1 /* Copyright (c) 2008, 2009, 2010 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 before pruning port entries that
79 * no longer exist. If set to zero, ports are never pruned. */
80 static int prune_timeout = 5000;
82 /* Shell command to execute (via popen()) to send a control command to the
83 * running ovs-vswitchd process. The string must contain one instance of %s,
84 * which is replaced by the control command. */
85 static char *appctl_command;
87 /* Netlink socket to listen for interface changes. */
88 static struct nl_sock *rtnl_sock;
90 /* Netlink socket to bridge compatibility kernel module. */
91 static struct nl_sock *brc_sock;
93 /* The Generic Netlink family number used for bridge compatibility. */
94 static int brc_family;
96 static const struct nl_policy brc_multicast_policy[] = {
97 [BRC_GENL_A_MC_GROUP] = {.type = NL_A_U32 }
100 static const struct nl_policy rtnlgrp_link_policy[] = {
101 [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false },
102 [IFLA_MASTER] = { .type = NL_A_U32, .optional = true },
106 lookup_brc_multicast_group(int *multicast_group)
108 struct nl_sock *sock;
109 struct ofpbuf request, *reply;
110 struct nlattr *attrs[ARRAY_SIZE(brc_multicast_policy)];
113 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
117 ofpbuf_init(&request, 0);
118 nl_msg_put_genlmsghdr(&request, sock, 0, brc_family,
119 NLM_F_REQUEST, BRC_GENL_C_QUERY_MC, 1);
120 retval = nl_sock_transact(sock, &request, &reply);
121 ofpbuf_uninit(&request);
123 nl_sock_destroy(sock);
126 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
127 brc_multicast_policy, attrs,
128 ARRAY_SIZE(brc_multicast_policy))) {
129 nl_sock_destroy(sock);
130 ofpbuf_delete(reply);
133 *multicast_group = nl_attr_get_u32(attrs[BRC_GENL_A_MC_GROUP]);
134 nl_sock_destroy(sock);
135 ofpbuf_delete(reply);
140 /* Opens a socket for brcompat notifications. Returns 0 if successful,
141 * otherwise a positive errno value. */
143 brc_open(struct nl_sock **sock)
145 int multicast_group = 0;
148 retval = nl_lookup_genl_family(BRC_GENL_FAMILY_NAME, &brc_family);
153 retval = lookup_brc_multicast_group(&multicast_group);
158 retval = nl_sock_create(NETLINK_GENERIC, multicast_group, 0, 0, sock);
166 static const struct nl_policy brc_dp_policy[] = {
167 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
170 static struct ovsrec_bridge *
171 find_bridge(const struct ovsrec_open_vswitch *ovs, const char *br_name)
175 for (i = 0; i < ovs->n_bridges; i++) {
176 if (!strcmp(br_name, ovs->bridges[i]->name)) {
177 return ovs->bridges[i];
185 execute_appctl_command(const char *unixctl_command, char **output)
187 char *stdout_log, *stderr_log;
193 argv[2] = xasprintf(appctl_command, unixctl_command);
196 /* Run process and log status. */
197 error = process_run_capture(argv, &stdout_log, &stderr_log, &status);
199 VLOG_ERR("failed to execute %s command via ovs-appctl: %s",
200 unixctl_command, strerror(error));
202 char *msg = process_status_msg(status);
203 VLOG_ERR("ovs-appctl exited with error (%s)", msg);
208 /* Deal with stdout_log. */
210 *output = stdout_log;
215 /* Deal with stderr_log */
216 if (stderr_log && *stderr_log) {
217 VLOG_INFO("ovs-appctl wrote to stderr:\n%s", stderr_log);
227 do_get_bridge_parts(const struct ovsrec_bridge *br, struct svec *parts,
228 int vlan, bool break_down_bonds)
234 for (i = 0; i < br->n_ports; i++) {
235 const struct ovsrec_port *port = br->ports[i];
237 svec_add(&ports, port->name);
239 int port_vlan = port->n_tag ? *port->tag : 0;
240 if (vlan != port_vlan) {
244 if (break_down_bonds) {
245 for (j = 0; j < port->n_interfaces; j++) {
246 const struct ovsrec_interface *iface = port->interfaces[j];
247 svec_add(parts, iface->name);
250 svec_add(parts, port->name);
253 svec_destroy(&ports);
256 /* Add all the interfaces for 'bridge' to 'ifaces', breaking bonded interfaces
257 * down into their constituent parts.
259 * If 'vlan' < 0, all interfaces on 'bridge' are reported. If 'vlan' == 0,
260 * then only interfaces for trunk ports or ports with implicit VLAN 0 are
261 * reported. If 'vlan' > 0, only interfaces with implicit VLAN 'vlan' are
264 get_bridge_ifaces(const struct ovsrec_bridge *br, struct svec *ifaces,
267 do_get_bridge_parts(br, ifaces, vlan, true);
270 /* Add all the ports for 'bridge' to 'ports'. Bonded ports are reported under
271 * the bond name, not broken down into their constituent interfaces.
273 * If 'vlan' < 0, all ports on 'bridge' are reported. If 'vlan' == 0, then
274 * only trunk ports or ports with implicit VLAN 0 are reported. If 'vlan' > 0,
275 * only port with implicit VLAN 'vlan' are reported. */
277 get_bridge_ports(const struct ovsrec_bridge *br, struct svec *ports,
280 do_get_bridge_parts(br, ports, vlan, false);
284 /* Go through the configuration file and remove any ports that no longer
285 * exist associated with a bridge. */
290 struct svec bridges, delete;
292 if (cfg_lock(NULL, 0)) {
293 /* Couldn't lock config file. */
299 cfg_get_subsections(&bridges, "bridge");
300 for (i=0; i<bridges.n; i++) {
301 const char *br_name = bridges.names[i];
304 /* Check that each bridge interface exists. */
306 get_bridge_ifaces(br_name, &ifaces, -1);
307 for (j = 0; j < ifaces.n; j++) {
308 const char *iface_name = ifaces.names[j];
310 /* The local port and internal ports are created and destroyed by
311 * ovs-vswitchd itself, so don't bother checking for them at all.
312 * In practice, they might not exist if ovs-vswitchd hasn't
313 * finished reloading since the configuration file was updated. */
314 if (!strcmp(iface_name, br_name)
315 || cfg_get_bool(0, "iface.%s.internal", iface_name)) {
319 if (!netdev_exists(iface_name)) {
320 VLOG_INFO_RL(&rl, "removing dead interface %s from %s",
321 iface_name, br_name);
322 svec_add(&delete, iface_name);
325 svec_destroy(&ifaces);
327 svec_destroy(&bridges);
332 for (i = 0; i < delete.n; i++) {
333 cfg_del_match("bridge.*.port=%s", delete.names[i]);
334 cfg_del_match("bonding.*.slave=%s", delete.names[i]);
341 svec_destroy(&delete);
345 static struct ovsdb_idl_txn *
346 txn_from_openvswitch(const struct ovsrec_open_vswitch *ovs)
348 return ovsdb_idl_txn_get(&ovs->header_);
352 port_is_fake_bridge(const struct ovsrec_port *port)
354 return (port->fake_bridge
356 && *port->tag >= 1 && *port->tag <= 4095);
360 ovs_insert_bridge(const struct ovsrec_open_vswitch *ovs,
361 struct ovsrec_bridge *bridge)
363 struct ovsrec_bridge **bridges;
366 bridges = xmalloc(sizeof *ovs->bridges * (ovs->n_bridges + 1));
367 for (i = 0; i < ovs->n_bridges; i++) {
368 bridges[i] = ovs->bridges[i];
370 bridges[ovs->n_bridges] = bridge;
371 ovsrec_open_vswitch_set_bridges(ovs, bridges, ovs->n_bridges + 1);
376 where_uuid_equals(const struct uuid *uuid)
381 json_string_create("_uuid"),
382 json_string_create("=="),
384 json_string_create("uuid"),
385 json_string_create_nocopy(
386 xasprintf(UUID_FMT, UUID_ARGS(uuid))))));
389 /* Commits 'txn'. If 'wait_for_reload' is true, also waits for Open vSwitch to
390 reload the configuration before returning.
392 Returns EAGAIN if the caller should try the operation again, 0 on success,
393 otherwise a positive errno value. */
395 commit_txn(struct ovsdb_idl_txn *txn, bool wait_for_reload)
397 struct ovsdb_idl *idl = ovsdb_idl_txn_get_idl (txn);
398 enum ovsdb_idl_txn_status status;
399 int64_t next_cfg = 0;
401 if (wait_for_reload) {
402 const struct ovsrec_open_vswitch *ovs = ovsrec_open_vswitch_first(idl);
403 struct json *where = where_uuid_equals(&ovs->header_.uuid);
404 ovsdb_idl_txn_increment(txn, "Open_vSwitch", "next_cfg", where);
407 status = ovsdb_idl_txn_commit_block(txn);
408 if (wait_for_reload && status == TXN_SUCCESS) {
409 next_cfg = ovsdb_idl_txn_get_increment_new_value(txn);
411 ovsdb_idl_txn_destroy(txn);
418 VLOG_ERR_RL(&rl, "OVSDB transaction unexpectedly aborted");
425 if (wait_for_reload) {
427 /* We can't use 'ovs' any longer because ovsdb_idl_run() can
429 const struct ovsrec_open_vswitch *ovs2;
432 OVSREC_OPEN_VSWITCH_FOR_EACH (ovs2, idl) {
433 if (ovs2->cur_cfg >= next_cfg) {
445 VLOG_ERR_RL(&rl, "OVSDB transaction needs retry");
449 VLOG_ERR_RL(&rl, "OVSDB transaction failed: %s",
450 ovsdb_idl_txn_get_error(txn));
459 add_bridge(struct ovsdb_idl *idl, const struct ovsrec_open_vswitch *ovs,
462 struct ovsrec_bridge *br;
463 struct ovsrec_port *port;
464 struct ovsrec_interface *iface;
465 struct ovsdb_idl_txn *txn;
467 if (find_bridge(ovs, br_name)) {
468 VLOG_WARN("addbr %s: bridge %s exists", br_name, br_name);
470 } else if (netdev_exists(br_name)) {
473 for (i = 0; i < ovs->n_bridges; i++) {
475 struct ovsrec_bridge *br_cfg = ovs->bridges[i];
477 for (j = 0; j < br_cfg->n_ports; j++) {
478 if (port_is_fake_bridge(br_cfg->ports[j])) {
479 VLOG_WARN("addbr %s: %s exists as a fake bridge",
486 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
487 "device named %s already exists",
488 br_name, br_name, br_name);
492 txn = ovsdb_idl_txn_create(idl);
494 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
495 ovsrec_interface_set_name(iface, br_name);
497 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
498 ovsrec_port_set_name(port, br_name);
499 ovsrec_port_set_interfaces(port, &iface, 1);
501 br = ovsrec_bridge_insert(txn_from_openvswitch(ovs));
502 ovsrec_bridge_set_name(br, br_name);
503 ovsrec_bridge_set_ports(br, &port, 1);
505 ovs_insert_bridge(ovs, br);
507 return commit_txn(txn, true);
511 add_port(const struct ovsrec_open_vswitch *ovs,
512 const struct ovsrec_bridge *br, const char *port_name)
514 struct ovsrec_interface *iface;
515 struct ovsrec_port *port;
516 struct ovsrec_port **ports;
519 /* xxx Check conflicts? */
520 iface = ovsrec_interface_insert(txn_from_openvswitch(ovs));
521 ovsrec_interface_set_name(iface, port_name);
523 port = ovsrec_port_insert(txn_from_openvswitch(ovs));
524 ovsrec_port_set_name(port, port_name);
525 ovsrec_port_set_interfaces(port, &iface, 1);
527 ports = xmalloc(sizeof *br->ports * (br->n_ports + 1));
528 for (i = 0; i < br->n_ports; i++) {
529 ports[i] = br->ports[i];
531 ports[br->n_ports] = port;
532 ovsrec_bridge_set_ports(br, ports, br->n_ports + 1);
537 del_port(const struct ovsrec_bridge *br, const char *port_name)
540 struct ovsrec_port *port_rec = NULL;
542 for (i = 0; i < br->n_ports; i++) {
543 struct ovsrec_port *port = br->ports[i];
544 if (!strcmp(port_name, port->name)) {
547 for (j = 0; j < port->n_interfaces; j++) {
548 struct ovsrec_interface *iface = port->interfaces[j];
549 if (!strcmp(port_name, iface->name)) {
550 ovsrec_interface_delete(iface);
555 /* xxx Probably can move this into the "for" loop. */
557 struct ovsrec_port **ports;
560 ports = xmalloc(sizeof *br->ports * br->n_ports);
561 for (i = n = 0; i < br->n_ports; i++) {
562 if (br->ports[i] != port_rec) {
563 ports[n++] = br->ports[i];
566 ovsrec_bridge_set_ports(br, ports, n);
569 ovsrec_port_delete(port_rec);
574 del_bridge(struct ovsdb_idl *idl,
575 const struct ovsrec_open_vswitch *ovs, const char *br_name)
577 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
578 struct ovsrec_bridge **bridges;
579 struct ovsdb_idl_txn *txn;
583 VLOG_WARN("delbr %s: no bridge named %s", br_name, br_name);
587 txn = ovsdb_idl_txn_create(idl);
589 del_port(br, br_name);
591 bridges = xmalloc(sizeof *ovs->bridges * ovs->n_bridges);
592 for (i = n = 0; i < ovs->n_bridges; i++) {
593 if (ovs->bridges[i] != br) {
594 bridges[n++] = ovs->bridges[i];
597 ovsrec_open_vswitch_set_bridges(ovs, bridges, n);
600 /* Delete the bridge itself. */
601 ovsrec_bridge_delete(br);
603 return commit_txn(txn, true);
607 parse_command(struct ofpbuf *buffer, uint32_t *seq, const char **br_name,
608 const char **port_name, uint64_t *count, uint64_t *skip)
610 static const struct nl_policy policy[] = {
611 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING, .optional = true },
612 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING, .optional = true },
613 [BRC_GENL_A_FDB_COUNT] = { .type = NL_A_U64, .optional = true },
614 [BRC_GENL_A_FDB_SKIP] = { .type = NL_A_U64, .optional = true },
616 struct nlattr *attrs[ARRAY_SIZE(policy)];
618 if (!nl_policy_parse(buffer, NLMSG_HDRLEN + GENL_HDRLEN, policy,
619 attrs, ARRAY_SIZE(policy))
620 || (br_name && !attrs[BRC_GENL_A_DP_NAME])
621 || (port_name && !attrs[BRC_GENL_A_PORT_NAME])
622 || (count && !attrs[BRC_GENL_A_FDB_COUNT])
623 || (skip && !attrs[BRC_GENL_A_FDB_SKIP])) {
627 *seq = ((struct nlmsghdr *) buffer->data)->nlmsg_seq;
629 *br_name = nl_attr_get_string(attrs[BRC_GENL_A_DP_NAME]);
632 *port_name = nl_attr_get_string(attrs[BRC_GENL_A_PORT_NAME]);
635 *count = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_COUNT]);
638 *skip = nl_attr_get_u64(attrs[BRC_GENL_A_FDB_SKIP]);
643 /* Composes and returns a reply to a request made by the datapath with Netlink
644 * sequence number 'seq' and error code 'error'. The caller may add additional
645 * attributes to the message, then it may send it with send_reply(). */
646 static struct ofpbuf *
647 compose_reply(uint32_t seq, int error)
649 struct ofpbuf *reply = ofpbuf_new(4096);
650 nl_msg_put_genlmsghdr(reply, brc_sock, 32, brc_family, NLM_F_REQUEST,
651 BRC_GENL_C_DP_RESULT, 1);
652 ((struct nlmsghdr *) reply->data)->nlmsg_seq = seq;
653 nl_msg_put_u32(reply, BRC_GENL_A_ERR_CODE, error);
657 /* Sends 'reply' to the datapath and frees it. */
659 send_reply(struct ofpbuf *reply)
661 int retval = nl_sock_send(brc_sock, reply, false);
663 VLOG_WARN_RL(&rl, "replying to brcompat request: %s",
666 ofpbuf_delete(reply);
669 /* Composes and sends a reply to a request made by the datapath with Netlink
670 * sequence number 'seq' and error code 'error'. */
672 send_simple_reply(uint32_t seq, int error)
674 send_reply(compose_reply(seq, error));
678 handle_bridge_cmd(struct ovsdb_idl *idl,
679 const struct ovsrec_open_vswitch *ovs,
680 struct ofpbuf *buffer, bool add)
686 error = parse_command(buffer, &seq, &br_name, NULL, NULL, NULL);
691 retval = (add ? add_bridge : del_bridge)(idl, ovs, br_name);
692 VLOG_INFO_RL(&rl, "%sbr %s: %s",
693 add ? "add" : "del", br_name, strerror(retval));
694 } while (retval == EAGAIN);
696 send_simple_reply(seq, error);
701 static const struct nl_policy brc_port_policy[] = {
702 [BRC_GENL_A_DP_NAME] = { .type = NL_A_STRING },
703 [BRC_GENL_A_PORT_NAME] = { .type = NL_A_STRING },
707 handle_port_cmd(struct ovsdb_idl *idl,
708 const struct ovsrec_open_vswitch *ovs,
709 struct ofpbuf *buffer, bool add)
711 const char *cmd_name = add ? "add-if" : "del-if";
712 const char *br_name, *port_name;
716 error = parse_command(buffer, &seq, &br_name, &port_name, NULL, NULL);
718 struct ovsrec_bridge *br = find_bridge(ovs, br_name);
721 VLOG_WARN("%s %s %s: no bridge named %s",
722 cmd_name, br_name, port_name, br_name);
724 } else if (!netdev_exists(port_name)) {
725 VLOG_WARN("%s %s %s: no network device named %s",
726 cmd_name, br_name, port_name, port_name);
730 struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl);
732 add_port(ovs, br, port_name);
734 del_port(br, port_name);
736 error = commit_txn(txn, true);
737 VLOG_INFO_RL(&rl, "%s %s %s: %s",
738 cmd_name, br_name, port_name, strerror(error));
739 } while (error == EAGAIN);
741 send_simple_reply(seq, error);
747 /* The caller is responsible for freeing '*ovs_name' if the call is
750 linux_bridge_to_ovs_bridge(const struct ovsrec_open_vswitch *ovs,
751 const char *linux_name,
752 const struct ovsrec_bridge **ovs_bridge,
755 *ovs_bridge = find_bridge(ovs, linux_name);
757 /* Bridge name is the same. We are interested in VLAN 0. */
761 /* No such Open vSwitch bridge 'linux_name', but there might be an
762 * internal port named 'linux_name' on some other bridge
763 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
764 * port 'linux_name' on the bridge named 'ovs_bridge'. */
767 for (i = 0; i < ovs->n_bridges; i++) {
768 const struct ovsrec_bridge *br = ovs->bridges[i];
770 for (j = 0; j < br->n_ports; j++) {
771 const struct ovsrec_port *port = br->ports[j];
773 if (!strcmp(port->name, linux_name)) {
775 *br_vlan = port->n_tag ? *port->tag : -1;
786 handle_fdb_query_cmd(const struct ovsrec_open_vswitch *ovs,
787 struct ofpbuf *buffer)
789 /* This structure is copied directly from the Linux 2.6.30 header files.
790 * It would be more straightforward to #include <linux/if_bridge.h>, but
791 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
792 * with old header files won't have it. */
797 __u32 ageing_timer_value;
806 struct mac *local_macs;
810 /* Impedance matching between the vswitchd and Linux kernel notions of what
811 * a bridge is. The kernel only handles a single VLAN per bridge, but
812 * vswitchd can deal with all the VLANs on a single bridge. We have to
813 * pretend that the former is the case even though the latter is the
815 const char *linux_name; /* Name used by brctl. */
816 const struct ovsrec_bridge *ovs_bridge; /* Bridge used by ovs-vswitchd. */
817 int br_vlan; /* VLAN tag. */
820 struct ofpbuf query_data;
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(ifaces.n * sizeof *local_macs);
858 for (i = 0; i < ifaces.n; i++) {
859 const char *iface_name = ifaces.names[i];
860 struct mac *mac = &local_macs[n_local_macs];
861 struct netdev *netdev;
863 error = netdev_open_default(iface_name, &netdev);
865 if (!netdev_get_etheraddr(netdev, mac->addr)) {
868 netdev_close(netdev);
871 svec_destroy(&ifaces);
873 /* Parse the response from ovs-appctl and convert it to binary format to
874 * pass back to the kernel. */
875 ofpbuf_init(&query_data, sizeof(struct __fdb_entry) * 8);
877 strtok_r(output, "\n", &save_ptr); /* Skip header line. */
879 struct __fdb_entry *entry;
881 uint8_t mac[ETH_ADDR_LEN];
885 line = strtok_r(NULL, "\n", &save_ptr);
890 if (sscanf(line, "%d %d "ETH_ADDR_SCAN_FMT" %d",
891 &port, &vlan, ETH_ADDR_SCAN_ARGS(mac), &age)
892 != 2 + ETH_ADDR_SCAN_COUNT + 1) {
893 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
894 VLOG_INFO_RL(&rl, "fdb/show output has invalid format: %s", line);
898 if (vlan != br_vlan) {
907 /* Is this the MAC address of an interface on the bridge? */
909 for (i = 0; i < n_local_macs; i++) {
910 if (eth_addr_equals(local_macs[i].addr, mac)) {
916 entry = ofpbuf_put_uninit(&query_data, sizeof *entry);
917 memcpy(entry->mac_addr, mac, ETH_ADDR_LEN);
918 entry->port_no = port & 0xff;
919 entry->is_local = is_local;
920 entry->ageing_timer_value = age * HZ;
921 entry->port_hi = (port & 0xff00) >> 8;
928 /* Compose and send reply to datapath. */
929 reply = compose_reply(seq, 0);
930 nl_msg_put_unspec(reply, BRC_GENL_A_FDB_DATA,
931 query_data.data, query_data.size);
935 ofpbuf_uninit(&query_data);
941 send_ifindex_reply(uint32_t seq, struct svec *ifaces)
943 struct ofpbuf *reply;
949 /* Make sure that any given interface only occurs once. This shouldn't
950 * happen, but who knows what people put into their configuration files. */
951 svec_sort_unique(ifaces);
953 /* Convert 'ifaces' into ifindexes. */
955 indices = xmalloc(ifaces->n * sizeof *indices);
956 SVEC_FOR_EACH (i, iface, ifaces) {
957 int ifindex = if_nametoindex(iface);
959 indices[n_indices++] = ifindex;
963 /* Compose and send reply. */
964 reply = compose_reply(seq, 0);
965 nl_msg_put_unspec(reply, BRC_GENL_A_IFINDEXES,
966 indices, n_indices * sizeof *indices);
974 handle_get_bridges_cmd(const struct ovsrec_open_vswitch *ovs,
975 struct ofpbuf *buffer)
984 /* Parse Netlink command.
986 * The command doesn't actually have any arguments, but we need the
987 * sequence number to send the reply. */
988 error = parse_command(buffer, &seq, NULL, NULL, NULL, NULL);
993 /* Get all the real bridges and all the fake ones. */
995 for (i = 0; i < ovs->n_bridges; i++) {
996 const struct ovsrec_bridge *br = ovs->bridges[i];
998 svec_add(&bridges, br->name);
999 for (j = 0; j < br->n_ports; j++) {
1000 const struct ovsrec_port *port = br->ports[j];
1002 if (port->fake_bridge) {
1003 svec_add(&bridges, port->name);
1008 send_ifindex_reply(seq, &bridges);
1009 svec_destroy(&bridges);
1015 handle_get_ports_cmd(const struct ovsrec_open_vswitch *ovs,
1016 struct ofpbuf *buffer)
1020 const char *linux_name;
1021 const struct ovsrec_bridge *ovs_bridge;
1028 /* Parse Netlink command. */
1029 error = parse_command(buffer, &seq, &linux_name, NULL, NULL, NULL);
1034 error = linux_bridge_to_ovs_bridge(ovs, linux_name,
1035 &ovs_bridge, &br_vlan);
1037 send_simple_reply(seq, error);
1042 get_bridge_ports(ovs_bridge, &ports, br_vlan);
1044 svec_del(&ports, linux_name);
1045 send_ifindex_reply(seq, &ports); /* XXX bonds won't show up */
1046 svec_destroy(&ports);
1052 brc_recv_update(struct ovsdb_idl *idl)
1055 struct ofpbuf *buffer;
1056 struct genlmsghdr *genlmsghdr;
1057 const struct ovsrec_open_vswitch *ovs;
1061 ofpbuf_delete(buffer);
1062 retval = nl_sock_recv(brc_sock, &buffer, false);
1063 } while (retval == ENOBUFS
1065 && (nl_msg_nlmsgerr(buffer, NULL)
1066 || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE)));
1068 if (retval != EAGAIN) {
1069 VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
1074 genlmsghdr = nl_msg_genlmsghdr(buffer);
1076 VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
1080 if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) {
1081 VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
1082 nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family);
1086 /* Get the Open vSwitch configuration. Just drop the request on the floor
1087 * if a valid configuration doesn't exist. (We could check this earlier,
1088 * but we want to drain pending Netlink messages even when there is no Open
1089 * vSwitch configuration.) */
1090 ovs = ovsrec_open_vswitch_first(idl);
1092 VLOG_WARN_RL(&rl, "could not find valid configuration to update");
1096 switch (genlmsghdr->cmd) {
1097 case BRC_GENL_C_DP_ADD:
1098 handle_bridge_cmd(idl, ovs, buffer, true);
1101 case BRC_GENL_C_DP_DEL:
1102 handle_bridge_cmd(idl, ovs, buffer, false);
1105 case BRC_GENL_C_PORT_ADD:
1106 handle_port_cmd(idl, ovs, buffer, true);
1109 case BRC_GENL_C_PORT_DEL:
1110 handle_port_cmd(idl, ovs, buffer, false);
1113 case BRC_GENL_C_FDB_QUERY:
1114 handle_fdb_query_cmd(ovs, buffer);
1117 case BRC_GENL_C_GET_BRIDGES:
1118 handle_get_bridges_cmd(ovs, buffer);
1121 case BRC_GENL_C_GET_PORTS:
1122 handle_get_ports_cmd(ovs, buffer);
1126 VLOG_WARN_RL(&rl, "received unknown brc netlink command: %d\n",
1132 ofpbuf_delete(buffer);
1136 /* Check for interface configuration changes announced through RTNL. */
1138 rtnl_recv_update(struct ovsdb_idl *idl,
1139 const struct ovsrec_open_vswitch *ovs)
1143 int error = nl_sock_recv(rtnl_sock, &buf, false);
1144 if (error == EAGAIN) {
1145 /* Nothing to do. */
1146 } else if (error == ENOBUFS) {
1147 VLOG_WARN_RL(&rl, "network monitor socket overflowed");
1149 VLOG_WARN_RL(&rl, "error on network monitor socket: %s",
1152 struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)];
1153 struct nlmsghdr *nlh;
1154 struct ifinfomsg *iim;
1156 nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN);
1157 iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim);
1159 VLOG_WARN_RL(&rl, "received bad rtnl message (no ifinfomsg)");
1164 if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
1165 rtnlgrp_link_policy,
1166 attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
1167 VLOG_WARN_RL(&rl,"received bad rtnl message (policy)");
1171 if (nlh->nlmsg_type == RTM_DELLINK && attrs[IFLA_MASTER]) {
1172 const char *port_name = nl_attr_get_string(attrs[IFLA_IFNAME]);
1173 char br_name[IFNAMSIZ];
1174 uint32_t br_idx = nl_attr_get_u32(attrs[IFLA_MASTER]);
1176 if (!if_indextoname(br_idx, br_name)) {
1181 if (!netdev_exists(port_name)) {
1182 /* Network device is really gone. */
1183 struct ovsdb_idl_txn *txn;
1184 struct ovsrec_bridge *br;
1186 VLOG_INFO("network device %s destroyed, "
1187 "removing from bridge %s", port_name, br_name);
1189 br = find_bridge(ovs, br_name);
1191 VLOG_WARN("no bridge named %s from which to remove %s",
1192 br_name, port_name);
1197 txn = ovsdb_idl_txn_create(idl);
1198 del_port(br, port_name);
1199 commit_txn(txn, false);
1201 /* A network device by that name exists even though the kernel
1202 * told us it had disappeared. Probably, what happened was
1205 * 1. Device destroyed.
1206 * 2. Notification sent to us.
1207 * 3. New device created with same name as old one.
1208 * 4. ovs-brcompatd notified, removes device from bridge.
1210 * There's no a priori reason that in this situation that the
1211 * new device with the same name should remain in the bridge;
1212 * on the contrary, that would be unexpected. *But* there is
1213 * one important situation where, if we do this, bad things
1214 * happen. This is the case of XenServer Tools version 5.0.0,
1215 * which on boot of a Windows VM cause something like this to
1216 * happen on the Xen host:
1218 * i. Create tap1.0 and vif1.0.
1219 * ii. Delete tap1.0.
1220 * iii. Delete vif1.0.
1221 * iv. Re-create vif1.0.
1223 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1224 * neither does a VM without Tools installed at all.@.)
1226 * Steps iii and iv happen within a few seconds of each other.
1227 * Step iv causes /etc/xensource/scripts/vif to run, which in
1228 * turn calls ovs-cfg-mod to add the new device to the bridge.
1229 * If step iv happens after step 4 (in our first list of
1230 * steps), then all is well, but if it happens between 3 and 4
1231 * (which can easily happen if ovs-brcompatd has to wait to
1232 * lock the configuration file), then we will remove the new
1233 * incarnation from the bridge instead of the old one!
1235 * So, to avoid this problem, we do nothing here. This is
1236 * strictly incorrect except for this one particular case, and
1237 * perhaps that will bite us someday. If that happens, then we
1238 * will have to somehow track network devices by ifindex, since
1239 * a new device will have a new ifindex even if it has the same
1240 * name as an old device.
1242 VLOG_INFO("kernel reported network device %s removed but "
1243 "a device by that name exists (XS Tools 5.0.0?)",
1252 main(int argc, char *argv[])
1254 struct unixctl_server *unixctl;
1256 struct ovsdb_idl *idl;
1259 proctitle_init(argc, argv);
1260 set_program_name(argv[0]);
1263 vlog_set_levels(VLM_ANY_MODULE, VLF_CONSOLE, VLL_WARN);
1264 vlog_set_levels(VLM_reconnect, VLF_ANY_FACILITY, VLL_WARN);
1266 remote = parse_options(argc, argv);
1267 signal(SIGPIPE, SIG_IGN);
1271 die_if_already_running();
1274 retval = unixctl_server_create(NULL, &unixctl);
1279 if (brc_open(&brc_sock)) {
1280 ovs_fatal(0, "could not open brcompat socket. Check "
1281 "\"brcompat\" kernel module.");
1284 if (prune_timeout) {
1285 if (nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &rtnl_sock)) {
1286 ovs_fatal(0, "could not create rtnetlink socket");
1290 daemonize_complete();
1292 idl = ovsdb_idl_create(remote, &ovsrec_idl_class);
1295 const struct ovsrec_open_vswitch *ovs;
1299 unixctl_server_run(unixctl);
1300 brc_recv_update(idl);
1302 ovs = ovsrec_open_vswitch_first(idl);
1303 if (!ovs && ovsdb_idl_has_ever_connected(idl)) {
1304 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1305 VLOG_WARN_RL(&rl, "%s: database does not contain any Open vSwitch "
1306 "configuration", remote);
1310 /* If 'prune_timeout' is non-zero, we actively prune from the
1311 * configuration of port entries that are no longer valid. We
1314 * 1) The kernel explicitly notifies us of removed ports
1315 * through the RTNL messages.
1317 * 2) We periodically check all ports associated with bridges
1318 * to see if they no longer exist.
1320 if (ovs && prune_timeout) {
1321 rtnl_recv_update(idl, ovs);
1326 nl_sock_wait(rtnl_sock, POLLIN);
1327 poll_timer_wait(prune_timeout);
1331 nl_sock_wait(brc_sock, POLLIN);
1332 ovsdb_idl_wait(idl);
1333 unixctl_server_wait(unixctl);
1338 ovsdb_idl_destroy(idl);
1344 validate_appctl_command(void)
1350 for (p = strchr(appctl_command, '%'); p; p = strchr(p + 2, '%')) {
1352 /* Nothing to do. */
1353 } else if (p[1] == 's') {
1356 ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
1360 ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
1365 parse_options(int argc, char *argv[])
1371 LEAK_CHECKER_OPTION_ENUMS
1373 static struct option long_options[] = {
1374 {"help", no_argument, 0, 'h'},
1375 {"version", no_argument, 0, 'V'},
1376 {"prune-timeout", required_argument, 0, OPT_PRUNE_TIMEOUT},
1377 {"appctl-command", required_argument, 0, OPT_APPCTL_COMMAND},
1378 DAEMON_LONG_OPTIONS,
1380 LEAK_CHECKER_LONG_OPTIONS,
1383 char *short_options = long_options_to_short_options(long_options);
1385 appctl_command = xasprintf("%s/ovs-appctl %%s", ovs_bindir);
1389 c = getopt_long(argc, argv, short_options, long_options, NULL);
1400 OVS_PRINT_VERSION(0, 0);
1403 case OPT_PRUNE_TIMEOUT:
1404 prune_timeout = atoi(optarg) * 1000;
1407 case OPT_APPCTL_COMMAND:
1408 appctl_command = optarg;
1411 VLOG_OPTION_HANDLERS
1412 DAEMON_OPTION_HANDLERS
1413 LEAK_CHECKER_OPTION_HANDLERS
1422 free(short_options);
1424 validate_appctl_command();
1430 ovs_fatal(0, "database socket is non-option argument; "
1431 "use --help for usage");
1440 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1441 "usage: %s [OPTIONS] CONFIG\n"
1442 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1443 program_name, program_name);
1444 printf("\nConfiguration options:\n"
1445 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1446 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1450 printf("\nOther options:\n"
1451 " -h, --help display this help message\n"
1452 " -V, --version display version information\n");
1453 leak_checker_usage();
1454 printf("\nThe default appctl command is:\n%s\n", appctl_command);