X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fbridge.c;h=c39670a3b2a0d90f01a470e26cf8fa40479a9fd4;hb=2e8873af89047bc4093ad8f2755867af632b9363;hp=057172e446d95edd1810621b142a9be47934391e;hpb=4f2cad2c3b1079ede2d3eed0dd25423552105255;p=openvswitch

diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
index 057172e4..c39670a3 100644
--- a/vswitchd/bridge.c
+++ b/vswitchd/bridge.c
@@ -130,6 +130,7 @@ struct port {
     tag_type active_iface_tag;  /* Tag for bcast flows. */
     tag_type no_ifaces_tag;     /* Tag for flows when all ifaces disabled. */
     int updelay, downdelay;     /* Delay before iface goes up/down, in ms. */
+    bool bond_compat_is_stale;  /* Need to call port_update_bond_compat()? */
 
     /* Port mirroring info. */
     mirror_mask_t src_mirrors;  /* Mirrors triggered when packet received. */
@@ -460,9 +461,23 @@ bridge_reconfigure(void)
         for (i = 0; i < add_ifaces.n; i++) {
             const char *if_name = add_ifaces.names[i];
             for (;;) {
-                int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
-                int error = dpif_port_add(&br->dpif, if_name, next_port_no++,
-                                          internal ? ODP_PORT_INTERNAL : 0);
+                bool internal;
+                int error;
+
+                /* It's an internal interface if it's marked that way, or if
+                 * it's a bonded interface for which we're faking up a network
+                 * device. */
+                internal = cfg_get_bool(0, "iface.%s.internal", if_name);
+                if (cfg_get_bool(0, "bonding.%s.fake-iface", if_name)) {
+                    struct port *port = port_lookup(br, if_name);
+                    if (port && port->n_ifaces > 1) {
+                        internal = true;
+                    }
+                }
+
+                /* Add to datapath. */
+                error = dpif_port_add(&br->dpif, if_name, next_port_no++,
+                                      internal ? ODP_PORT_INTERNAL : 0);
                 if (error != EEXIST) {
                     if (next_port_no >= 256) {
                         VLOG_ERR("ran out of valid port numbers on dp%u",
@@ -630,6 +645,18 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
         if (iface_ea_u64) {
             /* User specified explicitly. */
             eth_addr_from_uint64(iface_ea_u64, iface_ea);
+
+            /* Find the interface with this Ethernet address (if any) so that
+             * we can provide the correct devname to the caller. */
+            iface = NULL;
+            for (j = 0; j < port->n_ifaces; j++) {
+                struct iface *candidate = port->ifaces[j];
+                uint8_t candidate_ea[ETH_ADDR_LEN];
+                if (!netdev_nodev_get_etheraddr(candidate->name, candidate_ea)
+                    && eth_addr_equals(iface_ea, candidate_ea)) {
+                    iface = candidate;
+                }
+            }
         } else {
             /* Choose the interface whose MAC address will represent the port.
              * The Linux kernel bonding code always chooses the MAC address of
@@ -671,7 +698,7 @@ bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
             memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0)
         {
             memcpy(ea, iface_ea, ETH_ADDR_LEN);
-            *devname = iface->name;
+            *devname = iface ? iface->name : NULL;
         }
     }
     if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
@@ -1308,9 +1335,12 @@ bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
             struct iface *iface = port->ifaces[j];
             svec_add(ifaces, iface->name);
         }
+        if (port->n_ifaces > 1
+            && cfg_get_bool(0, "bonding.%s.fake-iface", port->name)) {
+            svec_add(ifaces, port->name);
+        }
     }
-    svec_sort(ifaces);
-    assert(svec_is_unique(ifaces));
+    svec_sort_unique(ifaces);
 }
 
 /* For robustness, in case the administrator moves around datapath ports behind
@@ -1405,6 +1435,7 @@ choose_output_iface(const struct port *port, const uint8_t *dl_src,
                 return false;
             }
             e->iface_tag = tag_create_random();
+            ((struct port *) port)->bond_compat_is_stale = true;
         }
         *tags |= e->iface_tag;
         iface = port->ifaces[e->iface_idx];
@@ -1494,6 +1525,7 @@ bond_enable_slave(struct iface *iface, bool enable)
         }
         iface->tag = tag_create_random();
     }
+    port_update_bond_compat(port);
 }
 
 static void
@@ -1503,6 +1535,12 @@ bond_run(struct bridge *br)
 
     for (i = 0; i < br->n_ports; i++) {
         struct port *port = br->ports[i];
+
+        if (port->bond_compat_is_stale) {
+            port->bond_compat_is_stale = false;
+            port_update_bond_compat(port);
+        }
+
         if (port->n_ifaces < 2) {
             continue;
         }
@@ -1673,12 +1711,14 @@ compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
                 for (i = 0; i < br->n_ports; i++) {
                     struct port *port = br->ports[i];
                     if (port_includes_vlan(port, m->out_vlan)
-                        && set_dst(dst, flow, in_port, port, tags)
-                        && !dst_is_duplicate(dsts, dst - dsts, dst))
+                        && set_dst(dst, flow, in_port, port, tags))
                     {
                         if (port->vlan < 0) {
                             dst->vlan = m->out_vlan;
                         }
+                        if (dst_is_duplicate(dsts, dst - dsts, dst)) {
+                            continue;
+                        }
                         if (dst->dp_ifidx == flow->in_port
                             && dst->vlan == vlan) {
                             /* Don't send out input port on same VLAN. */
@@ -1852,7 +1892,7 @@ process_flow(struct bridge *br, const flow_t *flow,
             goto done;
         } else {
             /* Drop all multicast packets for which we have learned a different
-             * input port, because we probably sent the packet on one slaves
+             * input port, because we probably sent the packet on one slave
              * and got it back on the active slave.  Broadcast ARP replies are
              * an exception to this rule: the host has moved to another
              * switch. */
@@ -2285,6 +2325,7 @@ bond_rebalance_port(struct port *port)
             } else {
                 from++;
             }
+            port->bond_compat_is_stale = true;
         }
     }
 
@@ -2551,6 +2592,7 @@ bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
     ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
     entry->iface_idx = iface->port_ifidx;
     entry->iface_tag = tag_create_random();
+    port->bond_compat_is_stale = true;
     unixctl_command_reply(conn, 200, "migrated");
 }
 
@@ -2815,6 +2857,7 @@ port_destroy(struct port *port)
         size_t i;
 
         proc_net_compat_update_vlan(port->name, NULL, 0);
+        proc_net_compat_update_bond(port->name, NULL);
 
         for (i = 0; i < MAX_MIRRORS; i++) {
             struct mirror *m = br->mirrors[i];
@@ -2881,7 +2924,7 @@ port_update_bonding(struct port *port)
         if (port->bond_hash) {
             free(port->bond_hash);
             port->bond_hash = NULL;
-            proc_net_compat_update_bond(port->name, NULL);
+            port->bond_compat_is_stale = true;
         }
     } else {
         if (!port->bond_hash) {
@@ -2896,36 +2939,52 @@ port_update_bonding(struct port *port)
             port->no_ifaces_tag = tag_create_random();
             bond_choose_active_iface(port);
         }
-        port_update_bond_compat(port);
+        port->bond_compat_is_stale = true;
     }
 }
 
 static void
 port_update_bond_compat(struct port *port)
 {
+    struct compat_bond_hash compat_hashes[BOND_MASK + 1];
     struct compat_bond bond;
     size_t i;
 
     if (port->n_ifaces < 2) {
+        proc_net_compat_update_bond(port->name, NULL);
         return;
     }
 
     bond.up = false;
     bond.updelay = port->updelay;
     bond.downdelay = port->downdelay;
+
+    bond.n_hashes = 0;
+    bond.hashes = compat_hashes;
+    if (port->bond_hash) {
+        const struct bond_entry *e;
+        for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) {
+            if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
+                struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++];
+                cbh->hash = e - port->bond_hash;
+                cbh->netdev_name = port->ifaces[e->iface_idx]->name;
+            }
+        }
+    }
+
     bond.n_slaves = port->n_ifaces;
     bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
     for (i = 0; i < port->n_ifaces; i++) {
         struct iface *iface = port->ifaces[i];
         struct compat_bond_slave *slave = &bond.slaves[i];
         slave->name = iface->name;
-        slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
-                     (!iface->enabled && iface->delay_expires != LLONG_MAX));
+        slave->up = iface->enabled;
         if (slave->up) {
             bond.up = true;
         }
         memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
     }
+
     proc_net_compat_update_bond(port->name, &bond);
     free(bond.slaves);
 }
@@ -2981,8 +3040,17 @@ iface_create(struct port *port, const char *name)
     iface->tag = tag_create_random();
     iface->delay_expires = LLONG_MAX;
 
-    netdev_nodev_get_etheraddr(name, iface->mac);
-    netdev_nodev_get_carrier(name, &iface->enabled);
+    if (!cfg_get_bool(0, "iface.%s.internal", iface->name)) {
+        netdev_nodev_get_etheraddr(name, iface->mac);
+        netdev_nodev_get_carrier(name, &iface->enabled);
+    } else {
+        /* Internal interfaces are created later by the call to dpif_port_add()
+         * in bridge_reconfigure().  Until then, we can't obtain any
+         * information about them.  (There's no real value in doing so, anyway,
+         * because the 'mac' and 'enabled' values are only used for interfaces
+         * that are bond slaves, and it doesn't normally make sense to bond an
+         * internal interface.) */
+    }
 
     if (port->n_ifaces >= port->allocated_ifaces) {
         port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
@@ -3251,6 +3319,7 @@ mirror_reconfigure_one(struct mirror *m)
     int *vlans;
     size_t i;
     bool mirror_all_ports;
+    bool any_ports_specified;
 
     /* Get output port. */
     out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
@@ -3289,11 +3358,18 @@ mirror_reconfigure_one(struct mirror *m)
     cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
     cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
     cfg_get_all_keys(&ports, "%s.select.port", pfx);
+    any_ports_specified = src_ports.n || dst_ports.n || ports.n;
     svec_append(&src_ports, &ports);
     svec_append(&dst_ports, &ports);
     svec_destroy(&ports);
     prune_ports(m, &src_ports);
     prune_ports(m, &dst_ports);
+    if (any_ports_specified && !src_ports.n && !dst_ports.n) {
+        VLOG_ERR("%s: none of the specified ports exist; "
+                 "disabling port mirror %s", pfx, pfx);
+        mirror_destroy(m);
+        goto exit;
+    }
 
     /* Get all the vlans, and drop duplicate and invalid vlans. */
     svec_init(&vlan_strings);
@@ -3345,6 +3421,7 @@ mirror_reconfigure_one(struct mirror *m)
     }
 
     /* Clean up. */
+exit:
     svec_destroy(&src_ports);
     svec_destroy(&dst_ports);
     free(pfx);