X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=vswitchd%2Fovs-brcompatd.c;h=39844f69f2420c5b3dcff0da44d73153dbe3f197;hb=8fef8c7121222233075a03d57db7e0b48d5f6be5;hp=70570e925ec42f1969f0813bf592ac47c1f98202;hpb=6dd3fad481b5d801695c2b0529c7d37cac2c9b19;p=sliver-openvswitch.git

diff --git a/vswitchd/ovs-brcompatd.c b/vswitchd/ovs-brcompatd.c
index 70570e925..39844f69f 100644
--- a/vswitchd/ovs-brcompatd.c
+++ b/vswitchd/ovs-brcompatd.c
@@ -239,21 +239,14 @@ rewrite_and_reload_config(void)
     return 0;
 }
 
-/* Get all the interfaces for 'bridge' as 'ifaces', breaking bonded interfaces
- * down into their constituent parts.
- *
- * If 'vlan' < 0, all interfaces on 'bridge' are reported.  If 'vlan' == 0,
- * then only interfaces for trunk ports or ports with implicit VLAN 0 are
- * reported.  If 'vlan' > 0, only interfaces with implict VLAN 'vlan' are
- * reported.  */
 static void
-get_bridge_ifaces(const char *bridge, struct svec *ifaces, int vlan)
+do_get_bridge_parts(const char *bridge, struct svec *parts, int vlan,
+                    bool break_down_bonds)
 {
     struct svec ports;
     int i;
 
     svec_init(&ports);
-    svec_init(ifaces);
     cfg_get_all_keys(&ports, "bridge.%s.port", bridge);
     for (i = 0; i < ports.n; i++) {
         const char *port_name = ports.names[i];
@@ -266,19 +259,44 @@ get_bridge_ifaces(const char *bridge, struct svec *ifaces, int vlan)
                 continue;
             }
         }
-        if (cfg_has_section("bonding.%s", port_name)) {
+        if (break_down_bonds && cfg_has_section("bonding.%s", port_name)) {
             struct svec slaves;
             svec_init(&slaves);
             cfg_get_all_keys(&slaves, "bonding.%s.slave", port_name);
-            svec_append(ifaces, &slaves);
+            svec_append(parts, &slaves);
             svec_destroy(&slaves);
         } else {
-            svec_add(ifaces, port_name);
+            svec_add(parts, port_name);
         }
     }
     svec_destroy(&ports);
 }
 
+/* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
+ * down into their constituent parts.
+ *
+ * If 'vlan' < 0, all interfaces on 'bridge' are reported.  If 'vlan' == 0,
+ * then only interfaces for trunk ports or ports with implicit VLAN 0 are
+ * reported.  If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
+ * reported.  */
+static void
+get_bridge_ifaces(const char *bridge, struct svec *ifaces, int vlan)
+{
+    do_get_bridge_parts(bridge, ifaces, vlan, true);
+}
+
+/* Add all the ports for 'bridge' to 'ports'.  Bonded ports are reported under
+ * the bond name, not broken down into their constituent interfaces.
+ *
+ * If 'vlan' < 0, all ports on 'bridge' are reported.  If 'vlan' == 0, then
+ * only trunk ports or ports with implicit VLAN 0 are reported.  If 'vlan' > 0,
+ * only port with implicit VLAN 'vlan' are reported.  */
+static void
+get_bridge_ports(const char *bridge, struct svec *ports, int vlan)
+{
+    do_get_bridge_parts(bridge, ports, vlan, false);
+}
+
 /* Go through the configuration file and remove any ports that no longer
  * exist associated with a bridge. */
 static void
@@ -300,6 +318,7 @@ prune_ports(void)
         struct svec ifaces;
 
         /* Check that each bridge interface exists. */
+        svec_init(&ifaces);
         get_bridge_ifaces(br_name, &ifaces, -1);
         for (j = 0; j < ifaces.n; j++) {
             const char *iface_name = ifaces.names[j];
@@ -382,7 +401,7 @@ parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
               const char **port_name, uint64_t *count, uint64_t *skip)
 {
     static const struct nl_policy policy[] = {
-        [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
+        [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
         [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
         [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
         [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
@@ -391,6 +410,7 @@ parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
 
     if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
                          attrs, ARRAY_SIZE(policy))
+        || (br_name && !attrs[BRC_GENL_A_DP_NAME])
         || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
         || (count && !attrs[BRC_GENL_A_FDB_COUNT])
         || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
@@ -398,7 +418,9 @@ parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
     }
 
     *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
-    *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
+    if (br_name) {
+        *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
+    }
     if (port_name) {
         *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
     }
@@ -411,30 +433,38 @@ parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
     return 0;
 }
 
-static void
-send_reply(uint32_t seq, int error, struct ofpbuf *fdb_query_data)
+/* Composes and returns a reply to a request made by the datapath with Netlink
+ * sequence number 'seq' and error code 'error'.  The caller may add additional
+ * attributes to the message, then it may send it with send_reply(). */
+static struct ofpbuf *
+compose_reply(uint32_t seq, int error)
 {
-    struct ofpbuf msg;
-    int retval;
-
-    /* Compose reply. */
-    ofpbuf_init(&msg, 0);
-    nl_msg_put_genlmsghdr(&msg, brc_sock, 32, brc_family, NLM_F_REQUEST,
+    struct ofpbuf *reply = ofpbuf_new(4096);
+    nl_msg_put_genlmsghdr(reply, brc_sock, 32, brc_family, NLM_F_REQUEST,
                           BRC_GENL_C_DP_RESULT, 1);
-    ((struct nlmsghdr *) msg.data)->nlmsg_seq = seq;
-    nl_msg_put_u32(&msg, BRC_GENL_A_ERR_CODE, error);
-    if (fdb_query_data) {
-        nl_msg_put_unspec(&msg, BRC_GENL_A_FDB_DATA,
-                          fdb_query_data->data, fdb_query_data->size);
-    }
+    ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
+    nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
+    return reply;
+}
 
-    /* Send reply. */
-    retval = nl_sock_send(brc_sock, &msg, false);
+/* Sends 'reply' to the datapath and frees it. */
+static void
+send_reply(struct ofpbuf *reply)
+{
+    int retval = nl_sock_send(brc_sock, reply, false);
     if (retval) {
         VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
                      strerror(retval));
     }
-    ofpbuf_uninit(&msg);
+    ofpbuf_delete(reply);
+}
+
+/* Composes and sends a reply to a request made by the datapath with Netlink
+ * sequence number 'seq' and error code 'error'. */
+static void
+send_simple_reply(uint32_t seq, int error)
+{
+    send_reply(compose_reply(seq, error));
 }
 
 static int
@@ -450,7 +480,7 @@ handle_bridge_cmd(struct ofpbuf *buffer, bool add)
         if (!error) {
             error = rewrite_and_reload_config();
         }
-        send_reply(seq, error, NULL);
+        send_simple_reply(seq, error);
     }
     return error;
 }
@@ -495,7 +525,7 @@ handle_port_cmd(struct ofpbuf *buffer, bool add)
             VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
             error = rewrite_and_reload_config();
         }
-        send_reply(seq, error, NULL);
+        send_simple_reply(seq, error);
     }
 
     return error;
@@ -522,6 +552,33 @@ get_bridge_containing_port(const char *port_name)
     return xmemdup0(start, end - start);
 }
 
+static int
+linux_bridge_to_ovs_bridge(const char *linux_bridge,
+                           char **ovs_bridge, int *br_vlan)
+{
+    if (bridge_exists(linux_bridge)) {
+        /* Bridge name is the same.  We are interested in VLAN 0. */
+        *ovs_bridge = xstrdup(linux_bridge);
+        *br_vlan = 0;
+        return 0;
+    } else {
+        /* No such Open vSwitch bridge 'linux_bridge', but there might be an
+         * internal port named 'linux_bridge' on some other bridge
+         * 'ovs_bridge'.  If so then we are interested in the VLAN assigned to
+         * port 'linux_bridge' on the bridge named 'ovs_bridge'. */
+        const char *port_name = linux_bridge;
+
+        *ovs_bridge = get_bridge_containing_port(port_name);
+        *br_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
+        if (*ovs_bridge && *br_vlan >= 0) {
+            return 0;
+        } else {
+            free(*ovs_bridge);
+            return ENODEV;
+        }
+    }
+}
+
 static int
 handle_fdb_query_cmd(struct ofpbuf *buffer)
 {
@@ -557,6 +614,7 @@ handle_fdb_query_cmd(struct ofpbuf *buffer)
     struct svec ifaces;
 
     struct ofpbuf query_data;
+    struct ofpbuf *reply;
     char *unixctl_command;
     uint64_t count, skip;
     char *output;
@@ -572,24 +630,10 @@ handle_fdb_query_cmd(struct ofpbuf *buffer)
 
     /* Figure out vswitchd bridge and VLAN. */
     cfg_read();
-    if (bridge_exists(linux_bridge)) {
-        /* Bridge name is the same.  We are interested in VLAN 0. */
-        ovs_bridge = xstrdup(linux_bridge);
-        br_vlan = 0;
-    } else {
-        /* No such Open vSwitch bridge 'linux_bridge', but there might be an
-         * internal port named 'linux_bridge' on some other bridge
-         * 'ovs_bridge'.  If so then we are interested in the VLAN assigned to
-         * port 'linux_bridge' on the bridge named 'ovs_bridge'. */
-        const char *port_name = linux_bridge;
-
-        ovs_bridge = get_bridge_containing_port(port_name);
-        br_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
-        if (!ovs_bridge || br_vlan < 0) {
-            free(ovs_bridge);
-            send_reply(seq, ENODEV, NULL);
-            return error;
-        }
+    error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
+    if (error) {
+        send_simple_reply(seq, error);
+        return error;
     }
 
     /* Fetch the forwarding database using ovs-appctl. */
@@ -598,12 +642,13 @@ handle_fdb_query_cmd(struct ofpbuf *buffer)
     free(unixctl_command);
     if (error) {
         free(ovs_bridge);
-        send_reply(seq, error, NULL);
+        send_simple_reply(seq, error);
         return error;
     }
 
     /* Fetch the MAC address for each interface on the bridge, so that we can
      * fill in the is_local field in the response. */
+    svec_init(&ifaces);
     get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
     local_macs = xmalloc(ifaces.n * sizeof *local_macs);
     n_local_macs = 0;
@@ -677,13 +722,132 @@ handle_fdb_query_cmd(struct ofpbuf *buffer)
     }
     free(output);
 
-    send_reply(seq, 0, &query_data);
+    /* Compose and send reply to datapath. */
+    reply = compose_reply(seq, 0);
+    nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
+                      query_data.data, query_data.size);
+    send_reply(reply);
+
+    /* Free memory. */
     ofpbuf_uninit(&query_data);
     free(ovs_bridge);
 
     return 0;
 }
 
+static void
+send_ifindex_reply(uint32_t seq, struct svec *ifaces)
+{
+    struct ofpbuf *reply;
+    const char *iface;
+    size_t n_indices;
+    int *indices;
+    size_t i;
+
+    /* Make sure that any given interface only occurs once.  This shouldn't
+     * happen, but who knows what people put into their configuration files. */
+    svec_sort_unique(ifaces);
+
+    /* Convert 'ifaces' into ifindexes. */
+    n_indices = 0;
+    indices = xmalloc(ifaces->n * sizeof *indices);
+    SVEC_FOR_EACH (i, iface, ifaces) {
+        int ifindex = if_nametoindex(iface);
+        if (ifindex) {
+            indices[n_indices++] = ifindex;
+        }
+    }
+
+    /* Compose and send reply. */
+    reply = compose_reply(seq, 0);
+    nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
+                      indices, n_indices * sizeof *indices);
+    send_reply(reply);
+
+    /* Free memory. */
+    free(indices);
+}
+
+static int
+handle_get_bridges_cmd(struct ofpbuf *buffer)
+{
+    struct svec bridges;
+    const char *br_name;
+    size_t i;
+
+    uint32_t seq;
+
+    int error;
+
+    /* Parse Netlink command.
+     *
+     * The command doesn't actually have any arguments, but we need the
+     * sequence number to send the reply. */
+    error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
+    if (error) {
+        return error;
+    }
+
+    /* Get all the real bridges and all the fake ones. */
+    cfg_read();
+    cfg_get_subsections(&bridges, "bridge");
+    SVEC_FOR_EACH (i, br_name, &bridges) {
+        const char *iface_name;
+        struct svec ifaces;
+        size_t j;
+
+        svec_init(&ifaces);
+        get_bridge_ifaces(br_name, &ifaces, -1);
+        SVEC_FOR_EACH (j, iface_name, &ifaces) {
+            if (cfg_get_bool(0, "iface.%s.fake-bridge", iface_name)) {
+                svec_add(&bridges, iface_name);
+            }
+        }
+        svec_destroy(&ifaces);
+    }
+
+    send_ifindex_reply(seq, &bridges);
+    svec_destroy(&bridges);
+
+    return 0;
+}
+
+static int
+handle_get_ports_cmd(struct ofpbuf *buffer)
+{
+    uint32_t seq;
+
+    const char *linux_bridge;
+    char *ovs_bridge;
+    int br_vlan;
+
+    struct svec ports;
+
+    int error;
+
+    /* Parse Netlink command. */
+    error = parse_command(buffer, &seq, &linux_bridge, NULL, NULL, NULL);
+    if (error) {
+        return error;
+    }
+
+    cfg_read();
+    error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
+    if (error) {
+        send_simple_reply(seq, error);
+        return error;
+    }
+
+    svec_init(&ports);
+    get_bridge_ports(ovs_bridge, &ports, br_vlan);
+    send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
+    svec_destroy(&ports);
+
+    free(ovs_bridge);
+
+    return 0;
+}
+
 static int
 brc_recv_update(void)
 {
@@ -746,6 +910,14 @@ brc_recv_update(void)
         retval = handle_fdb_query_cmd(buffer);
         break;
 
+    case BRC_GENL_C_GET_BRIDGES:
+        retval = handle_get_bridges_cmd(buffer);
+        break;
+
+    case BRC_GENL_C_GET_PORTS:
+        retval = handle_get_ports_cmd(buffer);
+        break;
+
     default:
         retval = EPROTO;
     }
@@ -905,7 +1077,10 @@ main(int argc, char *argv[])
         }
     }
 
-    cfg_read();
+    retval = cfg_read();
+    if (retval) {
+        ovs_fatal(retval, "could not read config file");
+    }
 
     for (;;) {
         unixctl_server_run(unixctl);
@@ -1040,6 +1215,7 @@ parse_options(int argc, char *argv[])
                 "use --help for usage");
     }
 
+    cfg_init();
     config_file = argv[0];
     error = cfg_set_file(config_file);
     if (error) {