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, 65536,
206 VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
207 unixctl_command, strerror(error));
209 char *msg = process_status_msg(status);
210 VLOG_ERR("ovs-appctl exited with error (%s)", msg);
215 /* Deal with stdout_log. */
217 *output = stdout_log;
222 /* Deal with stderr_log */
223 if (stderr_log && *stderr_log) {
224 VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
234 do_get_bridge_parts(const struct ovsrec_bridge *br, struct sset *parts,
235 int vlan, bool break_down_bonds)
239 for (i = 0; i < br->n_ports; i++) {
240 const struct ovsrec_port *port = br->ports[i];
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 sset_add(parts, iface->name);
254 sset_add(parts, port->name);
259 /* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
260 * down into their constituent parts.
262 * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
263 * then only interfaces for trunk ports or ports with implicit VLAN 0 are
264 * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
267 get_bridge_ifaces(const struct ovsrec_bridge *br, struct sset *ifaces,
270 do_get_bridge_parts(br, ifaces, vlan, true);
273 /* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
274 * the bond name, not broken down into their constituent interfaces.
276 * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
277 * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
278 * only port with implicit VLAN 'vlan' are reported. */
280 get_bridge_ports(const struct ovsrec_bridge *br, struct sset *ports,
283 do_get_bridge_parts(br, ports, vlan, false);
286 static struct ovsdb_idl_txn *
287 txn_from_openvswitch(const struct ovsrec_open_vswitch *ovs)
289 return ovsdb_idl_txn_get(&ovs->header_);
293 port_is_fake_bridge(const struct ovsrec_port *port)
295 return (port->fake_bridge
297 && *port->tag >= 1 && *port->tag <= 4095);
301 ovs_insert_bridge(const struct ovsrec_open_vswitch *ovs,
302 struct ovsrec_bridge *bridge)
304 struct ovsrec_bridge **bridges;
307 bridges = xmalloc(sizeof *ovs->bridges * (ovs->n_bridges + 1));
308 for (i = 0; i < ovs->n_bridges; i++) {
309 bridges[i] = ovs->bridges[i];
311 bridges[ovs->n_bridges] = bridge;
312 ovsrec_open_vswitch_set_bridges(ovs, bridges, ovs->n_bridges + 1);
317 where_uuid_equals(const struct uuid *uuid)
322 json_string_create("_uuid"),
323 json_string_create("=="),
325 json_string_create("uuid"),
326 json_string_create_nocopy(
327 xasprintf(UUID_FMT, UUID_ARGS(uuid))))));
330 /* Commits 'txn'. If 'wait_for_reload' is true, also waits for Open vSwitch to
331 reload the configuration before returning.
333 Returns EAGAIN if the caller should try the operation again, 0 on success,
334 otherwise a positive errno value. */
336 commit_txn(struct ovsdb_idl_txn *txn, bool wait_for_reload)
338 struct ovsdb_idl *idl = ovsdb_idl_txn_get_idl (txn);
339 enum ovsdb_idl_txn_status status;
340 int64_t next_cfg = 0;
342 if (wait_for_reload) {
343 const struct ovsrec_open_vswitch *ovs = ovsrec_open_vswitch_first(idl);
344 struct json *where = where_uuid_equals(&ovs->header_.uuid);
345 ovsdb_idl_txn_increment(txn, "Open_vSwitch", "next_cfg", where);
348 status = ovsdb_idl_txn_commit_block(txn);
349 if (wait_for_reload && status == TXN_SUCCESS) {
350 next_cfg = ovsdb_idl_txn_get_increment_new_value(txn);
352 ovsdb_idl_txn_destroy(txn);
359 VLOG_ERR_RL(&rl, "OVSDB transaction unexpectedly aborted");
366 if (wait_for_reload) {
368 /* We can't use 'ovs' any longer because ovsdb_idl_run() can
370 const struct ovsrec_open_vswitch *ovs2;
373 OVSREC_OPEN_VSWITCH_FOR_EACH (ovs2, idl) {
374 if (ovs2->cur_cfg >= next_cfg) {
386 VLOG_ERR_RL(&rl, "OVSDB transaction needs retry");
390 VLOG_ERR_RL(&rl, "OVSDB transaction failed: %s",
391 ovsdb_idl_txn_get_error(txn));
400 add_bridge(struct ovsdb_idl *idl, const struct ovsrec_open_vswitch *ovs,
403 struct ovsrec_bridge *br;
404 struct ovsrec_port *port;
405 struct ovsrec_interface *iface;
406 struct ovsdb_idl_txn *txn;
408 if (find_bridge(ovs, br_name)) {
409 VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
411 } else if (netdev_exists(br_name)) {
414 for (i = 0; i < ovs->n_bridges; i++) {
416 struct ovsrec_bridge *br_cfg = ovs->bridges[i];
418 for (j = 0; j < br_cfg->n_ports; j++) {
419 if (port_is_fake_bridge(br_cfg->ports[j])) {
420 VLOG_WARN("addbr %s: %s exists as a fake bridge",
427 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
428 "device named %s already exists",
429 br_name, br_name, br_name);
433 txn = ovsdb_idl_txn_create(idl);
435 ovsdb_idl_txn_add_comment(txn, "ovs-brcompatd: addbr %s", br_name);
437 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
438 ovsrec_interface_set_name(iface, br_name);
440 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
441 ovsrec_port_set_name(port, br_name);
442 ovsrec_port_set_interfaces(port, &iface, 1);
444 br = ovsrec_bridge_insert(txn_from_openvswitch(ovs));
445 ovsrec_bridge_set_name(br, br_name);
446 ovsrec_bridge_set_ports(br, &port, 1);
448 ovs_insert_bridge(ovs, br);
450 return commit_txn(txn, true);
454 add_port(const struct ovsrec_open_vswitch *ovs,
455 const struct ovsrec_bridge *br, const char *port_name)
457 struct ovsrec_interface *iface;
458 struct ovsrec_port *port;
459 struct ovsrec_port **ports;
462 /* xxx Check conflicts? */
463 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
464 ovsrec_interface_set_name(iface, port_name);
466 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
467 ovsrec_port_set_name(port, port_name);
468 ovsrec_port_set_interfaces(port, &iface, 1);
470 ports = xmalloc(sizeof *br->ports * (br->n_ports + 1));
471 for (i = 0; i < br->n_ports; i++) {
472 ports[i] = br->ports[i];
474 ports[br->n_ports] = port;
475 ovsrec_bridge_set_ports(br, ports, br->n_ports + 1);
479 /* Deletes 'port' from 'br'.
481 * After calling this function, 'port' must not be referenced again. */
483 del_port(const struct ovsrec_bridge *br, const struct ovsrec_port *port)
485 struct ovsrec_port **ports;
488 /* Remove 'port' from the bridge's list of ports. */
489 ports = xmalloc(sizeof *br->ports * br->n_ports);
490 for (i = n = 0; i < br->n_ports; i++) {
491 if (br->ports[i] != port) {
492 ports[n++] = br->ports[i];
495 ovsrec_bridge_set_ports(br, ports, n);
499 /* Delete 'iface' from 'port' (which must be within 'br'). If 'iface' was
500 * 'port''s only interface, delete 'port' from 'br' also.
502 * After calling this function, 'iface' must not be referenced again. */
504 del_interface(const struct ovsrec_bridge *br,
505 const struct ovsrec_port *port,
506 const struct ovsrec_interface *iface)
508 if (port->n_interfaces == 1) {
511 struct ovsrec_interface **ifaces;
514 ifaces = xmalloc(sizeof *port->interfaces * port->n_interfaces);
515 for (i = n = 0; i < port->n_interfaces; i++) {
516 if (port->interfaces[i] != iface) {
517 ifaces[n++] = port->interfaces[i];
520 ovsrec_port_set_interfaces(port, ifaces, n);
525 /* Find and return a port within 'br' named 'port_name'. */
526 static const struct ovsrec_port *
527 find_port(const struct ovsrec_bridge *br, const char *port_name)
531 for (i = 0; i < br->n_ports; i++) {
532 struct ovsrec_port *port = br->ports[i];
533 if (!strcmp(port_name, port->name)) {
540 /* Find and return an interface within 'br' named 'iface_name'. */
541 static const struct ovsrec_interface *
542 find_interface(const struct ovsrec_bridge *br, const char *iface_name,
543 struct ovsrec_port **portp)
547 for (i = 0; i < br->n_ports; i++) {
548 struct ovsrec_port *port = br->ports[i];
551 for (j = 0; j < port->n_interfaces; j++) {
552 struct ovsrec_interface *iface = port->interfaces[j];
553 if (!strcmp(iface->name, iface_name)) {
565 del_bridge(struct ovsdb_idl *idl,
566 const struct ovsrec_open_vswitch *ovs, const char *br_name)
568 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
569 struct ovsrec_bridge **bridges;
570 struct ovsdb_idl_txn *txn;
574 VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
578 txn = ovsdb_idl_txn_create(idl);
580 ovsdb_idl_txn_add_comment(txn, "ovs-brcompatd: delbr %s", br_name);
582 /* Remove 'br' from the vswitch's list of bridges. */
583 bridges = xmalloc(sizeof *ovs->bridges * ovs->n_bridges);
584 for (i = n = 0; i < ovs->n_bridges; i++) {
585 if (ovs->bridges[i] != br) {
586 bridges[n++] = ovs->bridges[i];
589 ovsrec_open_vswitch_set_bridges(ovs, bridges, n);
592 return commit_txn(txn, true);
596 parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
597 const char **port_name, uint64_t *count, uint64_t *skip)
599 static const struct nl_policy policy[] = {
600 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
601 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
602 [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
603 [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
605 struct nlattr *attrs[ARRAY_SIZE(policy)];
607 if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
608 attrs, ARRAY_SIZE(policy))
609 || (br_name && !attrs[BRC_GENL_A_DP_NAME])
610 || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
611 || (count && !attrs[BRC_GENL_A_FDB_COUNT])
612 || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
616 *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
618 *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
621 *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
624 *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
627 *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
632 /* Composes and returns a reply to a request made by the datapath with Netlink
633 * sequence number 'seq' and error code 'error'. The caller may add additional
634 * attributes to the message, then it may send it with send_reply(). */
635 static struct ofpbuf *
636 compose_reply(uint32_t seq, int error)
638 struct ofpbuf *reply = ofpbuf_new(4096);
639 nl_msg_put_genlmsghdr(reply, 32, brc_family, NLM_F_REQUEST,
640 BRC_GENL_C_DP_RESULT, 1);
641 ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
642 nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
646 /* Sends 'reply' to the datapath and frees it. */
648 send_reply(struct ofpbuf *reply)
650 int retval = nl_sock_send(brc_sock, reply, false);
652 VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
655 ofpbuf_delete(reply);
658 /* Composes and sends a reply to a request made by the datapath with Netlink
659 * sequence number 'seq' and error code 'error'. */
661 send_simple_reply(uint32_t seq, int error)
663 send_reply(compose_reply(seq, error));
667 handle_bridge_cmd(struct ovsdb_idl *idl,
668 const struct ovsrec_open_vswitch *ovs,
669 struct ofpbuf *buffer, bool add)
675 error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
680 retval = (add ? add_bridge : del_bridge)(idl, ovs, br_name);
681 VLOG_INFO_RL(&rl, "%sbr %s: %s",
682 add ? "add" : "del", br_name, strerror(retval));
683 } while (retval == EAGAIN);
685 send_simple_reply(seq, error);
690 static const struct nl_policy brc_port_policy[] = {
691 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
692 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
696 handle_port_cmd(struct ovsdb_idl *idl,
697 const struct ovsrec_open_vswitch *ovs,
698 struct ofpbuf *buffer, bool add)
700 const char *cmd_name = add ? "add-if" : "del-if";
701 const char *br_name, *port_name;
705 error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
707 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
710 VLOG_WARN("%s %s %s: no bridge named %s",
711 cmd_name, br_name, port_name, br_name);
713 } else if (!netdev_exists(port_name)) {
714 VLOG_WARN("%s %s %s: no network device named %s",
715 cmd_name, br_name, port_name, port_name);
719 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
722 ovsdb_idl_txn_add_comment(txn, "ovs-brcompatd: add-if %s",
724 add_port(ovs, br, port_name);
726 const struct ovsrec_port *port = find_port(br, port_name);
728 ovsdb_idl_txn_add_comment(txn,
729 "ovs-brcompatd: del-if %s",
735 error = commit_txn(txn, true);
736 VLOG_INFO_RL(&rl, "%s %s %s: %s",
737 cmd_name, br_name, port_name, strerror(error));
738 } while (error == EAGAIN);
740 send_simple_reply(seq, error);
746 /* The caller is responsible for freeing '*ovs_name' if the call is
749 linux_bridge_to_ovs_bridge(const struct ovsrec_open_vswitch *ovs,
750 const char *linux_name,
751 const struct ovsrec_bridge **ovs_bridge,
754 *ovs_bridge = find_bridge(ovs, linux_name);
756 /* Bridge name is the same. We are interested in VLAN 0. */
760 /* No such Open vSwitch bridge 'linux_name', but there might be an
761 * internal port named 'linux_name' on some other bridge
762 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
763 * port 'linux_name' on the bridge named 'ovs_bridge'. */
766 for (i = 0; i < ovs->n_bridges; i++) {
767 const struct ovsrec_bridge *br = ovs->bridges[i];
769 for (j = 0; j < br->n_ports; j++) {
770 const struct ovsrec_port *port = br->ports[j];
772 if (!strcmp(port->name, linux_name)) {
774 *br_vlan = port->n_tag ? *port->tag : -1;
785 handle_fdb_query_cmd(const struct ovsrec_open_vswitch *ovs,
786 struct ofpbuf *buffer)
788 /* This structure is copied directly from the Linux 2.6.30 header files.
789 * It would be more straightforward to #include <linux/if_bridge.h>, but
790 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
791 * with old header files won't have it. */
796 __u32 ageing_timer_value;
805 struct mac *local_macs;
809 /* Impedance matching between the vswitchd and Linux kernel notions of what
810 * a bridge is. The kernel only handles a single VLAN per bridge, but
811 * vswitchd can deal with all the VLANs on a single bridge. We have to
812 * pretend that the former is the case even though the latter is the
814 const char *linux_name; /* Name used by brctl. */
815 const struct ovsrec_bridge *ovs_bridge; /* Bridge used by ovs-vswitchd. */
816 int br_vlan; /* VLAN tag. */
819 struct ofpbuf query_data;
820 const char *iface_name;
821 struct ofpbuf *reply;
822 char *unixctl_command;
823 uint64_t count, skip;
829 /* Parse the command received from brcompat_mod. */
830 error = parse_command(buffer, &seq, &linux_name, NULL, &count, &skip);
835 /* Figure out vswitchd bridge and VLAN. */
836 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
837 &ovs_bridge, &br_vlan);
839 send_simple_reply(seq, error);
843 /* Fetch the forwarding database using ovs-appctl. */
844 unixctl_command = xasprintf("fdb/show %s", ovs_bridge->name);
845 error = execute_appctl_command(unixctl_command, &output);
846 free(unixctl_command);
848 send_simple_reply(seq, error);
852 /* Fetch the MAC address for each interface on the bridge, so that we can
853 * fill in the is_local field in the response. */
855 get_bridge_ifaces(ovs_bridge, &ifaces, br_vlan);
856 local_macs = xmalloc(sset_count(&ifaces) * sizeof *local_macs);
858 SSET_FOR_EACH (iface_name, &ifaces) {
859 struct mac *mac = &local_macs[n_local_macs];
860 struct netdev *netdev;
862 error = netdev_open_default(iface_name, &netdev);
864 if (!netdev_get_etheraddr(netdev, mac->addr)) {
867 netdev_close(netdev);
870 sset_destroy(&ifaces);
872 /* Parse the response from ovs-appctl and convert it to binary format to
873 * pass back to the kernel. */
874 ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
876 strtok_r(output, "\n", &save_ptr); /* Skip header line. */
878 struct __fdb_entry *entry;
880 uint8_t mac[ETH_ADDR_LEN];
884 line = strtok_r(NULL, "\n", &save_ptr);
889 if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
890 &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
891 != 2 + ETH_ADDR_SCAN_COUNT + 1) {
892 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
893 VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
897 if (vlan != br_vlan) {
906 /* Is this the MAC address of an interface on the bridge? */
908 for (i = 0; i < n_local_macs; i++) {
909 if (eth_addr_equals(local_macs[i].addr, mac)) {
915 entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
916 memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
917 entry->port_no = port & 0xff;
918 entry->is_local = is_local;
919 entry->ageing_timer_value = age * HZ;
920 entry->port_hi = (port & 0xff00) >> 8;
927 /* Compose and send reply to datapath. */
928 reply = compose_reply(seq, 0);
929 nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
930 query_data.data, query_data.size);
934 ofpbuf_uninit(&query_data);
941 send_ifindex_reply(uint32_t seq, struct sset *ifaces)
943 struct ofpbuf *reply;
948 /* Convert 'ifaces' into ifindexes. */
950 indices = xmalloc(sset_count(ifaces) * sizeof *indices);
951 SSET_FOR_EACH (iface, ifaces) {
952 int ifindex = if_nametoindex(iface);
954 indices[n_indices++] = ifindex;
958 /* Compose and send reply. */
959 reply = compose_reply(seq, 0);
960 nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
961 indices, n_indices * sizeof *indices);
969 handle_get_bridges_cmd(const struct ovsrec_open_vswitch *ovs,
970 struct ofpbuf *buffer)
979 /* Parse Netlink command.
981 * The command doesn't actually have any arguments, but we need the
982 * sequence number to send the reply. */
983 error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
988 /* Get all the real bridges and all the fake ones. */
990 for (i = 0; i < ovs->n_bridges; i++) {
991 const struct ovsrec_bridge *br = ovs->bridges[i];
993 sset_add(&bridges, br->name);
994 for (j = 0; j < br->n_ports; j++) {
995 const struct ovsrec_port *port = br->ports[j];
997 if (port->fake_bridge) {
998 sset_add(&bridges, port->name);
1003 send_ifindex_reply(seq, &bridges);
1004 sset_destroy(&bridges);
1010 handle_get_ports_cmd(const struct ovsrec_open_vswitch *ovs,
1011 struct ofpbuf *buffer)
1015 const char *linux_name;
1016 const struct ovsrec_bridge *ovs_bridge;
1023 /* Parse Netlink command. */
1024 error = parse_command(buffer, &seq, &linux_name, NULL, NULL, NULL);
1029 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
1030 &ovs_bridge, &br_vlan);
1032 send_simple_reply(seq, error);
1037 get_bridge_ports(ovs_bridge, &ports, br_vlan);
1038 sset_find_and_delete(&ports, linux_name);
1039 send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
1040 sset_destroy(&ports);
1045 static struct ofpbuf *
1046 brc_recv_update__(void)
1049 struct ofpbuf *buffer;
1052 retval = nl_sock_recv(brc_sock, &buffer, false);
1055 if (nl_msg_nlmsgerr(buffer, NULL)
1056 || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE) {
1068 VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
1071 ofpbuf_delete(buffer);
1076 brc_recv_update(struct ovsdb_idl *idl)
1078 struct ofpbuf *buffer;
1079 struct genlmsghdr *genlmsghdr;
1080 const struct ovsrec_open_vswitch *ovs;
1082 buffer = brc_recv_update__();
1087 genlmsghdr = nl_msg_genlmsghdr(buffer);
1089 VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
1093 if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
1094 VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
1095 nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
1099 /* Get the Open vSwitch configuration. Just drop the request on the floor
1100 * if a valid configuration doesn't exist. (We could check this earlier,
1101 * but we want to drain pending Netlink messages even when there is no Open
1102 * vSwitch configuration.) */
1103 ovs = ovsrec_open_vswitch_first(idl);
1105 VLOG_WARN_RL(&rl, "could not find valid configuration to update");
1109 switch (genlmsghdr->cmd) {
1110 case BRC_GENL_C_DP_ADD:
1111 handle_bridge_cmd(idl, ovs, buffer, true);
1114 case BRC_GENL_C_DP_DEL:
1115 handle_bridge_cmd(idl, ovs, buffer, false);
1118 case BRC_GENL_C_PORT_ADD:
1119 handle_port_cmd(idl, ovs, buffer, true);
1122 case BRC_GENL_C_PORT_DEL:
1123 handle_port_cmd(idl, ovs, buffer, false);
1126 case BRC_GENL_C_FDB_QUERY:
1127 handle_fdb_query_cmd(ovs, buffer);
1130 case BRC_GENL_C_GET_BRIDGES:
1131 handle_get_bridges_cmd(ovs, buffer);
1134 case BRC_GENL_C_GET_PORTS:
1135 handle_get_ports_cmd(ovs, buffer);
1139 VLOG_WARN_RL(&rl, "received unknown brc netlink command: %d\n",
1145 ofpbuf_delete(buffer);
1149 /* Check for interface configuration changes announced through RTNL. */
1151 rtnl_recv_update(struct ovsdb_idl *idl,
1152 const struct ovsrec_open_vswitch *ovs)
1156 int error = nl_sock_recv(rtnl_sock, &buf, false);
1157 if (error == EAGAIN) {
1158 /* Nothing to do. */
1159 } else if (error == ENOBUFS) {
1160 VLOG_WARN_RL(&rl, "network monitor socket overflowed");
1162 VLOG_WARN_RL(&rl, "error on network monitor socket: %s",
1165 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1166 struct nlmsghdr *nlh;
1167 struct ifinfomsg *iim;
1169 nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
1170 iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
1172 VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
1177 if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1178 rtnlgrp_link_policy,
1179 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1180 VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
1184 if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
1185 const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
1186 char br_name[IFNAMSIZ];
1187 uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
1189 if (!if_indextoname(br_idx, br_name)) {
1194 if (!netdev_exists(port_name)) {
1195 /* Network device is really gone. */
1196 struct ovsdb_idl_txn *txn;
1197 const struct ovsrec_interface *iface;
1198 struct ovsrec_port *port;
1199 struct ovsrec_bridge *br;
1201 VLOG_INFO("network device %s destroyed, "
1202 "removing from bridge %s", port_name, br_name);
1204 br = find_bridge(ovs, br_name);
1206 VLOG_WARN("no bridge named %s from which to remove %s",
1207 br_name, port_name);
1212 txn = ovsdb_idl_txn_create(idl);
1214 iface = find_interface(br, port_name, &port);
1216 del_interface(br, port, iface);
1217 ovsdb_idl_txn_add_comment(txn,
1218 "ovs-brcompatd: destroy port %s",
1222 commit_txn(txn, false);
1224 /* A network device by that name exists even though the kernel
1225 * told us it had disappeared. Probably, what happened was
1228 * 1. Device destroyed.
1229 * 2. Notification sent to us.
1230 * 3. New device created with same name as old one.
1231 * 4. ovs-brcompatd notified, removes device from bridge.
1233 * There's no a priori reason that in this situation that the
1234 * new device with the same name should remain in the bridge;
1235 * on the contrary, that would be unexpected. *But* there is
1236 * one important situation where, if we do this, bad things
1237 * happen. This is the case of XenServer Tools version 5.0.0,
1238 * which on boot of a Windows VM cause something like this to
1239 * happen on the Xen host:
1241 * i. Create tap1.0 and vif1.0.
1242 * ii. Delete tap1.0.
1243 * iii. Delete vif1.0.
1244 * iv. Re-create vif1.0.
1246 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1247 * neither does a VM without Tools installed at all.@.)
1249 * Steps iii and iv happen within a few seconds of each other.
1250 * Step iv causes /etc/xensource/scripts/vif to run, which in
1251 * turn calls ovs-cfg-mod to add the new device to the bridge.
1252 * If step iv happens after step 4 (in our first list of
1253 * steps), then all is well, but if it happens between 3 and 4
1254 * (which can easily happen if ovs-brcompatd has to wait to
1255 * lock the configuration file), then we will remove the new
1256 * incarnation from the bridge instead of the old one!
1258 * So, to avoid this problem, we do nothing here. This is
1259 * strictly incorrect except for this one particular case, and
1260 * perhaps that will bite us someday. If that happens, then we
1261 * will have to somehow track network devices by ifindex, since
1262 * a new device will have a new ifindex even if it has the same
1263 * name as an old device.
1265 VLOG_INFO("kernel reported network device %s removed but "
1266 "a device by that name exists (XS Tools 5.0.0?)",
1275 main(int argc, char *argv[])
1277 extern struct vlog_module VLM_reconnect;
1278 struct unixctl_server *unixctl;
1280 struct ovsdb_idl *idl;
1283 proctitle_init(argc, argv);
1284 set_program_name(argv[0]);
1285 vlog_set_levels(NULL, VLF_CONSOLE, VLL_WARN);
1286 vlog_set_levels(&VLM_reconnect, VLF_ANY_FACILITY, VLL_WARN);
1288 remote = parse_options(argc, argv);
1289 signal(SIGPIPE, SIG_IGN);
1295 retval = unixctl_server_create(NULL, &unixctl);
1300 if (brc_open(&brc_sock)) {
1301 VLOG_FATAL("could not open brcompat socket. Check "
1302 "\"brcompat\" kernel module.");
1305 if (prune_timeout) {
1308 error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
1310 VLOG_FATAL("could not create rtnetlink socket (%s)",
1314 error = nl_sock_join_mcgroup(rtnl_sock, RTNLGRP_LINK);
1316 VLOG_FATAL("could not join RTNLGRP_LINK multicast group (%s)",
1321 daemonize_complete();
1323 idl = ovsdb_idl_create(remote, &ovsrec_idl_class, true);
1326 const struct ovsrec_open_vswitch *ovs;
1330 unixctl_server_run(unixctl);
1331 brc_recv_update(idl);
1333 ovs = ovsrec_open_vswitch_first(idl);
1334 if (!ovs && ovsdb_idl_has_ever_connected(idl)) {
1335 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1336 VLOG_WARN_RL(&rl, "%s: database does not contain any Open vSwitch "
1337 "configuration", remote);
1341 /* If 'prune_timeout' is non-zero, we actively prune from the
1342 * configuration of port entries that are no longer valid. We
1345 * 1) The kernel explicitly notifies us of removed ports
1346 * through the RTNL messages.
1348 * 2) We periodically check all ports associated with bridges
1349 * to see if they no longer exist.
1351 if (ovs && prune_timeout) {
1352 rtnl_recv_update(idl, ovs);
1353 nl_sock_wait(rtnl_sock, POLLIN);
1354 poll_timer_wait(prune_timeout);
1358 nl_sock_wait(brc_sock, POLLIN);
1359 ovsdb_idl_wait(idl);
1360 unixctl_server_wait(unixctl);
1365 ovsdb_idl_destroy(idl);
1371 validate_appctl_command(void)
1377 for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
1379 /* Nothing to do. */
1380 } else if (p[1] == 's') {
1383 VLOG_FATAL("only '%%s' and '%%%%' allowed in --appctl-command");
1387 VLOG_FATAL("'%%s' must appear exactly once in --appctl-command");
1392 parse_options(int argc, char *argv[])
1398 LEAK_CHECKER_OPTION_ENUMS,
1401 static struct option long_options[] = {
1402 {"help", no_argument, NULL, 'h'},
1403 {"version", no_argument, NULL, 'V'},
1404 {"prune-timeout", required_argument, NULL, OPT_PRUNE_TIMEOUT},
1405 {"appctl-command", required_argument, NULL, OPT_APPCTL_COMMAND},
1406 DAEMON_LONG_OPTIONS,
1408 LEAK_CHECKER_LONG_OPTIONS,
1411 char *short_options = long_options_to_short_options(long_options);
1413 appctl_command = xasprintf("%s/ovs-appctl %%s", ovs_bindir());
1417 c = getopt_long(argc, argv, short_options, long_options, NULL);
1428 OVS_PRINT_VERSION(0, 0);
1431 case OPT_PRUNE_TIMEOUT:
1432 prune_timeout = atoi(optarg) * 1000;
1435 case OPT_APPCTL_COMMAND:
1436 appctl_command = optarg;
1439 VLOG_OPTION_HANDLERS
1440 DAEMON_OPTION_HANDLERS
1441 LEAK_CHECKER_OPTION_HANDLERS
1450 free(short_options);
1452 validate_appctl_command();
1458 VLOG_FATAL("database socket is non-option argument; "
1459 "use --help for usage");
1468 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1469 "usage: %s [OPTIONS] CONFIG\n"
1470 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1471 program_name, program_name);
1472 printf("\nConfiguration options:\n"
1473 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1474 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1478 printf("\nOther options:\n"
1479 " -h, --help display this help message\n"
1480 " -V, --version display version information\n");
1481 leak_checker_usage();
1482 printf("\nThe default appctl command is:\n%s\n", appctl_command);