diff mbox series

[ovs-dev,21.12] controller: make garp_max_timeout configurable

Message ID 34b004347e951638a6c4d55aa5eb371c9e5d3220.1701884492.git.lorenzo.bianconi@redhat.com
State Rejected
Delegated to: Dumitru Ceara
Headers show
Series [ovs-dev,21.12] controller: make garp_max_timeout configurable | expand

Checks

Context Check Description
ovsrobot/apply-robot fail apply and check: fail

Commit Message

Lorenzo Bianconi Dec. 6, 2023, 5:43 p.m. UTC
When using VLAN backed networks and OVN routers leveraging the
'ovn-chassis-mac-mappings' option for east-west traffic, the eth.src field is
replaced by the chassis mac address in order to not expose the router mac
address from different nodes and confuse the TOR switch. However doing so
the TOR switch is not able to learn the port/mac bindings for routed E/W
traffic and it is force to always flood it. Fix this issue adding the
capability to configure a given timeout for garp sent by ovn-controller
and not disable it after the exponential backoff in order to keep
refreshing the entries in TOR swtich fdb table.
More into about the issue can be found here [0].

[0] https://mail.openvswitch.org/pipermail/ovs-discuss/2020-September/050678.html
Reported-at: https://bugzilla.redhat.com/show_bug.cgi?id=2087779

Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>
Acked-by: Ales Musil <amusil@redhat.com>
Signed-off-by: Mark Michelson <mmichels@redhat.com>
---
 NEWS                            |  2 +
 controller/ovn-controller.8.xml | 11 +++++
 controller/ovn-controller.c     |  4 +-
 controller/pinctrl.c            | 76 +++++++++++++++++++++++++++------
 controller/pinctrl.h            |  4 +-
 5 files changed, 81 insertions(+), 16 deletions(-)
diff mbox series

Patch

diff --git a/NEWS b/NEWS
index 4c300ebc1..69ae85635 100644
--- a/NEWS
+++ b/NEWS
@@ -3,6 +3,8 @@  OVN v21.12.3 - xx xxx xxxx
   - A new ACL option, "log-related" has been added that allows for reply and
     related traffic to be logged for an ACL in addition to the traffic that
     directly matches the ACL.
+  - Add "garp-max-timeout-sec" config option to vswitchd external-ids to
+    cap the time between when ovn-controller sends gARP packets.
 
 OVN v21.12.2 - 03 Jun 2022
 --------------------------
diff --git a/controller/ovn-controller.8.xml b/controller/ovn-controller.8.xml
index 68e7de2d4..cb54862bf 100644
--- a/controller/ovn-controller.8.xml
+++ b/controller/ovn-controller.8.xml
@@ -338,6 +338,17 @@ 
         of how many entries there are in the cache.  By default this is set to
         30000 (30 seconds).
       </dd>
+      <dt><code>external_ids:garp-max-timeout-sec</code></dt>
+      <dd>
+        When used, this configuration value specifies the maximum timeout
+        (in seconds) between two consecutive GARP packets sent by
+        <code>ovn-controller</code>.
+        <code>ovn-controller</code> by default sends just 4 GARP packets
+        with an exponential backoff timeout.
+        Setting <code>external_ids:garp-max-timeout-sec</code> allows to
+        cap for the exponential backoff used by <code>ovn-controller</code>
+        to send GARPs packets.
+      </dd>
     </dl>
 
     <p>
diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c
index 797222895..671485a47 100644
--- a/controller/ovn-controller.c
+++ b/controller/ovn-controller.c
@@ -3888,7 +3888,9 @@  main(int argc, char *argv[])
                                     &runtime_data->local_datapaths,
                                     &runtime_data->active_tunnels,
                                     &runtime_data->local_active_ports_ipv6_pd,
-                                    &runtime_data->local_active_ports_ras);
+                                    &runtime_data->local_active_ports_ras,
+                                    ovsrec_open_vswitch_table_get(
+                                            ovs_idl_loop.idl));
                         stopwatch_stop(PINCTRL_RUN_STOPWATCH_NAME,
                                        time_msec());
                         /* Updating monitor conditions if runtime data or
diff --git a/controller/pinctrl.c b/controller/pinctrl.c
index 499da59da..829a22aa3 100644
--- a/controller/pinctrl.c
+++ b/controller/pinctrl.c
@@ -164,6 +164,10 @@  static struct ovs_mutex pinctrl_mutex = OVS_MUTEX_INITIALIZER;
 static struct seq *pinctrl_handler_seq;
 static struct seq *pinctrl_main_seq;
 
+#define GARP_RARP_DEF_MAX_TIMEOUT    16000
+static long long int garp_rarp_max_timeout = GARP_RARP_DEF_MAX_TIMEOUT;
+static bool garp_rarp_continuous;
+
 static void *pinctrl_handler(void *arg);
 
 struct pinctrl {
@@ -209,7 +213,8 @@  static void send_garp_rarp_prepare(
     const struct ovsrec_bridge *,
     const struct sbrec_chassis *,
     const struct hmap *local_datapaths,
-    const struct sset *active_tunnels)
+    const struct sset *active_tunnels,
+    const struct ovsrec_open_vswitch_table *ovs_table)
     OVS_REQUIRES(pinctrl_mutex);
 static void send_garp_rarp_run(struct rconn *swconn,
                                long long int *send_garp_rarp_time)
@@ -3475,7 +3480,8 @@  pinctrl_run(struct ovsdb_idl_txn *ovnsb_idl_txn,
             const struct hmap *local_datapaths,
             const struct sset *active_tunnels,
             const struct shash *local_active_ports_ipv6_pd,
-            const struct shash *local_active_ports_ras)
+            const struct shash *local_active_ports_ras,
+            const struct ovsrec_open_vswitch_table *ovs_table)
 {
     ovs_mutex_lock(&pinctrl_mutex);
     pinctrl_set_br_int_name_(br_int->name);
@@ -3487,7 +3493,7 @@  pinctrl_run(struct ovsdb_idl_txn *ovnsb_idl_txn,
     send_garp_rarp_prepare(ovnsb_idl_txn, sbrec_port_binding_by_datapath,
                            sbrec_port_binding_by_name,
                            sbrec_mac_binding_by_lport_ip, br_int, chassis,
-                           local_datapaths, active_tunnels);
+                           local_datapaths, active_tunnels, ovs_table);
     prepare_ipv6_ras(local_active_ports_ras, sbrec_port_binding_by_name);
     prepare_ipv6_prefixd(ovnsb_idl_txn, sbrec_port_binding_by_name,
                          local_active_ports_ipv6_pd, chassis,
@@ -4330,7 +4336,8 @@  struct garp_rarp_data {
     struct eth_addr ea;          /* Ethernet address of port. */
     ovs_be32 ipv4;               /* Ipv4 address of port. */
     long long int announce_time; /* Next announcement in ms. */
-    int backoff;                 /* Backoff for the next announcement. */
+    int backoff;                 /* Backoff timeout for the next
+                                  * announcement (in msecs). */
     uint32_t dp_key;             /* Datapath used to output this GARP. */
     uint32_t port_key;           /* Port to inject the GARP into. */
 };
@@ -4359,7 +4366,7 @@  add_garp_rarp(const char *name, const struct eth_addr ea, ovs_be32 ip,
     garp_rarp->ea = ea;
     garp_rarp->ipv4 = ip;
     garp_rarp->announce_time = time_msec() + 1000;
-    garp_rarp->backoff = 1;
+    garp_rarp->backoff = 1000; /* msec. */
     garp_rarp->dp_key = dp_key;
     garp_rarp->port_key = port_key;
     shash_add(&send_garp_rarp_data, name, garp_rarp);
@@ -4375,7 +4382,9 @@  send_garp_rarp_update(struct ovsdb_idl_txn *ovnsb_idl_txn,
                       struct ovsdb_idl_index *sbrec_mac_binding_by_lport_ip,
                       const struct hmap *local_datapaths,
                       const struct sbrec_port_binding *binding_rec,
-                      struct shash *nat_addresses)
+                      struct shash *nat_addresses,
+                      long long int garp_max_timeout,
+                      bool garp_continuous)
 {
     volatile struct garp_rarp_data *garp_rarp = NULL;
 
@@ -4401,6 +4410,12 @@  send_garp_rarp_update(struct ovsdb_idl_txn *ovnsb_idl_txn,
                 if (garp_rarp) {
                     garp_rarp->dp_key = binding_rec->datapath->tunnel_key;
                     garp_rarp->port_key = binding_rec->tunnel_key;
+                    if (garp_max_timeout != garp_rarp_max_timeout ||
+                        garp_continuous != garp_rarp_continuous) {
+                        /* reset backoff */
+                        garp_rarp->announce_time = time_msec() + 1000;
+                        garp_rarp->backoff = 1000; /* msec. */
+                    }
                 } else {
                     add_garp_rarp(name, laddrs->ea,
                                   laddrs->ipv4_addrs[i].addr,
@@ -4425,6 +4440,12 @@  send_garp_rarp_update(struct ovsdb_idl_txn *ovnsb_idl_txn,
                     if (garp_rarp) {
                         garp_rarp->dp_key = binding_rec->datapath->tunnel_key;
                         garp_rarp->port_key = binding_rec->tunnel_key;
+                        if (garp_max_timeout != garp_rarp_max_timeout ||
+                            garp_continuous != garp_rarp_continuous) {
+                            /* reset backoff */
+                            garp_rarp->announce_time = time_msec() + 1000;
+                            garp_rarp->backoff = 1000; /* msec. */
+                        }
                     } else {
                         add_garp_rarp(name, laddrs->ea,
                                       0, binding_rec->datapath->tunnel_key,
@@ -4444,6 +4465,12 @@  send_garp_rarp_update(struct ovsdb_idl_txn *ovnsb_idl_txn,
     if (garp_rarp) {
         garp_rarp->dp_key = binding_rec->datapath->tunnel_key;
         garp_rarp->port_key = binding_rec->tunnel_key;
+        if (garp_max_timeout != garp_rarp_max_timeout ||
+            garp_continuous != garp_rarp_continuous) {
+            /* reset backoff */
+            garp_rarp->announce_time = time_msec() + 1000;
+            garp_rarp->backoff = 1000; /* msec. */
+        }
         return;
     }
 
@@ -4529,13 +4556,15 @@  send_garp_rarp(struct rconn *swconn, struct garp_rarp_data *garp_rarp,
     ofpbuf_uninit(&ofpacts);
 
     /* Set the next announcement.  At most 5 announcements are sent for a
-     * vif. */
-    if (garp_rarp->backoff < 16) {
-        garp_rarp->backoff *= 2;
-        garp_rarp->announce_time = current_time + garp_rarp->backoff * 1000;
+     * vif if garp_rarp_max_timeout is not specified otherwise cap the max
+     * timeout to garp_rarp_max_timeout. */
+    if (garp_rarp_continuous || garp_rarp->backoff < garp_rarp_max_timeout) {
+        garp_rarp->announce_time = current_time + garp_rarp->backoff;
     } else {
         garp_rarp->announce_time = LLONG_MAX;
     }
+    garp_rarp->backoff = MIN(garp_rarp_max_timeout, garp_rarp->backoff * 2);
+
     return garp_rarp->announce_time;
 }
 
@@ -5778,13 +5807,26 @@  send_garp_rarp_prepare(struct ovsdb_idl_txn *ovnsb_idl_txn,
                        const struct ovsrec_bridge *br_int,
                        const struct sbrec_chassis *chassis,
                        const struct hmap *local_datapaths,
-                       const struct sset *active_tunnels)
+                       const struct sset *active_tunnels,
+                       const struct ovsrec_open_vswitch_table *ovs_table)
     OVS_REQUIRES(pinctrl_mutex)
 {
     struct sset localnet_vifs = SSET_INITIALIZER(&localnet_vifs);
     struct sset local_l3gw_ports = SSET_INITIALIZER(&local_l3gw_ports);
     struct sset nat_ip_keys = SSET_INITIALIZER(&nat_ip_keys);
     struct shash nat_addresses;
+    unsigned long long garp_max_timeout = GARP_RARP_DEF_MAX_TIMEOUT;
+    bool garp_continuous = false;
+    const struct ovsrec_open_vswitch *cfg =
+        ovsrec_open_vswitch_table_first(ovs_table);
+    if (cfg) {
+        garp_max_timeout = smap_get_ullong(
+                &cfg->external_ids, "garp-max-timeout-sec", 0) * 1000;
+        garp_continuous = !!garp_max_timeout;
+        if (!garp_max_timeout) {
+            garp_max_timeout = GARP_RARP_DEF_MAX_TIMEOUT;
+        }
+    }
 
     shash_init(&nat_addresses);
 
@@ -5813,8 +5855,10 @@  send_garp_rarp_prepare(struct ovsdb_idl_txn *ovnsb_idl_txn,
         const struct sbrec_port_binding *pb = lport_lookup_by_name(
             sbrec_port_binding_by_name, iface_id);
         if (pb) {
-            send_garp_rarp_update(ovnsb_idl_txn, sbrec_mac_binding_by_lport_ip,
-                                  local_datapaths, pb, &nat_addresses);
+            send_garp_rarp_update(ovnsb_idl_txn,
+                                  sbrec_mac_binding_by_lport_ip,
+                                  local_datapaths, pb, &nat_addresses,
+                                  garp_max_timeout, garp_continuous);
         }
     }
 
@@ -5825,7 +5869,8 @@  send_garp_rarp_prepare(struct ovsdb_idl_txn *ovnsb_idl_txn,
             = lport_lookup_by_name(sbrec_port_binding_by_name, gw_port);
         if (pb) {
             send_garp_rarp_update(ovnsb_idl_txn, sbrec_mac_binding_by_lport_ip,
-                                  local_datapaths, pb, &nat_addresses);
+                                  local_datapaths, pb, &nat_addresses,
+                                  garp_max_timeout, garp_continuous);
         }
     }
 
@@ -5843,6 +5888,9 @@  send_garp_rarp_prepare(struct ovsdb_idl_txn *ovnsb_idl_txn,
     shash_destroy(&nat_addresses);
 
     sset_destroy(&nat_ip_keys);
+
+    garp_rarp_max_timeout = garp_max_timeout;
+    garp_rarp_continuous = garp_continuous;
 }
 
 static bool
diff --git a/controller/pinctrl.h b/controller/pinctrl.h
index 88f18e983..b76e43626 100644
--- a/controller/pinctrl.h
+++ b/controller/pinctrl.h
@@ -28,6 +28,7 @@  struct lport_index;
 struct ovsdb_idl_index;
 struct ovsdb_idl_txn;
 struct ovsrec_bridge;
+struct ovsrec_open_vswitch_table;
 struct sbrec_chassis;
 struct sbrec_dns_table;
 struct sbrec_controller_event_table;
@@ -52,7 +53,8 @@  void pinctrl_run(struct ovsdb_idl_txn *ovnsb_idl_txn,
                  const struct hmap *local_datapaths,
                  const struct sset *active_tunnels,
                  const struct shash *local_active_ports_ipv6_pd,
-                 const struct shash *local_active_ports_ras);
+                 const struct shash *local_active_ports_ras,
+                 const struct ovsrec_open_vswitch_table *ovs_table);
 void pinctrl_wait(struct ovsdb_idl_txn *ovnsb_idl_txn);
 void pinctrl_destroy(void);
 void pinctrl_set_br_int_name(char *br_int_name);