1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include <asm/param.h>
25 #include <linux/genetlink.h>
26 #include <linux/rtnetlink.h>
30 #include <sys/types.h>
37 #include "command-line.h"
42 #include "dynamic-string.h"
43 #include "fatal-signal.h"
45 #include "leak-checker.h"
49 #include "openvswitch/brcompat-netlink.h"
51 #include "poll-loop.h"
60 #define THIS_MODULE VLM_brcompatd
63 /* xxx Just hangs if datapath is rmmod/insmod. Learn to reconnect? */
65 /* Actions to modify bridge compatibility configuration. */
73 static void parse_options(int argc, char *argv[]);
74 static void usage(void) NO_RETURN;
76 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 60);
78 /* Maximum number of milliseconds to wait for the config file to be
79 * unlocked. If set to zero, no waiting will occur. */
80 static int lock_timeout = 500;
82 /* Maximum number of milliseconds to wait before pruning port entries that
83 * no longer exist. If set to zero, ports are never pruned. */
84 static int prune_timeout = 5000;
86 /* Config file shared with ovs-vswitchd (usually ovs-vswitchd.conf). */
87 static char *config_file;
89 /* Shell command to execute (via popen()) to send a control command to the
90 * running ovs-vswitchd process. The string must contain one instance of %s,
91 * which is replaced by the control command. */
92 static char *appctl_command;
94 /* Netlink socket to listen for interface changes. */
95 static struct nl_sock *rtnl_sock;
97 /* Netlink socket to bridge compatibility kernel module. */
98 static struct nl_sock *brc_sock;
100 /* The Generic Netlink family number used for bridge compatibility. */
101 static int brc_family;
103 static const struct nl_policy brc_multicast_policy[] = {
104 [BRC_GENL_A_MC_GROUP] = {.type = NL_A_U32 }
107 static const struct nl_policy rtnlgrp_link_policy[] = {
108 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
109 [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
113 lookup_brc_multicast_group(int *multicast_group)
115 struct nl_sock *sock;
116 struct ofpbuf request, *reply;
117 struct nlattr *attrs[ARRAY_SIZE(brc_multicast_policy)];
120 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
124 ofpbuf_init(&request, 0);
125 nl_msg_put_genlmsghdr(&request, sock, 0, brc_family,
126 NLM_F_REQUEST, BRC_GENL_C_QUERY_MC, 1);
127 retval = nl_sock_transact(sock, &request, &reply);
128 ofpbuf_uninit(&request);
130 nl_sock_destroy(sock);
133 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
134 brc_multicast_policy, attrs,
135 ARRAY_SIZE(brc_multicast_policy))) {
136 nl_sock_destroy(sock);
137 ofpbuf_delete(reply);
140 *multicast_group = nl_attr_get_u32(attrs[BRC_GENL_A_MC_GROUP]);
141 nl_sock_destroy(sock);
142 ofpbuf_delete(reply);
147 /* Opens a socket for brcompat notifications. Returns 0 if successful,
148 * otherwise a positive errno value. */
150 brc_open(struct nl_sock **sock)
152 int multicast_group = 0;
155 retval = nl_lookup_genl_family(BRC_GENL_FAMILY_NAME, &brc_family);
160 retval = lookup_brc_multicast_group(&multicast_group);
165 retval = nl_sock_create(NETLINK_GENERIC, multicast_group, 0, 0, sock);
173 static const struct nl_policy brc_dp_policy[] = {
174 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
178 bridge_exists(const char *name)
180 return cfg_has_section("bridge.%s", name);
184 execute_appctl_command(const char *unixctl_command, char **output)
186 char *stdout_log, *stderr_log;
192 argv[2] = xasprintf(appctl_command, unixctl_command);
195 /* Run process and log status. */
196 error = process_run_capture(argv, &stdout_log, &stderr_log, &status);
198 VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
199 unixctl_command, strerror(error));
201 char *msg = process_status_msg(status);
202 VLOG_ERR("ovs-appctl exited with error (%s)", msg);
207 /* Deal with stdout_log. */
209 *output = stdout_log;
214 /* Deal with stderr_log */
215 if (stderr_log && *stderr_log) {
216 VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
226 rewrite_and_reload_config(void)
228 if (cfg_is_dirty()) {
229 int error1 = cfg_write();
230 int error2 = cfg_read();
231 long long int reload_start = time_msec();
232 int error3 = execute_appctl_command("vswitchd/reload", NULL);
233 long long int elapsed = time_msec() - reload_start;
234 COVERAGE_INC(brcompatd_reload);
236 VLOG_INFO("reload command executed in %lld ms", elapsed);
238 return error1 ? error1 : error2 ? error2 : error3;
244 do_get_bridge_parts(const char *bridge, struct svec *parts, int vlan,
245 bool break_down_bonds)
251 cfg_get_all_keys(&ports, "bridge.%s.port", bridge);
252 for (i = 0; i < ports.n; i++) {
253 const char *port_name = ports.names[i];
255 int port_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
259 if (vlan != port_vlan) {
263 if (break_down_bonds && cfg_has_section("bonding.%s", port_name)) {
266 cfg_get_all_keys(&slaves, "bonding.%s.slave", port_name);
267 svec_append(parts, &slaves);
268 svec_destroy(&slaves);
270 svec_add(parts, port_name);
273 svec_destroy(&ports);
276 /* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
277 * down into their constituent parts.
279 * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
280 * then only interfaces for trunk ports or ports with implicit VLAN 0 are
281 * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
284 get_bridge_ifaces(const char *bridge, struct svec *ifaces, int vlan)
286 do_get_bridge_parts(bridge, ifaces, vlan, true);
289 /* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
290 * the bond name, not broken down into their constituent interfaces.
292 * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
293 * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
294 * only port with implicit VLAN 'vlan' are reported. */
296 get_bridge_ports(const char *bridge, struct svec *ports, int vlan)
298 do_get_bridge_parts(bridge, ports, vlan, false);
301 /* Go through the configuration file and remove any ports that no longer
302 * exist associated with a bridge. */
308 struct svec bridges, delete;
310 if (cfg_lock(NULL, 0)) {
311 /* Couldn't lock config file. */
317 cfg_get_subsections(&bridges, "bridge");
318 for (i=0; i<bridges.n; i++) {
319 const char *br_name = bridges.names[i];
322 /* Check that each bridge interface exists. */
324 get_bridge_ifaces(br_name, &ifaces, -1);
325 for (j = 0; j < ifaces.n; j++) {
326 const char *iface_name = ifaces.names[j];
327 enum netdev_flags flags;
329 /* The local port and internal ports are created and destroyed by
330 * ovs-vswitchd itself, so don't bother checking for them at all.
331 * In practice, they might not exist if ovs-vswitchd hasn't
332 * finished reloading since the configuration file was updated. */
333 if (!strcmp(iface_name, br_name)
334 || cfg_get_bool(0, "iface.%s.internal", iface_name)) {
338 error = netdev_nodev_get_flags(iface_name, &flags);
339 if (error == ENODEV) {
340 VLOG_INFO_RL(&rl, "removing dead interface %s from %s",
341 iface_name, br_name);
342 svec_add(&delete, iface_name);
344 VLOG_INFO_RL(&rl, "unknown error %d on interface %s from %s",
345 error, iface_name, br_name);
348 svec_destroy(&ifaces);
350 svec_destroy(&bridges);
355 for (i = 0; i < delete.n; i++) {
356 cfg_del_match("bridge.*.port=%s", delete.names[i]);
357 cfg_del_match("bonding.*.slave=%s", delete.names[i]);
359 rewrite_and_reload_config();
364 svec_destroy(&delete);
367 /* Checks whether a network device named 'name' exists and returns true if so,
370 * XXX it is possible that this doesn't entirely accomplish what we want in
371 * context, since ovs-vswitchd.conf may cause vswitchd to create or destroy
372 * network devices based on iface.*.internal settings.
374 * XXX may want to move this to lib/netdev.
376 * XXX why not just use netdev_nodev_get_flags() or similar function? */
378 netdev_exists(const char *name)
384 filename = xasprintf("/sys/class/net/%s", name);
385 error = stat(filename, &s);
391 add_bridge(const char *br_name)
393 if (bridge_exists(br_name)) {
394 VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
396 } else if (netdev_exists(br_name)) {
397 if (cfg_get_bool(0, "iface.%s.fake-bridge", br_name)) {
398 VLOG_WARN("addbr %s: %s exists as a fake bridge",
402 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
403 "device named %s already exists",
404 br_name, br_name, br_name);
409 cfg_add_entry("bridge.%s.port=%s", br_name, br_name);
410 VLOG_INFO("addbr %s: success", br_name);
416 del_bridge(const char *br_name)
418 if (!bridge_exists(br_name)) {
419 VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
423 cfg_del_section("bridge.%s", br_name);
424 VLOG_INFO("delbr %s: success", br_name);
430 parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
431 const char **port_name, uint64_t *count, uint64_t *skip)
433 static const struct nl_policy policy[] = {
434 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
435 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
436 [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
437 [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
439 struct nlattr *attrs[ARRAY_SIZE(policy)];
441 if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
442 attrs, ARRAY_SIZE(policy))
443 || (br_name && !attrs[BRC_GENL_A_DP_NAME])
444 || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
445 || (count && !attrs[BRC_GENL_A_FDB_COUNT])
446 || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
450 *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
452 *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
455 *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
458 *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
461 *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
466 /* Composes and returns a reply to a request made by the datapath with Netlink
467 * sequence number 'seq' and error code 'error'. The caller may add additional
468 * attributes to the message, then it may send it with send_reply(). */
469 static struct ofpbuf *
470 compose_reply(uint32_t seq, int error)
472 struct ofpbuf *reply = ofpbuf_new(4096);
473 nl_msg_put_genlmsghdr(reply, brc_sock, 32, brc_family, NLM_F_REQUEST,
474 BRC_GENL_C_DP_RESULT, 1);
475 ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
476 nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
480 /* Sends 'reply' to the datapath and frees it. */
482 send_reply(struct ofpbuf *reply)
484 int retval = nl_sock_send(brc_sock, reply, false);
486 VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
489 ofpbuf_delete(reply);
492 /* Composes and sends a reply to a request made by the datapath with Netlink
493 * sequence number 'seq' and error code 'error'. */
495 send_simple_reply(uint32_t seq, int error)
497 send_reply(compose_reply(seq, error));
501 handle_bridge_cmd(struct ofpbuf *buffer, bool add)
507 error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
509 error = add ? add_bridge(br_name) : del_bridge(br_name);
511 error = rewrite_and_reload_config();
513 send_simple_reply(seq, error);
518 static const struct nl_policy brc_port_policy[] = {
519 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
520 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
524 del_port(const char *br_name, const char *port_name)
526 cfg_del_entry("bridge.%s.port=%s", br_name, port_name);
527 cfg_del_match("bonding.*.slave=%s", port_name);
528 cfg_del_match("vlan.%s.[!0-9]*", port_name);
532 handle_port_cmd(struct ofpbuf *buffer, bool add)
534 const char *cmd_name = add ? "add-if" : "del-if";
535 const char *br_name, *port_name;
539 error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
541 if (!bridge_exists(br_name)) {
542 VLOG_WARN("%s %s %s: no bridge named %s",
543 cmd_name, br_name, port_name, br_name);
545 } else if (!netdev_exists(port_name)) {
546 VLOG_WARN("%s %s %s: no network device named %s",
547 cmd_name, br_name, port_name, port_name);
551 cfg_add_entry("bridge.%s.port=%s", br_name, port_name);
553 del_port(br_name, port_name);
555 VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
556 error = rewrite_and_reload_config();
558 send_simple_reply(seq, error);
564 /* Returns the name of the bridge that contains a port named 'port_name', as a
565 * malloc'd string that the caller must free, or a null pointer if no bridge
566 * contains a port named 'port_name'. */
568 get_bridge_containing_port(const char *port_name)
571 const char *start, *end;
574 cfg_get_matches(&matches, "bridge.*.port=%s", port_name);
579 start = matches.names[0] + strlen("bridge.");
580 end = strstr(start, ".port=");
582 return xmemdup0(start, end - start);
586 linux_bridge_to_ovs_bridge(const char *linux_bridge,
587 char **ovs_bridge, int *br_vlan)
589 if (bridge_exists(linux_bridge)) {
590 /* Bridge name is the same. We are interested in VLAN 0. */
591 *ovs_bridge = xstrdup(linux_bridge);
595 /* No such Open vSwitch bridge 'linux_bridge', but there might be an
596 * internal port named 'linux_bridge' on some other bridge
597 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
598 * port 'linux_bridge' on the bridge named 'ovs_bridge'. */
599 const char *port_name = linux_bridge;
601 *ovs_bridge = get_bridge_containing_port(port_name);
602 *br_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
603 if (*ovs_bridge && *br_vlan >= 0) {
613 handle_fdb_query_cmd(struct ofpbuf *buffer)
615 /* This structure is copied directly from the Linux 2.6.30 header files.
616 * It would be more straightforward to #include <linux/if_bridge.h>, but
617 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
618 * with old header files won't have it. */
623 __u32 ageing_timer_value;
632 struct mac *local_macs;
636 /* Impedance matching between the vswitchd and Linux kernel notions of what
637 * a bridge is. The kernel only handles a single VLAN per bridge, but
638 * vswitchd can deal with all the VLANs on a single bridge. We have to
639 * pretend that the former is the case even though the latter is the
641 const char *linux_bridge; /* Name used by brctl. */
642 char *ovs_bridge; /* Name used by ovs-vswitchd. */
643 int br_vlan; /* VLAN tag. */
646 struct ofpbuf query_data;
647 struct ofpbuf *reply;
648 char *unixctl_command;
649 uint64_t count, skip;
655 /* Parse the command received from brcompat_mod. */
656 error = parse_command(buffer, &seq, &linux_bridge, NULL, &count, &skip);
661 /* Figure out vswitchd bridge and VLAN. */
663 error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
665 send_simple_reply(seq, error);
669 /* Fetch the forwarding database using ovs-appctl. */
670 unixctl_command = xasprintf("fdb/show %s", ovs_bridge);
671 error = execute_appctl_command(unixctl_command, &output);
672 free(unixctl_command);
675 send_simple_reply(seq, error);
679 /* Fetch the MAC address for each interface on the bridge, so that we can
680 * fill in the is_local field in the response. */
682 get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
683 local_macs = xmalloc(ifaces.n * sizeof *local_macs);
685 for (i = 0; i < ifaces.n; i++) {
686 const char *iface_name = ifaces.names[i];
687 struct mac *mac = &local_macs[n_local_macs];
688 if (!netdev_nodev_get_etheraddr(iface_name, mac->addr)) {
692 svec_destroy(&ifaces);
694 /* Parse the response from ovs-appctl and convert it to binary format to
695 * pass back to the kernel. */
696 ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
698 strtok_r(output, "\n", &save_ptr); /* Skip header line. */
700 struct __fdb_entry *entry;
702 uint8_t mac[ETH_ADDR_LEN];
706 line = strtok_r(NULL, "\n", &save_ptr);
711 if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
712 &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
713 != 2 + ETH_ADDR_SCAN_COUNT + 1) {
714 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
715 VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
719 if (vlan != br_vlan) {
728 /* Is this the MAC address of an interface on the bridge? */
730 for (i = 0; i < n_local_macs; i++) {
731 if (eth_addr_equals(local_macs[i].addr, mac)) {
737 entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
738 memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
739 entry->port_no = port & 0xff;
740 entry->is_local = is_local;
741 entry->ageing_timer_value = age * HZ;
742 entry->port_hi = (port & 0xff00) >> 8;
749 /* Compose and send reply to datapath. */
750 reply = compose_reply(seq, 0);
751 nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
752 query_data.data, query_data.size);
756 ofpbuf_uninit(&query_data);
763 send_ifindex_reply(uint32_t seq, struct svec *ifaces)
765 struct ofpbuf *reply;
771 /* Make sure that any given interface only occurs once. This shouldn't
772 * happen, but who knows what people put into their configuration files. */
773 svec_sort_unique(ifaces);
775 /* Convert 'ifaces' into ifindexes. */
777 indices = xmalloc(ifaces->n * sizeof *indices);
778 SVEC_FOR_EACH (i, iface, ifaces) {
779 int ifindex = if_nametoindex(iface);
781 indices[n_indices++] = ifindex;
785 /* Compose and send reply. */
786 reply = compose_reply(seq, 0);
787 nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
788 indices, n_indices * sizeof *indices);
796 handle_get_bridges_cmd(struct ofpbuf *buffer)
806 /* Parse Netlink command.
808 * The command doesn't actually have any arguments, but we need the
809 * sequence number to send the reply. */
810 error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
815 /* Get all the real bridges and all the fake ones. */
818 cfg_get_subsections(&bridges, "bridge");
819 SVEC_FOR_EACH (i, br_name, &bridges) {
820 const char *iface_name;
825 get_bridge_ifaces(br_name, &ifaces, -1);
826 SVEC_FOR_EACH (j, iface_name, &ifaces) {
827 if (cfg_get_bool(0, "iface.%s.fake-bridge", iface_name)) {
828 svec_add(&bridges, iface_name);
831 svec_destroy(&ifaces);
834 send_ifindex_reply(seq, &bridges);
835 svec_destroy(&bridges);
841 handle_get_ports_cmd(struct ofpbuf *buffer)
845 const char *linux_bridge;
853 /* Parse Netlink command. */
854 error = parse_command(buffer, &seq, &linux_bridge, NULL, NULL, NULL);
860 error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
862 send_simple_reply(seq, error);
867 get_bridge_ports(ovs_bridge, &ports, br_vlan);
869 svec_del(&ports, linux_bridge);
870 send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
871 svec_destroy(&ports);
879 brc_recv_update(void)
882 struct ofpbuf *buffer;
883 struct genlmsghdr *genlmsghdr;
888 ofpbuf_delete(buffer);
889 retval = nl_sock_recv(brc_sock, &buffer, false);
890 } while (retval == ENOBUFS
892 && (nl_msg_nlmsgerr(buffer, NULL)
893 || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE)));
895 if (retval != EAGAIN) {
896 VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
901 genlmsghdr = nl_msg_genlmsghdr(buffer);
903 VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
907 if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
908 VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
909 nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
913 if (cfg_lock(NULL, lock_timeout)) {
914 /* Couldn't lock config file. */
919 switch (genlmsghdr->cmd) {
920 case BRC_GENL_C_DP_ADD:
921 retval = handle_bridge_cmd(buffer, true);
924 case BRC_GENL_C_DP_DEL:
925 retval = handle_bridge_cmd(buffer, false);
928 case BRC_GENL_C_PORT_ADD:
929 retval = handle_port_cmd(buffer, true);
932 case BRC_GENL_C_PORT_DEL:
933 retval = handle_port_cmd(buffer, false);
936 case BRC_GENL_C_FDB_QUERY:
937 retval = handle_fdb_query_cmd(buffer);
940 case BRC_GENL_C_GET_BRIDGES:
941 retval = handle_get_bridges_cmd(buffer);
944 case BRC_GENL_C_GET_PORTS:
945 retval = handle_get_ports_cmd(buffer);
955 ofpbuf_delete(buffer);
959 /* Check for interface configuration changes announced through RTNL. */
961 rtnl_recv_update(void)
965 int error = nl_sock_recv(rtnl_sock, &buf, false);
966 if (error == EAGAIN) {
968 } else if (error == ENOBUFS) {
969 VLOG_WARN_RL(&rl, "network monitor socket overflowed");
971 VLOG_WARN_RL(&rl, "error on network monitor socket: %s",
974 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
975 struct nlmsghdr *nlh;
976 struct ifinfomsg *iim;
978 nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
979 iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
981 VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
986 if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
988 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
989 VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
993 if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
994 const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
995 char br_name[IFNAMSIZ];
996 uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
997 enum netdev_flags flags;
999 if (!if_indextoname(br_idx, br_name)) {
1004 if (cfg_lock(NULL, lock_timeout)) {
1005 /* Couldn't lock config file. */
1006 /* xxx this should try again and print error msg. */
1011 if (netdev_nodev_get_flags(port_name, &flags) == ENODEV) {
1012 /* Network device is really gone. */
1015 VLOG_INFO("network device %s destroyed, "
1016 "removing from bridge %s", port_name, br_name);
1019 cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
1021 if (svec_contains(&ports, port_name)) {
1022 del_port(br_name, port_name);
1023 rewrite_and_reload_config();
1025 svec_destroy(&ports);
1027 /* A network device by that name exists even though the kernel
1028 * told us it had disappeared. Probably, what happened was
1031 * 1. Device destroyed.
1032 * 2. Notification sent to us.
1033 * 3. New device created with same name as old one.
1034 * 4. ovs-brcompatd notified, removes device from bridge.
1036 * There's no a priori reason that in this situation that the
1037 * new device with the same name should remain in the bridge;
1038 * on the contrary, that would be unexpected. *But* there is
1039 * one important situation where, if we do this, bad things
1040 * happen. This is the case of XenServer Tools version 5.0.0,
1041 * which on boot of a Windows VM cause something like this to
1042 * happen on the Xen host:
1044 * i. Create tap1.0 and vif1.0.
1045 * ii. Delete tap1.0.
1046 * iii. Delete vif1.0.
1047 * iv. Re-create vif1.0.
1049 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1050 * neither does a VM without Tools installed at all.@.)
1052 * Steps iii and iv happen within a few seconds of each other.
1053 * Step iv causes /etc/xensource/scripts/vif to run, which in
1054 * turn calls ovs-cfg-mod to add the new device to the bridge.
1055 * If step iv happens after step 4 (in our first list of
1056 * steps), then all is well, but if it happens between 3 and 4
1057 * (which can easily happen if ovs-brcompatd has to wait to
1058 * lock the configuration file), then we will remove the new
1059 * incarnation from the bridge instead of the old one!
1061 * So, to avoid this problem, we do nothing here. This is
1062 * strictly incorrect except for this one particular case, and
1063 * perhaps that will bite us someday. If that happens, then we
1064 * will have to somehow track network devices by ifindex, since
1065 * a new device will have a new ifindex even if it has the same
1066 * name as an old device.
1068 VLOG_INFO("kernel reported network device %s removed but "
1069 "a device by that name exists (XS Tools 5.0.0?)",
1079 main(int argc, char *argv[])
1081 struct unixctl_server *unixctl;
1084 set_program_name(argv[0]);
1085 register_fault_handlers();
1088 parse_options(argc, argv);
1089 signal(SIGPIPE, SIG_IGN);
1092 die_if_already_running();
1095 retval = unixctl_server_create(NULL, &unixctl);
1097 ovs_fatal(retval, "could not listen for vlog connections");
1100 if (brc_open(&brc_sock)) {
1101 ovs_fatal(0, "could not open brcompat socket. Check "
1102 "\"brcompat\" kernel module.");
1105 if (prune_timeout) {
1106 if (nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &rtnl_sock)) {
1107 ovs_fatal(0, "could not create rtnetlink socket");
1111 retval = cfg_read();
1113 ovs_fatal(retval, "could not read config file");
1117 unixctl_server_run(unixctl);
1120 /* If 'prune_timeout' is non-zero, we actively prune from the
1121 * config file any 'bridge.<br_name>.port' entries that are no
1122 * longer valid. We use two methods:
1124 * 1) The kernel explicitly notifies us of removed ports
1125 * through the RTNL messages.
1127 * 2) We periodically check all ports associated with bridges
1128 * to see if they no longer exist.
1130 if (prune_timeout) {
1134 nl_sock_wait(rtnl_sock, POLLIN);
1135 poll_timer_wait(prune_timeout);
1138 nl_sock_wait(brc_sock, POLLIN);
1139 unixctl_server_wait(unixctl);
1147 validate_appctl_command(void)
1153 for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
1155 /* Nothing to do. */
1156 } else if (p[1] == 's') {
1159 ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
1163 ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
1168 parse_options(int argc, char *argv[])
1171 OPT_LOCK_TIMEOUT = UCHAR_MAX + 1,
1175 LEAK_CHECKER_OPTION_ENUMS
1177 static struct option long_options[] = {
1178 {"help", no_argument, 0, 'h'},
1179 {"version", no_argument, 0, 'V'},
1180 {"lock-timeout", required_argument, 0, OPT_LOCK_TIMEOUT},
1181 {"prune-timeout", required_argument, 0, OPT_PRUNE_TIMEOUT},
1182 {"appctl-command", required_argument, 0, OPT_APPCTL_COMMAND},
1183 DAEMON_LONG_OPTIONS,
1185 LEAK_CHECKER_LONG_OPTIONS,
1188 char *short_options = long_options_to_short_options(long_options);
1191 appctl_command = xasprintf("%s/ovs-appctl -t "
1192 "%s/ovs-vswitchd.`cat %s/ovs-vswitchd.pid`.ctl "
1194 ovs_bindir, ovs_rundir, ovs_rundir);
1198 c = getopt_long(argc, argv, short_options, long_options, NULL);
1209 OVS_PRINT_VERSION(0, 0);
1212 case OPT_LOCK_TIMEOUT:
1213 lock_timeout = atoi(optarg);
1216 case OPT_PRUNE_TIMEOUT:
1217 prune_timeout = atoi(optarg) * 1000;
1220 case OPT_APPCTL_COMMAND:
1221 appctl_command = optarg;
1224 VLOG_OPTION_HANDLERS
1225 DAEMON_OPTION_HANDLERS
1226 LEAK_CHECKER_OPTION_HANDLERS
1235 free(short_options);
1237 validate_appctl_command();
1243 ovs_fatal(0, "exactly one non-option argument required; "
1244 "use --help for usage");
1248 config_file = argv[0];
1249 error = cfg_set_file(config_file);
1251 ovs_fatal(error, "failed to add configuration file \"%s\"",
1259 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1260 "usage: %s [OPTIONS] CONFIG\n"
1261 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1262 program_name, program_name);
1263 printf("\nConfiguration options:\n"
1264 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1265 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1266 " --lock-timeout=MSECS wait at most MSECS for CONFIG to unlock\n"
1270 printf("\nOther options:\n"
1271 " -h, --help display this help message\n"
1272 " -V, --version display version information\n");
1273 leak_checker_usage();
1274 printf("\nThe default appctl command is:\n%s\n", appctl_command);