthop_num = nhr->nexthop_num;
446413

447414
/* notify bgp fsm if nbr ip goes from valid->invalid */
448415
UNSET_FLAG(bnc->flags, BGP_NEXTHOP_PEER_NOTIFIED);
@@ -454,6 +421,77 @@ void bgp_parse_nexthop_update(int command, vrf_id_t vrf_id)
454421
evaluate_paths(bnc);
455422
}
456423

424+
void bgp_parse_nexthop_update(int command, vrf_id_t vrf_id)
425+
{
426+
struct bgp_nexthop_cache_head *tree = NULL;
427+
struct bgp_nexthop_cache *bnc;
428+
struct bgp *bgp;
429+
struct zapi_route nhr;
430+
afi_t afi;
431+
432+
bgp = bgp_lookup_by_vrf_id(vrf_id);
433+
if (!bgp) {
434+
flog_err(
435+
EC_BGP_NH_UPD,
436+
"parse nexthop update: instance not found for vrf_id %u",
437+
vrf_id);
438+
return;
439+
}
440+
441+
if (!zapi_nexthop_update_decode(zclient->ibuf, &nhr)) {
442+
if (BGP_DEBUG(nht, NHT))
443+
zlog_debug("%s[%s]: Failure to decode nexthop update",
444+
__PRETTY_FUNCTION__, bgp->name_pretty);
445+
return;
446+
}
447+
448+
afi = family2afi(nhr.prefix.family);
449+
if (command == ZEBRA_NEXTHOP_UPDATE)
450+
tree = &bgp->nexthop_cache_table[afi];
451+
else if (command == ZEBRA_IMPORT_CHECK_UPDATE)
452+
tree = &bgp->import_check_table[afi];
453+
454+
bnc = bnc_find(tree, &nhr.prefix, nhr.srte_color);
455+
if (!bnc) {
456+
if (BGP_DEBUG(nht, NHT)) {
457+
char buf[PREFIX2STR_BUFFER];
458+
459+
prefix2str(&nhr.prefix, buf, sizeof(buf));
460+
zlog_debug(
461+
"parse nexthop update(%s(%u)(%s)): bnc info not found",
462+
buf, nhr.srte_color, bgp->name_pretty);
463+
}
464+
return;
465+
}
466+
467+
bgp_process_nexthop_update(bnc, &nhr);
468+
469+
/*
470+
* HACK: if any BGP route is dependant on an SR-policy that doesn't
471+
* exist, zebra will never send NH updates relative to that policy. In
472+
* that case, whenever we receive an update about a colorless NH, update
473+
* the corresponding colorful NHs that share the same endpoint but that
474+
* are inactive. This ugly hack should work around the problem at the
475+
* cost of a performance pernalty. Long term, what should be done is to
476+
* make zebra's RNH subsystem aware of SR-TE colors (like bgpd is),
477+
* which should provide a better infrastructure to solve this issue in
478+
* a more efficient and elegant way.
479+
*/
480+
if (nhr.srte_color == 0) {
481+
struct bgp_nexthop_cache *bnc_iter;
482+
483+
frr_each (bgp_nexthop_cache, &bgp->nexthop_cache_table[afi],
484+
bnc_iter) {
485+
if (!prefix_same(&bnc->prefix, &bnc_iter->prefix)
486+
|| bnc_iter->srte_color == 0
487+
|| CHECK_FLAG(bnc_iter->flags, BGP_NEXTHOP_VALID))
488+
continue;
489+
490+
bgp_process_nexthop_update(bnc_iter, &nhr);
491+
}
492+
}
493+
}
494+
457495
/*
458496
* Cleanup nexthop registration and status information for BGP nexthops
459497
* pertaining to this VRF. This is invoked upon VRF deletion.
@@ -667,8 +705,8 @@ static void evaluate_paths(struct bgp_nexthop_cache *bnc)
667705
char buf[PREFIX2STR_BUFFER];
668706
bnc_str(bnc, buf, PREFIX2STR_BUFFER);
669707
zlog_debug(
670-
"NH update for %s %s flags 0x%x chgflags 0x%x - evaluate paths",
671-
buf, bnc->bgp->name_pretty, bnc->flags,
708+
"NH update for %s(%u)(%s) - flags 0x%x chgflags 0x%x - evaluate paths",
709+
buf, bnc->srte_color, bnc->bgp->name_pretty, bnc->flags,
672710
bnc->change_flags);
673711
}
674712

@@ -756,7 +794,8 @@ static void evaluate_paths(struct bgp_nexthop_cache *bnc)
756794
path->extra->igpmetric = 0;
757795

758796
if (CHECK_FLAG(bnc->change_flags, BGP_NEXTHOP_METRIC_CHANGED)
759-
|| CHECK_FLAG(bnc->change_flags, BGP_NEXTHOP_CHANGED))
797+
|| CHECK_FLAG(bnc->change_flags, BGP_NEXTHOP_CHANGED)
798+
|| path->attr->srte_color != 0)
760799
SET_FLAG(path->flags, BGP_PATH_IGP_CHANGED);
761800

762801
path_valid = !!CHECK_FLAG(path->flags, BGP_PATH_VALID);
@@ -874,7 +913,7 @@ void bgp_nht_reg_enhe_cap_intfs(struct peer *peer)
874913
if (p.family != AF_INET6)
875914
return;
876915

877-
bnc = bnc_find(&bgp->nexthop_cache_table[AFI_IP6], &p);
916+
bnc = bnc_find(&bgp->nexthop_cache_table[AFI_IP6], &p, 0);
878917
if (!bnc)
879918
return;
880919

@@ -916,7 +955,7 @@ void bgp_nht_dereg_enhe_cap_intfs(struct peer *peer)
916955
if (p.family != AF_INET6)
917956
return;
918957

919-
bnc = bnc_find(&bgp->nexthop_cache_table[AFI_IP6], &p);
958+
bnc = bnc_find(&bgp->nexthop_cache_table[AFI_IP6], &p, 0);
920959
if (!bnc)
921960
return;
922961

0 commit comments

Comments
 (0)