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>
36 #include "command-line.h"
40 #include "dynamic-string.h"
41 #include "fatal-signal.h"
43 #include "leak-checker.h"
47 #include "openvswitch/brcompat-netlink.h"
48 #include "ovsdb-idl.h"
50 #include "poll-loop.h"
57 #include "vswitchd/vswitch-idl.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 const char *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 /* Shell command to execute (via popen()) to send a control command to the
87 * running ovs-vswitchd process. The string must contain one instance of %s,
88 * which is replaced by the control command. */
89 static char *appctl_command;
91 /* Netlink socket to listen for interface changes. */
92 static struct nl_sock *rtnl_sock;
94 /* Netlink socket to bridge compatibility kernel module. */
95 static struct nl_sock *brc_sock;
97 /* The Generic Netlink family number used for bridge compatibility. */
98 static int brc_family;
100 static const struct nl_policy brc_multicast_policy[] = {
101 [BRC_GENL_A_MC_GROUP] = {.type = NL_A_U32 }
104 static const struct nl_policy rtnlgrp_link_policy[] = {
105 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
106 [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
110 lookup_brc_multicast_group(int *multicast_group)
112 struct nl_sock *sock;
113 struct ofpbuf request, *reply;
114 struct nlattr *attrs[ARRAY_SIZE(brc_multicast_policy)];
117 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
121 ofpbuf_init(&request, 0);
122 nl_msg_put_genlmsghdr(&request, sock, 0, brc_family,
123 NLM_F_REQUEST, BRC_GENL_C_QUERY_MC, 1);
124 retval = nl_sock_transact(sock, &request, &reply);
125 ofpbuf_uninit(&request);
127 nl_sock_destroy(sock);
130 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
131 brc_multicast_policy, attrs,
132 ARRAY_SIZE(brc_multicast_policy))) {
133 nl_sock_destroy(sock);
134 ofpbuf_delete(reply);
137 *multicast_group = nl_attr_get_u32(attrs[BRC_GENL_A_MC_GROUP]);
138 nl_sock_destroy(sock);
139 ofpbuf_delete(reply);
144 /* Opens a socket for brcompat notifications. Returns 0 if successful,
145 * otherwise a positive errno value. */
147 brc_open(struct nl_sock **sock)
149 int multicast_group = 0;
152 retval = nl_lookup_genl_family(BRC_GENL_FAMILY_NAME, &brc_family);
157 retval = lookup_brc_multicast_group(&multicast_group);
162 retval = nl_sock_create(NETLINK_GENERIC, multicast_group, 0, 0, sock);
170 static const struct nl_policy brc_dp_policy[] = {
171 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
174 static struct ovsrec_bridge *
175 find_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
179 for (i = 0; i < ovs->n_bridges; i++) {
180 if (!strcmp(br_name, ovs->bridges[i]->name)) {
181 return ovs->bridges[i];
189 execute_appctl_command(const char *unixctl_command, char **output)
191 char *stdout_log, *stderr_log;
197 argv[2] = xasprintf(appctl_command, unixctl_command);
200 /* Run process and log status. */
201 error = process_run_capture(argv, &stdout_log, &stderr_log, &status);
203 VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
204 unixctl_command, strerror(error));
206 char *msg = process_status_msg(status);
207 VLOG_ERR("ovs-appctl exited with error (%s)", msg);
212 /* Deal with stdout_log. */
214 *output = stdout_log;
219 /* Deal with stderr_log */
220 if (stderr_log && *stderr_log) {
221 VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
231 do_get_bridge_parts(const struct ovsrec_bridge *br, struct svec *parts,
232 int vlan, bool break_down_bonds)
238 for (i = 0; i < br->n_ports; i++) {
239 const struct ovsrec_port *port = br->ports[i];
241 svec_add(&ports, port->name);
243 int port_vlan = port->n_tag ? *port->tag : 0;
244 if (vlan != port_vlan) {
248 if (break_down_bonds) {
249 for (j = 0; j < port->n_interfaces; j++) {
250 const struct ovsrec_interface *iface = port->interfaces[j];
251 svec_add(parts, iface->name);
254 svec_add(parts, port->name);
257 svec_destroy(&ports);
260 /* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
261 * down into their constituent parts.
263 * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
264 * then only interfaces for trunk ports or ports with implicit VLAN 0 are
265 * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
268 get_bridge_ifaces(const struct ovsrec_bridge *br, struct svec *ifaces,
271 do_get_bridge_parts(br, ifaces, vlan, true);
274 /* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
275 * the bond name, not broken down into their constituent interfaces.
277 * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
278 * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
279 * only port with implicit VLAN 'vlan' are reported. */
281 get_bridge_ports(const struct ovsrec_bridge *br, struct svec *ports,
284 do_get_bridge_parts(br, ports, vlan, false);
288 /* Go through the configuration file and remove any ports that no longer
289 * exist associated with a bridge. */
294 struct svec bridges, delete;
296 if (cfg_lock(NULL, 0)) {
297 /* Couldn't lock config file. */
303 cfg_get_subsections(&bridges, "bridge");
304 for (i=0; i<bridges.n; i++) {
305 const char *br_name = bridges.names[i];
308 /* Check that each bridge interface exists. */
310 get_bridge_ifaces(br_name, &ifaces, -1);
311 for (j = 0; j < ifaces.n; j++) {
312 const char *iface_name = ifaces.names[j];
314 /* The local port and internal ports are created and destroyed by
315 * ovs-vswitchd itself, so don't bother checking for them at all.
316 * In practice, they might not exist if ovs-vswitchd hasn't
317 * finished reloading since the configuration file was updated. */
318 if (!strcmp(iface_name, br_name)
319 || cfg_get_bool(0, "iface.%s.internal", iface_name)) {
323 if (!netdev_exists(iface_name)) {
324 VLOG_INFO_RL(&rl, "removing dead interface %s from %s",
325 iface_name, br_name);
326 svec_add(&delete, iface_name);
329 svec_destroy(&ifaces);
331 svec_destroy(&bridges);
336 for (i = 0; i < delete.n; i++) {
337 cfg_del_match("bridge.*.port=%s", delete.names[i]);
338 cfg_del_match("bonding.*.slave=%s", delete.names[i]);
345 svec_destroy(&delete);
349 static struct ovsdb_idl_txn *
350 txn_from_openvswitch(const struct ovsrec_open_vswitch *ovs)
352 return ovsdb_idl_txn_get(&ovs->header_);
356 port_is_fake_bridge(const struct ovsrec_port *port)
358 return (port->fake_bridge
360 && *port->tag >= 1 && *port->tag <= 4095);
364 ovs_insert_bridge(const struct ovsrec_open_vswitch *ovs,
365 struct ovsrec_bridge *bridge)
367 struct ovsrec_bridge **bridges;
370 bridges = xmalloc(sizeof *ovs->bridges * (ovs->n_bridges + 1));
371 for (i = 0; i < ovs->n_bridges; i++) {
372 bridges[i] = ovs->bridges[i];
374 bridges[ovs->n_bridges] = bridge;
375 ovsrec_open_vswitch_set_bridges(ovs, bridges, ovs->n_bridges + 1);
380 add_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
382 struct ovsrec_bridge *br;
383 struct ovsrec_port *port;
384 struct ovsrec_interface *iface;
386 if (find_bridge(ovs, br_name)) {
387 VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
389 } else if (netdev_exists(br_name)) {
392 for (i = 0; i < ovs->n_bridges; i++) {
394 struct ovsrec_bridge *br_cfg = ovs->bridges[i];
396 for (j = 0; j < br_cfg->n_ports; j++) {
397 if (port_is_fake_bridge(br_cfg->ports[j])) {
398 VLOG_WARN("addbr %s: %s exists as a fake bridge",
405 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
406 "device named %s already exists",
407 br_name, br_name, br_name);
411 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
412 ovsrec_interface_set_name(iface, br_name);
414 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
415 ovsrec_port_set_name(port, br_name);
416 ovsrec_port_set_interfaces(port, &iface, 1);
418 br = ovsrec_bridge_insert(txn_from_openvswitch(ovs));
419 ovsrec_bridge_set_name(br, br_name);
420 ovsrec_bridge_set_ports(br, &port, 1);
422 ovs_insert_bridge(ovs, br);
424 VLOG_INFO("addbr %s: success", br_name);
430 add_port(const struct ovsrec_open_vswitch *ovs,
431 const struct ovsrec_bridge *br, const char *port_name)
433 struct ovsrec_interface *iface;
434 struct ovsrec_port *port;
435 struct ovsrec_port **ports;
438 /* xxx Check conflicts? */
439 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
440 ovsrec_interface_set_name(iface, port_name);
442 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
443 ovsrec_port_set_name(port, port_name);
444 ovsrec_port_set_interfaces(port, &iface, 1);
446 ports = xmalloc(sizeof *br->ports * (br->n_ports + 1));
447 for (i = 0; i < br->n_ports; i++) {
448 ports[i] = br->ports[i];
450 ports[br->n_ports] = port;
451 ovsrec_bridge_set_ports(br, ports, br->n_ports + 1);
456 del_port(const struct ovsrec_bridge *br, const char *port_name)
459 struct ovsrec_port *port_rec = NULL;
461 for (i = 0; i < br->n_ports; i++) {
462 struct ovsrec_port *port = br->ports[i];
463 if (!strcmp(port_name, port->name)) {
466 for (j = 0; j < port->n_interfaces; j++) {
467 struct ovsrec_interface *iface = port->interfaces[j];
468 if (!strcmp(port_name, iface->name)) {
469 ovsrec_interface_delete(iface);
474 /* xxx Probably can move this into the "for" loop. */
476 struct ovsrec_port **ports;
479 ports = xmalloc(sizeof *br->ports * br->n_ports);
480 for (i = n = 0; i < br->n_ports; i++) {
481 if (br->ports[i] != port_rec) {
482 ports[n++] = br->ports[i];
485 ovsrec_bridge_set_ports(br, ports, n);
491 del_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
493 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
494 struct ovsrec_bridge **bridges;
498 VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
502 del_port(br, br_name);
504 ovsrec_bridge_delete(br);
506 bridges = xmalloc(sizeof *ovs->bridges * ovs->n_bridges);
507 for (i = n = 0; i < ovs->n_bridges; i++) {
508 if (ovs->bridges[i] != br) {
509 bridges[n++] = ovs->bridges[i];
512 ovsrec_open_vswitch_set_bridges(ovs, bridges, n);
515 VLOG_INFO("delbr %s: success", br_name);
521 parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
522 const char **port_name, uint64_t *count, uint64_t *skip)
524 static const struct nl_policy policy[] = {
525 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
526 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
527 [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
528 [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
530 struct nlattr *attrs[ARRAY_SIZE(policy)];
532 if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
533 attrs, ARRAY_SIZE(policy))
534 || (br_name && !attrs[BRC_GENL_A_DP_NAME])
535 || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
536 || (count && !attrs[BRC_GENL_A_FDB_COUNT])
537 || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
541 *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
543 *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
546 *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
549 *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
552 *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
557 /* Composes and returns a reply to a request made by the datapath with Netlink
558 * sequence number 'seq' and error code 'error'. The caller may add additional
559 * attributes to the message, then it may send it with send_reply(). */
560 static struct ofpbuf *
561 compose_reply(uint32_t seq, int error)
563 struct ofpbuf *reply = ofpbuf_new(4096);
564 nl_msg_put_genlmsghdr(reply, brc_sock, 32, brc_family, NLM_F_REQUEST,
565 BRC_GENL_C_DP_RESULT, 1);
566 ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
567 nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
571 /* Sends 'reply' to the datapath and frees it. */
573 send_reply(struct ofpbuf *reply)
575 int retval = nl_sock_send(brc_sock, reply, false);
577 VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
580 ofpbuf_delete(reply);
583 /* Composes and sends a reply to a request made by the datapath with Netlink
584 * sequence number 'seq' and error code 'error'. */
586 send_simple_reply(uint32_t seq, int error)
588 send_reply(compose_reply(seq, error));
592 handle_bridge_cmd(const struct ovsrec_open_vswitch *ovs,
593 struct ofpbuf *buffer, bool add)
599 error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
601 error = add ? add_bridge(ovs, br_name) : del_bridge(ovs, br_name);
602 send_simple_reply(seq, error);
607 static const struct nl_policy brc_port_policy[] = {
608 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
609 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
613 handle_port_cmd(const struct ovsrec_open_vswitch *ovs,
614 struct ofpbuf *buffer, bool add)
616 const char *cmd_name = add ? "add-if" : "del-if";
617 const char *br_name, *port_name;
621 error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
623 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
626 VLOG_WARN("%s %s %s: no bridge named %s",
627 cmd_name, br_name, port_name, br_name);
629 } else if (!netdev_exists(port_name)) {
630 VLOG_WARN("%s %s %s: no network device named %s",
631 cmd_name, br_name, port_name, port_name);
635 add_port(ovs, br, port_name);
637 del_port(br, port_name);
639 VLOG_INFO("%s %s %s: success", cmd_name, br_name, port_name);
641 send_simple_reply(seq, error);
647 /* The caller is responsible for freeing '*ovs_name' if the call is
650 linux_bridge_to_ovs_bridge(const struct ovsrec_open_vswitch *ovs,
651 const char *linux_name,
652 const struct ovsrec_bridge **ovs_bridge,
655 *ovs_bridge = find_bridge(ovs, linux_name);
657 /* Bridge name is the same. We are interested in VLAN 0. */
661 /* No such Open vSwitch bridge 'linux_name', but there might be an
662 * internal port named 'linux_name' on some other bridge
663 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
664 * port 'linux_name' on the bridge named 'ovs_bridge'. */
667 for (i = 0; i < ovs->n_bridges; i++) {
668 const struct ovsrec_bridge *br = ovs->bridges[i];
670 for (j = 0; j < br->n_ports; j++) {
671 const struct ovsrec_port *port = br->ports[j];
673 if (!strcmp(port->name, linux_name)) {
675 *br_vlan = port->n_tag ? *port->tag : -1;
686 handle_fdb_query_cmd(const struct ovsrec_open_vswitch *ovs,
687 struct ofpbuf *buffer)
689 /* This structure is copied directly from the Linux 2.6.30 header files.
690 * It would be more straightforward to #include <linux/if_bridge.h>, but
691 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
692 * with old header files won't have it. */
697 __u32 ageing_timer_value;
706 struct mac *local_macs;
710 /* Impedance matching between the vswitchd and Linux kernel notions of what
711 * a bridge is. The kernel only handles a single VLAN per bridge, but
712 * vswitchd can deal with all the VLANs on a single bridge. We have to
713 * pretend that the former is the case even though the latter is the
715 const char *linux_name; /* Name used by brctl. */
716 const struct ovsrec_bridge *ovs_bridge; /* Bridge used by ovs-vswitchd. */
717 int br_vlan; /* VLAN tag. */
720 struct ofpbuf query_data;
721 struct ofpbuf *reply;
722 char *unixctl_command;
723 uint64_t count, skip;
729 /* Parse the command received from brcompat_mod. */
730 error = parse_command(buffer, &seq, &linux_name, NULL, &count, &skip);
735 /* Figure out vswitchd bridge and VLAN. */
736 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
737 &ovs_bridge, &br_vlan);
739 send_simple_reply(seq, error);
743 /* Fetch the forwarding database using ovs-appctl. */
744 unixctl_command = xasprintf("fdb/show %s", ovs_bridge->name);
745 error = execute_appctl_command(unixctl_command, &output);
746 free(unixctl_command);
748 send_simple_reply(seq, error);
752 /* Fetch the MAC address for each interface on the bridge, so that we can
753 * fill in the is_local field in the response. */
755 get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
756 local_macs = xmalloc(ifaces.n * sizeof *local_macs);
758 for (i = 0; i < ifaces.n; i++) {
759 const char *iface_name = ifaces.names[i];
760 struct mac *mac = &local_macs[n_local_macs];
761 struct netdev *netdev;
763 error = netdev_open(iface_name, NETDEV_ETH_TYPE_NONE, &netdev);
765 if (!netdev_get_etheraddr(netdev, mac->addr)) {
768 netdev_close(netdev);
771 svec_destroy(&ifaces);
773 /* Parse the response from ovs-appctl and convert it to binary format to
774 * pass back to the kernel. */
775 ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
777 strtok_r(output, "\n", &save_ptr); /* Skip header line. */
779 struct __fdb_entry *entry;
781 uint8_t mac[ETH_ADDR_LEN];
785 line = strtok_r(NULL, "\n", &save_ptr);
790 if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
791 &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
792 != 2 + ETH_ADDR_SCAN_COUNT + 1) {
793 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
794 VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
798 if (vlan != br_vlan) {
807 /* Is this the MAC address of an interface on the bridge? */
809 for (i = 0; i < n_local_macs; i++) {
810 if (eth_addr_equals(local_macs[i].addr, mac)) {
816 entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
817 memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
818 entry->port_no = port & 0xff;
819 entry->is_local = is_local;
820 entry->ageing_timer_value = age * HZ;
821 entry->port_hi = (port & 0xff00) >> 8;
828 /* Compose and send reply to datapath. */
829 reply = compose_reply(seq, 0);
830 nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
831 query_data.data, query_data.size);
835 ofpbuf_uninit(&query_data);
841 send_ifindex_reply(uint32_t seq, struct svec *ifaces)
843 struct ofpbuf *reply;
849 /* Make sure that any given interface only occurs once. This shouldn't
850 * happen, but who knows what people put into their configuration files. */
851 svec_sort_unique(ifaces);
853 /* Convert 'ifaces' into ifindexes. */
855 indices = xmalloc(ifaces->n * sizeof *indices);
856 SVEC_FOR_EACH (i, iface, ifaces) {
857 int ifindex = if_nametoindex(iface);
859 indices[n_indices++] = ifindex;
863 /* Compose and send reply. */
864 reply = compose_reply(seq, 0);
865 nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
866 indices, n_indices * sizeof *indices);
874 handle_get_bridges_cmd(const struct ovsrec_open_vswitch *ovs,
875 struct ofpbuf *buffer)
884 /* Parse Netlink command.
886 * The command doesn't actually have any arguments, but we need the
887 * sequence number to send the reply. */
888 error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
893 /* Get all the real bridges and all the fake ones. */
895 for (i = 0; i < ovs->n_bridges; i++) {
896 const struct ovsrec_bridge *br = ovs->bridges[i];
898 svec_add(&bridges, br->name);
899 for (j = 0; j < br->n_ports; j++) {
900 const struct ovsrec_port *port = br->ports[j];
902 if (port->fake_bridge) {
903 svec_add(&bridges, port->name);
908 send_ifindex_reply(seq, &bridges);
909 svec_destroy(&bridges);
915 handle_get_ports_cmd(const struct ovsrec_open_vswitch *ovs,
916 struct ofpbuf *buffer)
920 const char *linux_name;
921 const struct ovsrec_bridge *ovs_bridge;
928 /* Parse Netlink command. */
929 error = parse_command(buffer, &seq, &linux_name, NULL, NULL, NULL);
934 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
935 &ovs_bridge, &br_vlan);
937 send_simple_reply(seq, error);
942 get_bridge_ports(ovs_bridge, &ports, br_vlan);
944 svec_del(&ports, linux_name);
945 send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
946 svec_destroy(&ports);
952 brc_recv_update(const struct ovsrec_open_vswitch *ovs)
955 struct ofpbuf *buffer;
956 struct genlmsghdr *genlmsghdr;
961 ofpbuf_delete(buffer);
962 retval = nl_sock_recv(brc_sock, &buffer, false);
963 } while (retval == ENOBUFS
965 && (nl_msg_nlmsgerr(buffer, NULL)
966 || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE)));
968 if (retval != EAGAIN) {
969 VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
974 genlmsghdr = nl_msg_genlmsghdr(buffer);
976 VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
980 if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
981 VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
982 nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
986 switch (genlmsghdr->cmd) {
987 case BRC_GENL_C_DP_ADD:
988 handle_bridge_cmd(ovs, buffer, true);
991 case BRC_GENL_C_DP_DEL:
992 handle_bridge_cmd(ovs, buffer, false);
995 case BRC_GENL_C_PORT_ADD:
996 handle_port_cmd(ovs, buffer, true);
999 case BRC_GENL_C_PORT_DEL:
1000 handle_port_cmd(ovs, buffer, false);
1003 case BRC_GENL_C_FDB_QUERY:
1004 handle_fdb_query_cmd(ovs, buffer);
1007 case BRC_GENL_C_GET_BRIDGES:
1008 handle_get_bridges_cmd(ovs, buffer);
1011 case BRC_GENL_C_GET_PORTS:
1012 handle_get_ports_cmd(ovs, buffer);
1016 VLOG_WARN_RL(&rl, "received unknown brc netlink command: %d\n",
1022 ofpbuf_delete(buffer);
1027 /* Check for interface configuration changes announced through RTNL. */
1029 rtnl_recv_update(void)
1033 int error = nl_sock_recv(rtnl_sock, &buf, false);
1034 if (error == EAGAIN) {
1035 /* Nothing to do. */
1036 } else if (error == ENOBUFS) {
1037 VLOG_WARN_RL(&rl, "network monitor socket overflowed");
1039 VLOG_WARN_RL(&rl, "error on network monitor socket: %s",
1042 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1043 struct nlmsghdr *nlh;
1044 struct ifinfomsg *iim;
1046 nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
1047 iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
1049 VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
1054 if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1055 rtnlgrp_link_policy,
1056 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1057 VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
1061 if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
1062 const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
1063 char br_name[IFNAMSIZ];
1064 uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
1066 if (!if_indextoname(br_idx, br_name)) {
1071 if (cfg_lock(NULL, lock_timeout)) {
1072 /* Couldn't lock config file. */
1073 /* xxx this should try again and print error msg. */
1078 if (!netdev_exists(port_name)) {
1079 /* Network device is really gone. */
1082 VLOG_INFO("network device %s destroyed, "
1083 "removing from bridge %s", port_name, br_name);
1086 cfg_get_all_keys(&ports, "bridge.%s.port", br_name);
1088 if (svec_contains(&ports, port_name)) {
1089 del_port(br_name, port_name);
1091 svec_destroy(&ports);
1093 /* A network device by that name exists even though the kernel
1094 * told us it had disappeared. Probably, what happened was
1097 * 1. Device destroyed.
1098 * 2. Notification sent to us.
1099 * 3. New device created with same name as old one.
1100 * 4. ovs-brcompatd notified, removes device from bridge.
1102 * There's no a priori reason that in this situation that the
1103 * new device with the same name should remain in the bridge;
1104 * on the contrary, that would be unexpected. *But* there is
1105 * one important situation where, if we do this, bad things
1106 * happen. This is the case of XenServer Tools version 5.0.0,
1107 * which on boot of a Windows VM cause something like this to
1108 * happen on the Xen host:
1110 * i. Create tap1.0 and vif1.0.
1111 * ii. Delete tap1.0.
1112 * iii. Delete vif1.0.
1113 * iv. Re-create vif1.0.
1115 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1116 * neither does a VM without Tools installed at all.@.)
1118 * Steps iii and iv happen within a few seconds of each other.
1119 * Step iv causes /etc/xensource/scripts/vif to run, which in
1120 * turn calls ovs-cfg-mod to add the new device to the bridge.
1121 * If step iv happens after step 4 (in our first list of
1122 * steps), then all is well, but if it happens between 3 and 4
1123 * (which can easily happen if ovs-brcompatd has to wait to
1124 * lock the configuration file), then we will remove the new
1125 * incarnation from the bridge instead of the old one!
1127 * So, to avoid this problem, we do nothing here. This is
1128 * strictly incorrect except for this one particular case, and
1129 * perhaps that will bite us someday. If that happens, then we
1130 * will have to somehow track network devices by ifindex, since
1131 * a new device will have a new ifindex even if it has the same
1132 * name as an old device.
1134 VLOG_INFO("kernel reported network device %s removed but "
1135 "a device by that name exists (XS Tools 5.0.0?)",
1146 main(int argc, char *argv[])
1148 struct unixctl_server *unixctl;
1150 struct ovsdb_idl *idl;
1151 unsigned int idl_seqno;
1154 set_program_name(argv[0]);
1155 register_fault_handlers();
1158 vlog_set_levels(VLM_ANY_MODULE, VLF_CONSOLE, VLL_WARN);
1159 vlog_set_levels(VLM_reconnect, VLF_ANY_FACILITY, VLL_WARN);
1161 remote = parse_options(argc, argv);
1162 signal(SIGPIPE, SIG_IGN);
1165 die_if_already_running();
1168 retval = unixctl_server_create(NULL, &unixctl);
1170 ovs_fatal(retval, "could not listen for vlog connections");
1173 if (brc_open(&brc_sock)) {
1174 ovs_fatal(0, "could not open brcompat socket. Check "
1175 "\"brcompat\" kernel module.");
1178 if (prune_timeout) {
1179 if (nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &rtnl_sock)) {
1180 ovs_fatal(0, "could not create rtnetlink socket");
1184 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
1185 idl_seqno = ovsdb_idl_get_seqno(idl);
1188 const struct ovsrec_open_vswitch *ovs;
1189 struct ovsdb_idl_txn *txn;
1191 unsigned int new_idl_seqno;
1195 /* xxx Complete hack to get around bad ovs! */
1196 new_idl_seqno = ovsdb_idl_get_seqno(idl);
1197 if (new_idl_seqno == idl_seqno) {
1198 ovsdb_idl_wait(idl);
1200 printf("xxx trying again...\n");
1201 idl_seqno = new_idl_seqno;
1205 ovs = ovsrec_open_vswitch_first(idl);
1207 /* XXX it would be more user-friendly to create a record ourselves
1208 * (while verifying that the table is empty before doing so). */
1209 ovs_fatal(0, "%s: database does not contain any Open vSwitch "
1210 "configuration", remote);
1213 txn = ovsdb_idl_txn_create(idl);
1215 unixctl_server_run(unixctl);
1216 brc_recv_update(ovs);
1220 /* If 'prune_timeout' is non-zero, we actively prune from the
1221 * config file any 'bridge.<br_name>.port' entries that are no
1222 * longer valid. We use two methods:
1224 * 1) The kernel explicitly notifies us of removed ports
1225 * through the RTNL messages.
1227 * 2) We periodically check all ports associated with bridges
1228 * to see if they no longer exist.
1230 if (prune_timeout) {
1234 nl_sock_wait(rtnl_sock, POLLIN);
1235 poll_timer_wait(prune_timeout);
1239 while ((status = ovsdb_idl_txn_commit(txn)) == TXN_INCOMPLETE) {
1241 ovsdb_idl_wait(idl);
1242 ovsdb_idl_txn_wait(txn);
1245 ovsdb_idl_txn_destroy(txn);
1248 case TXN_INCOMPLETE:
1252 /* Should not happen--we never call ovsdb_idl_txn_abort(). */
1253 ovs_fatal(0, "transaction aborted");
1259 /* xxx Handle this better! */
1260 printf("xxx We need to try again!\n");
1264 /* xxx Is this what we want to do? */
1265 ovs_fatal(0, "transaction error");
1271 nl_sock_wait(brc_sock, POLLIN);
1272 ovsdb_idl_wait(idl);
1273 unixctl_server_wait(unixctl);
1278 ovsdb_idl_destroy(idl);
1284 validate_appctl_command(void)
1290 for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
1292 /* Nothing to do. */
1293 } else if (p[1] == 's') {
1296 ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
1300 ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
1305 parse_options(int argc, char *argv[])
1308 OPT_LOCK_TIMEOUT = UCHAR_MAX + 1,
1312 LEAK_CHECKER_OPTION_ENUMS
1314 static struct option long_options[] = {
1315 {"help", no_argument, 0, 'h'},
1316 {"version", no_argument, 0, 'V'},
1317 {"lock-timeout", required_argument, 0, OPT_LOCK_TIMEOUT},
1318 {"prune-timeout", required_argument, 0, OPT_PRUNE_TIMEOUT},
1319 {"appctl-command", required_argument, 0, OPT_APPCTL_COMMAND},
1320 DAEMON_LONG_OPTIONS,
1322 LEAK_CHECKER_LONG_OPTIONS,
1325 char *short_options = long_options_to_short_options(long_options);
1327 appctl_command = xasprintf("%s/ovs-appctl %%s", ovs_bindir);
1331 c = getopt_long(argc, argv, short_options, long_options, NULL);
1342 OVS_PRINT_VERSION(0, 0);
1345 case OPT_LOCK_TIMEOUT:
1346 lock_timeout = atoi(optarg);
1349 case OPT_PRUNE_TIMEOUT:
1350 prune_timeout = atoi(optarg) * 1000;
1353 case OPT_APPCTL_COMMAND:
1354 appctl_command = optarg;
1357 VLOG_OPTION_HANDLERS
1358 DAEMON_OPTION_HANDLERS
1359 LEAK_CHECKER_OPTION_HANDLERS
1368 free(short_options);
1370 validate_appctl_command();
1376 ovs_fatal(0, "database socket is non-option argument; "
1377 "use --help for usage");
1386 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1387 "usage: %s [OPTIONS] CONFIG\n"
1388 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1389 program_name, program_name);
1390 printf("\nConfiguration options:\n"
1391 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1392 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1393 " --lock-timeout=MSECS wait at most MSECS for CONFIG to unlock\n"
1397 printf("\nOther options:\n"
1398 " -h, --help display this help message\n"
1399 " -V, --version display version information\n");
1400 leak_checker_usage();
1401 printf("\nThe default appctl command is:\n%s\n", appctl_command);