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);
283 static struct ovsdb_idl_txn
*
284 txn_from_openvswitch(const struct ovsrec_open_vswitch
*ovs
)
286 return ovsdb_idl_txn_get(&ovs
->header_
);
290 port_is_fake_bridge(const struct ovsrec_port
*port
)
292 return (port
->fake_bridge
294 && *port
->tag
>= 1 && *port
->tag
<= 4095);
298 ovs_insert_bridge(const struct ovsrec_open_vswitch
*ovs
,
299 struct ovsrec_bridge
*bridge
)
301 struct ovsrec_bridge
**bridges
;
304 bridges
= xmalloc(sizeof *ovs
->bridges
* (ovs
->n_bridges
+ 1));
305 for (i
= 0; i
< ovs
->n_bridges
; i
++) {
306 bridges
[i
] = ovs
->bridges
[i
];
308 bridges
[ovs
->n_bridges
] = bridge
;
309 ovsrec_open_vswitch_set_bridges(ovs
, bridges
, ovs
->n_bridges
+ 1);
314 where_uuid_equals(const struct uuid
*uuid
)
319 json_string_create("_uuid"),
320 json_string_create("=="),
322 json_string_create("uuid"),
323 json_string_create_nocopy(
324 xasprintf(UUID_FMT
, UUID_ARGS(uuid
))))));
327 /* Commits 'txn'. If 'wait_for_reload' is true, also waits for Open vSwitch to
328 reload the configuration before returning.
330 Returns EAGAIN if the caller should try the operation again, 0 on success,
331 otherwise a positive errno value. */
333 commit_txn(struct ovsdb_idl_txn
*txn
, bool wait_for_reload
)
335 struct ovsdb_idl
*idl
= ovsdb_idl_txn_get_idl (txn
);
336 enum ovsdb_idl_txn_status status
;
337 int64_t next_cfg
= 0;
339 if (wait_for_reload
) {
340 const struct ovsrec_open_vswitch
*ovs
= ovsrec_open_vswitch_first(idl
);
341 struct json
*where
= where_uuid_equals(&ovs
->header_
.uuid
);
342 ovsdb_idl_txn_increment(txn
, "Open_vSwitch", "next_cfg", where
);
345 status
= ovsdb_idl_txn_commit_block(txn
);
346 if (wait_for_reload
&& status
== TXN_SUCCESS
) {
347 next_cfg
= ovsdb_idl_txn_get_increment_new_value(txn
);
349 ovsdb_idl_txn_destroy(txn
);
356 VLOG_ERR_RL(&rl
, "OVSDB transaction unexpectedly aborted");
363 if (wait_for_reload
) {
365 /* We can't use 'ovs' any longer because ovsdb_idl_run() can
367 const struct ovsrec_open_vswitch
*ovs2
;
370 OVSREC_OPEN_VSWITCH_FOR_EACH (ovs2
, idl
) {
371 if (ovs2
->cur_cfg
>= next_cfg
) {
383 VLOG_ERR_RL(&rl
, "OVSDB transaction needs retry");
387 VLOG_ERR_RL(&rl
, "OVSDB transaction failed: %s",
388 ovsdb_idl_txn_get_error(txn
));
397 add_bridge(struct ovsdb_idl
*idl
, const struct ovsrec_open_vswitch
*ovs
,
400 struct ovsrec_bridge
*br
;
401 struct ovsrec_port
*port
;
402 struct ovsrec_interface
*iface
;
403 struct ovsdb_idl_txn
*txn
;
405 if (find_bridge(ovs
, br_name
)) {
406 VLOG_WARN("addbr %s: bridge %s exists", br_name
, br_name
);
408 } else if (netdev_exists(br_name
)) {
411 for (i
= 0; i
< ovs
->n_bridges
; i
++) {
413 struct ovsrec_bridge
*br_cfg
= ovs
->bridges
[i
];
415 for (j
= 0; j
< br_cfg
->n_ports
; j
++) {
416 if (port_is_fake_bridge(br_cfg
->ports
[j
])) {
417 VLOG_WARN("addbr %s: %s exists as a fake bridge",
424 VLOG_WARN("addbr %s: cannot create bridge %s because a network "
425 "device named %s already exists",
426 br_name
, br_name
, br_name
);
430 txn
= ovsdb_idl_txn_create(idl
);
432 ovsdb_idl_txn_add_comment(txn
, "ovs-brcompatd: addbr %s", br_name
);
434 iface
= ovsrec_interface_insert(txn_from_openvswitch(ovs
));
435 ovsrec_interface_set_name(iface
, br_name
);
437 port
= ovsrec_port_insert(txn_from_openvswitch(ovs
));
438 ovsrec_port_set_name(port
, br_name
);
439 ovsrec_port_set_interfaces(port
, &iface
, 1);
441 br
= ovsrec_bridge_insert(txn_from_openvswitch(ovs
));
442 ovsrec_bridge_set_name(br
, br_name
);
443 ovsrec_bridge_set_ports(br
, &port
, 1);
445 ovs_insert_bridge(ovs
, br
);
447 return commit_txn(txn
, true);
451 add_port(const struct ovsrec_open_vswitch
*ovs
,
452 const struct ovsrec_bridge
*br
, const char *port_name
)
454 struct ovsrec_interface
*iface
;
455 struct ovsrec_port
*port
;
456 struct ovsrec_port
**ports
;
459 /* xxx Check conflicts? */
460 iface
= ovsrec_interface_insert(txn_from_openvswitch(ovs
));
461 ovsrec_interface_set_name(iface
, port_name
);
463 port
= ovsrec_port_insert(txn_from_openvswitch(ovs
));
464 ovsrec_port_set_name(port
, port_name
);
465 ovsrec_port_set_interfaces(port
, &iface
, 1);
467 ports
= xmalloc(sizeof *br
->ports
* (br
->n_ports
+ 1));
468 for (i
= 0; i
< br
->n_ports
; i
++) {
469 ports
[i
] = br
->ports
[i
];
471 ports
[br
->n_ports
] = port
;
472 ovsrec_bridge_set_ports(br
, ports
, br
->n_ports
+ 1);
476 /* Deletes 'port' from 'br'.
478 * After calling this function, 'port' must not be referenced again. */
480 del_port(const struct ovsrec_bridge
*br
, const struct ovsrec_port
*port
)
482 struct ovsrec_port
**ports
;
485 /* Remove 'port' from the bridge's list of ports. */
486 ports
= xmalloc(sizeof *br
->ports
* br
->n_ports
);
487 for (i
= n
= 0; i
< br
->n_ports
; i
++) {
488 if (br
->ports
[i
] != port
) {
489 ports
[n
++] = br
->ports
[i
];
492 ovsrec_bridge_set_ports(br
, ports
, n
);
495 /* Delete all of the port's interfaces. */
496 for (i
= 0; i
< port
->n_interfaces
; i
++) {
497 ovsrec_interface_delete(port
->interfaces
[i
]);
500 /* Delete the port itself. */
501 ovsrec_port_delete(port
);
504 /* Delete 'iface' from 'port' (which must be within 'br'). If 'iface' was
505 * 'port''s only interface, delete 'port' from 'br' also.
507 * After calling this function, 'iface' must not be referenced again. */
509 del_interface(const struct ovsrec_bridge
*br
,
510 const struct ovsrec_port
*port
,
511 const struct ovsrec_interface
*iface
)
513 if (port
->n_interfaces
== 1) {
516 struct ovsrec_interface
**ifaces
;
519 ifaces
= xmalloc(sizeof *port
->interfaces
* port
->n_interfaces
);
520 for (i
= n
= 0; i
< port
->n_interfaces
; i
++) {
521 if (port
->interfaces
[i
] != iface
) {
522 ifaces
[n
++] = port
->interfaces
[i
];
525 ovsrec_port_set_interfaces(port
, ifaces
, n
);
527 ovsrec_interface_delete(iface
);
531 /* Find and return a port within 'br' named 'port_name'. */
532 static const struct ovsrec_port
*
533 find_port(const struct ovsrec_bridge
*br
, const char *port_name
)
537 for (i
= 0; i
< br
->n_ports
; i
++) {
538 struct ovsrec_port
*port
= br
->ports
[i
];
539 if (!strcmp(port_name
, port
->name
)) {
546 /* Find and return an interface within 'br' named 'iface_name'. */
547 static const struct ovsrec_interface
*
548 find_interface(const struct ovsrec_bridge
*br
, const char *iface_name
,
549 struct ovsrec_port
**portp
)
553 for (i
= 0; i
< br
->n_ports
; i
++) {
554 struct ovsrec_port
*port
= br
->ports
[i
];
557 for (j
= 0; j
< port
->n_interfaces
; j
++) {
558 struct ovsrec_interface
*iface
= port
->interfaces
[j
];
559 if (!strcmp(iface
->name
, iface_name
)) {
571 del_bridge(struct ovsdb_idl
*idl
,
572 const struct ovsrec_open_vswitch
*ovs
, const char *br_name
)
574 struct ovsrec_bridge
*br
= find_bridge(ovs
, br_name
);
575 struct ovsrec_bridge
**bridges
;
576 struct ovsdb_idl_txn
*txn
;
580 VLOG_WARN("delbr %s: no bridge named %s", br_name
, br_name
);
584 txn
= ovsdb_idl_txn_create(idl
);
586 ovsdb_idl_txn_add_comment(txn
, "ovs-brcompatd: delbr %s", br_name
);
588 /* Delete everything that the bridge points to, then delete the bridge
590 while (br
->n_ports
> 0) {
591 del_port(br
, br
->ports
[0]);
593 for (i
= 0; i
< br
->n_mirrors
; i
++) {
594 ovsrec_mirror_delete(br
->mirrors
[i
]);
597 ovsrec_netflow_delete(br
->netflow
);
600 ovsrec_sflow_delete(br
->sflow
);
602 for (i
= 0; i
< br
->n_controller
; i
++) {
603 ovsrec_controller_delete(br
->controller
[i
]);
606 /* Remove 'br' from the vswitch's list of bridges. */
607 bridges
= xmalloc(sizeof *ovs
->bridges
* ovs
->n_bridges
);
608 for (i
= n
= 0; i
< ovs
->n_bridges
; i
++) {
609 if (ovs
->bridges
[i
] != br
) {
610 bridges
[n
++] = ovs
->bridges
[i
];
613 ovsrec_open_vswitch_set_bridges(ovs
, bridges
, n
);
616 /* Delete the bridge itself. */
617 ovsrec_bridge_delete(br
);
619 return commit_txn(txn
, true);
623 parse_command(struct ofpbuf
*buffer
, uint32_t *seq
, const char **br_name
,
624 const char **port_name
, uint64_t *count
, uint64_t *skip
)
626 static const struct nl_policy policy
[] = {
627 [BRC_GENL_A_DP_NAME
] = { .type
= NL_A_STRING
, .optional
= true },
628 [BRC_GENL_A_PORT_NAME
] = { .type
= NL_A_STRING
, .optional
= true },
629 [BRC_GENL_A_FDB_COUNT
] = { .type
= NL_A_U64
, .optional
= true },
630 [BRC_GENL_A_FDB_SKIP
] = { .type
= NL_A_U64
, .optional
= true },
632 struct nlattr
*attrs
[ARRAY_SIZE(policy
)];
634 if (!nl_policy_parse(buffer
, NLMSG_HDRLEN
+ GENL_HDRLEN
, policy
,
635 attrs
, ARRAY_SIZE(policy
))
636 || (br_name
&& !attrs
[BRC_GENL_A_DP_NAME
])
637 || (port_name
&& !attrs
[BRC_GENL_A_PORT_NAME
])
638 || (count
&& !attrs
[BRC_GENL_A_FDB_COUNT
])
639 || (skip
&& !attrs
[BRC_GENL_A_FDB_SKIP
])) {
643 *seq
= ((struct nlmsghdr
*) buffer
->data
)->nlmsg_seq
;
645 *br_name
= nl_attr_get_string(attrs
[BRC_GENL_A_DP_NAME
]);
648 *port_name
= nl_attr_get_string(attrs
[BRC_GENL_A_PORT_NAME
]);
651 *count
= nl_attr_get_u64(attrs
[BRC_GENL_A_FDB_COUNT
]);
654 *skip
= nl_attr_get_u64(attrs
[BRC_GENL_A_FDB_SKIP
]);
659 /* Composes and returns a reply to a request made by the datapath with Netlink
660 * sequence number 'seq' and error code 'error'. The caller may add additional
661 * attributes to the message, then it may send it with send_reply(). */
662 static struct ofpbuf
*
663 compose_reply(uint32_t seq
, int error
)
665 struct ofpbuf
*reply
= ofpbuf_new(4096);
666 nl_msg_put_genlmsghdr(reply
, brc_sock
, 32, brc_family
, NLM_F_REQUEST
,
667 BRC_GENL_C_DP_RESULT
, 1);
668 ((struct nlmsghdr
*) reply
->data
)->nlmsg_seq
= seq
;
669 nl_msg_put_u32(reply
, BRC_GENL_A_ERR_CODE
, error
);
673 /* Sends 'reply' to the datapath and frees it. */
675 send_reply(struct ofpbuf
*reply
)
677 int retval
= nl_sock_send(brc_sock
, reply
, false);
679 VLOG_WARN_RL(&rl
, "replying to brcompat request: %s",
682 ofpbuf_delete(reply
);
685 /* Composes and sends a reply to a request made by the datapath with Netlink
686 * sequence number 'seq' and error code 'error'. */
688 send_simple_reply(uint32_t seq
, int error
)
690 send_reply(compose_reply(seq
, error
));
694 handle_bridge_cmd(struct ovsdb_idl
*idl
,
695 const struct ovsrec_open_vswitch
*ovs
,
696 struct ofpbuf
*buffer
, bool add
)
702 error
= parse_command(buffer
, &seq
, &br_name
, NULL
, NULL
, NULL
);
707 retval
= (add
? add_bridge
: del_bridge
)(idl
, ovs
, br_name
);
708 VLOG_INFO_RL(&rl
, "%sbr %s: %s",
709 add
? "add" : "del", br_name
, strerror(retval
));
710 } while (retval
== EAGAIN
);
712 send_simple_reply(seq
, error
);
717 static const struct nl_policy brc_port_policy
[] = {
718 [BRC_GENL_A_DP_NAME
] = { .type
= NL_A_STRING
},
719 [BRC_GENL_A_PORT_NAME
] = { .type
= NL_A_STRING
},
723 handle_port_cmd(struct ovsdb_idl
*idl
,
724 const struct ovsrec_open_vswitch
*ovs
,
725 struct ofpbuf
*buffer
, bool add
)
727 const char *cmd_name
= add
? "add-if" : "del-if";
728 const char *br_name
, *port_name
;
732 error
= parse_command(buffer
, &seq
, &br_name
, &port_name
, NULL
, NULL
);
734 struct ovsrec_bridge
*br
= find_bridge(ovs
, br_name
);
737 VLOG_WARN("%s %s %s: no bridge named %s",
738 cmd_name
, br_name
, port_name
, br_name
);
740 } else if (!netdev_exists(port_name
)) {
741 VLOG_WARN("%s %s %s: no network device named %s",
742 cmd_name
, br_name
, port_name
, port_name
);
746 struct ovsdb_idl_txn
*txn
= ovsdb_idl_txn_create(idl
);
749 ovsdb_idl_txn_add_comment(txn
, "ovs-brcompatd: add-if %s",
751 add_port(ovs
, br
, port_name
);
753 const struct ovsrec_port
*port
= find_port(br
, port_name
);
755 ovsdb_idl_txn_add_comment(txn
,
756 "ovs-brcompatd: del-if %s",
762 error
= commit_txn(txn
, true);
763 VLOG_INFO_RL(&rl
, "%s %s %s: %s",
764 cmd_name
, br_name
, port_name
, strerror(error
));
765 } while (error
== EAGAIN
);
767 send_simple_reply(seq
, error
);
773 /* The caller is responsible for freeing '*ovs_name' if the call is
776 linux_bridge_to_ovs_bridge(const struct ovsrec_open_vswitch
*ovs
,
777 const char *linux_name
,
778 const struct ovsrec_bridge
**ovs_bridge
,
781 *ovs_bridge
= find_bridge(ovs
, linux_name
);
783 /* Bridge name is the same. We are interested in VLAN 0. */
787 /* No such Open vSwitch bridge 'linux_name', but there might be an
788 * internal port named 'linux_name' on some other bridge
789 * 'ovs_bridge'. If so then we are interested in the VLAN assigned to
790 * port 'linux_name' on the bridge named 'ovs_bridge'. */
793 for (i
= 0; i
< ovs
->n_bridges
; i
++) {
794 const struct ovsrec_bridge
*br
= ovs
->bridges
[i
];
796 for (j
= 0; j
< br
->n_ports
; j
++) {
797 const struct ovsrec_port
*port
= br
->ports
[j
];
799 if (!strcmp(port
->name
, linux_name
)) {
801 *br_vlan
= port
->n_tag
? *port
->tag
: -1;
812 handle_fdb_query_cmd(const struct ovsrec_open_vswitch
*ovs
,
813 struct ofpbuf
*buffer
)
815 /* This structure is copied directly from the Linux 2.6.30 header files.
816 * It would be more straightforward to #include <linux/if_bridge.h>, but
817 * the 'port_hi' member was only introduced in Linux 2.6.26 and so systems
818 * with old header files won't have it. */
823 __u32 ageing_timer_value
;
832 struct mac
*local_macs
;
836 /* Impedance matching between the vswitchd and Linux kernel notions of what
837 * a bridge is. The kernel only handles a single VLAN per bridge, but
838 * vswitchd can deal with all the VLANs on a single bridge. We have to
839 * pretend that the former is the case even though the latter is the
841 const char *linux_name
; /* Name used by brctl. */
842 const struct ovsrec_bridge
*ovs_bridge
; /* Bridge used by ovs-vswitchd. */
843 int br_vlan
; /* VLAN tag. */
846 struct ofpbuf query_data
;
847 struct ofpbuf
*reply
;
848 char *unixctl_command
;
849 uint64_t count
, skip
;
855 /* Parse the command received from brcompat_mod. */
856 error
= parse_command(buffer
, &seq
, &linux_name
, NULL
, &count
, &skip
);
861 /* Figure out vswitchd bridge and VLAN. */
862 error
= linux_bridge_to_ovs_bridge(ovs
, linux_name
,
863 &ovs_bridge
, &br_vlan
);
865 send_simple_reply(seq
, error
);
869 /* Fetch the forwarding database using ovs-appctl. */
870 unixctl_command
= xasprintf("fdb/show %s", ovs_bridge
->name
);
871 error
= execute_appctl_command(unixctl_command
, &output
);
872 free(unixctl_command
);
874 send_simple_reply(seq
, error
);
878 /* Fetch the MAC address for each interface on the bridge, so that we can
879 * fill in the is_local field in the response. */
881 get_bridge_ifaces(ovs_bridge
, &ifaces
, br_vlan
);
882 local_macs
= xmalloc(ifaces
.n
* sizeof *local_macs
);
884 for (i
= 0; i
< ifaces
.n
; i
++) {
885 const char *iface_name
= ifaces
.names
[i
];
886 struct mac
*mac
= &local_macs
[n_local_macs
];
887 struct netdev
*netdev
;
889 error
= netdev_open_default(iface_name
, &netdev
);
891 if (!netdev_get_etheraddr(netdev
, mac
->addr
)) {
894 netdev_close(netdev
);
897 svec_destroy(&ifaces
);
899 /* Parse the response from ovs-appctl and convert it to binary format to
900 * pass back to the kernel. */
901 ofpbuf_init(&query_data
, sizeof(struct __fdb_entry
) * 8);
903 strtok_r(output
, "\n", &save_ptr
); /* Skip header line. */
905 struct __fdb_entry
*entry
;
907 uint8_t mac
[ETH_ADDR_LEN
];
911 line
= strtok_r(NULL
, "\n", &save_ptr
);
916 if (sscanf(line
, "%d %d "ETH_ADDR_SCAN_FMT
" %d",
917 &port
, &vlan
, ETH_ADDR_SCAN_ARGS(mac
), &age
)
918 != 2 + ETH_ADDR_SCAN_COUNT
+ 1) {
919 struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
920 VLOG_INFO_RL(&rl
, "fdb/show output has invalid format: %s", line
);
924 if (vlan
!= br_vlan
) {
933 /* Is this the MAC address of an interface on the bridge? */
935 for (i
= 0; i
< n_local_macs
; i
++) {
936 if (eth_addr_equals(local_macs
[i
].addr
, mac
)) {
942 entry
= ofpbuf_put_uninit(&query_data
, sizeof *entry
);
943 memcpy(entry
->mac_addr
, mac
, ETH_ADDR_LEN
);
944 entry
->port_no
= port
& 0xff;
945 entry
->is_local
= is_local
;
946 entry
->ageing_timer_value
= age
* HZ
;
947 entry
->port_hi
= (port
& 0xff00) >> 8;
954 /* Compose and send reply to datapath. */
955 reply
= compose_reply(seq
, 0);
956 nl_msg_put_unspec(reply
, BRC_GENL_A_FDB_DATA
,
957 query_data
.data
, query_data
.size
);
961 ofpbuf_uninit(&query_data
);
967 send_ifindex_reply(uint32_t seq
, struct svec
*ifaces
)
969 struct ofpbuf
*reply
;
975 /* Make sure that any given interface only occurs once. This shouldn't
976 * happen, but who knows what people put into their configuration files. */
977 svec_sort_unique(ifaces
);
979 /* Convert 'ifaces' into ifindexes. */
981 indices
= xmalloc(ifaces
->n
* sizeof *indices
);
982 SVEC_FOR_EACH (i
, iface
, ifaces
) {
983 int ifindex
= if_nametoindex(iface
);
985 indices
[n_indices
++] = ifindex
;
989 /* Compose and send reply. */
990 reply
= compose_reply(seq
, 0);
991 nl_msg_put_unspec(reply
, BRC_GENL_A_IFINDEXES
,
992 indices
, n_indices
* sizeof *indices
);
1000 handle_get_bridges_cmd(const struct ovsrec_open_vswitch
*ovs
,
1001 struct ofpbuf
*buffer
)
1003 struct svec bridges
;
1010 /* Parse Netlink command.
1012 * The command doesn't actually have any arguments, but we need the
1013 * sequence number to send the reply. */
1014 error
= parse_command(buffer
, &seq
, NULL
, NULL
, NULL
, NULL
);
1019 /* Get all the real bridges and all the fake ones. */
1020 svec_init(&bridges
);
1021 for (i
= 0; i
< ovs
->n_bridges
; i
++) {
1022 const struct ovsrec_bridge
*br
= ovs
->bridges
[i
];
1024 svec_add(&bridges
, br
->name
);
1025 for (j
= 0; j
< br
->n_ports
; j
++) {
1026 const struct ovsrec_port
*port
= br
->ports
[j
];
1028 if (port
->fake_bridge
) {
1029 svec_add(&bridges
, port
->name
);
1034 send_ifindex_reply(seq
, &bridges
);
1035 svec_destroy(&bridges
);
1041 handle_get_ports_cmd(const struct ovsrec_open_vswitch
*ovs
,
1042 struct ofpbuf
*buffer
)
1046 const char *linux_name
;
1047 const struct ovsrec_bridge
*ovs_bridge
;
1054 /* Parse Netlink command. */
1055 error
= parse_command(buffer
, &seq
, &linux_name
, NULL
, NULL
, NULL
);
1060 error
= linux_bridge_to_ovs_bridge(ovs
, linux_name
,
1061 &ovs_bridge
, &br_vlan
);
1063 send_simple_reply(seq
, error
);
1068 get_bridge_ports(ovs_bridge
, &ports
, br_vlan
);
1070 svec_del(&ports
, linux_name
);
1071 send_ifindex_reply(seq
, &ports
); /* XXX bonds won't show up */
1072 svec_destroy(&ports
);
1078 brc_recv_update(struct ovsdb_idl
*idl
)
1081 struct ofpbuf
*buffer
;
1082 struct genlmsghdr
*genlmsghdr
;
1083 const struct ovsrec_open_vswitch
*ovs
;
1087 ofpbuf_delete(buffer
);
1088 retval
= nl_sock_recv(brc_sock
, &buffer
, false);
1089 } while (retval
== ENOBUFS
1091 && (nl_msg_nlmsgerr(buffer
, NULL
)
1092 || nl_msg_nlmsghdr(buffer
)->nlmsg_type
== NLMSG_DONE
)));
1094 if (retval
!= EAGAIN
) {
1095 VLOG_WARN_RL(&rl
, "brc_recv_update: %s", strerror(retval
));
1100 genlmsghdr
= nl_msg_genlmsghdr(buffer
);
1102 VLOG_WARN_RL(&rl
, "received packet too short for generic NetLink");
1106 if (nl_msg_nlmsghdr(buffer
)->nlmsg_type
!= brc_family
) {
1107 VLOG_DBG_RL(&rl
, "received type (%"PRIu16
") != brcompat family (%d)",
1108 nl_msg_nlmsghdr(buffer
)->nlmsg_type
, brc_family
);
1112 /* Get the Open vSwitch configuration. Just drop the request on the floor
1113 * if a valid configuration doesn't exist. (We could check this earlier,
1114 * but we want to drain pending Netlink messages even when there is no Open
1115 * vSwitch configuration.) */
1116 ovs
= ovsrec_open_vswitch_first(idl
);
1118 VLOG_WARN_RL(&rl
, "could not find valid configuration to update");
1122 switch (genlmsghdr
->cmd
) {
1123 case BRC_GENL_C_DP_ADD
:
1124 handle_bridge_cmd(idl
, ovs
, buffer
, true);
1127 case BRC_GENL_C_DP_DEL
:
1128 handle_bridge_cmd(idl
, ovs
, buffer
, false);
1131 case BRC_GENL_C_PORT_ADD
:
1132 handle_port_cmd(idl
, ovs
, buffer
, true);
1135 case BRC_GENL_C_PORT_DEL
:
1136 handle_port_cmd(idl
, ovs
, buffer
, false);
1139 case BRC_GENL_C_FDB_QUERY
:
1140 handle_fdb_query_cmd(ovs
, buffer
);
1143 case BRC_GENL_C_GET_BRIDGES
:
1144 handle_get_bridges_cmd(ovs
, buffer
);
1147 case BRC_GENL_C_GET_PORTS
:
1148 handle_get_ports_cmd(ovs
, buffer
);
1152 VLOG_WARN_RL(&rl
, "received unknown brc netlink command: %d\n",
1158 ofpbuf_delete(buffer
);
1162 /* Check for interface configuration changes announced through RTNL. */
1164 rtnl_recv_update(struct ovsdb_idl
*idl
,
1165 const struct ovsrec_open_vswitch
*ovs
)
1169 int error
= nl_sock_recv(rtnl_sock
, &buf
, false);
1170 if (error
== EAGAIN
) {
1171 /* Nothing to do. */
1172 } else if (error
== ENOBUFS
) {
1173 VLOG_WARN_RL(&rl
, "network monitor socket overflowed");
1175 VLOG_WARN_RL(&rl
, "error on network monitor socket: %s",
1178 struct nlattr
*attrs
[ARRAY_SIZE(rtnlgrp_link_policy
)];
1179 struct nlmsghdr
*nlh
;
1180 struct ifinfomsg
*iim
;
1182 nlh
= ofpbuf_at(buf
, 0, NLMSG_HDRLEN
);
1183 iim
= ofpbuf_at(buf
, NLMSG_HDRLEN
, sizeof *iim
);
1185 VLOG_WARN_RL(&rl
, "received bad rtnl message (no ifinfomsg)");
1190 if (!nl_policy_parse(buf
, NLMSG_HDRLEN
+ sizeof(struct ifinfomsg
),
1191 rtnlgrp_link_policy
,
1192 attrs
, ARRAY_SIZE(rtnlgrp_link_policy
))) {
1193 VLOG_WARN_RL(&rl
,"received bad rtnl message (policy)");
1197 if (nlh
->nlmsg_type
== RTM_DELLINK
&& attrs
[IFLA_MASTER
]) {
1198 const char *port_name
= nl_attr_get_string(attrs
[IFLA_IFNAME
]);
1199 char br_name
[IFNAMSIZ
];
1200 uint32_t br_idx
= nl_attr_get_u32(attrs
[IFLA_MASTER
]);
1202 if (!if_indextoname(br_idx
, br_name
)) {
1207 if (!netdev_exists(port_name
)) {
1208 /* Network device is really gone. */
1209 struct ovsdb_idl_txn
*txn
;
1210 const struct ovsrec_interface
*iface
;
1211 struct ovsrec_port
*port
;
1212 struct ovsrec_bridge
*br
;
1214 VLOG_INFO("network device %s destroyed, "
1215 "removing from bridge %s", port_name
, br_name
);
1217 br
= find_bridge(ovs
, br_name
);
1219 VLOG_WARN("no bridge named %s from which to remove %s",
1220 br_name
, port_name
);
1225 txn
= ovsdb_idl_txn_create(idl
);
1227 iface
= find_interface(br
, port_name
, &port
);
1229 del_interface(br
, port
, iface
);
1230 ovsdb_idl_txn_add_comment(txn
,
1231 "ovs-brcompatd: destroy port %s",
1235 commit_txn(txn
, false);
1237 /* A network device by that name exists even though the kernel
1238 * told us it had disappeared. Probably, what happened was
1241 * 1. Device destroyed.
1242 * 2. Notification sent to us.
1243 * 3. New device created with same name as old one.
1244 * 4. ovs-brcompatd notified, removes device from bridge.
1246 * There's no a priori reason that in this situation that the
1247 * new device with the same name should remain in the bridge;
1248 * on the contrary, that would be unexpected. *But* there is
1249 * one important situation where, if we do this, bad things
1250 * happen. This is the case of XenServer Tools version 5.0.0,
1251 * which on boot of a Windows VM cause something like this to
1252 * happen on the Xen host:
1254 * i. Create tap1.0 and vif1.0.
1255 * ii. Delete tap1.0.
1256 * iii. Delete vif1.0.
1257 * iv. Re-create vif1.0.
1259 * (XenServer Tools 5.5.0 does not exhibit this behavior, and
1260 * neither does a VM without Tools installed at all.@.)
1262 * Steps iii and iv happen within a few seconds of each other.
1263 * Step iv causes /etc/xensource/scripts/vif to run, which in
1264 * turn calls ovs-cfg-mod to add the new device to the bridge.
1265 * If step iv happens after step 4 (in our first list of
1266 * steps), then all is well, but if it happens between 3 and 4
1267 * (which can easily happen if ovs-brcompatd has to wait to
1268 * lock the configuration file), then we will remove the new
1269 * incarnation from the bridge instead of the old one!
1271 * So, to avoid this problem, we do nothing here. This is
1272 * strictly incorrect except for this one particular case, and
1273 * perhaps that will bite us someday. If that happens, then we
1274 * will have to somehow track network devices by ifindex, since
1275 * a new device will have a new ifindex even if it has the same
1276 * name as an old device.
1278 VLOG_INFO("kernel reported network device %s removed but "
1279 "a device by that name exists (XS Tools 5.0.0?)",
1288 main(int argc
, char *argv
[])
1290 struct unixctl_server
*unixctl
;
1292 struct ovsdb_idl
*idl
;
1295 proctitle_init(argc
, argv
);
1296 set_program_name(argv
[0]);
1299 vlog_set_levels(VLM_ANY_MODULE
, VLF_CONSOLE
, VLL_WARN
);
1300 vlog_set_levels(VLM_reconnect
, VLF_ANY_FACILITY
, VLL_WARN
);
1302 remote
= parse_options(argc
, argv
);
1303 signal(SIGPIPE
, SIG_IGN
);
1307 die_if_already_running();
1310 retval
= unixctl_server_create(NULL
, &unixctl
);
1315 if (brc_open(&brc_sock
)) {
1316 ovs_fatal(0, "could not open brcompat socket. Check "
1317 "\"brcompat\" kernel module.");
1320 if (prune_timeout
) {
1321 if (nl_sock_create(NETLINK_ROUTE
, RTNLGRP_LINK
, 0, 0, &rtnl_sock
)) {
1322 ovs_fatal(0, "could not create rtnetlink socket");
1326 daemonize_complete();
1328 idl
= ovsdb_idl_create(remote
, &ovsrec_idl_class
);
1331 const struct ovsrec_open_vswitch
*ovs
;
1335 unixctl_server_run(unixctl
);
1336 brc_recv_update(idl
);
1338 ovs
= ovsrec_open_vswitch_first(idl
);
1339 if (!ovs
&& ovsdb_idl_has_ever_connected(idl
)) {
1340 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
1341 VLOG_WARN_RL(&rl
, "%s: database does not contain any Open vSwitch "
1342 "configuration", remote
);
1346 /* If 'prune_timeout' is non-zero, we actively prune from the
1347 * configuration of port entries that are no longer valid. We
1350 * 1) The kernel explicitly notifies us of removed ports
1351 * through the RTNL messages.
1353 * 2) We periodically check all ports associated with bridges
1354 * to see if they no longer exist.
1356 if (ovs
&& prune_timeout
) {
1357 rtnl_recv_update(idl
, ovs
);
1358 nl_sock_wait(rtnl_sock
, POLLIN
);
1359 poll_timer_wait(prune_timeout
);
1363 nl_sock_wait(brc_sock
, POLLIN
);
1364 ovsdb_idl_wait(idl
);
1365 unixctl_server_wait(unixctl
);
1370 ovsdb_idl_destroy(idl
);
1376 validate_appctl_command(void)
1382 for (p
= strchr(appctl_command
, '%'); p
; p
= strchr(p
+ 2, '%')) {
1384 /* Nothing to do. */
1385 } else if (p
[1] == 's') {
1388 ovs_fatal(0, "only '%%s' and '%%%%' allowed in --appctl-command");
1392 ovs_fatal(0, "'%%s' must appear exactly once in --appctl-command");
1397 parse_options(int argc
, char *argv
[])
1403 LEAK_CHECKER_OPTION_ENUMS
1405 static struct option long_options
[] = {
1406 {"help", no_argument
, 0, 'h'},
1407 {"version", no_argument
, 0, 'V'},
1408 {"prune-timeout", required_argument
, 0, OPT_PRUNE_TIMEOUT
},
1409 {"appctl-command", required_argument
, 0, OPT_APPCTL_COMMAND
},
1410 DAEMON_LONG_OPTIONS
,
1412 LEAK_CHECKER_LONG_OPTIONS
,
1415 char *short_options
= long_options_to_short_options(long_options
);
1417 appctl_command
= xasprintf("%s/ovs-appctl %%s", ovs_bindir
);
1421 c
= getopt_long(argc
, argv
, short_options
, long_options
, NULL
);
1432 OVS_PRINT_VERSION(0, 0);
1435 case OPT_PRUNE_TIMEOUT
:
1436 prune_timeout
= atoi(optarg
) * 1000;
1439 case OPT_APPCTL_COMMAND
:
1440 appctl_command
= optarg
;
1443 VLOG_OPTION_HANDLERS
1444 DAEMON_OPTION_HANDLERS
1445 LEAK_CHECKER_OPTION_HANDLERS
1454 free(short_options
);
1456 validate_appctl_command();
1462 ovs_fatal(0, "database socket is non-option argument; "
1463 "use --help for usage");
1472 printf("%s: bridge compatibility front-end for ovs-vswitchd\n"
1473 "usage: %s [OPTIONS] CONFIG\n"
1474 "CONFIG is the configuration file used by ovs-vswitchd.\n",
1475 program_name
, program_name
);
1476 printf("\nConfiguration options:\n"
1477 " --appctl-command=COMMAND shell command to run ovs-appctl\n"
1478 " --prune-timeout=SECS wait at most SECS before pruning ports\n"
1482 printf("\nOther options:\n"
1483 " -h, --help display this help message\n"
1484 " -V, --version display version information\n");
1485 leak_checker_usage();
1486 printf("\nThe default appctl command is:\n%s\n", appctl_command
);