When there is no incoming data traffic at the interface for a period,
BFD decay allows the bfd session to increase the min_rx. This is
helpful in that some interfaces usually idle for long time. And cpu
consumption can be reduced by processing fewer bfd control packets.
Signed-off-by: Alex Wang <al...@nicira.com>
---
lib/bfd.c | 110 +++++++++++++++++++++++++++++++++++++++++++++---
lib/bfd.h | 5 ++-
ofproto/ofproto-dpif.c | 7 ++-
vswitchd/vswitch.xml | 10 +++++
4 files changed, 125 insertions(+), 7 deletions(-)
diff --git a/lib/bfd.c b/lib/bfd.c
index d4ac489..6420bf9 100644
--- a/lib/bfd.c
+++ b/lib/bfd.c
@@ -26,6 +26,7 @@
#include "hash.h"
#include "hmap.h"
#include "list.h"
+#include "netdev.h"
#include "netlink.h"
#include "odp-util.h"
#include "ofpbuf.h"
@@ -149,6 +150,9 @@ struct bfd {
bool cpath_down; /* Concatenated Path Down. */
uint8_t mult; /* bfd.DetectMult. */
+ struct netdev *netdev;
+ uint64_t rx_packets; /* Packets received by 'netdev'. */
+
enum state state; /* bfd.SessionState. */
enum state rmt_state; /* bfd.RemoteSessionState. */
@@ -184,6 +188,10 @@ struct bfd {
atomic_bool check_tnl_key; /* Verify tunnel key of inbound packets? */
atomic_int ref_cnt;
+
+ /* BFD decay related variables. */
+ int decay_min_rx;
+ long long int decay_detect_time; /* Decay detection time. */
};
static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
@@ -206,6 +214,8 @@ static void bfd_set_state(struct bfd *, enum state, enum
diag)
static uint32_t generate_discriminator(void) OVS_REQ_WRLOCK(&mutex);
static void bfd_put_details(struct ds *, const struct bfd *)
OVS_REQ_WRLOCK(&mutex);
+static uint64_t bfd_rx_packets(const struct bfd *) OVS_REQ_WRLOCK(&mutex);
+static void bfd_decay(struct bfd *) OVS_REQ_WRLOCK(&mutex);
static void bfd_unixctl_show(struct unixctl_conn *, int argc,
const char *argv[], void *aux OVS_UNUSED);
static void bfd_unixctl_set_forwarding_override(struct unixctl_conn *,
@@ -253,12 +263,13 @@ bfd_get_status(const struct bfd *bfd, struct smap *smap)
* handle for the session, or NULL if BFD is not enabled according to 'cfg'.
* Also returns NULL if cfg is NULL. */
struct bfd *
-bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg)
- OVS_EXCLUDED(mutex)
+bfd_configure(struct bfd *bfd, const char *name, const struct smap *cfg,
+ struct netdev *netdev) OVS_EXCLUDED(mutex)
{
static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
static atomic_uint16_t udp_src = ATOMIC_VAR_INIT(0);
+ int decay_min_rx;
long long int min_tx, min_rx;
bool cpath_down;
const char *hwaddr;
@@ -290,6 +301,9 @@ bfd_configure(struct bfd *bfd, const char *name, const
struct smap *cfg)
bfd->min_tx = 1000;
bfd->mult = 3;
atomic_init(&bfd->ref_cnt, 1);
+ bfd->netdev = netdev_ref(netdev);
+ bfd->decay_detect_time = 0;
+ bfd->rx_packets = bfd_rx_packets(bfd);
/* RFC 5881 section 4
* The source port MUST be in the range 49152 through 65535. The same
@@ -322,9 +336,26 @@ bfd_configure(struct bfd *bfd, const char *name, const
struct smap *cfg)
if (bfd->cfg_min_rx != min_rx) {
bfd->cfg_min_rx = min_rx;
if (bfd->state != STATE_UP
- || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)) {
+ || (!bfd_in_poll(bfd) && bfd->cfg_min_rx > bfd->min_rx)
+ || bfd->min_rx == bfd->decay_min_rx) {
bfd->min_rx = bfd->cfg_min_rx;
}
+ bfd->decay_min_rx = 0;
+ bfd_poll(bfd);
+ }
+
+ decay_min_rx = smap_get_int(cfg, "decay_min_rx", 0);
+ if (bfd->decay_min_rx != decay_min_rx ) {
+ if (decay_min_rx > 0 && decay_min_rx < bfd->cfg_min_rx) {
+ VLOG_WARN("%s: decay_min_rx cannot be less than %lld ms",
+ bfd->name, bfd->cfg_min_rx);
+ bfd->decay_min_rx = 0;
+ } else {
+ bfd->decay_min_rx = decay_min_rx;
+ }
+ bfd->min_rx = bfd->cfg_min_rx;
+ bfd->decay_detect_time = (bfd->decay_min_rx < 2000 ?
+ 2000 : bfd->decay_min_rx) + time_msec();
bfd_poll(bfd);
}
@@ -373,6 +404,7 @@ bfd_unref(struct bfd *bfd) OVS_EXCLUDED(mutex)
if (orig == 1) {
ovs_mutex_lock(&mutex);
hmap_remove(all_bfds, &bfd->node);
+ netdev_close(bfd->netdev);
free(bfd->name);
free(bfd);
ovs_mutex_unlock(&mutex);
@@ -398,14 +430,29 @@ bfd_wait(const struct bfd *bfd) OVS_EXCLUDED(mutex)
void
bfd_run(struct bfd *bfd) OVS_EXCLUDED(mutex)
{
+ long long int now;
+
ovs_mutex_lock(&mutex);
- if (bfd->state > STATE_DOWN && time_msec() >= bfd->detect_time) {
+ now = time_msec();
+
+ if (bfd->state > STATE_DOWN && now >= bfd->detect_time) {
bfd_set_state(bfd, STATE_DOWN, DIAG_EXPIRED);
}
+ if (bfd->state == STATE_UP && bfd->decay_min_rx > 0
+ && now >= bfd->decay_detect_time) {
+ bfd_decay(bfd);
+ }
+
if (bfd->min_tx != bfd->cfg_min_tx || bfd->min_rx != bfd->cfg_min_rx) {
+ /* Do not poll if already decayed to decay_min_rx. */
+ if (bfd->state == STATE_UP && bfd->poll_min_rx == bfd->decay_min_rx
+ && bfd->min_tx == bfd->cfg_min_tx) {
+ goto exit;
+ }
bfd_poll(bfd);
}
+exit:
ovs_mutex_unlock(&mutex);
}
@@ -680,6 +727,20 @@ bfd_process_packet(struct bfd *bfd, const struct flow
*flow,
out:
ovs_mutex_unlock(&mutex);
}
+
+/* Must be called when the netdev owned by 'bfd' should change. */
+void
+bfd_set_netdev(struct bfd *bfd, const struct netdev *netdev)
+ OVS_EXCLUDED(mutex)
+{
+ ovs_mutex_lock(&mutex);
+ if (bfd->netdev != netdev) {
+ netdev_close(bfd->netdev);
+ bfd->netdev = netdev_ref(netdev);
+ }
+ ovs_mutex_unlock(&mutex);
+}
+
static bool
bfd_forwarding__(const struct bfd *bfd) OVS_REQ_WRLOCK(mutex)
@@ -707,7 +768,8 @@ bfd_poll(struct bfd *bfd) OVS_REQ_WRLOCK(mutex)
if (bfd->state > STATE_DOWN && !bfd_in_poll(bfd)
&& !(bfd->flags & FLAG_FINAL)) {
bfd->poll_min_tx = bfd->cfg_min_tx;
- bfd->poll_min_rx = bfd->cfg_min_rx;
+ bfd->poll_min_rx = bfd->min_rx == bfd->decay_min_rx
+ ? bfd->decay_min_rx : bfd->cfg_min_rx;
bfd->flags |= FLAG_POLL;
bfd->next_tx = 0;
VLOG_INFO_RL(&rl, "%s: Initiating poll sequence", bfd->name);
@@ -882,6 +944,44 @@ bfd_set_state(struct bfd *bfd, enum state state, enum diag
diag)
}
}
+static uint64_t
+bfd_rx_packets(const struct bfd *bfd) OVS_REQ_WRLOCK(mutex)
+{
+ struct netdev_stats stats;
+
+ if (!netdev_get_stats(bfd->netdev, &stats)) {
+ return stats.rx_packets;
+ } else {
+ return 0;
+ }
+}
+
+static void
+bfd_decay(struct bfd *bfd) OVS_REQ_WRLOCK(mutex)
+{
+ uint64_t rx_packets = bfd_rx_packets(bfd);
+ int64_t diff, measure;
+
+ diff = rx_packets - bfd->rx_packets;
+ bfd->rx_packets = rx_packets;
+ bfd->decay_detect_time = (bfd->decay_min_rx < 2000 ?
+ 2000 : bfd->decay_min_rx) + time_msec();
+ measure = (bfd->decay_min_rx < 2000 ? 2000 : bfd->decay_min_rx)
+ / bfd->min_rx + 5;
+
+ if (diff <= measure) {
+ /* Decay when there is no obvious data traffic. */
+ if (bfd->min_rx != bfd->decay_min_rx) {
+ bfd->min_rx = bfd->decay_min_rx;
+ }
+ } else {
+ /* Restore the min_rx. */
+ if (bfd->min_rx != bfd->cfg_min_rx) {
+ bfd->min_rx = bfd->cfg_min_rx;
+ }
+ }
+}
+
static uint32_t
generate_discriminator(void)
{
diff --git a/lib/bfd.h b/lib/bfd.h
index 67d012e..0e1e33d 100644
--- a/lib/bfd.h
+++ b/lib/bfd.h
@@ -24,6 +24,7 @@
struct bfd;
struct flow;
struct flow_wildcards;
+struct netdev;
struct ofpbuf;
struct smap;
@@ -40,11 +41,13 @@ void bfd_process_packet(struct bfd *, const struct flow *,
const struct ofpbuf *);
struct bfd *bfd_configure(struct bfd *, const char *name,
- const struct smap *smap);
+ const struct smap *smap,
+ struct netdev *netdev);
struct bfd *bfd_ref(const struct bfd *);
void bfd_unref(struct bfd *);
bool bfd_forwarding(const struct bfd *);
void bfd_get_status(const struct bfd *, struct smap *);
+void bfd_set_netdev(struct bfd *, const struct netdev *);
#endif /* bfd.h */
diff --git a/ofproto/ofproto-dpif.c b/ofproto/ofproto-dpif.c
index a8e5cd5..75fd96c 100644
--- a/ofproto/ofproto-dpif.c
+++ b/ofproto/ofproto-dpif.c
@@ -1862,6 +1862,10 @@ port_modified(struct ofport *port_)
cfm_set_netdev(port->cfm, port->up.netdev);
}
+ if (port->bfd) {
+ bfd_set_netdev(port->bfd, port->up.netdev);
+ }
+
if (port->is_tunnel && tnl_port_reconfigure(port, port->up.netdev,
port->odp_port)) {
ofproto_dpif_cast(port->up.ofproto)->backer->need_revalidate =
@@ -1996,7 +2000,8 @@ set_bfd(struct ofport *ofport_, const struct smap *cfg)
struct bfd *old;
old = ofport->bfd;
- ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
+ ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev),
+ cfg, ofport->up.netdev);
if (ofport->bfd != old) {
ofproto->backer->need_revalidate = REV_RECONFIGURE;
}
diff --git a/vswitchd/vswitch.xml b/vswitchd/vswitch.xml
index b89d58c..b73a612 100644
--- a/vswitchd/vswitch.xml
+++ b/vswitchd/vswitch.xml
@@ -1880,6 +1880,16 @@
specified. Defaults to <code>100</code>.
</column>
+ <column name="bfd" key="decay_min_rx" type='{"type": "integer"}'>
+ <code>decay_min_rx</code> is used to set the <code>min_rx</code>,
+ when there is no obvious incoming data traffic at the interface.
+ It cannot be less than the <code>min_rx</code>. The decay feature
+ is disable by setting the <code>decay_min_rx</code> to 0. And the
+ feature is reset everytime itself or <code>min_rx</code> is
+ reconfigured.
+ </column>
+
+
<column name="bfd" key="cpath_down" type='{"type": "boolean"}'>
Concatenated path down may be used when the local system should not
have traffic forwarded to it for some reason other than a connectivty
--
1.7.9.5
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
