[ovs-dev] [PATCH ovn 4/4] ovn-northd: Minimize number of ARP/NS responder flows for DNAT.

Dumitru Ceara dceara at redhat.com
Wed Jun 24 15:51:44 UTC 2020


Most ARP/NS responder flows can be configured per datapath instead of per
router port.

The only exception is with distributed gateway router ports which need
special treatment. This patch changes the ARP/NS responder behavior and adds:
- Priority 92 flows to reply to ARP requests on distributed gateway router
  ports, on the chassis where the DNAT entry is bound.
- Priority 91 flows to drop ARP requests on distributed gateway router ports,
  on chassis where the DNAT entry is not bound.
- Priority 90 flows to reply to ARP requests on all other router ports. This
  last type of flows is programmed exactly once per logical router limiting
  the total number of required logical flows.

Suggested-by: Han Zhou <hzhou at ovn.org>
Reported-by: Girish Moodalbail <gmoodalbail at gmail.com>
Reported-at: https://mail.openvswitch.org/pipermail/ovs-discuss/2020-June/050186.html
Signed-off-by: Dumitru Ceara <dceara at redhat.com>
---
 northd/ovn-northd.8.xml |   16 +++-
 northd/ovn-northd.c     |  203 ++++++++++++++++++++++++++++++++---------------
 tests/ovn.at            |    8 +-
 3 files changed, 153 insertions(+), 74 deletions(-)

diff --git a/northd/ovn-northd.8.xml b/northd/ovn-northd.8.xml
index ec89a63..df66caa 100644
--- a/northd/ovn-northd.8.xml
+++ b/northd/ovn-northd.8.xml
@@ -1857,9 +1857,8 @@ nd_na_router {
           IPv4: For a configured DNAT IP address or a load balancer
           IPv4 VIP <var>A</var>, for each router port <var>P</var> with
           Ethernet address <var>E</var>, a priority-90 flow matches
-          <code>inport == <var>P</var> && arp.op == 1 &&
-          arp.tpa == <var>A</var></code> (ARP request)
-          with the following actions:
+          <code>arp.op == 1 && arp.tpa == <var>A</var></code>
+          (ARP request) with the following actions:
         </p>
 
         <pre>
@@ -1876,6 +1875,11 @@ output;
         </pre>
 
         <p>
+          IPv4: For a configured load balancer IPv4 VIP, a similar flow is
+          added with the additional match <code>inport == <var>P</var></code>.
+        </p>
+
+        <p>
           If the router port <var>P</var> is a distributed gateway router
           port, then the <code>is_chassis_resident(<var>P</var>)</code> is
           also added in the match condition for the load balancer IPv4
@@ -1922,9 +1926,11 @@ nd_na {
         <ul>
           <li>
             If the corresponding NAT rule cannot be handled in a
-            distributed manner, then this flow is only programmed on
+            distributed manner, then a priority-92 flow is programmed on
             the gateway port instance on the
-            <code>redirect-chassis</code>.  This behavior avoids
+            <code>redirect-chassis</code>.  A priority-91 drop flow is
+            programmed on the other chassis when ARP requests/NS packets
+            are received on the gateway port. This behavior avoids
             generation of multiple ARP responses from different chassis,
             and allows upstream MAC learning to point to the
             <code>redirect-chassis</code>.
diff --git a/northd/ovn-northd.c b/northd/ovn-northd.c
index 6b22316..82783f4 100644
--- a/northd/ovn-northd.c
+++ b/northd/ovn-northd.c
@@ -7961,7 +7961,7 @@ lrouter_nat_is_stateless(const struct nbrec_nat *nat)
 static void
 build_lrouter_arp_flow(struct ovn_datapath *od, struct ovn_port *op,
                        const char *ip_address, const char *eth_addr,
-                       struct ds *extra_match, uint16_t priority,
+                       struct ds *extra_match, bool drop, uint16_t priority,
                        struct hmap *lflows, const struct ovsdb_idl_row *hint)
 {
     struct ds match = DS_EMPTY_INITIALIZER;
@@ -7976,20 +7976,24 @@ build_lrouter_arp_flow(struct ovn_datapath *od, struct ovn_port *op,
     if (extra_match && ds_last(extra_match) != EOF) {
         ds_put_format(&match, " && %s", ds_cstr(extra_match));
     }
-    ds_put_format(&actions,
-                  "eth.dst = eth.src; "
-                  "eth.src = %s; "
-                  "arp.op = 2; /* ARP reply */ "
-                  "arp.tha = arp.sha; "
-                  "arp.sha = %s; "
-                  "arp.tpa = arp.spa; "
-                  "arp.spa = %s; "
-                  "outport = inport; "
-                  "flags.loopback = 1; "
-                  "output;",
-                  eth_addr,
-                  eth_addr,
-                  ip_address);
+    if (drop) {
+        ds_put_format(&actions, "drop;");
+    } else {
+        ds_put_format(&actions,
+                      "eth.dst = eth.src; "
+                      "eth.src = %s; "
+                      "arp.op = 2; /* ARP reply */ "
+                      "arp.tha = arp.sha; "
+                      "arp.sha = %s; "
+                      "arp.tpa = arp.spa; "
+                      "arp.spa = %s; "
+                      "outport = inport; "
+                      "flags.loopback = 1; "
+                      "output;",
+                      eth_addr,
+                      eth_addr,
+                      ip_address);
+    }
 
     ovn_lflow_add_with_hint(lflows, od, S_ROUTER_IN_IP_INPUT, priority,
                             ds_cstr(&match), ds_cstr(&actions), hint);
@@ -8008,7 +8012,7 @@ static void
 build_lrouter_nd_flow(struct ovn_datapath *od, struct ovn_port *op,
                       const char *action, const char *ip_address,
                       const char *sn_ip_address, const char *eth_addr,
-                      struct ds *extra_match, uint16_t priority,
+                      struct ds *extra_match, bool drop, uint16_t priority,
                       struct hmap *lflows,
                       const struct ovsdb_idl_row *hint)
 {
@@ -8030,21 +8034,25 @@ build_lrouter_nd_flow(struct ovn_datapath *od, struct ovn_port *op,
         ds_put_format(&match, " && %s", ds_cstr(extra_match));
     }
 
-    ds_put_format(&actions,
-                  "%s { "
-                    "eth.src = %s; "
-                    "ip6.src = %s; "
-                    "nd.target = %s; "
-                    "nd.tll = %s; "
-                    "outport = inport; "
-                    "flags.loopback = 1; "
-                    "output; "
-                  "};",
-                  action,
-                  eth_addr,
-                  ip_address,
-                  ip_address,
-                  eth_addr);
+    if (drop) {
+        ds_put_format(&actions, "drop;");
+    } else {
+        ds_put_format(&actions,
+                      "%s { "
+                        "eth.src = %s; "
+                        "ip6.src = %s; "
+                        "nd.target = %s; "
+                        "nd.tll = %s; "
+                        "outport = inport; "
+                        "flags.loopback = 1; "
+                        "output; "
+                      "};",
+                      action,
+                      eth_addr,
+                      ip_address,
+                      ip_address,
+                      eth_addr);
+    }
 
     ovn_lflow_add_with_hint(lflows, od, S_ROUTER_IN_IP_INPUT, priority,
                             ds_cstr(&match), ds_cstr(&actions), hint);
@@ -8234,7 +8242,41 @@ build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
                       "ip4.dst == 0.0.0.0/8",
                       "drop;");
 
-        /* Priority-90 flows reply to ARP requests and ND packets. */
+        /* Priority-90-92 flows handle ARP requests and ND packets. Most are
+         * per logical port but DNAT addresses can be handled per datapath
+         * for non gateway router ports.
+         */
+        for (int i = 0; i < od->nbr->n_nat; i++) {
+            struct ovn_nat *nat_entry = &od->nat_entries[i];
+            const struct nbrec_nat *nat = nat_entry->nb;
+
+            /* Skip entries we failed to parse. */
+            if (!nat_entry_is_valid(nat_entry)) {
+                continue;
+            }
+
+            if (!strcmp(nat->type, "snat")) {
+                continue;
+            }
+
+            /* Priority 91 and 92 flows are added for each gateway router
+             * port to handle the special cases. In case we get the packet
+             * on a regular port, just reply with the port's ETH address.
+             */
+            struct lport_addresses *ext_addrs = &nat_entry->ext_addrs;
+            if (nat_entry_is_v6(nat_entry)) {
+                build_lrouter_nd_flow(od, NULL, "nd_na",
+                                      ext_addrs->ipv6_addrs[0].addr_s,
+                                      ext_addrs->ipv6_addrs[0].sn_addr_s,
+                                      REG_INPORT_ETH_ADDR, NULL, false, 90,
+                                      lflows, &nat->header_);
+            } else {
+                build_lrouter_arp_flow(od, NULL,
+                                       ext_addrs->ipv4_addrs[0].addr_s,
+                                       REG_INPORT_ETH_ADDR, NULL, false, 90,
+                                       lflows, &nat->header_);
+            }
+        }
 
         /* Drop ARP packets (priority 85). ARP request packets for router's own
          * IPs are handled with priority-90 flows.
@@ -8384,8 +8426,8 @@ build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
 
             build_lrouter_arp_flow(op->od, op,
                                    op->lrp_networks.ipv4_addrs[i].addr_s,
-                                   REG_INPORT_ETH_ADDR, &match, 90, lflows,
-                                   &op->nbrp->header_);
+                                   REG_INPORT_ETH_ADDR, &match, false, 90,
+                                   lflows, &op->nbrp->header_);
         }
 
         /* A set to hold all load-balancer vips that need ARP responses. */
@@ -8403,7 +8445,7 @@ build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
 
             build_lrouter_arp_flow(op->od, op,
                                    ip_address, REG_INPORT_ETH_ADDR,
-                                   &match, 90, lflows, NULL);
+                                   &match, false, 90, lflows, NULL);
         }
 
         SSET_FOR_EACH (ip_address, &all_ips_v6) {
@@ -8415,7 +8457,7 @@ build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
 
             build_lrouter_nd_flow(op->od, op, "nd_na",
                                   ip_address, NULL, REG_INPORT_ETH_ADDR,
-                                  &match, 90, lflows, NULL);
+                                  &match, false, 90, lflows, NULL);
         }
 
         sset_destroy(&all_ips_v4);
@@ -8468,53 +8510,84 @@ build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
             /* Mac address to use when replying to ARP/NS. */
             const char *mac_s = REG_INPORT_ETH_ADDR;
 
+            /* ARP/NS packets are taken care of per router. The only exception
+             * is on the l3dgw_port where we might need to use a different
+             * ETH address.
+             */
+            if (op != op->od->l3dgw_port) {
+                continue;
+            }
+
             /* ARP / ND handling for external IP addresses.
              *
              * DNAT IP addresses are external IP addresses that need ARP
              * handling. */
             ds_clear(&match);
 
-            if (op->od->l3dgw_port && op == op->od->l3dgw_port) {
-                struct eth_addr mac;
-                if (nat->external_mac &&
-                    eth_addr_from_string(nat->external_mac, &mac)
-                    && nat->logical_port) {
-                    /* distributed NAT case, use nat->external_mac */
-                    mac_s = nat->external_mac;
-                    /* Traffic with eth.src = nat->external_mac should only be
-                     * sent from the chassis where nat->logical_port is
-                     * resident, so that upstream MAC learning points to the
-                     * correct chassis.  Also need to avoid generation of
-                     * multiple ARP responses from different chassis. */
-                    ds_put_format(&match, "is_chassis_resident(\"%s\")",
-                                  nat->logical_port);
-                } else {
-                    mac_s = REG_INPORT_ETH_ADDR;
-                    /* Traffic with eth.src = l3dgw_port->lrp_networks.ea_s
-                     * should only be sent from the "redirect-chassis", so that
-                     * upstream MAC learning points to the "redirect-chassis".
-                     * Also need to avoid generation of multiple ARP responses
-                     * from different chassis. */
-                    if (op->od->l3redirect_port) {
-                        ds_put_format(&match, "is_chassis_resident(%s)",
-                                      op->od->l3redirect_port->json_key);
-                    }
+            struct ds match_cr_port = DS_EMPTY_INITIALIZER;
+            struct ds match_non_cr_port = DS_EMPTY_INITIALIZER;
+
+            struct eth_addr mac;
+            if (nat->external_mac &&
+                eth_addr_from_string(nat->external_mac, &mac)
+                && nat->logical_port) {
+                /* distributed NAT case, use nat->external_mac */
+                mac_s = nat->external_mac;
+                /* Traffic with eth.src = nat->external_mac should only be
+                 * sent from the chassis where nat->logical_port is
+                 * resident, so that upstream MAC learning points to the
+                 * correct chassis.  Also need to avoid generation of
+                 * multiple ARP responses from different chassis. */
+                ds_put_format(&match_cr_port,
+                              "is_chassis_resident(\"%s\")",
+                              nat->logical_port);
+                ds_put_format(&match_non_cr_port,
+                              "!is_chassis_resident(\"%s\")",
+                              nat->logical_port);
+            } else {
+                mac_s = REG_INPORT_ETH_ADDR;
+                /* Traffic with eth.src = l3dgw_port->lrp_networks.ea_s
+                 * should only be sent from the "redirect-chassis", so that
+                 * upstream MAC learning points to the "redirect-chassis".
+                 * Also need to avoid generation of multiple ARP responses
+                 * from different chassis. */
+                if (op->od->l3redirect_port) {
+                    ds_put_format(&match_cr_port,
+                                  "is_chassis_resident(\"%s\")",
+                                  op->od->l3redirect_port->json_key);
+                    ds_put_format(&match_non_cr_port,
+                                  "!is_chassis_resident(\"%s\")",
+                                  op->od->l3redirect_port->json_key);
                 }
             }
 
+            /* Respond to ARP/NS requests on the chassis that binds the gw
+             * port. Drop the ARP/NS requests on other chassis.
+             */
             struct lport_addresses *ext_addrs = &nat_entry->ext_addrs;
             if (nat_entry_is_v6(nat_entry)) {
                 build_lrouter_nd_flow(op->od, op, "nd_na",
                                       ext_addrs->ipv6_addrs[0].addr_s,
                                       ext_addrs->ipv6_addrs[0].sn_addr_s,
-                                      mac_s, &match, 90,
+                                      mac_s, &match_cr_port, false, 92,
+                                      lflows, &nat->header_);
+                build_lrouter_nd_flow(op->od, op, "nd_na",
+                                      ext_addrs->ipv6_addrs[0].addr_s,
+                                      ext_addrs->ipv6_addrs[0].sn_addr_s,
+                                      mac_s, &match_non_cr_port, true, 91,
                                       lflows, &nat->header_);
             } else {
                 build_lrouter_arp_flow(op->od, op,
                                        ext_addrs->ipv4_addrs[0].addr_s,
-                                       mac_s, &match, 90,
+                                       mac_s, &match_cr_port, false, 92,
+                                       lflows, &nat->header_);
+                build_lrouter_arp_flow(op->od, op,
+                                       ext_addrs->ipv4_addrs[0].addr_s,
+                                       mac_s, &match_non_cr_port, true, 91,
                                        lflows, &nat->header_);
             }
+            ds_destroy(&match_cr_port);
+            ds_destroy(&match_non_cr_port);
         }
 
         if (!smap_get(&op->od->nbr->options, "chassis")
@@ -8700,8 +8773,8 @@ build_lrouter_flows(struct hmap *datapaths, struct hmap *ports,
             build_lrouter_nd_flow(op->od, op, "nd_na_router",
                                   op->lrp_networks.ipv6_addrs[i].addr_s,
                                   op->lrp_networks.ipv6_addrs[i].sn_addr_s,
-                                  REG_INPORT_ETH_ADDR, &match, 90, lflows,
-                                  &op->nbrp->header_);
+                                  REG_INPORT_ETH_ADDR, &match, false, 90,
+                                  lflows, &op->nbrp->header_);
         }
 
         /* UDP/TCP port unreachable */
diff --git a/tests/ovn.at b/tests/ovn.at
index 1ff7952..15139a1 100644
--- a/tests/ovn.at
+++ b/tests/ovn.at
@@ -19169,7 +19169,7 @@ OVS_WAIT_UNTIL([
 send_arp_request 1 0 ${src_mac} $(ip_to_hex 10 0 0 254) $(ip_to_hex 10 0 0 121)
 
 # Verify that the ARP request is replied to from hv1 and not hv2.
-match_arp_req="priority=90.*${match_r1_metadata}.*arp_tpa=10.0.0.121,arp_op=1"
+match_arp_req="priority=92.*${match_r1_metadata}.*arp_tpa=10.0.0.121,arp_op=1"
 
 as hv1
 OVS_WAIT_UNTIL([
@@ -19189,7 +19189,7 @@ OVS_WAIT_UNTIL([
 send_arp_request 1 0 ${src_mac} $(ip_to_hex 10 0 0 254) $(ip_to_hex 10 0 0 122)
 
 # Verify that the ARP request is replied to from hv2 and not hv1.
-match_arp_req="priority=90.*${match_r1_metadata}.*arp_tpa=10.0.0.122,arp_op=1"
+match_arp_req="priority=92.*${match_r1_metadata}.*arp_tpa=10.0.0.122,arp_op=1"
 
 as hv2
 OVS_WAIT_UNTIL([
@@ -19233,7 +19233,7 @@ dst_ipv6=00100000000000000000000000000121
 send_nd_ns 1 0 ${src_mac} ${src_ipv6} ${dst_ipv6} 72dd
 
 # Verify that the ND_NS is replied to from hv1 and not hv2.
-match_nd_ns="priority=90.*${match_r1_metadata}.*icmp_type=135.*nd_target=10::121"
+match_nd_ns="priority=92.*${match_r1_metadata}.*icmp_type=135.*nd_target=10::121"
 
 as hv1
 OVS_WAIT_UNTIL([
@@ -19255,7 +19255,7 @@ dst_ipv6=00100000000000000000000000000122
 send_nd_ns 1 0 ${src_mac} ${src_ipv6} ${dst_ipv6} 72db
 
 # Verify that the ND_NS is replied to from hv2 and not hv1.
-match_nd_ns="priority=90.*${match_r1_metadata}.*icmp_type=135.*nd_target=10::122"
+match_nd_ns="priority=92.*${match_r1_metadata}.*icmp_type=135.*nd_target=10::122"
 
 as hv2
 OVS_WAIT_UNTIL([



More information about the dev mailing list