X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=vswitchd%2Fovs-brcompatd.c;h=df9332f3aaf38bc6e73b4f293392b4eed93864c2;hb=2702241699d0539d6ff4a2d44c5bc0ab9d7f7fc8;hp=93d9469bda82836a6bcddda132de540c0fe9a984;hpb=064af42167bf4fc9aaea2702d80ce08074b889c0;p=openvswitch

diff --git a/vswitchd/ovs-brcompatd.c b/vswitchd/ovs-brcompatd.c
index 93d9469b..df9332f3 100644
--- a/vswitchd/ovs-brcompatd.c
+++ b/vswitchd/ovs-brcompatd.c
@@ -1,22 +1,21 @@
-/* Copyright (c) 2008, 2009 Nicira Networks
- * 
- * This program is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
+/* Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
  *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
  *
- * You should have received a copy of the GNU General Public License
- * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *     http://www.apache.org/licenses/LICENSE-2.0
  *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
  */
 
 #include <config.h>
 
+#include <asm/param.h>
 #include <assert.h>
 #include <errno.h>
 #include <getopt.h>
@@ -29,67 +28,55 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/types.h>
+#include <sys/wait.h>
 #include <sys/stat.h>
+#include <time.h>
 #include <fcntl.h>
 #include <unistd.h>
 
-#include "cfg.h"
 #include "command-line.h"
 #include "coverage.h"
 #include "daemon.h"
 #include "dirs.h"
-#include "dpif.h"
+#include "dynamic-string.h"
 #include "fatal-signal.h"
-#include "fault.h"
+#include "json.h"
 #include "leak-checker.h"
 #include "netdev.h"
 #include "netlink.h"
+#include "netlink-notifier.h"
+#include "netlink-socket.h"
 #include "ofpbuf.h"
 #include "openvswitch/brcompat-netlink.h"
+#include "packets.h"
 #include "poll-loop.h"
 #include "process.h"
+#include "rtnetlink-link.h"
 #include "signals.h"
+#include "sset.h"
 #include "svec.h"
 #include "timeval.h"
 #include "unixctl.h"
 #include "util.h"
-
 #include "vlog.h"
-#define THIS_MODULE VLM_brcompatd
 
+VLOG_DEFINE_THIS_MODULE(brcompatd);
 
 /* xxx Just hangs if datapath is rmmod/insmod.  Learn to reconnect? */
 
-/* Actions to modify bridge compatibility configuration. */
-enum bmc_action {
-    BMC_ADD_DP,
-    BMC_DEL_DP,
-    BMC_ADD_PORT,
-    BMC_DEL_PORT
-};
-
 static void parse_options(int argc, char *argv[]);
 static void usage(void) NO_RETURN;
 
 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 60);
 
-/* Maximum number of milliseconds to wait for the config file to be
- * unlocked.  If set to zero, no waiting will occur. */
-static int lock_timeout = 500;
-
-/* Maximum number of milliseconds to wait before pruning port entries that 
- * no longer exist.  If set to zero, ports are never pruned. */
-static int prune_timeout = 5000;
+/* --appctl: Absolute path to ovs-appctl. */
+static char *appctl_program;
 
-/* Config file shared with ovs-vswitchd (usually ovs-vswitchd.conf). */
-static char *config_file;
+/* --vsctl: Absolute path to ovs-vsctl. */
+static char *vsctl_program;
 
-/* Command to run (via system()) to reload the ovs-vswitchd configuration
- * file. */
-static char *reload_command;
-
-/* Netlink socket to listen for interface changes. */
-static struct nl_sock *rtnl_sock;
+/* Options that we should generally pass to ovs-vsctl. */
+#define VSCTL_OPTIONS "--timeout=5", "-vconsole:warn"
 
 /* Netlink socket to bridge compatibility kernel module. */
 static struct nl_sock *brc_sock;
@@ -101,10 +88,90 @@ static const struct nl_policy brc_multicast_policy[] = {
     [BRC_GENL_A_MC_GROUP] = {.type = NL_A_U32 }
 };
 
-static const struct nl_policy rtnlgrp_link_policy[] = {
-    [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
-    [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
-};
+static char *
+capture_vsctl_valist(const char *arg0, va_list args)
+{
+    char *stdout_log, *stderr_log;
+    enum vlog_level log_level;
+    struct svec argv;
+    int status;
+    char *msg;
+
+    /* Compose arguments. */
+    svec_init(&argv);
+    svec_add(&argv, arg0);
+    for (;;) {
+        const char *arg = va_arg(args, const char *);
+        if (!arg) {
+            break;
+        }
+        svec_add(&argv, arg);
+    }
+    svec_terminate(&argv);
+
+    /* Run process. */
+    if (process_run_capture(argv.names, &stdout_log, &stderr_log, SIZE_MAX,
+                            &status)) {
+        svec_destroy(&argv);
+        return NULL;
+    }
+
+    /* Log results. */
+    if (WIFEXITED(status)) {
+        int code = WEXITSTATUS(status);
+        log_level = code == 0 ? VLL_DBG : code == 1 ? VLL_WARN : VLL_ERR;
+    } else {
+        log_level = VLL_ERR;
+    }
+    msg = process_status_msg(status);
+    VLOG(log_level, "ovs-vsctl exited (%s)", msg);
+    if (stdout_log && *stdout_log) {
+        VLOG(log_level, "ovs-vsctl wrote to stdout:\n%s\n", stdout_log);
+    }
+    if (stderr_log && *stderr_log) {
+        VLOG(log_level, "ovs-vsctl wrote to stderr:\n%s\n", stderr_log);
+    }
+    free(msg);
+
+    svec_destroy(&argv);
+
+    free(stderr_log);
+    if (WIFEXITED(status) && !WEXITSTATUS(status)) {
+        return stdout_log;
+    } else {
+        free(stdout_log);
+        return NULL;
+    }
+}
+
+static char * SENTINEL(0)
+capture_vsctl(const char *arg0, ...)
+{
+    char *stdout_log;
+    va_list args;
+
+    va_start(args, arg0);
+    stdout_log = capture_vsctl_valist(arg0, args);
+    va_end(args);
+
+    return stdout_log;
+}
+
+static bool SENTINEL(0)
+run_vsctl(const char *arg0, ...)
+{
+    char *stdout_log;
+    va_list args;
+    bool ok;
+
+    va_start(args, arg0);
+    stdout_log = capture_vsctl_valist(arg0, args);
+    va_end(args);
+
+    ok = stdout_log != NULL;
+    free(stdout_log);
+    return ok;
+}
 
 static int
 lookup_brc_multicast_group(int *multicast_group)
@@ -114,12 +181,12 @@ lookup_brc_multicast_group(int *multicast_group)
     struct nlattr *attrs[ARRAY_SIZE(brc_multicast_policy)];
     int retval;
 
-    retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
+    retval = nl_sock_create(NETLINK_GENERIC, &sock);
     if (retval) {
         return retval;
     }
     ofpbuf_init(&request, 0);
-    nl_msg_put_genlmsghdr(&request, sock, 0, brc_family,
+    nl_msg_put_genlmsghdr(&request, 0, brc_family,
             NLM_F_REQUEST, BRC_GENL_C_QUERY_MC, 1);
     retval = nl_sock_transact(sock, &request, &reply);
     ofpbuf_uninit(&request);
@@ -159,500 +226,621 @@ brc_open(struct nl_sock **sock)
         return retval;
     }
 
-    retval = nl_sock_create(NETLINK_GENERIC, multicast_group, 0, 0, sock);
+    retval = nl_sock_create(NETLINK_GENERIC, sock);
     if (retval) {
         return retval;
     }
 
-    return 0;
+    retval = nl_sock_join_mcgroup(*sock, multicast_group);
+    if (retval) {
+        nl_sock_destroy(*sock);
+        *sock = NULL;
+    }
+    return retval;
 }
 
 static const struct nl_policy brc_dp_policy[] = {
     [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
 };
 
-static bool
-bridge_exists(const char *name)
-{
-    return cfg_has_section("bridge.%s", name);
-}
-
 static int
-rewrite_and_reload_config(void)
+parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
+              const char **port_name, uint64_t *count, uint64_t *skip)
 {
-    if (cfg_is_dirty()) {
-        int error1 = cfg_write();
-        int error2 = cfg_read();
-        long long int reload_start = time_msec();
-        int error3 = system(reload_command);
-        long long int elapsed = time_msec() - reload_start;
-        COVERAGE_INC(brcompatd_reload);
-        if (elapsed > 0) {
-            VLOG_INFO("reload command executed in %lld ms", elapsed);
-        }
-        if (error3 == -1) {
-            VLOG_ERR("failed to execute reload command: %s", strerror(errno));
-        } else if (error3 != 0) {
-            char *msg = process_status_msg(error3);
-            VLOG_ERR("reload command exited with error (%s)", msg);
-            free(msg);
-        }
-        return error1 ? error1 : error2 ? error2 : error3 ? ECHILD : 0;
+    static const struct nl_policy policy[] = {
+        [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 },
+    };
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+
+    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])) {
+        return EINVAL;
+    }
+
+    *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
+    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]);
+    }
+    if (count) {
+        *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
+    }
+    if (skip) {
+        *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
     }
     return 0;
 }
 
-/* Go through the configuration file and remove any ports that no longer
- * exist associated with a bridge. */
-static void
-prune_ports(void)
+/* Composes and returns a reply to a request made by the datapath with 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(int error)
 {
-    int i, j;
-    int error;
-    struct svec bridges, delete;
+    struct ofpbuf *reply = ofpbuf_new(4096);
+    nl_msg_put_genlmsghdr(reply, 32, brc_family, NLM_F_REQUEST,
+                          BRC_GENL_C_DP_RESULT, 1);
+    nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
+    return reply;
+}
 
-    if (cfg_lock(NULL, 0)) {
-        /* Couldn't lock config file. */
-        return;
+/* Sends 'reply' to the datapath, using sequence number 'nlmsg_seq', and frees
+ * it. */
+static void
+send_reply(struct ofpbuf *reply, uint32_t nlmsg_seq)
+{
+    int retval = nl_sock_send_seq(brc_sock, reply, nlmsg_seq, false);
+    if (retval) {
+        VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
+                     strerror(retval));
     }
+    ofpbuf_delete(reply);
+}
 
-    svec_init(&bridges);
-    svec_init(&delete);
-    cfg_get_subsections(&bridges, "bridge");
-    for (i=0; i<bridges.n; i++) {
-        const char *br_name = bridges.names[i];
-        struct svec ports, ifaces;
-
-        svec_init(&ports);
-
-        /* Get all the interfaces for the given bridge, breaking bonded
-         * interfaces down into their constituent parts. */
-        svec_init(&ifaces);
-        cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
-        for (j=0; j<ports.n; j++) {
-            const char *port_name = ports.names[j];
-            if (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_destroy(&slaves);
-            } else {
-                svec_add(&ifaces, port_name);
-            }
-        }
-        svec_destroy(&ports);
-
-        /* Check that the interfaces exist. */
-        for (j = 0; j < ifaces.n; j++) {
-            const char *iface_name = ifaces.names[j];
-            enum netdev_flags flags;
-
-            /* The local port and internal ports are created and destroyed by
-             * ovs-vswitchd itself, so don't bother checking for them at all.
-             * In practice, they might not exist if ovs-vswitchd hasn't
-             * finished reloading since the configuration file was updated. */
-            if (!strcmp(iface_name, br_name)
-                || cfg_get_bool(0, "iface.%s.internal", iface_name)) {
-                continue;
-            }
-
-            error = netdev_nodev_get_flags(iface_name, &flags);
-            if (error == ENODEV) {
-                VLOG_DBG_RL(&rl, "removing dead interface %s from %s", 
-                        iface_name, br_name);
-                svec_add(&delete, iface_name);
-            } else if (error) {
-                VLOG_DBG_RL(&rl, "unknown error %d on interface %s from %s", 
-                        error, iface_name, br_name);
-            }
-        }
-        svec_destroy(&ifaces);
-    }
-    svec_destroy(&bridges);
+/* 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(error), seq);
+}
 
-    if (delete.n) {
-        size_t i;
+static int
+handle_bridge_cmd(struct ofpbuf *buffer, bool add)
+{
+    const char *br_name;
+    uint32_t seq;
+    int error;
 
-        for (i = 0; i < delete.n; i++) {
-            cfg_del_match("bridge.*.port=%s", delete.names[i]);
-            cfg_del_match("bonding.*.slave=%s", delete.names[i]);
+    error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
+    if (!error) {
+        const char *vsctl_cmd = add ? "add-br" : "del-br";
+        const char *brctl_cmd = add ? "addbr" : "delbr";
+        if (!run_vsctl(vsctl_program, VSCTL_OPTIONS,
+                       "--", vsctl_cmd, br_name,
+                       "--", "comment", "ovs-brcompatd:", brctl_cmd, br_name,
+                       (char *) NULL)) {
+            error = add ? EEXIST : ENXIO;
         }
-        rewrite_and_reload_config();
-        cfg_unlock();
-    } else {
-        cfg_unlock();
+        send_simple_reply(seq, error);
     }
-    svec_destroy(&delete);
+    return error;
 }
 
+static const struct nl_policy brc_port_policy[] = {
+    [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
+    [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
+};
 
-/* Checks whether a network device named 'name' exists and returns true if so,
- * false otherwise.
- *
- * XXX it is possible that this doesn't entirely accomplish what we want in
- * context, since ovs-vswitchd.conf may cause vswitchd to create or destroy
- * network devices based on iface.*.internal settings.
- *
- * XXX may want to move this to lib/netdev. */
-static bool
-netdev_exists(const char *name)
+static int
+handle_port_cmd(struct ofpbuf *buffer, bool add)
 {
-    struct stat s;
-    char *filename;
+    const char *br_name, *port_name;
+    uint32_t seq;
     int error;
 
-    filename = xasprintf("/sys/class/net/%s", name);
-    error = stat(filename, &s);
-    free(filename);
-    return !error;
+    error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
+    if (!error) {
+        const char *vsctl_cmd = add ? "add-port" : "del-port";
+        const char *brctl_cmd = add ? "addif" : "delif";
+        if (!run_vsctl(vsctl_program, VSCTL_OPTIONS,
+                       "--", vsctl_cmd, br_name, port_name,
+                       "--", "comment", "ovs-brcompatd:", brctl_cmd,
+                       br_name, port_name, (char *) NULL)) {
+            error = EINVAL;
+        }
+        send_simple_reply(seq, error);
+    }
+    return error;
 }
 
-static int
-add_bridge(const char *br_name)
+static char *
+linux_bridge_to_ovs_bridge(const char *linux_name, int *br_vlanp)
 {
-    if (bridge_exists(br_name)) {
-        VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
-        return EEXIST;
-    } else if (netdev_exists(br_name)) {
-        if (cfg_get_bool(0, "iface.%s.fake-bridge", br_name)) {
-            VLOG_WARN("addbr %s: %s exists as a fake bridge",
-                      br_name, br_name);
-            return 0;
-        } else {
-            VLOG_WARN("addbr %s: cannot create bridge %s because a network "
-                      "device named %s already exists",
-                      br_name, br_name, br_name);
-            return EEXIST;
-        }
+    char *save_ptr = NULL;
+    const char *br_name, *br_vlan;
+    char *br_name_copy;
+    char *output;
+
+    output = capture_vsctl(vsctl_program, VSCTL_OPTIONS,
+                           "--", "br-to-parent", linux_name,
+                           "--", "br-to-vlan", linux_name,
+                           (char *) NULL);
+    if (!output) {
+        return NULL;
     }
 
-    cfg_add_entry("bridge.%s.port=%s", br_name, br_name);
-    VLOG_INFO("addbr %s: success", br_name);
+    br_name = strtok_r(output, " \t\r\n", &save_ptr);
+    br_vlan = strtok_r(NULL, " \t\r\n", &save_ptr);
+    if (!br_name || !br_vlan) {
+        free(output);
+        return NULL;
+    }
+    br_name_copy = xstrdup(br_name);
+    *br_vlanp = atoi(br_vlan);
 
-    return 0;
+    free(output);
+
+    return br_name_copy;
 }
 
-static int 
-del_bridge(const char *br_name)
+static void
+get_bridge_ifaces(const char *br_name, struct sset *ifaces)
 {
-    if (!bridge_exists(br_name)) {
-        VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
-        return ENXIO;
-    }
+    char *save_ptr = NULL;
+    char *output;
+    char *iface;
 
-    cfg_del_section("bridge.%s", br_name);
-    VLOG_INFO("delbr %s: success", br_name);
+    output = capture_vsctl(vsctl_program, VSCTL_OPTIONS, "list-ifaces",
+                           br_name, (char *) NULL);
+    if (!output) {
+        return;
+    }
 
-    return 0;
+    for (iface = strtok_r(output, " \t\r\n", &save_ptr); iface;
+         iface = strtok_r(NULL, " \t\r\n", &save_ptr)) {
+        sset_add(ifaces, iface);
+    }
+    free(output);
 }
 
 static int
-parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
-              const char **port_name)
+handle_fdb_query_cmd(struct ofpbuf *buffer)
 {
-    static const struct nl_policy policy[] = {
-        [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
-        [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
+    /* This structure is copied directly from the Linux 2.6.30 header files.
+     * It would be more straightforward to #include <linux/if_bridge.h>, but
+     * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
+     * with old header files won't have it. */
+    struct __fdb_entry {
+        __u8 mac_addr[6];
+        __u8 port_no;
+        __u8 is_local;
+        __u32 ageing_timer_value;
+        __u8 port_hi;
+        __u8 pad0;
+        __u16 unused;
     };
-    struct nlattr *attrs[ARRAY_SIZE(policy)];
 
-    if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
-                         attrs, ARRAY_SIZE(policy))
-        || (port_name && !attrs[BRC_GENL_A_PORT_NAME])) {
-        return EINVAL;
+    struct mac {
+        uint8_t addr[6];
+    };
+    struct mac *local_macs;
+    int n_local_macs;
+    int i;
+
+    /* Impedance matching between the vswitchd and Linux kernel notions of what
+     * a bridge is.  The kernel only handles a single VLAN per bridge, but
+     * vswitchd can deal with all the VLANs on a single bridge.  We have to
+     * pretend that the former is the case even though the latter is the
+     * implementation. */
+    const char *linux_name;   /* Name used by brctl. */
+    int br_vlan;                /* VLAN tag. */
+    struct sset ifaces;
+
+    struct ofpbuf query_data;
+    const char *iface_name;
+    struct ofpbuf *reply;
+    uint64_t count, skip;
+    char *br_name;
+    char *output;
+    char *save_ptr;
+    uint32_t seq;
+    int error;
+
+    /* Parse the command received from brcompat. */
+    error = parse_command(buffer, &seq, &linux_name, NULL, &count, &skip);
+    if (error) {
+        return error;
     }
 
-    *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
-    *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]);
+    /* Figure out vswitchd bridge and VLAN. */
+    br_name = linux_bridge_to_ovs_bridge(linux_name, &br_vlan);
+    if (!br_name) {
+        error = EINVAL;
+        send_simple_reply(seq, error);
+        return error;
     }
+
+    /* Fetch the forwarding database using ovs-appctl. */
+    output = capture_vsctl(appctl_program, "fdb/show", br_name,
+                           (char *) NULL);
+    if (!output) {
+        error = ECHILD;
+        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. */
+    sset_init(&ifaces);
+    get_bridge_ifaces(linux_name, &ifaces);
+    local_macs = xmalloc(sset_count(&ifaces) * sizeof *local_macs);
+    n_local_macs = 0;
+    SSET_FOR_EACH (iface_name, &ifaces) {
+        struct mac *mac = &local_macs[n_local_macs];
+        struct netdev *netdev;
+
+        error = netdev_open(iface_name, "system", &netdev);
+        if (!error) {
+            if (!netdev_get_etheraddr(netdev, mac->addr)) {
+                n_local_macs++;
+            }
+            netdev_close(netdev);
+        }
+    }
+    sset_destroy(&ifaces);
+
+    /* Parse the response from ovs-appctl and convert it to binary format to
+     * pass back to the kernel. */
+    ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
+    save_ptr = NULL;
+    strtok_r(output, "\n", &save_ptr); /* Skip header line. */
+    while (count > 0) {
+        struct __fdb_entry *entry;
+        int port, vlan, age;
+        uint8_t mac[ETH_ADDR_LEN];
+        char *line;
+        bool is_local;
+
+        line = strtok_r(NULL, "\n", &save_ptr);
+        if (!line) {
+            break;
+        }
+
+        if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
+                   &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
+            != 2 + ETH_ADDR_SCAN_COUNT + 1) {
+            static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+            VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
+            continue;
+        }
+
+        if (vlan != br_vlan) {
+            continue;
+        }
+
+        if (skip > 0) {
+            skip--;
+            continue;
+        }
+
+        /* Is this the MAC address of an interface on the bridge? */
+        is_local = false;
+        for (i = 0; i < n_local_macs; i++) {
+            if (eth_addr_equals(local_macs[i].addr, mac)) {
+                is_local = true;
+                break;
+            }
+        }
+
+        entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
+        memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
+        entry->port_no = port & 0xff;
+        entry->is_local = is_local;
+        entry->ageing_timer_value = age * HZ;
+        entry->port_hi = (port & 0xff00) >> 8;
+        entry->pad0 = 0;
+        entry->unused = 0;
+        count--;
+    }
+    free(output);
+
+    /* Compose and send reply to datapath. */
+    reply = compose_reply(0);
+    nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
+                      query_data.data, query_data.size);
+    send_reply(reply, seq);
+
+    /* Free memory. */
+    ofpbuf_uninit(&query_data);
+    free(local_macs);
+
     return 0;
 }
 
 static void
-send_reply(uint32_t seq, int error)
+send_ifindex_reply(uint32_t seq, char *output)
 {
-    struct ofpbuf msg;
-    int retval;
-
-    /* Compose reply. */
-    ofpbuf_init(&msg, 0);
-    nl_msg_put_genlmsghdr(&msg, 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);
+    size_t allocated_indices;
+    char *save_ptr = NULL;
+    struct ofpbuf *reply;
+    const char *iface;
+    size_t n_indices;
+    int *indices;
+
+    indices = NULL;
+    n_indices = allocated_indices = 0;
+    for (iface = strtok_r(output, " \t\r\n", &save_ptr); iface;
+         iface = strtok_r(NULL, " \t\r\n", &save_ptr)) {
+        int ifindex;
+
+        if (n_indices >= allocated_indices) {
+            indices = x2nrealloc(indices, &allocated_indices, sizeof *indices);
+        }
 
-    /* Send reply. */
-    retval = nl_sock_send(brc_sock, &msg, false);
-    if (retval) {
-        VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
-                     strerror(retval));
+        ifindex = if_nametoindex(iface);
+        if (ifindex) {
+            indices[n_indices++] = ifindex;
+        }
     }
-    ofpbuf_uninit(&msg);
+
+    /* Compose and send reply. */
+    reply = compose_reply(0);
+    nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
+                      indices, n_indices * sizeof *indices);
+    send_reply(reply, seq);
+
+    /* Free memory. */
+    free(indices);
 }
 
 static int
-handle_bridge_cmd(struct ofpbuf *buffer, bool add)
+handle_get_bridges_cmd(struct ofpbuf *buffer)
 {
-    const char *br_name;
+    char *output;
     uint32_t seq;
     int error;
 
-    error = parse_command(buffer, &seq, &br_name, NULL);
-    if (!error) {
-        error = add ? add_bridge(br_name) : del_bridge(br_name);
-        if (!error) {
-            error = rewrite_and_reload_config();
-        }
-        send_reply(seq, 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;
     }
-    return error;
-}
 
-static const struct nl_policy brc_port_policy[] = {
-    [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
-    [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
-};
+    output = capture_vsctl(vsctl_program, VSCTL_OPTIONS, "list-br", (char *) NULL);
+    if (!output) {
+        return ENODEV;
+    }
 
-static void
-del_port(const char *br_name, const char *port_name)
-{
-    cfg_del_entry("bridge.%s.port=%s", br_name, port_name);
-    cfg_del_match("bonding.*.slave=%s", port_name);
-    cfg_del_match("vlan.%s.*", port_name);
+    send_ifindex_reply(seq, output);
+    free(output);
+    return 0;
 }
 
 static int
-handle_port_cmd(struct ofpbuf *buffer, bool add)
+handle_get_ports_cmd(struct ofpbuf *buffer)
 {
-    const char *cmd_name = add ? "add-if" : "del-if";
-    const char *br_name, *port_name;
+    const char *linux_name;
     uint32_t seq;
+    char *output;
     int error;
 
-    error = parse_command(buffer, &seq, &br_name, &port_name);
-    if (!error) {
-        if (!bridge_exists(br_name)) {
-            VLOG_WARN("%s %s %s: no bridge named %s",
-                      cmd_name, br_name, port_name, br_name);
-            error = EINVAL;
-        } else if (!netdev_exists(port_name)) {
-            VLOG_WARN("%s %s %s: no network device named %s",
-                      cmd_name, br_name, port_name, port_name);
-            error = EINVAL;
-        } else {
-            if (add) {
-                cfg_add_entry("bridge.%s.port=%s", br_name, port_name);
-            } else {
-                del_port(br_name, port_name);
-            }
-            VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
-            error = rewrite_and_reload_config();
-        }
-        send_reply(seq, error);
+    /* Parse Netlink command. */
+    error = parse_command(buffer, &seq, &linux_name, NULL, NULL, NULL);
+    if (error) {
+        return error;
     }
 
-    return error;
+    output = capture_vsctl(vsctl_program, VSCTL_OPTIONS, "list-ports", linux_name,
+                           (char *) NULL);
+    if (!output) {
+        return ENODEV;
+    }
+
+    send_ifindex_reply(seq, output);
+    free(output);
+    return 0;
 }
 
-static int
-brc_recv_update(void)
+static bool
+brc_recv_update__(struct ofpbuf *buffer)
 {
-    int retval;
-    struct ofpbuf *buffer;
-    struct genlmsghdr *genlmsghdr;
+    for (;;) {
+        int retval = nl_sock_recv(brc_sock, buffer, false);
+        switch (retval) {
+        case 0:
+            if (nl_msg_nlmsgerr(buffer, NULL)
+                || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE) {
+                break;
+            }
+            return true;
 
+        case ENOBUFS:
+            break;
 
-    buffer = NULL;
-    do {
-        ofpbuf_delete(buffer);
-        retval = nl_sock_recv(brc_sock, &buffer, false);
-    } while (retval == ENOBUFS
-            || (!retval
-                && (nl_msg_nlmsgerr(buffer, NULL)
-                    || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE)));
-    if (retval) {
-        if (retval != EAGAIN) {
+        case EAGAIN:
+            return false;
+
+        default:
             VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
+            return false;
         }
-        return retval;
     }
+}
 
-    genlmsghdr = nl_msg_genlmsghdr(buffer);
+static void
+brc_recv_update(void)
+{
+    struct genlmsghdr *genlmsghdr;
+    uint64_t buffer_stub[1024 / 8];
+    struct ofpbuf buffer;
+
+    ofpbuf_use_stub(&buffer, buffer_stub, sizeof buffer_stub);
+    if (!brc_recv_update__(&buffer)) {
+        goto error;
+    }
+
+    genlmsghdr = nl_msg_genlmsghdr(&buffer);
     if (!genlmsghdr) {
         VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
         goto error;
     }
 
-    if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
+    if (nl_msg_nlmsghdr(&buffer)->nlmsg_type != brc_family) {
         VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
-                nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
+                nl_msg_nlmsghdr(&buffer)->nlmsg_type, brc_family);
         goto error;
     }
 
-    if (cfg_lock(NULL, lock_timeout)) {
-        /* Couldn't lock config file. */
-        retval = EAGAIN;
-        goto error;
-    }
+    /* Service all pending network device notifications before executing the
+     * command.  This is very important to avoid a race in a scenario like the
+     * following, which is what happens with XenServer Tools version 5.0.0
+     * during boot of a Windows VM:
+     *
+     *      1. Create tap1.0 and vif1.0.
+     *      2. Delete tap1.0.
+     *      3. Delete vif1.0.
+     *      4. Re-create vif1.0.
+     *
+     * We must process the network device notification from step 3 before we
+     * process the brctl command from step 4.  If we process them in the
+     * reverse order, then step 4 completes as a no-op but step 3 then deletes
+     * the port that was just added.
+     *
+     * (XenServer Tools 5.5.0 does not exhibit this behavior, and neither does
+     * a VM without Tools installed at all.)
+     */
+    rtnetlink_link_run();
 
     switch (genlmsghdr->cmd) {
     case BRC_GENL_C_DP_ADD:
-        retval = handle_bridge_cmd(buffer, true);
+        handle_bridge_cmd(&buffer, true);
         break;
 
     case BRC_GENL_C_DP_DEL:
-        retval = handle_bridge_cmd(buffer, false);
+        handle_bridge_cmd(&buffer, false);
         break;
 
     case BRC_GENL_C_PORT_ADD:
-        retval = handle_port_cmd(buffer, true);
+        handle_port_cmd(&buffer, true);
         break;
 
     case BRC_GENL_C_PORT_DEL:
-        retval = handle_port_cmd(buffer, false);
+        handle_port_cmd(&buffer, false);
+        break;
+
+    case BRC_GENL_C_FDB_QUERY:
+        handle_fdb_query_cmd(&buffer);
+        break;
+
+    case BRC_GENL_C_GET_BRIDGES:
+        handle_get_bridges_cmd(&buffer);
+        break;
+
+    case BRC_GENL_C_GET_PORTS:
+        handle_get_ports_cmd(&buffer);
         break;
 
     default:
-        retval = EPROTO;
+        VLOG_WARN_RL(&rl, "received unknown brc netlink command: %d\n",
+                     genlmsghdr->cmd);
+        break;
     }
 
-    cfg_unlock();
-
 error:
-    ofpbuf_delete(buffer);
-    return retval;
+    ofpbuf_uninit(&buffer);
 }
 
-/* Check for interface configuration changes announced through RTNL. */
 static void
-rtnl_recv_update(void)
+netdev_changed_cb(const struct rtnetlink_link_change *change,
+                  void *aux OVS_UNUSED)
 {
-    struct ofpbuf *buf;
+    char br_name[IFNAMSIZ];
+    const char *port_name;
 
-    int error = nl_sock_recv(rtnl_sock, &buf, false);
-    if (error == EAGAIN) {
-        /* Nothing to do. */
-    } else if (error == ENOBUFS) {
+    if (!change) {
         VLOG_WARN_RL(&rl, "network monitor socket overflowed");
-    } else if (error) {
-        VLOG_WARN_RL(&rl, "error on network monitor socket: %s", 
-                strerror(error));
-    } else {
-        struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
-        struct nlmsghdr *nlh;
-        struct ifinfomsg *iim;
-
-        nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
-        iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
-        if (!iim) {
-            VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
-            ofpbuf_delete(buf);
-            return;
-        } 
-    
-        if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
-                             rtnlgrp_link_policy,
-                             attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
-            VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
-            ofpbuf_delete(buf);
-            return;
-        }
-        if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
-            const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
-            char br_name[IFNAMSIZ];
-            uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
-            struct svec ports;
-
-            if (!if_indextoname(br_idx, br_name)) {
-                ofpbuf_delete(buf);
-                return;
-            }
+        return;
+    }
 
-            if (cfg_lock(NULL, lock_timeout)) {
-                /* Couldn't lock config file. */
-                /* xxx this should try again and print error msg. */
-                ofpbuf_delete(buf);
-                return;
-            }
+    if (change->nlmsg_type != RTM_DELLINK || !change->master_ifindex) {
+        return;
+    }
 
-            svec_init(&ports);
-            cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
-            svec_sort(&ports);
-            if (svec_contains(&ports, port_name)) {
-                del_port(br_name, port_name);
-                rewrite_and_reload_config();
-            }
-            cfg_unlock();
-        }
-        ofpbuf_delete(buf);
+    port_name = change->ifname;
+    if (!if_indextoname(change->master_ifindex, br_name)) {
+        return;
     }
+
+    VLOG_INFO("network device %s destroyed, removing from bridge %s",
+              port_name, br_name);
+
+    run_vsctl(vsctl_program, VSCTL_OPTIONS,
+              "--", "--if-exists", "del-port", port_name,
+              "--", "comment", "ovs-brcompatd:", port_name, "disappeared",
+              (char *) NULL);
 }
 
 int
 main(int argc, char *argv[])
 {
+    extern struct vlog_module VLM_reconnect;
+    struct nln_notifier *link_notifier;
     struct unixctl_server *unixctl;
     int retval;
 
+    proctitle_init(argc, argv);
     set_program_name(argv[0]);
-    register_fault_handlers();
-    time_init();
-    vlog_init();
+    vlog_set_levels(&VLM_reconnect, VLF_ANY_FACILITY, VLL_WARN);
+
     parse_options(argc, argv);
     signal(SIGPIPE, SIG_IGN);
     process_init();
 
-    die_if_already_running();
-    daemonize();
+    daemonize_start();
 
     retval = unixctl_server_create(NULL, &unixctl);
     if (retval) {
-        ovs_fatal(retval, "could not listen for vlog connections");
+        exit(EXIT_FAILURE);
     }
 
     if (brc_open(&brc_sock)) {
-        ovs_fatal(0, "could not open brcompat socket.  Check "
-                "\"brcompat\" kernel module.");
+        VLOG_FATAL("could not open brcompat socket.  Check "
+                   "\"brcompat\" kernel module.");
     }
 
-    if (prune_timeout) {
-        if (nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &rtnl_sock)) {
-            ovs_fatal(0, "could not create rtnetlink socket");
-        }
-    }
+    link_notifier = rtnetlink_link_notifier_create(netdev_changed_cb, NULL);
 
-    cfg_read();
+    daemonize_complete();
 
     for (;;) {
         unixctl_server_run(unixctl);
+        rtnetlink_link_run();
         brc_recv_update();
 
-        /* If 'prune_timeout' is non-zero, we actively prune from the
-         * config file any 'bridge.<br_name>.port' entries that are no 
-         * longer valid.  We use two methods: 
-         *
-         *   1) The kernel explicitly notifies us of removed ports
-         *      through the RTNL messages.
-         *
-         *   2) We periodically check all ports associated with bridges
-         *      to see if they no longer exist.
-         */
-        if (prune_timeout) {
-            rtnl_recv_update();
-            prune_ports();
-
-            nl_sock_wait(rtnl_sock, POLLIN);
-            poll_timer_wait(prune_timeout);
-        }
+        netdev_run();
 
         nl_sock_wait(brc_sock, POLLIN);
         unixctl_server_wait(unixctl);
+        rtnetlink_link_wait();
+        netdev_wait();
         poll_block();
     }
 
+    rtnetlink_link_notifier_destroy(link_notifier);
+
     return 0;
 }
 
@@ -660,31 +848,26 @@ static void
 parse_options(int argc, char *argv[])
 {
     enum {
-        OPT_LOCK_TIMEOUT = UCHAR_MAX + 1,
-        OPT_PRUNE_TIMEOUT,
-        OPT_RELOAD_COMMAND,
+        OPT_APPCTL,
+        OPT_VSCTL,
         VLOG_OPTION_ENUMS,
-        LEAK_CHECKER_OPTION_ENUMS
+        LEAK_CHECKER_OPTION_ENUMS,
+        DAEMON_OPTION_ENUMS
     };
     static struct option long_options[] = {
-        {"help",             no_argument, 0, 'h'},
-        {"version",          no_argument, 0, 'V'},
-        {"lock-timeout",     required_argument, 0, OPT_LOCK_TIMEOUT},
-        {"prune-timeout",    required_argument, 0, OPT_PRUNE_TIMEOUT},
-        {"reload-command",   required_argument, 0, OPT_RELOAD_COMMAND},
+        {"help",             no_argument, NULL, 'h'},
+        {"version",          no_argument, NULL, 'V'},
+        {"appctl",           required_argument, NULL, OPT_APPCTL},
+        {"vsctl",            required_argument, NULL, OPT_VSCTL},
         DAEMON_LONG_OPTIONS,
         VLOG_LONG_OPTIONS,
         LEAK_CHECKER_LONG_OPTIONS,
-        {0, 0, 0, 0},
+        {NULL, 0, NULL, 0},
     };
     char *short_options = long_options_to_short_options(long_options);
-    int error;
+    const char *appctl = "ovs-appctl";
+    const char *vsctl = "ovs-vsctl";
 
-    reload_command = xasprintf("%s/ovs-appctl -t "
-                               "%s/ovs-vswitchd.`cat %s/ovs-vswitchd.pid`.ctl "
-                               "-e vswitchd/reload 2>&1 "
-                               "| /usr/bin/logger -t brcompatd-reload",
-                               ovs_bindir, ovs_rundir, ovs_rundir);
     for (;;) {
         int c;
 
@@ -694,24 +877,19 @@ parse_options(int argc, char *argv[])
         }
 
         switch (c) {
-        case 'H':
         case 'h':
             usage();
 
         case 'V':
-            OVS_PRINT_VERSION(0, 0);
+            ovs_print_version(0, 0);
             exit(EXIT_SUCCESS);
 
-        case OPT_LOCK_TIMEOUT:
-            lock_timeout = atoi(optarg);
+        case OPT_APPCTL:
+            appctl = optarg;
             break;
 
-        case OPT_PRUNE_TIMEOUT:
-            prune_timeout = atoi(optarg) * 1000;
-            break;
-
-        case OPT_RELOAD_COMMAND:
-            reload_command = optarg;
+        case OPT_VSCTL:
+            vsctl = optarg;
             break;
 
         VLOG_OPTION_HANDLERS
@@ -727,19 +905,21 @@ parse_options(int argc, char *argv[])
     }
     free(short_options);
 
-    argc -= optind;
-    argv += optind;
+    appctl_program = process_search_path(appctl);
+    if (!appctl_program) {
+        VLOG_FATAL("%s: not found in $PATH (use --appctl to specify an "
+                   "alternate location)", appctl);
+    }
 
-    if (argc != 1) {
-        ovs_fatal(0, "exactly one non-option argument required; "
-                "use --help for usage");
+    vsctl_program = process_search_path(vsctl);
+    if (!vsctl_program) {
+        VLOG_FATAL("%s: not found in $PATH (use --vsctl to specify an "
+                   "alternate location)", vsctl);
     }
 
-    config_file = argv[0];
-    error = cfg_set_file(config_file);
-    if (error) {
-        ovs_fatal(error, "failed to add configuration file \"%s\"", 
-                config_file);
+    if (argc != optind) {
+        VLOG_FATAL("no non-option arguments are supported; "
+                   "use --help for usage");
     }
 }
 
@@ -747,13 +927,11 @@ static void
 usage(void)
 {
     printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
-           "usage: %s [OPTIONS] CONFIG\n"
-           "CONFIG is the configuration file used by ovs-vswitchd.\n",
+           "usage: %s [OPTIONS]\n",
            program_name, program_name);
     printf("\nConfiguration options:\n"
-           "  --reload-command=COMMAND  shell command to reload ovs-vswitchd\n"
-           "  --prune-timeout=SECS    wait at most SECS before pruning ports\n"
-           "  --lock-timeout=MSECS    wait at most MSECS for CONFIG to unlock\n"
+           "  --appctl=PROGRAM        overrides $PATH for finding ovs-appctl\n"
+           "  --vsctl=PROGRAM         overrides $PATH for finding ovs-vsctl\n"
           );
     daemon_usage();
     vlog_usage();
@@ -761,6 +939,5 @@ usage(void)
            "  -h, --help              display this help message\n"
            "  -V, --version           display version information\n");
     leak_checker_usage();
-    printf("\nThe default reload command is:\n%s\n", reload_command);
     exit(EXIT_SUCCESS);
 }