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"
41 #include "dynamic-string.h"
42 #include "fatal-signal.h"
44 #include "leak-checker.h"
48 #include "openvswitch/brcompat-netlink.h"
50 #include "poll-loop.h"
59 #define THIS_MODULE VLM_brcompatd
62 /* xxx Just hangs if datapath is rmmod/insmod. Learn to reconnect? */
64 /* Actions to modify bridge compatibility configuration. */
72 static void parse_options(int argc, char *argv[]);
73 static void usage(void) NO_RETURN;
75 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 60);
77 /* Maximum number of milliseconds to wait for the config file to be
78 * unlocked. If set to zero, no waiting will occur. */
79 static int lock_timeout = 500;
81 /* Maximum number of milliseconds to wait before pruning port entries that
82 * no longer exist. If set to zero, ports are never pruned. */
83 static int prune_timeout = 5000;
85 /* Config file shared with ovs-vswitchd (usually ovs-vswitchd.conf). */
86 static char *config_file;
88 /* Shell command to execute (via popen()) to send a control command to the
89 * running ovs-vswitchd process. The string must contain one instance of %s,
90 * which is replaced by the control command. */
91 static char *appctl_command;
93 /* Netlink socket to listen for interface changes. */
94 static struct nl_sock *rtnl_sock;
96 /* Netlink socket to bridge compatibility kernel module. */
97 static struct nl_sock *brc_sock;
99 /* The Generic Netlink family number used for bridge compatibility. */
100 static int brc_family;
102 static const struct nl_policy brc_multicast_policy[] = {
103 [BRC_GENL_A_MC_GROUP] = {.type = NL_A_U32 }
106 static const struct nl_policy rtnlgrp_link_policy[] = {
107 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
108 [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
112 lookup_brc_multicast_group(int *multicast_group)
114 struct nl_sock *sock;
115 struct ofpbuf request, *reply;
116 struct nlattr *attrs[ARRAY_SIZE(brc_multicast_policy)];
119 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
123 ofpbuf_init(&request, 0);
124 nl_msg_put_genlmsghdr(&request, sock, 0, brc_family,
125 NLM_F_REQUEST, BRC_GENL_C_QUERY_MC, 1);
126 retval = nl_sock_transact(sock, &request, &reply);
127 ofpbuf_uninit(&request);
129 nl_sock_destroy(sock);
132 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
133 brc_multicast_policy, attrs,
134 ARRAY_SIZE(brc_multicast_policy))) {
135 nl_sock_destroy(sock);
136 ofpbuf_delete(reply);
139 *multicast_group = nl_attr_get_u32(attrs[BRC_GENL_A_MC_GROUP]);
140 nl_sock_destroy(sock);
141 ofpbuf_delete(reply);
146 /* Opens a socket for brcompat notifications. Returns 0 if successful,
147 * otherwise a positive errno value. */
149 brc_open(struct nl_sock **sock)
151 int multicast_group = 0;
154 retval = nl_lookup_genl_family(BRC_GENL_FAMILY_NAME, &brc_family);
159 retval = lookup_brc_multicast_group(&multicast_group);
164 retval = nl_sock_create(NETLINK_GENERIC, multicast_group, 0, 0, sock);
172 static const struct nl_policy brc_dp_policy[] = {
173 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
177 bridge_exists(const char *name)
179 return cfg_has_section("bridge.%s", name);
183 execute_appctl_command(const char *unixctl_command, char **output)
185 char *stdout_log, *stderr_log;
191 argv[2] = xasprintf(appctl_command, unixctl_command);
194 /* Run process and log status. */
195 error = process_run_capture(argv, &stdout_log, &stderr_log, &status);
197 VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
198 unixctl_command, strerror(error));
200 char *msg = process_status_msg(status);
201 VLOG_ERR("ovs-appctl exited with error (%s)", msg);
206 /* Deal with stdout_log. */
208 *output = stdout_log;
213 /* Deal with stderr_log */
214 if (stderr_log && *stderr_log) {
215 VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
225 rewrite_and_reload_config(void)
227 if (cfg_is_dirty()) {
228 int error1 = cfg_write();
229 int error2 = cfg_read();
230 long long int reload_start = time_msec();
231 int error3 = execute_appctl_command("vswitchd/reload", NULL);
232 long long int elapsed = time_msec() - reload_start;
233 COVERAGE_INC(brcompatd_reload);
235 VLOG_INFO("reload command executed in %lld ms", elapsed);
237 return error1 ? error1 : error2 ? error2 : error3;
243 do_get_bridge_parts(const char *bridge, struct svec *parts, int vlan,
244 bool break_down_bonds)
250 cfg_get_all_keys(&ports, "bridge.%s.port", bridge);
251 for (i = 0; i < ports.n; i++) {
252 const char *port_name = ports.names[i];
254 int port_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
258 if (vlan != port_vlan) {
262 if (break_down_bonds && cfg_has_section("bonding.%s", port_name)) {
265 cfg_get_all_keys(&slaves, "bonding.%s.slave", port_name);
266 svec_append(parts, &slaves);
267 svec_destroy(&slaves);
269 svec_add(parts, port_name);
272 svec_destroy(&ports);
275 /* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
276 * down into their constituent parts.
278 * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
279 * then only interfaces for trunk ports or ports with implicit VLAN 0 are
280 * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
283 get_bridge_ifaces(const char *bridge, struct svec *ifaces, int vlan)
285 do_get_bridge_parts(bridge, ifaces, vlan, true);
288 /* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
289 * the bond name, not broken down into their constituent interfaces.
291 * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
292 * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
293 * only port with implicit VLAN 'vlan' are reported. */
295 get_bridge_ports(const char *bridge, struct svec *ports, int vlan)
297 do_get_bridge_parts(bridge, ports, vlan, false);
300 /* Go through the configuration file and remove any ports that no longer
301 * exist associated with a bridge. */
306 struct svec bridges, delete;
308 if (cfg_lock(NULL, 0)) {
309 /* Couldn't lock config file. */
315 cfg_get_subsections(&bridges, "bridge");
316 for (i=0; i<bridges.n; i++) {
317 const char *br_name = bridges.names[i];
320 /* Check that each bridge interface exists. */
322 get_bridge_ifaces(br_name, &ifaces, -1);
323 for (j = 0; j < ifaces.n; j++) {
324 const char *iface_name = ifaces.names[j];
326 /* The local port and internal ports are created and destroyed by
327 * ovs-vswitchd itself, so don't bother checking for them at all.
328 * In practice, they might not exist if ovs-vswitchd hasn't
329 * finished reloading since the configuration file was updated. */
330 if (!strcmp(iface_name, br_name)
331 || cfg_get_bool(0, "iface.%s.internal", iface_name)) {
335 if (!netdev_exists(iface_name)) {
336 VLOG_INFO_RL(&rl, "removing dead interface %s from %s",
337 iface_name, br_name);
338 svec_add(&delete, iface_name);
341 svec_destroy(&ifaces);
343 svec_destroy(&bridges);
348 for (i = 0; i < delete.n; i++) {
349 cfg_del_match("bridge.*.port=%s", delete.names[i]);
350 cfg_del_match("bonding.*.slave=%s", delete.names[i]);
352 rewrite_and_reload_config();
357 svec_destroy(&delete);
361 add_bridge(const char *br_name)
363 if (bridge_exists(br_name)) {
364 VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
366 } else if (netdev_exists(br_name)) {
367 if (cfg_get_bool(0, "iface.%s.fake-bridge", br_name)) {
368 VLOG_WARN("addbr %s: %s exists as a fake bridge",
372 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
373 "device named %s already exists",
374 br_name, br_name, br_name);
379 cfg_add_entry("bridge.%s.port=%s", br_name, br_name);
380 VLOG_INFO("addbr %s: success", br_name);
386 del_bridge(const char *br_name)
388 if (!bridge_exists(br_name)) {
389 VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
393 cfg_del_section("bridge.%s", br_name);
394 VLOG_INFO("delbr %s: success", br_name);
400 parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
401 const char **port_name, uint64_t *count, uint64_t *skip)
403 static const struct nl_policy policy[] = {
404 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
405 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
406 [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
407 [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
409 struct nlattr *attrs[ARRAY_SIZE(policy)];
411 if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
412 attrs, ARRAY_SIZE(policy))
413 || (br_name && !attrs[BRC_GENL_A_DP_NAME])
414 || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
415 || (count && !attrs[BRC_GENL_A_FDB_COUNT])
416 || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
420 *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
422 *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
425 *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
428 *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
431 *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
436 /* Composes and returns a reply to a request made by the datapath with Netlink
437 * sequence number 'seq' and error code 'error'. The caller may add additional
438 * attributes to the message, then it may send it with send_reply(). */
439 static struct ofpbuf *
440 compose_reply(uint32_t seq, int error)
442 struct ofpbuf *reply = ofpbuf_new(4096);
443 nl_msg_put_genlmsghdr(reply, brc_sock, 32, brc_family, NLM_F_REQUEST,
444 BRC_GENL_C_DP_RESULT, 1);
445 ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
446 nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
450 /* Sends 'reply' to the datapath and frees it. */
452 send_reply(struct ofpbuf *reply)
454 int retval = nl_sock_send(brc_sock, reply, false);
456 VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
459 ofpbuf_delete(reply);
462 /* Composes and sends a reply to a request made by the datapath with Netlink
463 * sequence number 'seq' and error code 'error'. */
465 send_simple_reply(uint32_t seq, int error)
467 send_reply(compose_reply(seq, error));
471 handle_bridge_cmd(struct ofpbuf *buffer, bool add)
477 error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
479 error = add ? add_bridge(br_name) : del_bridge(br_name);
481 error = rewrite_and_reload_config();
483 send_simple_reply(seq, error);
488 static const struct nl_policy brc_port_policy[] = {
489 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
490 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
494 del_port(const char *br_name, const char *port_name)
496 cfg_del_entry("bridge.%s.port=%s", br_name, port_name);
497 cfg_del_match("bonding.*.slave=%s", port_name);
498 cfg_del_match("vlan.%s.*", port_name);
502 handle_port_cmd(struct ofpbuf *buffer, bool add)
504 const char *cmd_name = add ? "add-if" : "del-if";
505 const char *br_name, *port_name;
509 error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
511 if (!bridge_exists(br_name)) {
512 VLOG_WARN("%s %s %s: no bridge named %s",
513 cmd_name, br_name, port_name, br_name);
515 } else if (!netdev_exists(port_name)) {
516 VLOG_WARN("%s %s %s: no network device named %s",
517 cmd_name, br_name, port_name, port_name);
521 cfg_add_entry("bridge.%s.port=%s", br_name, port_name);
523 del_port(br_name, port_name);
525 VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
526 error = rewrite_and_reload_config();
528 send_simple_reply(seq, error);
534 /* Returns the name of the bridge that contains a port named 'port_name', as a
535 * malloc'd string that the caller must free, or a null pointer if no bridge
536 * contains a port named 'port_name'. */
538 get_bridge_containing_port(const char *port_name)
541 const char *start, *end;
544 cfg_get_matches(&matches, "bridge.*.port=%s", port_name);
549 start = matches.names[0] + strlen("bridge.");
550 end = strstr(start, ".port=");
552 return xmemdup0(start, end - start);
556 linux_bridge_to_ovs_bridge(const char *linux_bridge,
557 char **ovs_bridge, int *br_vlan)
559 if (bridge_exists(linux_bridge)) {
560 /* Bridge name is the same. We are interested in VLAN 0. */
561 *ovs_bridge = xstrdup(linux_bridge);
565 /* No such Open vSwitch bridge 'linux_bridge', but there might be an
566 * internal port named 'linux_bridge' on some other bridge
567 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
568 * port 'linux_bridge' on the bridge named 'ovs_bridge'. */
569 const char *port_name = linux_bridge;
571 *ovs_bridge = get_bridge_containing_port(port_name);
572 *br_vlan = cfg_get_vlan(0, "vlan.%s.tag", port_name);
573 if (*ovs_bridge && *br_vlan >= 0) {
583 handle_fdb_query_cmd(struct ofpbuf *buffer)
585 /* This structure is copied directly from the Linux 2.6.30 header files.
586 * It would be more straightforward to #include <linux/if_bridge.h>, but
587 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
588 * with old header files won't have it. */
593 __u32 ageing_timer_value;
602 struct mac *local_macs;
606 /* Impedance matching between the vswitchd and Linux kernel notions of what
607 * a bridge is. The kernel only handles a single VLAN per bridge, but
608 * vswitchd can deal with all the VLANs on a single bridge. We have to
609 * pretend that the former is the case even though the latter is the
611 const char *linux_bridge; /* Name used by brctl. */
612 char *ovs_bridge; /* Name used by ovs-vswitchd. */
613 int br_vlan; /* VLAN tag. */
616 struct ofpbuf query_data;
617 struct ofpbuf *reply;
618 char *unixctl_command;
619 uint64_t count, skip;
625 /* Parse the command received from brcompat_mod. */
626 error = parse_command(buffer, &seq, &linux_bridge, NULL, &count, &skip);
631 /* Figure out vswitchd bridge and VLAN. */
633 error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
635 send_simple_reply(seq, error);
639 /* Fetch the forwarding database using ovs-appctl. */
640 unixctl_command = xasprintf("fdb/show %s", ovs_bridge);
641 error = execute_appctl_command(unixctl_command, &output);
642 free(unixctl_command);
645 send_simple_reply(seq, error);
649 /* Fetch the MAC address for each interface on the bridge, so that we can
650 * fill in the is_local field in the response. */
652 get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
653 local_macs = xmalloc(ifaces.n * sizeof *local_macs);
655 for (i = 0; i < ifaces.n; i++) {
656 const char *iface_name = ifaces.names[i];
657 struct mac *mac = &local_macs[n_local_macs];
658 struct netdev *netdev;
660 error = netdev_open(iface_name, NETDEV_ETH_TYPE_NONE, &netdev);
662 if (!netdev_get_etheraddr(netdev, mac->addr)) {
665 netdev_close(netdev);
668 svec_destroy(&ifaces);
670 /* Parse the response from ovs-appctl and convert it to binary format to
671 * pass back to the kernel. */
672 ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
674 strtok_r(output, "\n", &save_ptr); /* Skip header line. */
676 struct __fdb_entry *entry;
678 uint8_t mac[ETH_ADDR_LEN];
682 line = strtok_r(NULL, "\n", &save_ptr);
687 if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
688 &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
689 != 2 + ETH_ADDR_SCAN_COUNT + 1) {
690 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
691 VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
695 if (vlan != br_vlan) {
704 /* Is this the MAC address of an interface on the bridge? */
706 for (i = 0; i < n_local_macs; i++) {
707 if (eth_addr_equals(local_macs[i].addr, mac)) {
713 entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
714 memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
715 entry->port_no = port & 0xff;
716 entry->is_local = is_local;
717 entry->ageing_timer_value = age * HZ;
718 entry->port_hi = (port & 0xff00) >> 8;
725 /* Compose and send reply to datapath. */
726 reply = compose_reply(seq, 0);
727 nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
728 query_data.data, query_data.size);
732 ofpbuf_uninit(&query_data);
739 send_ifindex_reply(uint32_t seq, struct svec *ifaces)
741 struct ofpbuf *reply;
747 /* Make sure that any given interface only occurs once. This shouldn't
748 * happen, but who knows what people put into their configuration files. */
749 svec_sort_unique(ifaces);
751 /* Convert 'ifaces' into ifindexes. */
753 indices = xmalloc(ifaces->n * sizeof *indices);
754 SVEC_FOR_EACH (i, iface, ifaces) {
755 int ifindex = if_nametoindex(iface);
757 indices[n_indices++] = ifindex;
761 /* Compose and send reply. */
762 reply = compose_reply(seq, 0);
763 nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
764 indices, n_indices * sizeof *indices);
772 handle_get_bridges_cmd(struct ofpbuf *buffer)
782 /* Parse Netlink command.
784 * The command doesn't actually have any arguments, but we need the
785 * sequence number to send the reply. */
786 error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
791 /* Get all the real bridges and all the fake ones. */
794 cfg_get_subsections(&bridges, "bridge");
795 SVEC_FOR_EACH (i, br_name, &bridges) {
796 const char *iface_name;
801 get_bridge_ifaces(br_name, &ifaces, -1);
802 SVEC_FOR_EACH (j, iface_name, &ifaces) {
803 if (cfg_get_bool(0, "iface.%s.fake-bridge", iface_name)) {
804 svec_add(&bridges, iface_name);
807 svec_destroy(&ifaces);
810 send_ifindex_reply(seq, &bridges);
811 svec_destroy(&bridges);
817 handle_get_ports_cmd(struct ofpbuf *buffer)
821 const char *linux_bridge;
829 /* Parse Netlink command. */
830 error = parse_command(buffer, &seq, &linux_bridge, NULL, NULL, NULL);
836 error = linux_bridge_to_ovs_bridge(linux_bridge, &ovs_bridge, &br_vlan);
838 send_simple_reply(seq, error);
843 get_bridge_ports(ovs_bridge, &ports, br_vlan);
845 svec_del(&ports, linux_bridge);
846 send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
847 svec_destroy(&ports);
855 brc_recv_update(void)
858 struct ofpbuf *buffer;
859 struct genlmsghdr *genlmsghdr;
864 ofpbuf_delete(buffer);
865 retval = nl_sock_recv(brc_sock, &buffer, false);
866 } while (retval == ENOBUFS
868 && (nl_msg_nlmsgerr(buffer, NULL)
869 || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE)));
871 if (retval != EAGAIN) {
872 VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
877 genlmsghdr = nl_msg_genlmsghdr(buffer);
879 VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
883 if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
884 VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
885 nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
889 if (cfg_lock(NULL, lock_timeout)) {
890 /* Couldn't lock config file. */
895 switch (genlmsghdr->cmd) {
896 case BRC_GENL_C_DP_ADD:
897 retval = handle_bridge_cmd(buffer, true);
900 case BRC_GENL_C_DP_DEL:
901 retval = handle_bridge_cmd(buffer, false);
904 case BRC_GENL_C_PORT_ADD:
905 retval = handle_port_cmd(buffer, true);
908 case BRC_GENL_C_PORT_DEL:
909 retval = handle_port_cmd(buffer, false);
912 case BRC_GENL_C_FDB_QUERY:
913 retval = handle_fdb_query_cmd(buffer);
916 case BRC_GENL_C_GET_BRIDGES:
917 retval = handle_get_bridges_cmd(buffer);
920 case BRC_GENL_C_GET_PORTS:
921 retval = handle_get_ports_cmd(buffer);
931 ofpbuf_delete(buffer);
935 /* Check for interface configuration changes announced through RTNL. */
937 rtnl_recv_update(void)
941 int error = nl_sock_recv(rtnl_sock, &buf, false);
942 if (error == EAGAIN) {
944 } else if (error == ENOBUFS) {
945 VLOG_WARN_RL(&rl, "network monitor socket overflowed");
947 VLOG_WARN_RL(&rl, "error on network monitor socket: %s",
950 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
951 struct nlmsghdr *nlh;
952 struct ifinfomsg *iim;
954 nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
955 iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
957 VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
962 if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
964 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
965 VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
969 if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
970 const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
971 char br_name[IFNAMSIZ];
972 uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
974 if (!if_indextoname(br_idx, br_name)) {
979 if (cfg_lock(NULL, lock_timeout)) {
980 /* Couldn't lock config file. */
981 /* xxx this should try again and print error msg. */
986 if (!netdev_exists(port_name)) {
987 /* Network device is really gone. */
990 VLOG_INFO("network device %s destroyed, "
991 "removing from bridge %s", port_name, br_name);
994 cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
996 if (svec_contains(&ports, port_name)) {
997 del_port(br_name, port_name);
998 rewrite_and_reload_config();
1000 svec_destroy(&ports);
1002 /* A network device by that name exists even though the kernel
1003 * told us it had disappeared. Probably, what happened was
1006 * 1. Device destroyed.
1007 * 2. Notification sent to us.
1008 * 3. New device created with same name as old one.
1009 * 4. ovs-brcompatd notified, removes device from bridge.
1011 * There's no a priori reason that in this situation that the
1012 * new device with the same name should remain in the bridge;
1013 * on the contrary, that would be unexpected. *But* there is
1014 * one important situation where, if we do this, bad things
1015 * happen. This is the case of XenServer Tools version 5.0.0,
1016 * which on boot of a Windows VM cause something like this to
1017 * happen on the Xen host:
1019 * i. Create tap1.0 and vif1.0.
1020 * ii. Delete tap1.0.
1021 * iii. Delete vif1.0.
1022 * iv. Re-create vif1.0.
1024 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1025 * neither does a VM without Tools installed at all.@.)
1027 * Steps iii and iv happen within a few seconds of each other.
1028 * Step iv causes /etc/xensource/scripts/vif to run, which in
1029 * turn calls ovs-cfg-mod to add the new device to the bridge.
1030 * If step iv happens after step 4 (in our first list of
1031 * steps), then all is well, but if it happens between 3 and 4
1032 * (which can easily happen if ovs-brcompatd has to wait to
1033 * lock the configuration file), then we will remove the new
1034 * incarnation from the bridge instead of the old one!
1036 * So, to avoid this problem, we do nothing here. This is
1037 * strictly incorrect except for this one particular case, and
1038 * perhaps that will bite us someday. If that happens, then we
1039 * will have to somehow track network devices by ifindex, since
1040 * a new device will have a new ifindex even if it has the same
1041 * name as an old device.
1043 VLOG_INFO("kernel reported network device %s removed but "
1044 "a device by that name exists (XS Tools 5.0.0?)",
1054 main(int argc, char *argv[])
1056 struct unixctl_server *unixctl;
1059 set_program_name(argv[0]);
1060 register_fault_handlers();
1063 parse_options(argc, argv);
1064 signal(SIGPIPE, SIG_IGN);
1067 die_if_already_running();
1070 retval = unixctl_server_create(NULL, &unixctl);
1072 ovs_fatal(retval, "could not listen for vlog connections");
1075 if (brc_open(&brc_sock)) {
1076 ovs_fatal(0, "could not open brcompat socket. Check "
1077 "\"brcompat\" kernel module.");
1080 if (prune_timeout) {
1081 if (nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &rtnl_sock)) {
1082 ovs_fatal(0, "could not create rtnetlink socket");
1086 retval = cfg_read();
1088 ovs_fatal(retval, "could not read config file");
1092 unixctl_server_run(unixctl);
1096 /* If 'prune_timeout' is non-zero, we actively prune from the
1097 * config file any 'bridge.<br_name>.port' entries that are no
1098 * longer valid. We use two methods:
1100 * 1) The kernel explicitly notifies us of removed ports
1101 * through the RTNL messages.
1103 * 2) We periodically check all ports associated with bridges
1104 * to see if they no longer exist.
1106 if (prune_timeout) {
1110 nl_sock_wait(rtnl_sock, POLLIN);
1111 poll_timer_wait(prune_timeout);
1114 nl_sock_wait(brc_sock, POLLIN);
1115 unixctl_server_wait(unixctl);
1124 validate_appctl_command(void)
1130 for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
1132 /* Nothing to do. */
1133 } else if (p[1] == 's') {
1136 ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
1140 ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
1145 parse_options(int argc, char *argv[])
1148 OPT_LOCK_TIMEOUT = UCHAR_MAX + 1,
1152 LEAK_CHECKER_OPTION_ENUMS
1154 static struct option long_options[] = {
1155 {"help", no_argument, 0, 'h'},
1156 {"version", no_argument, 0, 'V'},
1157 {"lock-timeout", required_argument, 0, OPT_LOCK_TIMEOUT},
1158 {"prune-timeout", required_argument, 0, OPT_PRUNE_TIMEOUT},
1159 {"appctl-command", required_argument, 0, OPT_APPCTL_COMMAND},
1160 DAEMON_LONG_OPTIONS,
1162 LEAK_CHECKER_LONG_OPTIONS,
1165 char *short_options = long_options_to_short_options(long_options);
1168 appctl_command = xasprintf("%s/ovs-appctl -t "
1169 "%s/ovs-vswitchd.`cat %s/ovs-vswitchd.pid`.ctl "
1171 ovs_bindir, ovs_rundir, ovs_rundir);
1175 c = getopt_long(argc, argv, short_options, long_options, NULL);
1186 OVS_PRINT_VERSION(0, 0);
1189 case OPT_LOCK_TIMEOUT:
1190 lock_timeout = atoi(optarg);
1193 case OPT_PRUNE_TIMEOUT:
1194 prune_timeout = atoi(optarg) * 1000;
1197 case OPT_APPCTL_COMMAND:
1198 appctl_command = optarg;
1201 VLOG_OPTION_HANDLERS
1202 DAEMON_OPTION_HANDLERS
1203 LEAK_CHECKER_OPTION_HANDLERS
1212 free(short_options);
1214 validate_appctl_command();
1220 ovs_fatal(0, "exactly one non-option argument required; "
1221 "use --help for usage");
1225 config_file = argv[0];
1226 error = cfg_set_file(config_file);
1228 ovs_fatal(error, "failed to add configuration file \"%s\"",
1236 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1237 "usage: %s [OPTIONS] CONFIG\n"
1238 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1239 program_name, program_name);
1240 printf("\nConfiguration options:\n"
1241 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1242 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1243 " --lock-timeout=MSECS wait at most MSECS for CONFIG to unlock\n"
1247 printf("\nOther options:\n"
1248 " -h, --help display this help message\n"
1249 " -V, --version display version information\n");
1250 leak_checker_usage();
1251 printf("\nThe default appctl command is:\n%s\n", appctl_command);