Skip to content

Commit

Permalink
bgpd: Validate only affected RPKI prefixes instead of a full RIB
Browse files Browse the repository at this point in the history 
Before this fix, if rpki_sync_socket_rtr socket returns EAGAIN, then ALL routes
in the RIB are revalidated which takes lots of CPU and some unnecessary traffic,
e.g. if using BMP servers. With a full feed it would waste 50-80Mbps.

Instead we should try to drain an existing pipe (another end), and revalidate
only affected prefixes.

Signed-off-by: Donatas Abraitis <[email protected]>
(cherry picked from commit b0800bf)
  • Loading branch information
@ton31337 @mergify
ton31337 authored and mergify[bot] committed Dec 20, 2024
1 parent 1437880 commit 410eb0d
Show file tree
Hide file tree
Showing 3 changed files with 53 additions and 122 deletions.
170 changes: 53 additions & 117 deletions bgpd/bgp_rpki.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -155,7 +155,6 @@ static enum route_map_cmd_result_t route_match(void *rule,
void *object);
static void *route_match_compile(const char *arg);
static void revalidate_bgp_node(struct bgp_dest *dest, afi_t afi, safi_t safi);
static void revalidate_all_routes(struct rpki_vrf *rpki_vrf);

static bool rpki_debug_conf, rpki_debug_term;

Expand Down Expand Up @@ -586,48 +585,10 @@ static void rpki_revalidate_prefix(struct event *thread)
XFREE(MTYPE_BGP_RPKI_REVALIDATE, rrp);
}

static void bgpd_sync_callback(struct event *thread)
static void revalidate_single_prefix(struct vrf *vrf, struct prefix prefix, afi_t afi)
{
struct bgp *bgp;
struct listnode *node;
struct prefix prefix;
struct pfx_record rec;
struct rpki_vrf *rpki_vrf = EVENT_ARG(thread);
struct vrf *vrf = NULL;

event_add_read(bm->master, bgpd_sync_callback, rpki_vrf,
rpki_vrf->rpki_sync_socket_bgpd, NULL);

if (atomic_load_explicit(&rpki_vrf->rtr_update_overflow,
memory_order_seq_cst)) {
while (read(rpki_vrf->rpki_sync_socket_bgpd, &rec,
sizeof(struct pfx_record)) != -1)
;

atomic_store_explicit(&rpki_vrf->rtr_update_overflow, 0,
memory_order_seq_cst);
revalidate_all_routes(rpki_vrf);
return;
}

int retval = read(rpki_vrf->rpki_sync_socket_bgpd, &rec,
sizeof(struct pfx_record));
if (retval != sizeof(struct pfx_record)) {
RPKI_DEBUG("Could not read from rpki_sync_socket_bgpd");
return;
}
pfx_record_to_prefix(&rec, &prefix);

afi_t afi = (rec.prefix.ver == LRTR_IPV4) ? AFI_IP : AFI_IP6;

if (rpki_vrf->vrfname) {
vrf = vrf_lookup_by_name(rpki_vrf->vrfname);
if (!vrf) {
zlog_err("%s(): vrf for rpki %s not found", __func__,
rpki_vrf->vrfname);
return;
}
}

for (ALL_LIST_ELEMENTS_RO(bm->bgp, node, bgp)) {
safi_t safi;
Expand Down Expand Up @@ -655,101 +616,76 @@ static void bgpd_sync_callback(struct event *thread)
}
}

static void revalidate_bgp_node(struct bgp_dest *bgp_dest, afi_t afi,
safi_t safi)
static void bgpd_sync_callback(struct event *thread)
{
struct bgp_adj_in *ain;
mpls_label_t *label;
uint8_t num_labels;

for (ain = bgp_dest->adj_in; ain; ain = ain->next) {
struct bgp_path_info *path =
bgp_dest_get_bgp_path_info(bgp_dest);

num_labels = BGP_PATH_INFO_NUM_LABELS(path);
label = num_labels ? path->extra->labels->label : NULL;

(void)bgp_update(ain->peer, bgp_dest_get_prefix(bgp_dest),
ain->addpath_rx_id, ain->attr, afi, safi,
ZEBRA_ROUTE_BGP, BGP_ROUTE_NORMAL, NULL, label,
num_labels, 1, NULL);
}
}

/*
* The act of a soft reconfig in revalidation is really expensive
* coupled with the fact that the download of a full rpki state
* from a rpki server can be expensive, let's break up the revalidation
* to a point in time in the future to allow other bgp events
* to take place too.
*/
struct rpki_revalidate_peer {
struct prefix prefix;
struct pfx_record rec;
struct rpki_vrf *rpki_vrf = EVENT_ARG(thread);
struct vrf *vrf = NULL;
afi_t afi;
safi_t safi;
struct peer *peer;
};
int retval;

static void bgp_rpki_revalidate_peer(struct event *thread)
{
struct rpki_revalidate_peer *rvp = EVENT_ARG(thread);

/*
* Here's the expensive bit of gnomish deviousness
*/
bgp_soft_reconfig_in(rvp->peer, rvp->afi, rvp->safi);

XFREE(MTYPE_BGP_RPKI_REVALIDATE, rvp);
}

static void revalidate_all_routes(struct rpki_vrf *rpki_vrf)
{
struct bgp *bgp;
struct listnode *node;
struct vrf *vrf = NULL;
event_add_read(bm->master, bgpd_sync_callback, rpki_vrf, rpki_vrf->rpki_sync_socket_bgpd,
NULL);

if (rpki_vrf->vrfname) {
vrf = vrf_lookup_by_name(rpki_vrf->vrfname);
if (!vrf) {
zlog_err("%s(): vrf for rpki %s not found", __func__,
rpki_vrf->vrfname);
zlog_err("%s(): vrf for rpki %s not found", __func__, rpki_vrf->vrfname);
return;
}
}

for (ALL_LIST_ELEMENTS_RO(bm->bgp, node, bgp)) {
struct peer *peer;
struct listnode *peer_listnode;
if (atomic_load_explicit(&rpki_vrf->rtr_update_overflow, memory_order_seq_cst)) {
ssize_t size = 0;

if (!vrf && bgp->vrf_id != VRF_DEFAULT)
continue;
if (vrf && bgp->vrf_id != vrf->vrf_id)
continue;
retval = read(rpki_vrf->rpki_sync_socket_bgpd, &rec, sizeof(struct pfx_record));
while (retval != -1) {
if (retval != sizeof(struct pfx_record))
break;

for (ALL_LIST_ELEMENTS_RO(bgp->peer, peer_listnode, peer)) {
afi_t afi;
safi_t safi;
size += retval;
pfx_record_to_prefix(&rec, &prefix);
afi = (rec.prefix.ver == LRTR_IPV4) ? AFI_IP : AFI_IP6;
revalidate_single_prefix(vrf, prefix, afi);

FOREACH_AFI_SAFI (afi, safi) {
struct rpki_revalidate_peer *rvp;
retval = read(rpki_vrf->rpki_sync_socket_bgpd, &rec,
sizeof(struct pfx_record));
}

if (!bgp->rib[afi][safi])
continue;
RPKI_DEBUG("Socket overflow detected (%zu), revalidating affected prefixes", size);

if (!peer_established(peer->connection))
continue;
atomic_store_explicit(&rpki_vrf->rtr_update_overflow, 0, memory_order_seq_cst);
return;
}

rvp = XCALLOC(MTYPE_BGP_RPKI_REVALIDATE,
sizeof(*rvp));
rvp->peer = peer;
rvp->afi = afi;
rvp->safi = safi;
retval = read(rpki_vrf->rpki_sync_socket_bgpd, &rec, sizeof(struct pfx_record));
if (retval != sizeof(struct pfx_record)) {
RPKI_DEBUG("Could not read from rpki_sync_socket_bgpd");
return;
}
pfx_record_to_prefix(&rec, &prefix);

event_add_event(
bm->master, bgp_rpki_revalidate_peer,
rvp, 0,
&peer->t_revalidate_all[afi][safi]);
}
}
afi = (rec.prefix.ver == LRTR_IPV4) ? AFI_IP : AFI_IP6;

revalidate_single_prefix(vrf, prefix, afi);
}

static void revalidate_bgp_node(struct bgp_dest *bgp_dest, afi_t afi, safi_t safi)
{
struct bgp_adj_in *ain;
mpls_label_t *label;
uint8_t num_labels;

for (ain = bgp_dest->adj_in; ain; ain = ain->next) {
struct bgp_path_info *path = bgp_dest_get_bgp_path_info(bgp_dest);

num_labels = BGP_PATH_INFO_NUM_LABELS(path);
label = num_labels ? path->extra->labels->label : NULL;

(void)bgp_update(ain->peer, bgp_dest_get_prefix(bgp_dest), ain->addpath_rx_id,
ain->attr, afi, safi, ZEBRA_ROUTE_BGP, BGP_ROUTE_NORMAL, NULL,
label, num_labels, 1, NULL);
}
}

Expand Down
4 changes: 0 additions & 4 deletions bgpd/bgpd.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1255,8 +1255,6 @@ static void peer_free(struct peer *peer)
bgp_reads_off(peer->connection);
bgp_writes_off(peer->connection);
event_cancel_event_ready(bm->master, peer->connection);
FOREACH_AFI_SAFI (afi, safi)
EVENT_OFF(peer->t_revalidate_all[afi][safi]);
assert(!peer->connection->t_write);
assert(!peer->connection->t_read);

Expand Down Expand Up @@ -2723,8 +2721,6 @@ int peer_delete(struct peer *peer)
bgp_reads_off(peer->connection);
bgp_writes_off(peer->connection);
event_cancel_event_ready(bm->master, peer->connection);
FOREACH_AFI_SAFI (afi, safi)
EVENT_OFF(peer->t_revalidate_all[afi][safi]);
assert(!CHECK_FLAG(peer->connection->thread_flags,
PEER_THREAD_WRITES_ON));
assert(!CHECK_FLAG(peer->connection->thread_flags,
Expand Down
1 change: 0 additions & 1 deletion bgpd/bgpd.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1646,7 +1646,6 @@ struct peer {

/* Threads. */
struct event *t_llgr_stale[AFI_MAX][SAFI_MAX];
struct event *t_revalidate_all[AFI_MAX][SAFI_MAX];
struct event *t_refresh_stalepath;

/* Thread flags. */
Expand Down

0 comments on commit 410eb0d

Please sign in to comment.