1 /* Copyright (c) 2008, 2009, 2010, 2011 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"
46 #include "netlink-socket.h"
48 #include "openvswitch/brcompat-netlink.h"
49 #include "ovsdb-idl.h"
51 #include "poll-loop.h"
59 #include "vswitchd/vswitch-idl.h"
61 VLOG_DEFINE_THIS_MODULE(brcompatd);
64 /* xxx Just hangs if datapath is rmmod/insmod. Learn to reconnect? */
66 /* Actions to modify bridge compatibility configuration. */
74 static const char *parse_options(int argc, char *argv[]);
75 static void usage(void) NO_RETURN;
77 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 60);
79 /* Maximum number of milliseconds to wait before pruning port entries that
80 * no longer exist. If set to zero, ports are never pruned. */
81 static int prune_timeout = 5000;
83 /* Shell command to execute (via popen()) to send a control command to the
84 * running ovs-vswitchd process. The string must contain one instance of %s,
85 * which is replaced by the control command. */
86 static char *appctl_command;
88 /* Netlink socket to listen for interface changes. */
89 static struct nl_sock *rtnl_sock;
91 /* Netlink socket to bridge compatibility kernel module. */
92 static struct nl_sock *brc_sock;
94 /* The Generic Netlink family number used for bridge compatibility. */
95 static int brc_family;
97 static const struct nl_policy brc_multicast_policy[] = {
98 [BRC_GENL_A_MC_GROUP] = {.type = NL_A_U32 }
101 static const struct nl_policy rtnlgrp_link_policy[] = {
102 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
103 [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
107 lookup_brc_multicast_group(int *multicast_group)
109 struct nl_sock *sock;
110 struct ofpbuf request, *reply;
111 struct nlattr *attrs[ARRAY_SIZE(brc_multicast_policy)];
114 retval = nl_sock_create(NETLINK_GENERIC, &sock);
118 ofpbuf_init(&request, 0);
119 nl_msg_put_genlmsghdr(&request, 0, brc_family,
120 NLM_F_REQUEST, BRC_GENL_C_QUERY_MC, 1);
121 retval = nl_sock_transact(sock, &request, &reply);
122 ofpbuf_uninit(&request);
124 nl_sock_destroy(sock);
127 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
128 brc_multicast_policy, attrs,
129 ARRAY_SIZE(brc_multicast_policy))) {
130 nl_sock_destroy(sock);
131 ofpbuf_delete(reply);
134 *multicast_group = nl_attr_get_u32(attrs[BRC_GENL_A_MC_GROUP]);
135 nl_sock_destroy(sock);
136 ofpbuf_delete(reply);
141 /* Opens a socket for brcompat notifications. Returns 0 if successful,
142 * otherwise a positive errno value. */
144 brc_open(struct nl_sock **sock)
146 int multicast_group = 0;
149 retval = nl_lookup_genl_family(BRC_GENL_FAMILY_NAME, &brc_family);
154 retval = lookup_brc_multicast_group(&multicast_group);
159 retval = nl_sock_create(NETLINK_GENERIC, sock);
164 retval = nl_sock_join_mcgroup(*sock, multicast_group);
166 nl_sock_destroy(*sock);
172 static const struct nl_policy brc_dp_policy[] = {
173 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
176 static struct ovsrec_bridge *
177 find_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
181 for (i = 0; i < ovs->n_bridges; i++) {
182 if (!strcmp(br_name, ovs->bridges[i]->name)) {
183 return ovs->bridges[i];
191 execute_appctl_command(const char *unixctl_command, char **output)
193 char *stdout_log, *stderr_log;
199 argv[2] = xasprintf(appctl_command, unixctl_command);
202 /* Run process and log status. */
203 error = process_run_capture(argv, &stdout_log, &stderr_log, &status);
205 VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
206 unixctl_command, strerror(error));
208 char *msg = process_status_msg(status);
209 VLOG_ERR("ovs-appctl exited with error (%s)", msg);
214 /* Deal with stdout_log. */
216 *output = stdout_log;
221 /* Deal with stderr_log */
222 if (stderr_log && *stderr_log) {
223 VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
233 do_get_bridge_parts(const struct ovsrec_bridge *br, struct sset *parts,
234 int vlan, bool break_down_bonds)
238 for (i = 0; i < br->n_ports; i++) {
239 const struct ovsrec_port *port = br->ports[i];
242 int port_vlan = port->n_tag ? *port->tag : 0;
243 if (vlan != port_vlan) {
247 if (break_down_bonds) {
248 for (j = 0; j < port->n_interfaces; j++) {
249 const struct ovsrec_interface *iface = port->interfaces[j];
250 sset_add(parts, iface->name);
253 sset_add(parts, port->name);
258 /* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
259 * down into their constituent parts.
261 * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
262 * then only interfaces for trunk ports or ports with implicit VLAN 0 are
263 * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
266 get_bridge_ifaces(const struct ovsrec_bridge *br, struct sset *ifaces,
269 do_get_bridge_parts(br, ifaces, vlan, true);
272 /* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
273 * the bond name, not broken down into their constituent interfaces.
275 * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
276 * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
277 * only port with implicit VLAN 'vlan' are reported. */
279 get_bridge_ports(const struct ovsrec_bridge *br, struct sset *ports,
282 do_get_bridge_parts(br, ports, vlan, false);
285 static struct ovsdb_idl_txn *
286 txn_from_openvswitch(const struct ovsrec_open_vswitch *ovs)
288 return ovsdb_idl_txn_get(&ovs->header_);
292 port_is_fake_bridge(const struct ovsrec_port *port)
294 return (port->fake_bridge
296 && *port->tag >= 1 && *port->tag <= 4095);
300 ovs_insert_bridge(const struct ovsrec_open_vswitch *ovs,
301 struct ovsrec_bridge *bridge)
303 struct ovsrec_bridge **bridges;
306 bridges = xmalloc(sizeof *ovs->bridges * (ovs->n_bridges + 1));
307 for (i = 0; i < ovs->n_bridges; i++) {
308 bridges[i] = ovs->bridges[i];
310 bridges[ovs->n_bridges] = bridge;
311 ovsrec_open_vswitch_set_bridges(ovs, bridges, ovs->n_bridges + 1);
316 where_uuid_equals(const struct uuid *uuid)
321 json_string_create("_uuid"),
322 json_string_create("=="),
324 json_string_create("uuid"),
325 json_string_create_nocopy(
326 xasprintf(UUID_FMT, UUID_ARGS(uuid))))));
329 /* Commits 'txn'. If 'wait_for_reload' is true, also waits for Open vSwitch to
330 reload the configuration before returning.
332 Returns EAGAIN if the caller should try the operation again, 0 on success,
333 otherwise a positive errno value. */
335 commit_txn(struct ovsdb_idl_txn *txn, bool wait_for_reload)
337 struct ovsdb_idl *idl = ovsdb_idl_txn_get_idl (txn);
338 enum ovsdb_idl_txn_status status;
339 int64_t next_cfg = 0;
341 if (wait_for_reload) {
342 const struct ovsrec_open_vswitch *ovs = ovsrec_open_vswitch_first(idl);
343 struct json *where = where_uuid_equals(&ovs->header_.uuid);
344 ovsdb_idl_txn_increment(txn, "Open_vSwitch", "next_cfg", where);
347 status = ovsdb_idl_txn_commit_block(txn);
348 if (wait_for_reload && status == TXN_SUCCESS) {
349 next_cfg = ovsdb_idl_txn_get_increment_new_value(txn);
351 ovsdb_idl_txn_destroy(txn);
358 VLOG_ERR_RL(&rl, "OVSDB transaction unexpectedly aborted");
365 if (wait_for_reload) {
367 /* We can't use 'ovs' any longer because ovsdb_idl_run() can
369 const struct ovsrec_open_vswitch *ovs2;
372 OVSREC_OPEN_VSWITCH_FOR_EACH (ovs2, idl) {
373 if (ovs2->cur_cfg >= next_cfg) {
385 VLOG_ERR_RL(&rl, "OVSDB transaction needs retry");
389 VLOG_ERR_RL(&rl, "OVSDB transaction failed: %s",
390 ovsdb_idl_txn_get_error(txn));
399 add_bridge(struct ovsdb_idl *idl, const struct ovsrec_open_vswitch *ovs,
402 struct ovsrec_bridge *br;
403 struct ovsrec_port *port;
404 struct ovsrec_interface *iface;
405 struct ovsdb_idl_txn *txn;
407 if (find_bridge(ovs, br_name)) {
408 VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
410 } else if (netdev_exists(br_name)) {
413 for (i = 0; i < ovs->n_bridges; i++) {
415 struct ovsrec_bridge *br_cfg = ovs->bridges[i];
417 for (j = 0; j < br_cfg->n_ports; j++) {
418 if (port_is_fake_bridge(br_cfg->ports[j])) {
419 VLOG_WARN("addbr %s: %s exists as a fake bridge",
426 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
427 "device named %s already exists",
428 br_name, br_name, br_name);
432 txn = ovsdb_idl_txn_create(idl);
434 ovsdb_idl_txn_add_comment(txn, "ovs-brcompatd: addbr %s", br_name);
436 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
437 ovsrec_interface_set_name(iface, br_name);
439 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
440 ovsrec_port_set_name(port, br_name);
441 ovsrec_port_set_interfaces(port, &iface, 1);
443 br = ovsrec_bridge_insert(txn_from_openvswitch(ovs));
444 ovsrec_bridge_set_name(br, br_name);
445 ovsrec_bridge_set_ports(br, &port, 1);
447 ovs_insert_bridge(ovs, br);
449 return commit_txn(txn, true);
453 add_port(const struct ovsrec_open_vswitch *ovs,
454 const struct ovsrec_bridge *br, const char *port_name)
456 struct ovsrec_interface *iface;
457 struct ovsrec_port *port;
458 struct ovsrec_port **ports;
461 /* xxx Check conflicts? */
462 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
463 ovsrec_interface_set_name(iface, port_name);
465 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
466 ovsrec_port_set_name(port, port_name);
467 ovsrec_port_set_interfaces(port, &iface, 1);
469 ports = xmalloc(sizeof *br->ports * (br->n_ports + 1));
470 for (i = 0; i < br->n_ports; i++) {
471 ports[i] = br->ports[i];
473 ports[br->n_ports] = port;
474 ovsrec_bridge_set_ports(br, ports, br->n_ports + 1);
478 /* Deletes 'port' from 'br'.
480 * After calling this function, 'port' must not be referenced again. */
482 del_port(const struct ovsrec_bridge *br, const struct ovsrec_port *port)
484 struct ovsrec_port **ports;
487 /* Remove 'port' from the bridge's list of ports. */
488 ports = xmalloc(sizeof *br->ports * br->n_ports);
489 for (i = n = 0; i < br->n_ports; i++) {
490 if (br->ports[i] != port) {
491 ports[n++] = br->ports[i];
494 ovsrec_bridge_set_ports(br, ports, n);
498 /* Delete 'iface' from 'port' (which must be within 'br'). If 'iface' was
499 * 'port''s only interface, delete 'port' from 'br' also.
501 * After calling this function, 'iface' must not be referenced again. */
503 del_interface(const struct ovsrec_bridge *br,
504 const struct ovsrec_port *port,
505 const struct ovsrec_interface *iface)
507 if (port->n_interfaces == 1) {
510 struct ovsrec_interface **ifaces;
513 ifaces = xmalloc(sizeof *port->interfaces * port->n_interfaces);
514 for (i = n = 0; i < port->n_interfaces; i++) {
515 if (port->interfaces[i] != iface) {
516 ifaces[n++] = port->interfaces[i];
519 ovsrec_port_set_interfaces(port, ifaces, n);
524 /* Find and return a port within 'br' named 'port_name'. */
525 static const struct ovsrec_port *
526 find_port(const struct ovsrec_bridge *br, const char *port_name)
530 for (i = 0; i < br->n_ports; i++) {
531 struct ovsrec_port *port = br->ports[i];
532 if (!strcmp(port_name, port->name)) {
539 /* Find and return an interface within 'br' named 'iface_name'. */
540 static const struct ovsrec_interface *
541 find_interface(const struct ovsrec_bridge *br, const char *iface_name,
542 struct ovsrec_port **portp)
546 for (i = 0; i < br->n_ports; i++) {
547 struct ovsrec_port *port = br->ports[i];
550 for (j = 0; j < port->n_interfaces; j++) {
551 struct ovsrec_interface *iface = port->interfaces[j];
552 if (!strcmp(iface->name, iface_name)) {
564 del_bridge(struct ovsdb_idl *idl,
565 const struct ovsrec_open_vswitch *ovs, const char *br_name)
567 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
568 struct ovsrec_bridge **bridges;
569 struct ovsdb_idl_txn *txn;
573 VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
577 txn = ovsdb_idl_txn_create(idl);
579 ovsdb_idl_txn_add_comment(txn, "ovs-brcompatd: delbr %s", br_name);
581 /* Remove 'br' from the vswitch's list of bridges. */
582 bridges = xmalloc(sizeof *ovs->bridges * ovs->n_bridges);
583 for (i = n = 0; i < ovs->n_bridges; i++) {
584 if (ovs->bridges[i] != br) {
585 bridges[n++] = ovs->bridges[i];
588 ovsrec_open_vswitch_set_bridges(ovs, bridges, n);
591 return commit_txn(txn, true);
595 parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
596 const char **port_name, uint64_t *count, uint64_t *skip)
598 static const struct nl_policy policy[] = {
599 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
600 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
601 [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
602 [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
604 struct nlattr *attrs[ARRAY_SIZE(policy)];
606 if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
607 attrs, ARRAY_SIZE(policy))
608 || (br_name && !attrs[BRC_GENL_A_DP_NAME])
609 || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
610 || (count && !attrs[BRC_GENL_A_FDB_COUNT])
611 || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
615 *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
617 *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
620 *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
623 *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
626 *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
631 /* Composes and returns a reply to a request made by the datapath with Netlink
632 * sequence number 'seq' and error code 'error'. The caller may add additional
633 * attributes to the message, then it may send it with send_reply(). */
634 static struct ofpbuf *
635 compose_reply(uint32_t seq, int error)
637 struct ofpbuf *reply = ofpbuf_new(4096);
638 nl_msg_put_genlmsghdr(reply, 32, brc_family, NLM_F_REQUEST,
639 BRC_GENL_C_DP_RESULT, 1);
640 ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
641 nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
645 /* Sends 'reply' to the datapath and frees it. */
647 send_reply(struct ofpbuf *reply)
649 int retval = nl_sock_send(brc_sock, reply, false);
651 VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
654 ofpbuf_delete(reply);
657 /* Composes and sends a reply to a request made by the datapath with Netlink
658 * sequence number 'seq' and error code 'error'. */
660 send_simple_reply(uint32_t seq, int error)
662 send_reply(compose_reply(seq, error));
666 handle_bridge_cmd(struct ovsdb_idl *idl,
667 const struct ovsrec_open_vswitch *ovs,
668 struct ofpbuf *buffer, bool add)
674 error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
679 retval = (add ? add_bridge : del_bridge)(idl, ovs, br_name);
680 VLOG_INFO_RL(&rl, "%sbr %s: %s",
681 add ? "add" : "del", br_name, strerror(retval));
682 } while (retval == EAGAIN);
684 send_simple_reply(seq, error);
689 static const struct nl_policy brc_port_policy[] = {
690 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
691 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
695 handle_port_cmd(struct ovsdb_idl *idl,
696 const struct ovsrec_open_vswitch *ovs,
697 struct ofpbuf *buffer, bool add)
699 const char *cmd_name = add ? "add-if" : "del-if";
700 const char *br_name, *port_name;
704 error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
706 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
709 VLOG_WARN("%s %s %s: no bridge named %s",
710 cmd_name, br_name, port_name, br_name);
712 } else if (!netdev_exists(port_name)) {
713 VLOG_WARN("%s %s %s: no network device named %s",
714 cmd_name, br_name, port_name, port_name);
718 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
721 ovsdb_idl_txn_add_comment(txn, "ovs-brcompatd: add-if %s",
723 add_port(ovs, br, port_name);
725 const struct ovsrec_port *port = find_port(br, port_name);
727 ovsdb_idl_txn_add_comment(txn,
728 "ovs-brcompatd: del-if %s",
734 error = commit_txn(txn, true);
735 VLOG_INFO_RL(&rl, "%s %s %s: %s",
736 cmd_name, br_name, port_name, strerror(error));
737 } while (error == EAGAIN);
739 send_simple_reply(seq, error);
745 /* The caller is responsible for freeing '*ovs_name' if the call is
748 linux_bridge_to_ovs_bridge(const struct ovsrec_open_vswitch *ovs,
749 const char *linux_name,
750 const struct ovsrec_bridge **ovs_bridge,
753 *ovs_bridge = find_bridge(ovs, linux_name);
755 /* Bridge name is the same. We are interested in VLAN 0. */
759 /* No such Open vSwitch bridge 'linux_name', but there might be an
760 * internal port named 'linux_name' on some other bridge
761 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
762 * port 'linux_name' on the bridge named 'ovs_bridge'. */
765 for (i = 0; i < ovs->n_bridges; i++) {
766 const struct ovsrec_bridge *br = ovs->bridges[i];
768 for (j = 0; j < br->n_ports; j++) {
769 const struct ovsrec_port *port = br->ports[j];
771 if (!strcmp(port->name, linux_name)) {
773 *br_vlan = port->n_tag ? *port->tag : -1;
784 handle_fdb_query_cmd(const struct ovsrec_open_vswitch *ovs,
785 struct ofpbuf *buffer)
787 /* This structure is copied directly from the Linux 2.6.30 header files.
788 * It would be more straightforward to #include <linux/if_bridge.h>, but
789 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
790 * with old header files won't have it. */
795 __u32 ageing_timer_value;
804 struct mac *local_macs;
808 /* Impedance matching between the vswitchd and Linux kernel notions of what
809 * a bridge is. The kernel only handles a single VLAN per bridge, but
810 * vswitchd can deal with all the VLANs on a single bridge. We have to
811 * pretend that the former is the case even though the latter is the
813 const char *linux_name; /* Name used by brctl. */
814 const struct ovsrec_bridge *ovs_bridge; /* Bridge used by ovs-vswitchd. */
815 int br_vlan; /* VLAN tag. */
818 struct ofpbuf query_data;
819 const char *iface_name;
820 struct ofpbuf *reply;
821 char *unixctl_command;
822 uint64_t count, skip;
828 /* Parse the command received from brcompat_mod. */
829 error = parse_command(buffer, &seq, &linux_name, NULL, &count, &skip);
834 /* Figure out vswitchd bridge and VLAN. */
835 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
836 &ovs_bridge, &br_vlan);
838 send_simple_reply(seq, error);
842 /* Fetch the forwarding database using ovs-appctl. */
843 unixctl_command = xasprintf("fdb/show %s", ovs_bridge->name);
844 error = execute_appctl_command(unixctl_command, &output);
845 free(unixctl_command);
847 send_simple_reply(seq, error);
851 /* Fetch the MAC address for each interface on the bridge, so that we can
852 * fill in the is_local field in the response. */
854 get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
855 local_macs = xmalloc(sset_count(&ifaces) * sizeof *local_macs);
857 SSET_FOR_EACH (iface_name, &ifaces) {
858 struct mac *mac = &local_macs[n_local_macs];
859 struct netdev *netdev;
861 error = netdev_open_default(iface_name, &netdev);
863 if (!netdev_get_etheraddr(netdev, mac->addr)) {
866 netdev_close(netdev);
869 sset_destroy(&ifaces);
871 /* Parse the response from ovs-appctl and convert it to binary format to
872 * pass back to the kernel. */
873 ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
875 strtok_r(output, "\n", &save_ptr); /* Skip header line. */
877 struct __fdb_entry *entry;
879 uint8_t mac[ETH_ADDR_LEN];
883 line = strtok_r(NULL, "\n", &save_ptr);
888 if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
889 &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
890 != 2 + ETH_ADDR_SCAN_COUNT + 1) {
891 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
892 VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
896 if (vlan != br_vlan) {
905 /* Is this the MAC address of an interface on the bridge? */
907 for (i = 0; i < n_local_macs; i++) {
908 if (eth_addr_equals(local_macs[i].addr, mac)) {
914 entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
915 memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
916 entry->port_no = port & 0xff;
917 entry->is_local = is_local;
918 entry->ageing_timer_value = age * HZ;
919 entry->port_hi = (port & 0xff00) >> 8;
926 /* Compose and send reply to datapath. */
927 reply = compose_reply(seq, 0);
928 nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
929 query_data.data, query_data.size);
933 ofpbuf_uninit(&query_data);
940 send_ifindex_reply(uint32_t seq, struct sset *ifaces)
942 struct ofpbuf *reply;
947 /* Convert 'ifaces' into ifindexes. */
949 indices = xmalloc(sset_count(ifaces) * sizeof *indices);
950 SSET_FOR_EACH (iface, ifaces) {
951 int ifindex = if_nametoindex(iface);
953 indices[n_indices++] = ifindex;
957 /* Compose and send reply. */
958 reply = compose_reply(seq, 0);
959 nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
960 indices, n_indices * sizeof *indices);
968 handle_get_bridges_cmd(const struct ovsrec_open_vswitch *ovs,
969 struct ofpbuf *buffer)
978 /* Parse Netlink command.
980 * The command doesn't actually have any arguments, but we need the
981 * sequence number to send the reply. */
982 error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
987 /* Get all the real bridges and all the fake ones. */
989 for (i = 0; i < ovs->n_bridges; i++) {
990 const struct ovsrec_bridge *br = ovs->bridges[i];
992 sset_add(&bridges, br->name);
993 for (j = 0; j < br->n_ports; j++) {
994 const struct ovsrec_port *port = br->ports[j];
996 if (port->fake_bridge) {
997 sset_add(&bridges, port->name);
1002 send_ifindex_reply(seq, &bridges);
1003 sset_destroy(&bridges);
1009 handle_get_ports_cmd(const struct ovsrec_open_vswitch *ovs,
1010 struct ofpbuf *buffer)
1014 const char *linux_name;
1015 const struct ovsrec_bridge *ovs_bridge;
1022 /* Parse Netlink command. */
1023 error = parse_command(buffer, &seq, &linux_name, NULL, NULL, NULL);
1028 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
1029 &ovs_bridge, &br_vlan);
1031 send_simple_reply(seq, error);
1036 get_bridge_ports(ovs_bridge, &ports, br_vlan);
1037 sset_find_and_delete(&ports, linux_name);
1038 send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
1039 sset_destroy(&ports);
1044 static struct ofpbuf *
1045 brc_recv_update__(void)
1048 struct ofpbuf *buffer;
1051 retval = nl_sock_recv(brc_sock, &buffer, false);
1054 if (nl_msg_nlmsgerr(buffer, NULL)
1055 || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE) {
1067 VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
1070 ofpbuf_delete(buffer);
1075 brc_recv_update(struct ovsdb_idl *idl)
1077 struct ofpbuf *buffer;
1078 struct genlmsghdr *genlmsghdr;
1079 const struct ovsrec_open_vswitch *ovs;
1081 buffer = brc_recv_update__();
1086 genlmsghdr = nl_msg_genlmsghdr(buffer);
1088 VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
1092 if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
1093 VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
1094 nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
1098 /* Get the Open vSwitch configuration. Just drop the request on the floor
1099 * if a valid configuration doesn't exist. (We could check this earlier,
1100 * but we want to drain pending Netlink messages even when there is no Open
1101 * vSwitch configuration.) */
1102 ovs = ovsrec_open_vswitch_first(idl);
1104 VLOG_WARN_RL(&rl, "could not find valid configuration to update");
1108 switch (genlmsghdr->cmd) {
1109 case BRC_GENL_C_DP_ADD:
1110 handle_bridge_cmd(idl, ovs, buffer, true);
1113 case BRC_GENL_C_DP_DEL:
1114 handle_bridge_cmd(idl, ovs, buffer, false);
1117 case BRC_GENL_C_PORT_ADD:
1118 handle_port_cmd(idl, ovs, buffer, true);
1121 case BRC_GENL_C_PORT_DEL:
1122 handle_port_cmd(idl, ovs, buffer, false);
1125 case BRC_GENL_C_FDB_QUERY:
1126 handle_fdb_query_cmd(ovs, buffer);
1129 case BRC_GENL_C_GET_BRIDGES:
1130 handle_get_bridges_cmd(ovs, buffer);
1133 case BRC_GENL_C_GET_PORTS:
1134 handle_get_ports_cmd(ovs, buffer);
1138 VLOG_WARN_RL(&rl, "received unknown brc netlink command: %d\n",
1144 ofpbuf_delete(buffer);
1148 /* Check for interface configuration changes announced through RTNL. */
1150 rtnl_recv_update(struct ovsdb_idl *idl,
1151 const struct ovsrec_open_vswitch *ovs)
1155 int error = nl_sock_recv(rtnl_sock, &buf, false);
1156 if (error == EAGAIN) {
1157 /* Nothing to do. */
1158 } else if (error == ENOBUFS) {
1159 VLOG_WARN_RL(&rl, "network monitor socket overflowed");
1161 VLOG_WARN_RL(&rl, "error on network monitor socket: %s",
1164 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1165 struct nlmsghdr *nlh;
1166 struct ifinfomsg *iim;
1168 nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
1169 iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
1171 VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
1176 if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1177 rtnlgrp_link_policy,
1178 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1179 VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
1183 if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
1184 const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
1185 char br_name[IFNAMSIZ];
1186 uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
1188 if (!if_indextoname(br_idx, br_name)) {
1193 if (!netdev_exists(port_name)) {
1194 /* Network device is really gone. */
1195 struct ovsdb_idl_txn *txn;
1196 const struct ovsrec_interface *iface;
1197 struct ovsrec_port *port;
1198 struct ovsrec_bridge *br;
1200 VLOG_INFO("network device %s destroyed, "
1201 "removing from bridge %s", port_name, br_name);
1203 br = find_bridge(ovs, br_name);
1205 VLOG_WARN("no bridge named %s from which to remove %s",
1206 br_name, port_name);
1211 txn = ovsdb_idl_txn_create(idl);
1213 iface = find_interface(br, port_name, &port);
1215 del_interface(br, port, iface);
1216 ovsdb_idl_txn_add_comment(txn,
1217 "ovs-brcompatd: destroy port %s",
1221 commit_txn(txn, false);
1223 /* A network device by that name exists even though the kernel
1224 * told us it had disappeared. Probably, what happened was
1227 * 1. Device destroyed.
1228 * 2. Notification sent to us.
1229 * 3. New device created with same name as old one.
1230 * 4. ovs-brcompatd notified, removes device from bridge.
1232 * There's no a priori reason that in this situation that the
1233 * new device with the same name should remain in the bridge;
1234 * on the contrary, that would be unexpected. *But* there is
1235 * one important situation where, if we do this, bad things
1236 * happen. This is the case of XenServer Tools version 5.0.0,
1237 * which on boot of a Windows VM cause something like this to
1238 * happen on the Xen host:
1240 * i. Create tap1.0 and vif1.0.
1241 * ii. Delete tap1.0.
1242 * iii. Delete vif1.0.
1243 * iv. Re-create vif1.0.
1245 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1246 * neither does a VM without Tools installed at all.@.)
1248 * Steps iii and iv happen within a few seconds of each other.
1249 * Step iv causes /etc/xensource/scripts/vif to run, which in
1250 * turn calls ovs-cfg-mod to add the new device to the bridge.
1251 * If step iv happens after step 4 (in our first list of
1252 * steps), then all is well, but if it happens between 3 and 4
1253 * (which can easily happen if ovs-brcompatd has to wait to
1254 * lock the configuration file), then we will remove the new
1255 * incarnation from the bridge instead of the old one!
1257 * So, to avoid this problem, we do nothing here. This is
1258 * strictly incorrect except for this one particular case, and
1259 * perhaps that will bite us someday. If that happens, then we
1260 * will have to somehow track network devices by ifindex, since
1261 * a new device will have a new ifindex even if it has the same
1262 * name as an old device.
1264 VLOG_INFO("kernel reported network device %s removed but "
1265 "a device by that name exists (XS Tools 5.0.0?)",
1274 main(int argc, char *argv[])
1276 extern struct vlog_module VLM_reconnect;
1277 struct unixctl_server *unixctl;
1279 struct ovsdb_idl *idl;
1282 proctitle_init(argc, argv);
1283 set_program_name(argv[0]);
1284 vlog_set_levels(NULL, VLF_CONSOLE, VLL_WARN);
1285 vlog_set_levels(&VLM_reconnect, VLF_ANY_FACILITY, VLL_WARN);
1287 remote = parse_options(argc, argv);
1288 signal(SIGPIPE, SIG_IGN);
1292 die_if_already_running();
1295 retval = unixctl_server_create(NULL, &unixctl);
1300 if (brc_open(&brc_sock)) {
1301 ovs_fatal(0, "could not open brcompat socket. Check "
1302 "\"brcompat\" kernel module.");
1305 if (prune_timeout) {
1308 error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
1310 ovs_fatal(error, "could not create rtnetlink socket");
1313 error = nl_sock_join_mcgroup(rtnl_sock, RTNLGRP_LINK);
1315 ovs_fatal(error, "could not join RTNLGRP_LINK multicast group");
1319 daemonize_complete();
1321 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
1324 const struct ovsrec_open_vswitch *ovs;
1328 unixctl_server_run(unixctl);
1329 brc_recv_update(idl);
1331 ovs = ovsrec_open_vswitch_first(idl);
1332 if (!ovs && ovsdb_idl_has_ever_connected(idl)) {
1333 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1334 VLOG_WARN_RL(&rl, "%s: database does not contain any Open vSwitch "
1335 "configuration", remote);
1339 /* If 'prune_timeout' is non-zero, we actively prune from the
1340 * configuration of port entries that are no longer valid. We
1343 * 1) The kernel explicitly notifies us of removed ports
1344 * through the RTNL messages.
1346 * 2) We periodically check all ports associated with bridges
1347 * to see if they no longer exist.
1349 if (ovs && prune_timeout) {
1350 rtnl_recv_update(idl, ovs);
1351 nl_sock_wait(rtnl_sock, POLLIN);
1352 poll_timer_wait(prune_timeout);
1356 nl_sock_wait(brc_sock, POLLIN);
1357 ovsdb_idl_wait(idl);
1358 unixctl_server_wait(unixctl);
1363 ovsdb_idl_destroy(idl);
1369 validate_appctl_command(void)
1375 for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
1377 /* Nothing to do. */
1378 } else if (p[1] == 's') {
1381 ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
1385 ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
1390 parse_options(int argc, char *argv[])
1396 LEAK_CHECKER_OPTION_ENUMS,
1399 static struct option long_options[] = {
1400 {"help", no_argument, 0, 'h'},
1401 {"version", no_argument, 0, 'V'},
1402 {"prune-timeout", required_argument, 0, OPT_PRUNE_TIMEOUT},
1403 {"appctl-command", required_argument, 0, OPT_APPCTL_COMMAND},
1404 DAEMON_LONG_OPTIONS,
1406 LEAK_CHECKER_LONG_OPTIONS,
1409 char *short_options = long_options_to_short_options(long_options);
1411 appctl_command = xasprintf("%s/ovs-appctl %%s", ovs_bindir());
1415 c = getopt_long(argc, argv, short_options, long_options, NULL);
1426 OVS_PRINT_VERSION(0, 0);
1429 case OPT_PRUNE_TIMEOUT:
1430 prune_timeout = atoi(optarg) * 1000;
1433 case OPT_APPCTL_COMMAND:
1434 appctl_command = optarg;
1437 VLOG_OPTION_HANDLERS
1438 DAEMON_OPTION_HANDLERS
1439 LEAK_CHECKER_OPTION_HANDLERS
1448 free(short_options);
1450 validate_appctl_command();
1456 ovs_fatal(0, "database socket is non-option argument; "
1457 "use --help for usage");
1466 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1467 "usage: %s [OPTIONS] CONFIG\n"
1468 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1469 program_name, program_name);
1470 printf("\nConfiguration options:\n"
1471 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1472 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1476 printf("\nOther options:\n"
1477 " -h, --help display this help message\n"
1478 " -V, --version display version information\n");
1479 leak_checker_usage();
1480 printf("\nThe default appctl command is:\n%s\n", appctl_command);