1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id
;
74 /* These members are always valid. */
75 struct port
*port
; /* Containing port. */
76 size_t port_ifidx
; /* Index within containing port. */
77 char *name
; /* Host network device name. */
78 tag_type tag
; /* Tag associated with this interface. */
79 long long delay_expires
; /* Time after which 'enabled' may change. */
81 /* These members are valid only after bridge_reconfigure() causes them to
83 int dp_ifidx
; /* Index within kernel datapath. */
84 struct netdev
*netdev
; /* Network device. */
85 bool enabled
; /* May be chosen for flows? */
88 #define BOND_MASK 0xff
90 int iface_idx
; /* Index of assigned iface, or -1 if none. */
91 uint64_t tx_bytes
; /* Count of bytes recently transmitted. */
92 tag_type iface_tag
; /* Tag associated with iface_idx. */
95 #define MAX_MIRRORS 32
96 typedef uint32_t mirror_mask_t
;
97 #define MIRROR_MASK_C(X) UINT32_C(X)
98 BUILD_ASSERT_DECL(sizeof(mirror_mask_t
) * CHAR_BIT
>= MAX_MIRRORS
);
100 struct bridge
*bridge
;
104 /* Selection criteria. */
105 struct svec src_ports
;
106 struct svec dst_ports
;
111 struct port
*out_port
;
115 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
117 struct bridge
*bridge
;
119 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
120 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
123 /* An ordinary bridge port has 1 interface.
124 * A bridge port for bonding has at least 2 interfaces. */
125 struct iface
**ifaces
;
126 size_t n_ifaces
, allocated_ifaces
;
129 struct bond_entry
*bond_hash
; /* An array of (BOND_MASK + 1) elements. */
130 int active_iface
; /* Ifidx on which bcasts accepted, or -1. */
131 tag_type active_iface_tag
; /* Tag for bcast flows. */
132 tag_type no_ifaces_tag
; /* Tag for flows when all ifaces disabled. */
133 int updelay
, downdelay
; /* Delay before iface goes up/down, in ms. */
134 bool bond_compat_is_stale
; /* Need to call port_update_bond_compat()? */
136 /* Port mirroring info. */
137 mirror_mask_t src_mirrors
; /* Mirrors triggered when packet received. */
138 mirror_mask_t dst_mirrors
; /* Mirrors triggered when packet sent. */
139 bool is_mirror_output_port
; /* Does port mirroring send frames here? */
141 /* Spanning tree info. */
142 enum stp_state stp_state
; /* Always STP_FORWARDING if STP not in use. */
143 tag_type stp_state_tag
; /* Tag for STP state change. */
146 #define DP_MAX_PORTS 255
148 struct list node
; /* Node in global list of bridges. */
149 char *name
; /* User-specified arbitrary name. */
150 struct mac_learning
*ml
; /* MAC learning table, or null not to learn. */
151 bool sent_config_request
; /* Successfully sent config request? */
152 uint8_t default_ea
[ETH_ADDR_LEN
]; /* Default MAC. */
154 /* Support for remote controllers. */
155 char *controller
; /* NULL if there is no remote controller;
156 * "discover" to do controller discovery;
157 * otherwise a vconn name. */
159 /* OpenFlow switch processing. */
160 struct ofproto
*ofproto
; /* OpenFlow switch. */
162 /* Kernel datapath information. */
163 struct dpif
*dpif
; /* Datapath. */
164 struct port_array ifaces
; /* Indexed by kernel datapath port number. */
168 size_t n_ports
, allocated_ports
;
171 bool has_bonded_ports
;
172 long long int bond_next_rebalance
;
177 /* Flow statistics gathering. */
178 time_t next_stats_request
;
180 /* Port mirroring. */
181 struct mirror
*mirrors
[MAX_MIRRORS
];
185 long long int stp_last_tick
;
188 /* List of all bridges. */
189 static struct list all_bridges
= LIST_INITIALIZER(&all_bridges
);
191 /* Maximum number of datapaths. */
192 enum { DP_MAX
= 256 };
194 static struct bridge
*bridge_create(const char *name
);
195 static void bridge_destroy(struct bridge
*);
196 static struct bridge
*bridge_lookup(const char *name
);
197 static void bridge_unixctl_dump_flows(struct unixctl_conn
*, const char *);
198 static int bridge_run_one(struct bridge
*);
199 static void bridge_reconfigure_one(struct bridge
*);
200 static void bridge_reconfigure_controller(struct bridge
*);
201 static void bridge_get_all_ifaces(const struct bridge
*, struct svec
*ifaces
);
202 static void bridge_fetch_dp_ifaces(struct bridge
*);
203 static void bridge_flush(struct bridge
*);
204 static void bridge_pick_local_hw_addr(struct bridge
*,
205 uint8_t ea
[ETH_ADDR_LEN
],
206 struct iface
**hw_addr_iface
);
207 static uint64_t bridge_pick_datapath_id(struct bridge
*,
208 const uint8_t bridge_ea
[ETH_ADDR_LEN
],
209 struct iface
*hw_addr_iface
);
210 static struct iface
*bridge_get_local_iface(struct bridge
*);
211 static uint64_t dpid_from_hash(const void *, size_t nbytes
);
213 static void bridge_unixctl_fdb_show(struct unixctl_conn
*, const char *args
);
215 static void bond_init(void);
216 static void bond_run(struct bridge
*);
217 static void bond_wait(struct bridge
*);
218 static void bond_rebalance_port(struct port
*);
219 static void bond_send_learning_packets(struct port
*);
221 static void port_create(struct bridge
*, const char *name
);
222 static void port_reconfigure(struct port
*);
223 static void port_destroy(struct port
*);
224 static struct port
*port_lookup(const struct bridge
*, const char *name
);
225 static struct iface
*port_lookup_iface(const struct port
*, const char *name
);
226 static struct port
*port_from_dp_ifidx(const struct bridge
*,
228 static void port_update_bond_compat(struct port
*);
229 static void port_update_vlan_compat(struct port
*);
230 static void port_update_bonding(struct port
*);
232 static void mirror_create(struct bridge
*, const char *name
);
233 static void mirror_destroy(struct mirror
*);
234 static void mirror_reconfigure(struct bridge
*);
235 static void mirror_reconfigure_one(struct mirror
*);
236 static bool vlan_is_mirrored(const struct mirror
*, int vlan
);
238 static void brstp_reconfigure(struct bridge
*);
239 static void brstp_adjust_timers(struct bridge
*);
240 static void brstp_run(struct bridge
*);
241 static void brstp_wait(struct bridge
*);
243 static void iface_create(struct port
*, const char *name
);
244 static void iface_destroy(struct iface
*);
245 static struct iface
*iface_lookup(const struct bridge
*, const char *name
);
246 static struct iface
*iface_from_dp_ifidx(const struct bridge
*,
249 /* Hooks into ofproto processing. */
250 static struct ofhooks bridge_ofhooks
;
252 /* Public functions. */
254 /* Adds the name of each interface used by a bridge, including local and
255 * internal ports, to 'svec'. */
257 bridge_get_ifaces(struct svec
*svec
)
259 struct bridge
*br
, *next
;
262 LIST_FOR_EACH_SAFE (br
, next
, struct bridge
, node
, &all_bridges
) {
263 for (i
= 0; i
< br
->n_ports
; i
++) {
264 struct port
*port
= br
->ports
[i
];
266 for (j
= 0; j
< port
->n_ifaces
; j
++) {
267 struct iface
*iface
= port
->ifaces
[j
];
268 if (iface
->dp_ifidx
< 0) {
269 VLOG_ERR("%s interface not in datapath %s, ignoring",
270 iface
->name
, dpif_name(br
->dpif
));
272 if (iface
->dp_ifidx
!= ODPP_LOCAL
) {
273 svec_add(svec
, iface
->name
);
281 /* The caller must already have called cfg_read(). */
285 struct svec dpif_names
;
288 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show
);
290 svec_init(&dpif_names
);
291 dp_enumerate(&dpif_names
);
292 for (i
= 0; i
< dpif_names
.n
; i
++) {
293 const char *dpif_name
= dpif_names
.names
[i
];
297 retval
= dpif_open(dpif_name
, &dpif
);
299 struct svec all_names
;
302 svec_init(&all_names
);
303 dpif_get_all_names(dpif
, &all_names
);
304 for (j
= 0; j
< all_names
.n
; j
++) {
305 if (cfg_has("bridge.%s.port", all_names
.names
[j
])) {
311 svec_destroy(&all_names
);
316 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows
);
319 bridge_reconfigure();
324 config_string_change(const char *key
, char **valuep
)
326 const char *value
= cfg_get_string(0, "%s", key
);
327 if (value
&& (!*valuep
|| strcmp(value
, *valuep
))) {
329 *valuep
= xstrdup(value
);
337 bridge_configure_ssl(void)
339 /* XXX SSL should be configurable on a per-bridge basis.
340 * XXX should be possible to de-configure SSL. */
341 static char *private_key_file
;
342 static char *certificate_file
;
343 static char *cacert_file
;
346 if (config_string_change("ssl.private-key", &private_key_file
)) {
347 vconn_ssl_set_private_key_file(private_key_file
);
350 if (config_string_change("ssl.certificate", &certificate_file
)) {
351 vconn_ssl_set_certificate_file(certificate_file
);
354 /* We assume that even if the filename hasn't changed, if the CA cert
355 * file has been removed, that we want to move back into
356 * boot-strapping mode. This opens a small security hole, because
357 * the old certificate will still be trusted until vSwitch is
358 * restarted. We may want to address this in vconn's SSL library. */
359 if (config_string_change("ssl.ca-cert", &cacert_file
)
360 || (cacert_file
&& stat(cacert_file
, &s
) && errno
== ENOENT
)) {
361 vconn_ssl_set_ca_cert_file(cacert_file
,
362 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
367 /* iterate_and_prune_ifaces() callback function that opens the network device
368 * for 'iface', if it is not already open, and retrieves the interface's MAC
369 * address and carrier status. */
371 init_iface_netdev(struct bridge
*br UNUSED
, struct iface
*iface
,
376 } else if (!netdev_open(iface
->name
, NETDEV_ETH_TYPE_NONE
,
378 netdev_get_carrier(iface
->netdev
, &iface
->enabled
);
381 /* If the network device can't be opened, then we're not going to try
382 * to do anything with this interface. */
388 check_iface_dp_ifidx(struct bridge
*br
, struct iface
*iface
, void *aux UNUSED
)
390 if (iface
->dp_ifidx
>= 0) {
391 VLOG_DBG("%s has interface %s on port %d",
393 iface
->name
, iface
->dp_ifidx
);
396 VLOG_ERR("%s interface not in %s, dropping",
397 iface
->name
, dpif_name(br
->dpif
));
403 set_iface_policing(struct bridge
*br UNUSED
, struct iface
*iface
,
406 int rate
= cfg_get_int(0, "port.%s.ingress.policing-rate", iface
->name
);
407 int burst
= cfg_get_int(0, "port.%s.ingress.policing-burst", iface
->name
);
408 netdev_set_policing(iface
->netdev
, rate
, burst
);
412 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
413 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
414 * deletes from 'br' any ports that no longer have any interfaces. */
416 iterate_and_prune_ifaces(struct bridge
*br
,
417 bool (*cb
)(struct bridge
*, struct iface
*,
423 for (i
= 0; i
< br
->n_ports
; ) {
424 struct port
*port
= br
->ports
[i
];
425 for (j
= 0; j
< port
->n_ifaces
; ) {
426 struct iface
*iface
= port
->ifaces
[j
];
427 if (cb(br
, iface
, aux
)) {
430 iface_destroy(iface
);
434 if (port
->n_ifaces
) {
437 VLOG_ERR("%s port has no interfaces, dropping", port
->name
);
444 bridge_reconfigure(void)
446 struct svec old_br
, new_br
;
447 struct bridge
*br
, *next
;
450 COVERAGE_INC(bridge_reconfigure
);
452 /* Collect old and new bridges. */
455 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
456 svec_add(&old_br
, br
->name
);
458 cfg_get_subsections(&new_br
, "bridge");
460 /* Get rid of deleted bridges and add new bridges. */
463 assert(svec_is_unique(&old_br
));
464 assert(svec_is_unique(&new_br
));
465 LIST_FOR_EACH_SAFE (br
, next
, struct bridge
, node
, &all_bridges
) {
466 if (!svec_contains(&new_br
, br
->name
)) {
470 for (i
= 0; i
< new_br
.n
; i
++) {
471 const char *name
= new_br
.names
[i
];
472 if (!svec_contains(&old_br
, name
)) {
476 svec_destroy(&old_br
);
477 svec_destroy(&new_br
);
481 bridge_configure_ssl();
484 /* Reconfigure all bridges. */
485 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
486 bridge_reconfigure_one(br
);
489 /* Add and delete ports on all datapaths.
491 * The kernel will reject any attempt to add a given port to a datapath if
492 * that port already belongs to a different datapath, so we must do all
493 * port deletions before any port additions. */
494 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
495 struct odp_port
*dpif_ports
;
497 struct svec want_ifaces
;
499 dpif_port_list(br
->dpif
, &dpif_ports
, &n_dpif_ports
);
500 bridge_get_all_ifaces(br
, &want_ifaces
);
501 for (i
= 0; i
< n_dpif_ports
; i
++) {
502 const struct odp_port
*p
= &dpif_ports
[i
];
503 if (!svec_contains(&want_ifaces
, p
->devname
)
504 && strcmp(p
->devname
, br
->name
)) {
505 int retval
= dpif_port_del(br
->dpif
, p
->port
);
507 VLOG_ERR("failed to remove %s interface from %s: %s",
508 p
->devname
, dpif_name(br
->dpif
),
513 svec_destroy(&want_ifaces
);
516 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
517 struct odp_port
*dpif_ports
;
519 struct svec cur_ifaces
, want_ifaces
, add_ifaces
;
521 dpif_port_list(br
->dpif
, &dpif_ports
, &n_dpif_ports
);
522 svec_init(&cur_ifaces
);
523 for (i
= 0; i
< n_dpif_ports
; i
++) {
524 svec_add(&cur_ifaces
, dpif_ports
[i
].devname
);
527 svec_sort_unique(&cur_ifaces
);
528 bridge_get_all_ifaces(br
, &want_ifaces
);
529 svec_diff(&want_ifaces
, &cur_ifaces
, &add_ifaces
, NULL
, NULL
);
531 for (i
= 0; i
< add_ifaces
.n
; i
++) {
532 const char *if_name
= add_ifaces
.names
[i
];
536 /* It's an internal interface if it's marked that way, or if
537 * it's a bonded interface for which we're faking up a network
539 internal
= cfg_get_bool(0, "iface.%s.internal", if_name
);
540 if (cfg_get_bool(0, "bonding.%s.fake-iface", if_name
)) {
541 struct port
*port
= port_lookup(br
, if_name
);
542 if (port
&& port
->n_ifaces
> 1) {
547 /* Add to datapath. */
548 error
= dpif_port_add(br
->dpif
, if_name
,
549 internal
? ODP_PORT_INTERNAL
: 0, NULL
);
550 if (error
== EXFULL
) {
551 VLOG_ERR("ran out of valid port numbers on %s",
552 dpif_name(br
->dpif
));
555 VLOG_ERR("failed to add %s interface to %s: %s",
556 if_name
, dpif_name(br
->dpif
), strerror(error
));
559 svec_destroy(&cur_ifaces
);
560 svec_destroy(&want_ifaces
);
561 svec_destroy(&add_ifaces
);
563 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
566 struct iface
*local_iface
;
567 struct iface
*hw_addr_iface
;
568 uint8_t engine_type
, engine_id
;
569 bool add_id_to_iface
= false;
570 struct svec nf_hosts
;
572 bridge_fetch_dp_ifaces(br
);
573 iterate_and_prune_ifaces(br
, init_iface_netdev
, NULL
);
575 iterate_and_prune_ifaces(br
, check_iface_dp_ifidx
, NULL
);
577 /* Pick local port hardware address, datapath ID. */
578 bridge_pick_local_hw_addr(br
, ea
, &hw_addr_iface
);
579 local_iface
= bridge_get_local_iface(br
);
581 int error
= netdev_set_etheraddr(local_iface
->netdev
, ea
);
583 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
584 VLOG_ERR_RL(&rl
, "bridge %s: failed to set bridge "
585 "Ethernet address: %s",
586 br
->name
, strerror(error
));
590 dpid
= bridge_pick_datapath_id(br
, ea
, hw_addr_iface
);
591 ofproto_set_datapath_id(br
->ofproto
, dpid
);
593 /* Set NetFlow configuration on this bridge. */
594 dpif_get_netflow_ids(br
->dpif
, &engine_type
, &engine_id
);
595 if (cfg_has("netflow.%s.engine-type", br
->name
)) {
596 engine_type
= cfg_get_int(0, "netflow.%s.engine-type",
599 if (cfg_has("netflow.%s.engine-id", br
->name
)) {
600 engine_id
= cfg_get_int(0, "netflow.%s.engine-id", br
->name
);
602 if (cfg_has("netflow.%s.add-id-to-iface", br
->name
)) {
603 add_id_to_iface
= cfg_get_bool(0, "netflow.%s.add-id-to-iface",
606 if (add_id_to_iface
&& engine_id
> 0x7f) {
607 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
608 "another vswitch, choose an engine id less than 128",
611 if (add_id_to_iface
&& br
->n_ports
> 0x1ff) {
612 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
613 "another port when 512 or more ports are used",
616 svec_init(&nf_hosts
);
617 cfg_get_all_keys(&nf_hosts
, "netflow.%s.host", br
->name
);
618 if (ofproto_set_netflow(br
->ofproto
, &nf_hosts
, engine_type
,
619 engine_id
, add_id_to_iface
)) {
620 VLOG_ERR("bridge %s: problem setting netflow collectors",
624 /* Update the controller and related settings. It would be more
625 * straightforward to call this from bridge_reconfigure_one(), but we
626 * can't do it there for two reasons. First, and most importantly, at
627 * that point we don't know the dp_ifidx of any interfaces that have
628 * been added to the bridge (because we haven't actually added them to
629 * the datapath). Second, at that point we haven't set the datapath ID
630 * yet; when a controller is configured, resetting the datapath ID will
631 * immediately disconnect from the controller, so it's better to set
632 * the datapath ID before the controller. */
633 bridge_reconfigure_controller(br
);
635 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
636 for (i
= 0; i
< br
->n_ports
; i
++) {
637 struct port
*port
= br
->ports
[i
];
638 port_update_vlan_compat(port
);
639 port_update_bonding(port
);
642 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
643 brstp_reconfigure(br
);
644 iterate_and_prune_ifaces(br
, set_iface_policing
, NULL
);
649 bridge_pick_local_hw_addr(struct bridge
*br
, uint8_t ea
[ETH_ADDR_LEN
],
650 struct iface
**hw_addr_iface
)
652 uint64_t requested_ea
;
656 *hw_addr_iface
= NULL
;
658 /* Did the user request a particular MAC? */
659 requested_ea
= cfg_get_mac(0, "bridge.%s.mac", br
->name
);
661 eth_addr_from_uint64(requested_ea
, ea
);
662 if (eth_addr_is_multicast(ea
)) {
663 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
664 "address "ETH_ADDR_FMT
, br
->name
, ETH_ADDR_ARGS(ea
));
665 } else if (eth_addr_is_zero(ea
)) {
666 VLOG_ERR("bridge %s: cannot set MAC address to zero", br
->name
);
672 /* Otherwise choose the minimum MAC address among all of the interfaces.
673 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
674 * MAC of the physical interface in such an environment.) */
675 memset(ea
, 0xff, sizeof ea
);
676 for (i
= 0; i
< br
->n_ports
; i
++) {
677 struct port
*port
= br
->ports
[i
];
678 uint8_t iface_ea
[ETH_ADDR_LEN
];
679 uint64_t iface_ea_u64
;
682 /* Mirror output ports don't participate. */
683 if (port
->is_mirror_output_port
) {
687 /* Choose the MAC address to represent the port. */
688 iface_ea_u64
= cfg_get_mac(0, "port.%s.mac", port
->name
);
690 /* User specified explicitly. */
691 eth_addr_from_uint64(iface_ea_u64
, iface_ea
);
693 /* Find the interface with this Ethernet address (if any) so that
694 * we can provide the correct devname to the caller. */
696 for (j
= 0; j
< port
->n_ifaces
; j
++) {
697 struct iface
*candidate
= port
->ifaces
[j
];
698 uint8_t candidate_ea
[ETH_ADDR_LEN
];
699 if (!netdev_get_etheraddr(candidate
->netdev
, candidate_ea
)
700 && eth_addr_equals(iface_ea
, candidate_ea
)) {
705 /* Choose the interface whose MAC address will represent the port.
706 * The Linux kernel bonding code always chooses the MAC address of
707 * the first slave added to a bond, and the Fedora networking
708 * scripts always add slaves to a bond in alphabetical order, so
709 * for compatibility we choose the interface with the name that is
710 * first in alphabetical order. */
711 iface
= port
->ifaces
[0];
712 for (j
= 1; j
< port
->n_ifaces
; j
++) {
713 struct iface
*candidate
= port
->ifaces
[j
];
714 if (strcmp(candidate
->name
, iface
->name
) < 0) {
719 /* The local port doesn't count (since we're trying to choose its
720 * MAC address anyway). Other internal ports don't count because
721 * we really want a physical MAC if we can get it, and internal
722 * ports typically have randomly generated MACs. */
723 if (iface
->dp_ifidx
== ODPP_LOCAL
724 || cfg_get_bool(0, "iface.%s.internal", iface
->name
)) {
729 error
= netdev_get_etheraddr(iface
->netdev
, iface_ea
);
731 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
732 VLOG_ERR_RL(&rl
, "failed to obtain Ethernet address of %s: %s",
733 iface
->name
, strerror(error
));
738 /* Compare against our current choice. */
739 if (!eth_addr_is_multicast(iface_ea
) &&
740 !eth_addr_is_reserved(iface_ea
) &&
741 !eth_addr_is_zero(iface_ea
) &&
742 memcmp(iface_ea
, ea
, ETH_ADDR_LEN
) < 0)
744 *hw_addr_iface
= iface
;
747 if (eth_addr_is_multicast(ea
) || eth_addr_is_vif(ea
)) {
748 memcpy(ea
, br
->default_ea
, ETH_ADDR_LEN
);
749 *hw_addr_iface
= NULL
;
750 VLOG_WARN("bridge %s: using default bridge Ethernet "
751 "address "ETH_ADDR_FMT
, br
->name
, ETH_ADDR_ARGS(ea
));
753 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT
,
754 br
->name
, ETH_ADDR_ARGS(ea
));
758 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
759 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
760 * an interface on 'br', then that interface must be passed in as
761 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
762 * 'hw_addr_iface' must be passed in as a null pointer. */
764 bridge_pick_datapath_id(struct bridge
*br
,
765 const uint8_t bridge_ea
[ETH_ADDR_LEN
],
766 struct iface
*hw_addr_iface
)
769 * The procedure for choosing a bridge MAC address will, in the most
770 * ordinary case, also choose a unique MAC that we can use as a datapath
771 * ID. In some special cases, though, multiple bridges will end up with
772 * the same MAC address. This is OK for the bridges, but it will confuse
773 * the OpenFlow controller, because each datapath needs a unique datapath
776 * Datapath IDs must be unique. It is also very desirable that they be
777 * stable from one run to the next, so that policy set on a datapath
782 dpid
= cfg_get_dpid(0, "bridge.%s.datapath-id", br
->name
);
789 if (!netdev_get_vlan_vid(hw_addr_iface
->netdev
, &vlan
)) {
791 * A bridge whose MAC address is taken from a VLAN network device
792 * (that is, a network device created with vconfig(8) or similar
793 * tool) will have the same MAC address as a bridge on the VLAN
794 * device's physical network device.
796 * Handle this case by hashing the physical network device MAC
797 * along with the VLAN identifier.
799 uint8_t buf
[ETH_ADDR_LEN
+ 2];
800 memcpy(buf
, bridge_ea
, ETH_ADDR_LEN
);
801 buf
[ETH_ADDR_LEN
] = vlan
>> 8;
802 buf
[ETH_ADDR_LEN
+ 1] = vlan
;
803 return dpid_from_hash(buf
, sizeof buf
);
806 * Assume that this bridge's MAC address is unique, since it
807 * doesn't fit any of the cases we handle specially.
812 * A purely internal bridge, that is, one that has no non-virtual
813 * network devices on it at all, is more difficult because it has no
814 * natural unique identifier at all.
816 * When the host is a XenServer, we handle this case by hashing the
817 * host's UUID with the name of the bridge. Names of bridges are
818 * persistent across XenServer reboots, although they can be reused if
819 * an internal network is destroyed and then a new one is later
820 * created, so this is fairly effective.
822 * When the host is not a XenServer, we punt by using a random MAC
823 * address on each run.
825 const char *host_uuid
= xenserver_get_host_uuid();
827 char *combined
= xasprintf("%s,%s", host_uuid
, br
->name
);
828 dpid
= dpid_from_hash(combined
, strlen(combined
));
834 return eth_addr_to_uint64(bridge_ea
);
838 dpid_from_hash(const void *data
, size_t n
)
840 uint8_t hash
[SHA1_DIGEST_SIZE
];
842 BUILD_ASSERT_DECL(sizeof hash
>= ETH_ADDR_LEN
);
843 sha1_bytes(data
, n
, hash
);
844 eth_addr_mark_random(hash
);
845 return eth_addr_to_uint64(hash
);
851 struct bridge
*br
, *next
;
855 LIST_FOR_EACH_SAFE (br
, next
, struct bridge
, node
, &all_bridges
) {
856 int error
= bridge_run_one(br
);
858 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
859 VLOG_ERR_RL(&rl
, "bridge %s: datapath was destroyed externally, "
860 "forcing reconfiguration", br
->name
);
874 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
875 ofproto_wait(br
->ofproto
);
876 if (br
->controller
) {
881 mac_learning_wait(br
->ml
);
888 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
889 * configuration changes. */
891 bridge_flush(struct bridge
*br
)
893 COVERAGE_INC(bridge_flush
);
896 mac_learning_flush(br
->ml
);
900 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
902 static struct iface
*
903 bridge_get_local_iface(struct bridge
*br
)
907 for (i
= 0; i
< br
->n_ports
; i
++) {
908 struct port
*port
= br
->ports
[i
];
909 for (j
= 0; j
< port
->n_ifaces
; j
++) {
910 struct iface
*iface
= port
->ifaces
[j
];
911 if (iface
->dp_ifidx
== ODPP_LOCAL
) {
920 /* Bridge unixctl user interface functions. */
922 bridge_unixctl_fdb_show(struct unixctl_conn
*conn
, const char *args
)
924 struct ds ds
= DS_EMPTY_INITIALIZER
;
925 const struct bridge
*br
;
927 br
= bridge_lookup(args
);
929 unixctl_command_reply(conn
, 501, "no such bridge");
933 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
935 const struct mac_entry
*e
;
936 LIST_FOR_EACH (e
, struct mac_entry
, lru_node
, &br
->ml
->lrus
) {
937 if (e
->port
< 0 || e
->port
>= br
->n_ports
) {
940 ds_put_format(&ds
, "%5d %4d "ETH_ADDR_FMT
" %3d\n",
941 br
->ports
[e
->port
]->ifaces
[0]->dp_ifidx
,
942 e
->vlan
, ETH_ADDR_ARGS(e
->mac
), mac_entry_age(e
));
945 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
949 /* Bridge reconfiguration functions. */
951 static struct bridge
*
952 bridge_create(const char *name
)
957 assert(!bridge_lookup(name
));
958 br
= xcalloc(1, sizeof *br
);
960 error
= dpif_create(name
, &br
->dpif
);
961 if (error
== EEXIST
|| error
== EBUSY
) {
962 error
= dpif_open(name
, &br
->dpif
);
964 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
965 name
, strerror(error
));
969 dpif_flow_flush(br
->dpif
);
971 VLOG_ERR("failed to create datapath %s: %s", name
, strerror(error
));
976 error
= ofproto_create(name
, &bridge_ofhooks
, br
, &br
->ofproto
);
978 VLOG_ERR("failed to create switch %s: %s", name
, strerror(error
));
979 dpif_delete(br
->dpif
);
980 dpif_close(br
->dpif
);
985 br
->name
= xstrdup(name
);
986 br
->ml
= mac_learning_create();
987 br
->sent_config_request
= false;
988 eth_addr_random(br
->default_ea
);
990 port_array_init(&br
->ifaces
);
993 br
->bond_next_rebalance
= time_msec() + 10000;
995 list_push_back(&all_bridges
, &br
->node
);
997 VLOG_INFO("created bridge %s on %s", br
->name
, dpif_name(br
->dpif
));
1003 bridge_destroy(struct bridge
*br
)
1008 while (br
->n_ports
> 0) {
1009 port_destroy(br
->ports
[br
->n_ports
- 1]);
1011 list_remove(&br
->node
);
1012 error
= dpif_delete(br
->dpif
);
1013 if (error
&& error
!= ENOENT
) {
1014 VLOG_ERR("failed to delete %s: %s",
1015 dpif_name(br
->dpif
), strerror(error
));
1017 dpif_close(br
->dpif
);
1018 ofproto_destroy(br
->ofproto
);
1019 free(br
->controller
);
1020 mac_learning_destroy(br
->ml
);
1021 port_array_destroy(&br
->ifaces
);
1028 static struct bridge
*
1029 bridge_lookup(const char *name
)
1033 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
1034 if (!strcmp(br
->name
, name
)) {
1042 bridge_exists(const char *name
)
1044 return bridge_lookup(name
) ? true : false;
1048 bridge_get_datapathid(const char *name
)
1050 struct bridge
*br
= bridge_lookup(name
);
1051 return br
? ofproto_get_datapath_id(br
->ofproto
) : 0;
1054 /* Handle requests for a listing of all flows known by the OpenFlow
1055 * stack, including those normally hidden. */
1057 bridge_unixctl_dump_flows(struct unixctl_conn
*conn
, const char *args
)
1062 br
= bridge_lookup(args
);
1064 unixctl_command_reply(conn
, 501, "Unknown bridge");
1069 ofproto_get_all_flows(br
->ofproto
, &results
);
1071 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
1072 ds_destroy(&results
);
1076 bridge_run_one(struct bridge
*br
)
1080 error
= ofproto_run1(br
->ofproto
);
1086 mac_learning_run(br
->ml
, ofproto_get_revalidate_set(br
->ofproto
));
1091 error
= ofproto_run2(br
->ofproto
, br
->flush
);
1098 bridge_get_controller(const struct bridge
*br
)
1100 const char *controller
;
1102 controller
= cfg_get_string(0, "bridge.%s.controller", br
->name
);
1104 controller
= cfg_get_string(0, "mgmt.controller");
1106 return controller
&& controller
[0] ? controller
: NULL
;
1110 check_duplicate_ifaces(struct bridge
*br
, struct iface
*iface
, void *ifaces_
)
1112 struct svec
*ifaces
= ifaces_
;
1113 if (!svec_contains(ifaces
, iface
->name
)) {
1114 svec_add(ifaces
, iface
->name
);
1118 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1120 br
->name
, iface
->name
, iface
->port
->name
);
1126 bridge_reconfigure_one(struct bridge
*br
)
1128 struct svec old_ports
, new_ports
, ifaces
;
1129 struct svec listeners
, old_listeners
;
1130 struct svec snoops
, old_snoops
;
1133 /* Collect old ports. */
1134 svec_init(&old_ports
);
1135 for (i
= 0; i
< br
->n_ports
; i
++) {
1136 svec_add(&old_ports
, br
->ports
[i
]->name
);
1138 svec_sort(&old_ports
);
1139 assert(svec_is_unique(&old_ports
));
1141 /* Collect new ports. */
1142 svec_init(&new_ports
);
1143 cfg_get_all_keys(&new_ports
, "bridge.%s.port", br
->name
);
1144 svec_sort(&new_ports
);
1145 if (bridge_get_controller(br
)) {
1146 char local_name
[IF_NAMESIZE
];
1149 error
= dpif_port_get_name(br
->dpif
, ODPP_LOCAL
,
1150 local_name
, sizeof local_name
);
1151 if (!error
&& !svec_contains(&new_ports
, local_name
)) {
1152 svec_add(&new_ports
, local_name
);
1153 svec_sort(&new_ports
);
1156 if (!svec_is_unique(&new_ports
)) {
1157 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1158 br
->name
, svec_get_duplicate(&new_ports
));
1159 svec_unique(&new_ports
);
1162 ofproto_set_mgmt_id(br
->ofproto
, mgmt_id
);
1164 /* Get rid of deleted ports and add new ports. */
1165 for (i
= 0; i
< br
->n_ports
; ) {
1166 struct port
*port
= br
->ports
[i
];
1167 if (!svec_contains(&new_ports
, port
->name
)) {
1173 for (i
= 0; i
< new_ports
.n
; i
++) {
1174 const char *name
= new_ports
.names
[i
];
1175 if (!svec_contains(&old_ports
, name
)) {
1176 port_create(br
, name
);
1179 svec_destroy(&old_ports
);
1180 svec_destroy(&new_ports
);
1182 /* Reconfigure all ports. */
1183 for (i
= 0; i
< br
->n_ports
; i
++) {
1184 port_reconfigure(br
->ports
[i
]);
1187 /* Check and delete duplicate interfaces. */
1189 iterate_and_prune_ifaces(br
, check_duplicate_ifaces
, &ifaces
);
1190 svec_destroy(&ifaces
);
1192 /* Delete all flows if we're switching from connected to standalone or vice
1193 * versa. (XXX Should we delete all flows if we are switching from one
1194 * controller to another?) */
1196 /* Configure OpenFlow management listeners. */
1197 svec_init(&listeners
);
1198 cfg_get_all_strings(&listeners
, "bridge.%s.openflow.listeners", br
->name
);
1200 svec_add_nocopy(&listeners
, xasprintf("punix:%s/%s.mgmt",
1201 ovs_rundir
, br
->name
));
1202 } else if (listeners
.n
== 1 && !strcmp(listeners
.names
[0], "none")) {
1203 svec_clear(&listeners
);
1205 svec_sort_unique(&listeners
);
1207 svec_init(&old_listeners
);
1208 ofproto_get_listeners(br
->ofproto
, &old_listeners
);
1209 svec_sort_unique(&old_listeners
);
1211 if (!svec_equal(&listeners
, &old_listeners
)) {
1212 ofproto_set_listeners(br
->ofproto
, &listeners
);
1214 svec_destroy(&listeners
);
1215 svec_destroy(&old_listeners
);
1217 /* Configure OpenFlow controller connection snooping. */
1219 cfg_get_all_strings(&snoops
, "bridge.%s.openflow.snoops", br
->name
);
1221 svec_add_nocopy(&snoops
, xasprintf("punix:%s/%s.snoop",
1222 ovs_rundir
, br
->name
));
1223 } else if (snoops
.n
== 1 && !strcmp(snoops
.names
[0], "none")) {
1224 svec_clear(&snoops
);
1226 svec_sort_unique(&snoops
);
1228 svec_init(&old_snoops
);
1229 ofproto_get_snoops(br
->ofproto
, &old_snoops
);
1230 svec_sort_unique(&old_snoops
);
1232 if (!svec_equal(&snoops
, &old_snoops
)) {
1233 ofproto_set_snoops(br
->ofproto
, &snoops
);
1235 svec_destroy(&snoops
);
1236 svec_destroy(&old_snoops
);
1238 mirror_reconfigure(br
);
1242 bridge_reconfigure_controller(struct bridge
*br
)
1244 char *pfx
= xasprintf("bridge.%s.controller", br
->name
);
1245 const char *controller
;
1247 controller
= bridge_get_controller(br
);
1248 if ((br
->controller
!= NULL
) != (controller
!= NULL
)) {
1249 ofproto_flush_flows(br
->ofproto
);
1251 free(br
->controller
);
1252 br
->controller
= controller
? xstrdup(controller
) : NULL
;
1255 const char *fail_mode
;
1256 int max_backoff
, probe
;
1257 int rate_limit
, burst_limit
;
1259 if (!strcmp(controller
, "discover")) {
1260 bool update_resolv_conf
= true;
1262 if (cfg_has("%s.update-resolv.conf", pfx
)) {
1263 update_resolv_conf
= cfg_get_bool(0, "%s.update-resolv.conf",
1266 ofproto_set_discovery(br
->ofproto
, true,
1267 cfg_get_string(0, "%s.accept-regex", pfx
),
1268 update_resolv_conf
);
1270 struct iface
*local_iface
;
1273 in_band
= (!cfg_is_valid(CFG_BOOL
| CFG_REQUIRED
,
1275 || cfg_get_bool(0, "%s.in-band", pfx
));
1276 ofproto_set_discovery(br
->ofproto
, false, NULL
, NULL
);
1277 ofproto_set_in_band(br
->ofproto
, in_band
);
1279 local_iface
= bridge_get_local_iface(br
);
1281 && cfg_is_valid(CFG_IP
| CFG_REQUIRED
, "%s.ip", pfx
)) {
1282 struct netdev
*netdev
= local_iface
->netdev
;
1283 struct in_addr ip
, mask
, gateway
;
1284 ip
.s_addr
= cfg_get_ip(0, "%s.ip", pfx
);
1285 mask
.s_addr
= cfg_get_ip(0, "%s.netmask", pfx
);
1286 gateway
.s_addr
= cfg_get_ip(0, "%s.gateway", pfx
);
1288 netdev_turn_flags_on(netdev
, NETDEV_UP
, true);
1290 mask
.s_addr
= guess_netmask(ip
.s_addr
);
1292 if (!netdev_set_in4(netdev
, ip
, mask
)) {
1293 VLOG_INFO("bridge %s: configured IP address "IP_FMT
", "
1295 br
->name
, IP_ARGS(&ip
.s_addr
),
1296 IP_ARGS(&mask
.s_addr
));
1299 if (gateway
.s_addr
) {
1300 if (!netdev_add_router(netdev
, gateway
)) {
1301 VLOG_INFO("bridge %s: configured gateway "IP_FMT
,
1302 br
->name
, IP_ARGS(&gateway
.s_addr
));
1308 fail_mode
= cfg_get_string(0, "%s.fail-mode", pfx
);
1310 fail_mode
= cfg_get_string(0, "mgmt.fail-mode");
1312 ofproto_set_failure(br
->ofproto
,
1314 || !strcmp(fail_mode
, "standalone")
1315 || !strcmp(fail_mode
, "open")));
1317 probe
= cfg_get_int(0, "%s.inactivity-probe", pfx
);
1319 probe
= cfg_get_int(0, "mgmt.inactivity-probe");
1324 ofproto_set_probe_interval(br
->ofproto
, probe
);
1326 max_backoff
= cfg_get_int(0, "%s.max-backoff", pfx
);
1328 max_backoff
= cfg_get_int(0, "mgmt.max-backoff");
1333 ofproto_set_max_backoff(br
->ofproto
, max_backoff
);
1335 rate_limit
= cfg_get_int(0, "%s.rate-limit", pfx
);
1337 rate_limit
= cfg_get_int(0, "mgmt.rate-limit");
1339 burst_limit
= cfg_get_int(0, "%s.burst-limit", pfx
);
1341 burst_limit
= cfg_get_int(0, "mgmt.burst-limit");
1343 ofproto_set_rate_limit(br
->ofproto
, rate_limit
, burst_limit
);
1345 ofproto_set_stp(br
->ofproto
, cfg_get_bool(0, "%s.stp", pfx
));
1347 if (cfg_has("%s.commands.acl", pfx
)) {
1348 struct svec command_acls
;
1351 svec_init(&command_acls
);
1352 cfg_get_all_strings(&command_acls
, "%s.commands.acl", pfx
);
1353 command_acl
= svec_join(&command_acls
, ",", "");
1355 ofproto_set_remote_execution(br
->ofproto
, command_acl
,
1356 cfg_get_string(0, "%s.commands.dir",
1359 svec_destroy(&command_acls
);
1362 ofproto_set_remote_execution(br
->ofproto
, NULL
, NULL
);
1365 union ofp_action action
;
1368 /* Set up a flow that matches every packet and directs them to
1369 * OFPP_NORMAL (which goes to us). */
1370 memset(&action
, 0, sizeof action
);
1371 action
.type
= htons(OFPAT_OUTPUT
);
1372 action
.output
.len
= htons(sizeof action
);
1373 action
.output
.port
= htons(OFPP_NORMAL
);
1374 memset(&flow
, 0, sizeof flow
);
1375 ofproto_add_flow(br
->ofproto
, &flow
, OFPFW_ALL
, 0,
1378 ofproto_set_in_band(br
->ofproto
, false);
1379 ofproto_set_max_backoff(br
->ofproto
, 1);
1380 ofproto_set_probe_interval(br
->ofproto
, 5);
1381 ofproto_set_failure(br
->ofproto
, false);
1382 ofproto_set_stp(br
->ofproto
, false);
1386 ofproto_set_controller(br
->ofproto
, br
->controller
);
1390 bridge_get_all_ifaces(const struct bridge
*br
, struct svec
*ifaces
)
1395 for (i
= 0; i
< br
->n_ports
; i
++) {
1396 struct port
*port
= br
->ports
[i
];
1397 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1398 struct iface
*iface
= port
->ifaces
[j
];
1399 svec_add(ifaces
, iface
->name
);
1401 if (port
->n_ifaces
> 1
1402 && cfg_get_bool(0, "bonding.%s.fake-iface", port
->name
)) {
1403 svec_add(ifaces
, port
->name
);
1406 svec_sort_unique(ifaces
);
1409 /* For robustness, in case the administrator moves around datapath ports behind
1410 * our back, we re-check all the datapath port numbers here.
1412 * This function will set the 'dp_ifidx' members of interfaces that have
1413 * disappeared to -1, so only call this function from a context where those
1414 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1415 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1416 * datapath, which doesn't support UINT16_MAX+1 ports. */
1418 bridge_fetch_dp_ifaces(struct bridge
*br
)
1420 struct odp_port
*dpif_ports
;
1421 size_t n_dpif_ports
;
1424 /* Reset all interface numbers. */
1425 for (i
= 0; i
< br
->n_ports
; i
++) {
1426 struct port
*port
= br
->ports
[i
];
1427 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1428 struct iface
*iface
= port
->ifaces
[j
];
1429 iface
->dp_ifidx
= -1;
1432 port_array_clear(&br
->ifaces
);
1434 dpif_port_list(br
->dpif
, &dpif_ports
, &n_dpif_ports
);
1435 for (i
= 0; i
< n_dpif_ports
; i
++) {
1436 struct odp_port
*p
= &dpif_ports
[i
];
1437 struct iface
*iface
= iface_lookup(br
, p
->devname
);
1439 if (iface
->dp_ifidx
>= 0) {
1440 VLOG_WARN("%s reported interface %s twice",
1441 dpif_name(br
->dpif
), p
->devname
);
1442 } else if (iface_from_dp_ifidx(br
, p
->port
)) {
1443 VLOG_WARN("%s reported interface %"PRIu16
" twice",
1444 dpif_name(br
->dpif
), p
->port
);
1446 port_array_set(&br
->ifaces
, p
->port
, iface
);
1447 iface
->dp_ifidx
= p
->port
;
1454 /* Bridge packet processing functions. */
1457 bond_hash(const uint8_t mac
[ETH_ADDR_LEN
])
1459 return hash_bytes(mac
, ETH_ADDR_LEN
, 0) & BOND_MASK
;
1462 static struct bond_entry
*
1463 lookup_bond_entry(const struct port
*port
, const uint8_t mac
[ETH_ADDR_LEN
])
1465 return &port
->bond_hash
[bond_hash(mac
)];
1469 bond_choose_iface(const struct port
*port
)
1472 for (i
= 0; i
< port
->n_ifaces
; i
++) {
1473 if (port
->ifaces
[i
]->enabled
) {
1481 choose_output_iface(const struct port
*port
, const uint8_t *dl_src
,
1482 uint16_t *dp_ifidx
, tag_type
*tags
)
1484 struct iface
*iface
;
1486 assert(port
->n_ifaces
);
1487 if (port
->n_ifaces
== 1) {
1488 iface
= port
->ifaces
[0];
1490 struct bond_entry
*e
= lookup_bond_entry(port
, dl_src
);
1491 if (e
->iface_idx
< 0 || e
->iface_idx
>= port
->n_ifaces
1492 || !port
->ifaces
[e
->iface_idx
]->enabled
) {
1493 /* XXX select interface properly. The current interface selection
1494 * is only good for testing the rebalancing code. */
1495 e
->iface_idx
= bond_choose_iface(port
);
1496 if (e
->iface_idx
< 0) {
1497 *tags
|= port
->no_ifaces_tag
;
1500 e
->iface_tag
= tag_create_random();
1501 ((struct port
*) port
)->bond_compat_is_stale
= true;
1503 *tags
|= e
->iface_tag
;
1504 iface
= port
->ifaces
[e
->iface_idx
];
1506 *dp_ifidx
= iface
->dp_ifidx
;
1507 *tags
|= iface
->tag
; /* Currently only used for bonding. */
1512 bond_link_status_update(struct iface
*iface
, bool carrier
)
1514 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1515 struct port
*port
= iface
->port
;
1517 if ((carrier
== iface
->enabled
) == (iface
->delay_expires
== LLONG_MAX
)) {
1518 /* Nothing to do. */
1521 VLOG_INFO_RL(&rl
, "interface %s: carrier %s",
1522 iface
->name
, carrier
? "detected" : "dropped");
1523 if (carrier
== iface
->enabled
) {
1524 iface
->delay_expires
= LLONG_MAX
;
1525 VLOG_INFO_RL(&rl
, "interface %s: will not be %s",
1526 iface
->name
, carrier
? "disabled" : "enabled");
1527 } else if (carrier
&& port
->updelay
&& port
->active_iface
< 0) {
1528 iface
->delay_expires
= time_msec();
1529 VLOG_INFO_RL(&rl
, "interface %s: skipping %d ms updelay since no "
1530 "other interface is up", iface
->name
, port
->updelay
);
1532 int delay
= carrier
? port
->updelay
: port
->downdelay
;
1533 iface
->delay_expires
= time_msec() + delay
;
1536 "interface %s: will be %s if it stays %s for %d ms",
1538 carrier
? "enabled" : "disabled",
1539 carrier
? "up" : "down",
1546 bond_choose_active_iface(struct port
*port
)
1548 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1550 port
->active_iface
= bond_choose_iface(port
);
1551 port
->active_iface_tag
= tag_create_random();
1552 if (port
->active_iface
>= 0) {
1553 VLOG_INFO_RL(&rl
, "port %s: active interface is now %s",
1554 port
->name
, port
->ifaces
[port
->active_iface
]->name
);
1556 VLOG_WARN_RL(&rl
, "port %s: all ports disabled, no active interface",
1562 bond_enable_slave(struct iface
*iface
, bool enable
)
1564 struct port
*port
= iface
->port
;
1565 struct bridge
*br
= port
->bridge
;
1567 iface
->delay_expires
= LLONG_MAX
;
1568 if (enable
== iface
->enabled
) {
1572 iface
->enabled
= enable
;
1573 if (!iface
->enabled
) {
1574 VLOG_WARN("interface %s: disabled", iface
->name
);
1575 ofproto_revalidate(br
->ofproto
, iface
->tag
);
1576 if (iface
->port_ifidx
== port
->active_iface
) {
1577 ofproto_revalidate(br
->ofproto
,
1578 port
->active_iface_tag
);
1579 bond_choose_active_iface(port
);
1581 bond_send_learning_packets(port
);
1583 VLOG_WARN("interface %s: enabled", iface
->name
);
1584 if (port
->active_iface
< 0) {
1585 ofproto_revalidate(br
->ofproto
, port
->no_ifaces_tag
);
1586 bond_choose_active_iface(port
);
1587 bond_send_learning_packets(port
);
1589 iface
->tag
= tag_create_random();
1594 bond_run(struct bridge
*br
)
1598 for (i
= 0; i
< br
->n_ports
; i
++) {
1599 struct port
*port
= br
->ports
[i
];
1601 if (port
->bond_compat_is_stale
) {
1602 port
->bond_compat_is_stale
= false;
1603 port_update_bond_compat(port
);
1606 if (port
->n_ifaces
< 2) {
1609 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1610 struct iface
*iface
= port
->ifaces
[j
];
1611 if (time_msec() >= iface
->delay_expires
) {
1612 bond_enable_slave(iface
, !iface
->enabled
);
1619 bond_wait(struct bridge
*br
)
1623 for (i
= 0; i
< br
->n_ports
; i
++) {
1624 struct port
*port
= br
->ports
[i
];
1625 if (port
->n_ifaces
< 2) {
1628 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1629 struct iface
*iface
= port
->ifaces
[j
];
1630 if (iface
->delay_expires
!= LLONG_MAX
) {
1631 poll_timer_wait(iface
->delay_expires
- time_msec());
1638 set_dst(struct dst
*p
, const flow_t
*flow
,
1639 const struct port
*in_port
, const struct port
*out_port
,
1644 * XXX This uses too many tags: any broadcast flow will get one tag per
1645 * destination port, and thus a broadcast on a switch of any size is likely
1646 * to have all tag bits set. We should figure out a way to be smarter.
1648 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1649 *tags
|= out_port
->stp_state_tag
;
1650 if (!(out_port
->stp_state
& (STP_DISABLED
| STP_FORWARDING
))) {
1654 p
->vlan
= (out_port
->vlan
>= 0 ? OFP_VLAN_NONE
1655 : in_port
->vlan
>= 0 ? in_port
->vlan
1656 : ntohs(flow
->dl_vlan
));
1657 return choose_output_iface(out_port
, flow
->dl_src
, &p
->dp_ifidx
, tags
);
1661 swap_dst(struct dst
*p
, struct dst
*q
)
1663 struct dst tmp
= *p
;
1668 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1669 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1670 * that we push to the datapath. We could in fact fully sort the array by
1671 * vlan, but in most cases there are at most two different vlan tags so that's
1672 * possibly overkill.) */
1674 partition_dsts(struct dst
*dsts
, size_t n_dsts
, int vlan
)
1676 struct dst
*first
= dsts
;
1677 struct dst
*last
= dsts
+ n_dsts
;
1679 while (first
!= last
) {
1681 * - All dsts < first have vlan == 'vlan'.
1682 * - All dsts >= last have vlan != 'vlan'.
1683 * - first < last. */
1684 while (first
->vlan
== vlan
) {
1685 if (++first
== last
) {
1690 /* Same invariants, plus one additional:
1691 * - first->vlan != vlan.
1693 while (last
[-1].vlan
!= vlan
) {
1694 if (--last
== first
) {
1699 /* Same invariants, plus one additional:
1700 * - last[-1].vlan == vlan.*/
1701 swap_dst(first
++, --last
);
1706 mirror_mask_ffs(mirror_mask_t mask
)
1708 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask
));
1713 dst_is_duplicate(const struct dst
*dsts
, size_t n_dsts
,
1714 const struct dst
*test
)
1717 for (i
= 0; i
< n_dsts
; i
++) {
1718 if (dsts
[i
].vlan
== test
->vlan
&& dsts
[i
].dp_ifidx
== test
->dp_ifidx
) {
1726 port_trunks_vlan(const struct port
*port
, uint16_t vlan
)
1728 return port
->vlan
< 0 && bitmap_is_set(port
->trunks
, vlan
);
1732 port_includes_vlan(const struct port
*port
, uint16_t vlan
)
1734 return vlan
== port
->vlan
|| port_trunks_vlan(port
, vlan
);
1738 compose_dsts(const struct bridge
*br
, const flow_t
*flow
, uint16_t vlan
,
1739 const struct port
*in_port
, const struct port
*out_port
,
1740 struct dst dsts
[], tag_type
*tags
)
1742 mirror_mask_t mirrors
= in_port
->src_mirrors
;
1743 struct dst
*dst
= dsts
;
1746 *tags
|= in_port
->stp_state_tag
;
1747 if (out_port
== FLOOD_PORT
) {
1748 /* XXX use ODP_FLOOD if no vlans or bonding. */
1749 /* XXX even better, define each VLAN as a datapath port group */
1750 for (i
= 0; i
< br
->n_ports
; i
++) {
1751 struct port
*port
= br
->ports
[i
];
1752 if (port
!= in_port
&& port_includes_vlan(port
, vlan
)
1753 && !port
->is_mirror_output_port
1754 && set_dst(dst
, flow
, in_port
, port
, tags
)) {
1755 mirrors
|= port
->dst_mirrors
;
1759 } else if (out_port
&& set_dst(dst
, flow
, in_port
, out_port
, tags
)) {
1760 mirrors
|= out_port
->dst_mirrors
;
1765 struct mirror
*m
= br
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
1766 if (!m
->n_vlans
|| vlan_is_mirrored(m
, vlan
)) {
1768 if (set_dst(dst
, flow
, in_port
, m
->out_port
, tags
)
1769 && !dst_is_duplicate(dsts
, dst
- dsts
, dst
)) {
1773 for (i
= 0; i
< br
->n_ports
; i
++) {
1774 struct port
*port
= br
->ports
[i
];
1775 if (port_includes_vlan(port
, m
->out_vlan
)
1776 && set_dst(dst
, flow
, in_port
, port
, tags
)
1777 && !dst_is_duplicate(dsts
, dst
- dsts
, dst
))
1779 if (port
->vlan
< 0) {
1780 dst
->vlan
= m
->out_vlan
;
1782 if (dst
->dp_ifidx
== flow
->in_port
1783 && dst
->vlan
== vlan
) {
1784 /* Don't send out input port on same VLAN. */
1792 mirrors
&= mirrors
- 1;
1795 partition_dsts(dsts
, dst
- dsts
, ntohs(flow
->dl_vlan
));
1800 print_dsts(const struct dst
*dsts
, size_t n
)
1802 for (; n
--; dsts
++) {
1803 printf(">p%"PRIu16
, dsts
->dp_ifidx
);
1804 if (dsts
->vlan
!= OFP_VLAN_NONE
) {
1805 printf("v%"PRIu16
, dsts
->vlan
);
1811 compose_actions(struct bridge
*br
, const flow_t
*flow
, uint16_t vlan
,
1812 const struct port
*in_port
, const struct port
*out_port
,
1813 tag_type
*tags
, struct odp_actions
*actions
)
1815 struct dst dsts
[DP_MAX_PORTS
* (MAX_MIRRORS
+ 1)];
1817 const struct dst
*p
;
1820 n_dsts
= compose_dsts(br
, flow
, vlan
, in_port
, out_port
, dsts
, tags
);
1822 cur_vlan
= ntohs(flow
->dl_vlan
);
1823 for (p
= dsts
; p
< &dsts
[n_dsts
]; p
++) {
1824 union odp_action
*a
;
1825 if (p
->vlan
!= cur_vlan
) {
1826 if (p
->vlan
== OFP_VLAN_NONE
) {
1827 odp_actions_add(actions
, ODPAT_STRIP_VLAN
);
1829 a
= odp_actions_add(actions
, ODPAT_SET_VLAN_VID
);
1830 a
->vlan_vid
.vlan_vid
= htons(p
->vlan
);
1834 a
= odp_actions_add(actions
, ODPAT_OUTPUT
);
1835 a
->output
.port
= p
->dp_ifidx
;
1840 is_bcast_arp_reply(const flow_t
*flow
, const struct ofpbuf
*packet
)
1842 struct arp_eth_header
*arp
= (struct arp_eth_header
*) packet
->data
;
1843 return (flow
->dl_type
== htons(ETH_TYPE_ARP
)
1844 && eth_addr_is_broadcast(flow
->dl_dst
)
1845 && packet
->size
>= sizeof(struct arp_eth_header
)
1846 && arp
->ar_op
== ARP_OP_REQUEST
);
1849 /* If the composed actions may be applied to any packet in the given 'flow',
1850 * returns true. Otherwise, the actions should only be applied to 'packet', or
1851 * not at all, if 'packet' was NULL. */
1853 process_flow(struct bridge
*br
, const flow_t
*flow
,
1854 const struct ofpbuf
*packet
, struct odp_actions
*actions
,
1857 struct iface
*in_iface
;
1858 struct port
*in_port
;
1859 struct port
*out_port
= NULL
; /* By default, drop the packet/flow. */
1862 /* Find the interface and port structure for the received packet. */
1863 in_iface
= iface_from_dp_ifidx(br
, flow
->in_port
);
1865 /* No interface? Something fishy... */
1866 if (packet
!= NULL
) {
1867 /* Odd. A few possible reasons here:
1869 * - We deleted an interface but there are still a few packets
1870 * queued up from it.
1872 * - Someone externally added an interface (e.g. with "ovs-dpctl
1873 * add-if") that we don't know about.
1875 * - Packet arrived on the local port but the local port is not
1876 * one of our bridge ports.
1878 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1880 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
1881 "interface %"PRIu16
, br
->name
, flow
->in_port
);
1884 /* Return without adding any actions, to drop packets on this flow. */
1887 in_port
= in_iface
->port
;
1889 /* Figure out what VLAN this packet belongs to.
1891 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1892 * belongs to VLAN 0, so we should treat both cases identically. (In the
1893 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1894 * presumably to allow a priority to be specified. In the latter case, the
1895 * packet does not have any 802.1Q header.) */
1896 vlan
= ntohs(flow
->dl_vlan
);
1897 if (vlan
== OFP_VLAN_NONE
) {
1900 if (in_port
->vlan
>= 0) {
1902 /* XXX support double tagging? */
1903 if (packet
!= NULL
) {
1904 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1905 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" tagged "
1906 "packet received on port %s configured with "
1907 "implicit VLAN %"PRIu16
,
1908 br
->name
, ntohs(flow
->dl_vlan
),
1909 in_port
->name
, in_port
->vlan
);
1913 vlan
= in_port
->vlan
;
1915 if (!port_includes_vlan(in_port
, vlan
)) {
1916 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1917 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %d tagged "
1918 "packet received on port %s not configured for "
1920 br
->name
, vlan
, in_port
->name
, vlan
);
1925 /* Drop frames for ports that STP wants entirely killed (both for
1926 * forwarding and for learning). Later, after we do learning, we'll drop
1927 * the frames that STP wants to do learning but not forwarding on. */
1928 if (in_port
->stp_state
& (STP_LISTENING
| STP_BLOCKING
)) {
1932 /* Drop frames for reserved multicast addresses. */
1933 if (eth_addr_is_reserved(flow
->dl_dst
)) {
1937 /* Drop frames on ports reserved for mirroring. */
1938 if (in_port
->is_mirror_output_port
) {
1939 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1940 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port %s, "
1941 "which is reserved exclusively for mirroring",
1942 br
->name
, in_port
->name
);
1946 /* Multicast (and broadcast) packets on bonds need special attention, to
1947 * avoid receiving duplicates. */
1948 if (in_port
->n_ifaces
> 1 && eth_addr_is_multicast(flow
->dl_dst
)) {
1949 *tags
|= in_port
->active_iface_tag
;
1950 if (in_port
->active_iface
!= in_iface
->port_ifidx
) {
1951 /* Drop all multicast packets on inactive slaves. */
1954 /* Drop all multicast packets for which we have learned a different
1955 * input port, because we probably sent the packet on one slaves
1956 * and got it back on the active slave. Broadcast ARP replies are
1957 * an exception to this rule: the host has moved to another
1959 int src_idx
= mac_learning_lookup(br
->ml
, flow
->dl_src
, vlan
);
1960 if (src_idx
!= -1 && src_idx
!= in_port
->port_idx
) {
1962 if (!is_bcast_arp_reply(flow
, packet
)) {
1966 /* No way to know whether it's an ARP reply, because the
1967 * flow entry doesn't include enough information and we
1968 * don't have a packet. Punt. */
1976 out_port
= FLOOD_PORT
;
1980 /* Learn source MAC (but don't try to learn from revalidation). */
1982 tag_type rev_tag
= mac_learning_learn(br
->ml
, flow
->dl_src
,
1983 vlan
, in_port
->port_idx
);
1985 /* The log messages here could actually be useful in debugging,
1986 * so keep the rate limit relatively high. */
1987 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30,
1989 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
1990 "on port %s in VLAN %d",
1991 br
->name
, ETH_ADDR_ARGS(flow
->dl_src
),
1992 in_port
->name
, vlan
);
1993 ofproto_revalidate(br
->ofproto
, rev_tag
);
1997 /* Determine output port. */
1998 out_port_idx
= mac_learning_lookup_tag(br
->ml
, flow
->dl_dst
, vlan
,
2000 if (out_port_idx
>= 0 && out_port_idx
< br
->n_ports
) {
2001 out_port
= br
->ports
[out_port_idx
];
2005 /* Don't send packets out their input ports. Don't forward frames that STP
2006 * wants us to discard. */
2007 if (in_port
== out_port
|| in_port
->stp_state
== STP_LEARNING
) {
2012 compose_actions(br
, flow
, vlan
, in_port
, out_port
, tags
, actions
);
2015 * We send out only a single packet, instead of setting up a flow, if the
2016 * packet is an ARP directed to broadcast that arrived on a bonded
2017 * interface. In such a situation ARP requests and replies must be handled
2018 * differently, but OpenFlow unfortunately can't distinguish them.
2020 return (in_port
->n_ifaces
< 2
2021 || flow
->dl_type
!= htons(ETH_TYPE_ARP
)
2022 || !eth_addr_is_broadcast(flow
->dl_dst
));
2025 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
2028 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason
,
2029 const struct ofp_phy_port
*opp
,
2032 struct bridge
*br
= br_
;
2033 struct iface
*iface
;
2036 iface
= iface_from_dp_ifidx(br
, ofp_port_to_odp_port(opp
->port_no
));
2042 if (reason
== OFPPR_DELETE
) {
2043 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
2044 br
->name
, iface
->name
);
2045 iface_destroy(iface
);
2046 if (!port
->n_ifaces
) {
2047 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
2048 br
->name
, port
->name
);
2054 if (port
->n_ifaces
> 1) {
2055 bool up
= !(opp
->state
& OFPPS_LINK_DOWN
);
2056 bond_link_status_update(iface
, up
);
2057 port_update_bond_compat(port
);
2063 bridge_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
2064 struct odp_actions
*actions
, tag_type
*tags
, void *br_
)
2066 struct bridge
*br
= br_
;
2069 if (flow
->dl_type
== htons(OFP_DL_TYPE_NOT_ETH_TYPE
)
2070 && eth_addr_equals(flow
->dl_dst
, stp_eth_addr
)) {
2071 brstp_receive(br
, flow
, payload
);
2076 COVERAGE_INC(bridge_process_flow
);
2077 return process_flow(br
, flow
, packet
, actions
, tags
);
2081 bridge_account_flow_ofhook_cb(const flow_t
*flow
,
2082 const union odp_action
*actions
,
2083 size_t n_actions
, unsigned long long int n_bytes
,
2086 struct bridge
*br
= br_
;
2087 const union odp_action
*a
;
2089 if (!br
->has_bonded_ports
) {
2093 for (a
= actions
; a
< &actions
[n_actions
]; a
++) {
2094 if (a
->type
== ODPAT_OUTPUT
) {
2095 struct port
*port
= port_from_dp_ifidx(br
, a
->output
.port
);
2096 if (port
&& port
->n_ifaces
>= 2) {
2097 struct bond_entry
*e
= lookup_bond_entry(port
, flow
->dl_src
);
2098 e
->tx_bytes
+= n_bytes
;
2105 bridge_account_checkpoint_ofhook_cb(void *br_
)
2107 struct bridge
*br
= br_
;
2110 if (!br
->has_bonded_ports
) {
2114 /* The current ofproto implementation calls this callback at least once a
2115 * second, so this timer implementation is sufficient. */
2116 if (time_msec() < br
->bond_next_rebalance
) {
2119 br
->bond_next_rebalance
= time_msec() + 10000;
2121 for (i
= 0; i
< br
->n_ports
; i
++) {
2122 struct port
*port
= br
->ports
[i
];
2123 if (port
->n_ifaces
> 1) {
2124 bond_rebalance_port(port
);
2129 static struct ofhooks bridge_ofhooks
= {
2130 bridge_port_changed_ofhook_cb
,
2131 bridge_normal_ofhook_cb
,
2132 bridge_account_flow_ofhook_cb
,
2133 bridge_account_checkpoint_ofhook_cb
,
2136 /* Bonding functions. */
2138 /* Statistics for a single interface on a bonded port, used for load-based
2139 * bond rebalancing. */
2140 struct slave_balance
{
2141 struct iface
*iface
; /* The interface. */
2142 uint64_t tx_bytes
; /* Sum of hashes[*]->tx_bytes. */
2144 /* All the "bond_entry"s that are assigned to this interface, in order of
2145 * increasing tx_bytes. */
2146 struct bond_entry
**hashes
;
2150 /* Sorts pointers to pointers to bond_entries in ascending order by the
2151 * interface to which they are assigned, and within a single interface in
2152 * ascending order of bytes transmitted. */
2154 compare_bond_entries(const void *a_
, const void *b_
)
2156 const struct bond_entry
*const *ap
= a_
;
2157 const struct bond_entry
*const *bp
= b_
;
2158 const struct bond_entry
*a
= *ap
;
2159 const struct bond_entry
*b
= *bp
;
2160 if (a
->iface_idx
!= b
->iface_idx
) {
2161 return a
->iface_idx
> b
->iface_idx
? 1 : -1;
2162 } else if (a
->tx_bytes
!= b
->tx_bytes
) {
2163 return a
->tx_bytes
> b
->tx_bytes
? 1 : -1;
2169 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2170 * *descending* order by number of bytes transmitted. */
2172 compare_slave_balance(const void *a_
, const void *b_
)
2174 const struct slave_balance
*a
= a_
;
2175 const struct slave_balance
*b
= b_
;
2176 if (a
->iface
->enabled
!= b
->iface
->enabled
) {
2177 return a
->iface
->enabled
? -1 : 1;
2178 } else if (a
->tx_bytes
!= b
->tx_bytes
) {
2179 return a
->tx_bytes
> b
->tx_bytes
? -1 : 1;
2186 swap_bals(struct slave_balance
*a
, struct slave_balance
*b
)
2188 struct slave_balance tmp
= *a
;
2193 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2194 * given that 'p' (and only 'p') might be in the wrong location.
2196 * This function invalidates 'p', since it might now be in a different memory
2199 resort_bals(struct slave_balance
*p
,
2200 struct slave_balance bals
[], size_t n_bals
)
2203 for (; p
> bals
&& p
->tx_bytes
> p
[-1].tx_bytes
; p
--) {
2204 swap_bals(p
, p
- 1);
2206 for (; p
< &bals
[n_bals
- 1] && p
->tx_bytes
< p
[1].tx_bytes
; p
++) {
2207 swap_bals(p
, p
+ 1);
2213 log_bals(const struct slave_balance
*bals
, size_t n_bals
, struct port
*port
)
2215 if (VLOG_IS_DBG_ENABLED()) {
2216 struct ds ds
= DS_EMPTY_INITIALIZER
;
2217 const struct slave_balance
*b
;
2219 for (b
= bals
; b
< bals
+ n_bals
; b
++) {
2223 ds_put_char(&ds
, ',');
2225 ds_put_format(&ds
, " %s %"PRIu64
"kB",
2226 b
->iface
->name
, b
->tx_bytes
/ 1024);
2228 if (!b
->iface
->enabled
) {
2229 ds_put_cstr(&ds
, " (disabled)");
2231 if (b
->n_hashes
> 0) {
2232 ds_put_cstr(&ds
, " (");
2233 for (i
= 0; i
< b
->n_hashes
; i
++) {
2234 const struct bond_entry
*e
= b
->hashes
[i
];
2236 ds_put_cstr(&ds
, " + ");
2238 ds_put_format(&ds
, "h%td: %"PRIu64
"kB",
2239 e
- port
->bond_hash
, e
->tx_bytes
/ 1024);
2241 ds_put_cstr(&ds
, ")");
2244 VLOG_DBG("bond %s:%s", port
->name
, ds_cstr(&ds
));
2249 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2251 bond_shift_load(struct slave_balance
*from
, struct slave_balance
*to
,
2252 struct bond_entry
*hash
)
2254 struct port
*port
= from
->iface
->port
;
2255 uint64_t delta
= hash
->tx_bytes
;
2257 VLOG_INFO("bond %s: shift %"PRIu64
"kB of load (with hash %td) "
2258 "from %s to %s (now carrying %"PRIu64
"kB and "
2259 "%"PRIu64
"kB load, respectively)",
2260 port
->name
, delta
/ 1024, hash
- port
->bond_hash
,
2261 from
->iface
->name
, to
->iface
->name
,
2262 (from
->tx_bytes
- delta
) / 1024,
2263 (to
->tx_bytes
+ delta
) / 1024);
2265 /* Delete element from from->hashes.
2267 * We don't bother to add the element to to->hashes because not only would
2268 * it require more work, the only purpose it would be to allow that hash to
2269 * be migrated to another slave in this rebalancing run, and there is no
2270 * point in doing that. */
2271 if (from
->hashes
[0] == hash
) {
2274 int i
= hash
- from
->hashes
[0];
2275 memmove(from
->hashes
+ i
, from
->hashes
+ i
+ 1,
2276 (from
->n_hashes
- (i
+ 1)) * sizeof *from
->hashes
);
2280 /* Shift load away from 'from' to 'to'. */
2281 from
->tx_bytes
-= delta
;
2282 to
->tx_bytes
+= delta
;
2284 /* Arrange for flows to be revalidated. */
2285 ofproto_revalidate(port
->bridge
->ofproto
, hash
->iface_tag
);
2286 hash
->iface_idx
= to
->iface
->port_ifidx
;
2287 hash
->iface_tag
= tag_create_random();
2291 bond_rebalance_port(struct port
*port
)
2293 struct slave_balance bals
[DP_MAX_PORTS
];
2295 struct bond_entry
*hashes
[BOND_MASK
+ 1];
2296 struct slave_balance
*b
, *from
, *to
;
2297 struct bond_entry
*e
;
2300 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2301 * descending order of tx_bytes, so that bals[0] represents the most
2302 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2305 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2306 * array for each slave_balance structure, we sort our local array of
2307 * hashes in order by slave, so that all of the hashes for a given slave
2308 * become contiguous in memory, and then we point each 'hashes' members of
2309 * a slave_balance structure to the start of a contiguous group. */
2310 n_bals
= port
->n_ifaces
;
2311 for (b
= bals
; b
< &bals
[n_bals
]; b
++) {
2312 b
->iface
= port
->ifaces
[b
- bals
];
2317 for (i
= 0; i
<= BOND_MASK
; i
++) {
2318 hashes
[i
] = &port
->bond_hash
[i
];
2320 qsort(hashes
, BOND_MASK
+ 1, sizeof *hashes
, compare_bond_entries
);
2321 for (i
= 0; i
<= BOND_MASK
; i
++) {
2323 if (e
->iface_idx
>= 0 && e
->iface_idx
< port
->n_ifaces
) {
2324 b
= &bals
[e
->iface_idx
];
2325 b
->tx_bytes
+= e
->tx_bytes
;
2327 b
->hashes
= &hashes
[i
];
2332 qsort(bals
, n_bals
, sizeof *bals
, compare_slave_balance
);
2333 log_bals(bals
, n_bals
, port
);
2335 /* Discard slaves that aren't enabled (which were sorted to the back of the
2336 * array earlier). */
2337 while (!bals
[n_bals
- 1].iface
->enabled
) {
2344 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2345 to
= &bals
[n_bals
- 1];
2346 for (from
= bals
; from
< to
; ) {
2347 uint64_t overload
= from
->tx_bytes
- to
->tx_bytes
;
2348 if (overload
< to
->tx_bytes
>> 5 || overload
< 100000) {
2349 /* The extra load on 'from' (and all less-loaded slaves), compared
2350 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2351 * it is less than ~1Mbps. No point in rebalancing. */
2353 } else if (from
->n_hashes
== 1) {
2354 /* 'from' only carries a single MAC hash, so we can't shift any
2355 * load away from it, even though we want to. */
2358 /* 'from' is carrying significantly more load than 'to', and that
2359 * load is split across at least two different hashes. Pick a hash
2360 * to migrate to 'to' (the least-loaded slave), given that doing so
2361 * must not cause 'to''s load to exceed 'from''s load.
2363 * The sort order we use means that we prefer to shift away the
2364 * smallest hashes instead of the biggest ones. There is little
2365 * reason behind this decision; we could use the opposite sort
2366 * order to shift away big hashes ahead of small ones. */
2369 for (i
= 0; i
< from
->n_hashes
; i
++) {
2370 uint64_t delta
= from
->hashes
[i
]->tx_bytes
;
2371 if (to
->tx_bytes
+ delta
< from
->tx_bytes
- delta
) {
2375 if (i
< from
->n_hashes
) {
2376 bond_shift_load(from
, to
, from
->hashes
[i
]);
2378 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2379 * point to different slave_balance structures. It is only
2380 * valid to do these two operations in a row at all because we
2381 * know that 'from' will not move past 'to' and vice versa. */
2382 resort_bals(from
, bals
, n_bals
);
2383 resort_bals(to
, bals
, n_bals
);
2387 port
->bond_compat_is_stale
= true;
2391 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2392 * historical data to decay to <1% in 7 rebalancing runs. */
2393 for (e
= &port
->bond_hash
[0]; e
<= &port
->bond_hash
[BOND_MASK
]; e
++) {
2399 bond_send_learning_packets(struct port
*port
)
2401 struct bridge
*br
= port
->bridge
;
2402 struct mac_entry
*e
;
2403 struct ofpbuf packet
;
2404 int error
, n_packets
, n_errors
;
2406 if (!port
->n_ifaces
|| port
->active_iface
< 0 || !br
->ml
) {
2410 ofpbuf_init(&packet
, 128);
2411 error
= n_packets
= n_errors
= 0;
2412 LIST_FOR_EACH (e
, struct mac_entry
, lru_node
, &br
->ml
->lrus
) {
2413 static const char s
[] = "Open vSwitch Bond Failover";
2414 union ofp_action actions
[2], *a
;
2415 struct eth_header
*eth
;
2416 struct llc_snap_header
*llc_snap
;
2422 if (e
->port
== port
->port_idx
2423 || !choose_output_iface(port
, e
->mac
, &dp_ifidx
, &tags
)) {
2427 /* Compose packet to send. */
2428 ofpbuf_clear(&packet
);
2429 eth
= ofpbuf_put_zeros(&packet
, ETH_HEADER_LEN
);
2430 llc_snap
= ofpbuf_put_zeros(&packet
, LLC_SNAP_HEADER_LEN
);
2431 ofpbuf_put(&packet
, s
, sizeof s
); /* Includes null byte. */
2432 ofpbuf_put(&packet
, e
->mac
, ETH_ADDR_LEN
);
2434 memcpy(eth
->eth_dst
, eth_addr_broadcast
, ETH_ADDR_LEN
);
2435 memcpy(eth
->eth_src
, e
->mac
, ETH_ADDR_LEN
);
2436 eth
->eth_type
= htons(packet
.size
- ETH_HEADER_LEN
);
2438 llc_snap
->llc
.llc_dsap
= LLC_DSAP_SNAP
;
2439 llc_snap
->llc
.llc_ssap
= LLC_SSAP_SNAP
;
2440 llc_snap
->llc
.llc_cntl
= LLC_CNTL_SNAP
;
2441 memcpy(llc_snap
->snap
.snap_org
, "\x00\x23\x20", 3);
2442 llc_snap
->snap
.snap_type
= htons(0xf177); /* Random number. */
2444 /* Compose actions. */
2445 memset(actions
, 0, sizeof actions
);
2448 a
->vlan_vid
.type
= htons(OFPAT_SET_VLAN_VID
);
2449 a
->vlan_vid
.len
= htons(sizeof *a
);
2450 a
->vlan_vid
.vlan_vid
= htons(e
->vlan
);
2453 a
->output
.type
= htons(OFPAT_OUTPUT
);
2454 a
->output
.len
= htons(sizeof *a
);
2455 a
->output
.port
= htons(odp_port_to_ofp_port(dp_ifidx
));
2460 flow_extract(&packet
, ODPP_NONE
, &flow
);
2461 retval
= ofproto_send_packet(br
->ofproto
, &flow
, actions
, a
- actions
,
2468 ofpbuf_uninit(&packet
);
2471 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2472 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2473 "packets, last error was: %s",
2474 port
->name
, n_errors
, n_packets
, strerror(error
));
2476 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2477 port
->name
, n_packets
);
2481 /* Bonding unixctl user interface functions. */
2484 bond_unixctl_list(struct unixctl_conn
*conn
, const char *args UNUSED
)
2486 struct ds ds
= DS_EMPTY_INITIALIZER
;
2487 const struct bridge
*br
;
2489 ds_put_cstr(&ds
, "bridge\tbond\tslaves\n");
2491 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
2494 for (i
= 0; i
< br
->n_ports
; i
++) {
2495 const struct port
*port
= br
->ports
[i
];
2496 if (port
->n_ifaces
> 1) {
2499 ds_put_format(&ds
, "%s\t%s\t", br
->name
, port
->name
);
2500 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2501 const struct iface
*iface
= port
->ifaces
[j
];
2503 ds_put_cstr(&ds
, ", ");
2505 ds_put_cstr(&ds
, iface
->name
);
2507 ds_put_char(&ds
, '\n');
2511 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
2515 static struct port
*
2516 bond_find(const char *name
)
2518 const struct bridge
*br
;
2520 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
2523 for (i
= 0; i
< br
->n_ports
; i
++) {
2524 struct port
*port
= br
->ports
[i
];
2525 if (!strcmp(port
->name
, name
) && port
->n_ifaces
> 1) {
2534 bond_unixctl_show(struct unixctl_conn
*conn
, const char *args
)
2536 struct ds ds
= DS_EMPTY_INITIALIZER
;
2537 const struct port
*port
;
2540 port
= bond_find(args
);
2542 unixctl_command_reply(conn
, 501, "no such bond");
2546 ds_put_format(&ds
, "updelay: %d ms\n", port
->updelay
);
2547 ds_put_format(&ds
, "downdelay: %d ms\n", port
->downdelay
);
2548 ds_put_format(&ds
, "next rebalance: %lld ms\n",
2549 port
->bridge
->bond_next_rebalance
- time_msec());
2550 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2551 const struct iface
*iface
= port
->ifaces
[j
];
2552 struct bond_entry
*be
;
2555 ds_put_format(&ds
, "slave %s: %s\n",
2556 iface
->name
, iface
->enabled
? "enabled" : "disabled");
2557 if (j
== port
->active_iface
) {
2558 ds_put_cstr(&ds
, "\tactive slave\n");
2560 if (iface
->delay_expires
!= LLONG_MAX
) {
2561 ds_put_format(&ds
, "\t%s expires in %lld ms\n",
2562 iface
->enabled
? "downdelay" : "updelay",
2563 iface
->delay_expires
- time_msec());
2567 for (be
= port
->bond_hash
; be
<= &port
->bond_hash
[BOND_MASK
]; be
++) {
2568 int hash
= be
- port
->bond_hash
;
2569 struct mac_entry
*me
;
2571 if (be
->iface_idx
!= j
) {
2575 ds_put_format(&ds
, "\thash %d: %lld kB load\n",
2576 hash
, be
->tx_bytes
/ 1024);
2579 if (!port
->bridge
->ml
) {
2583 LIST_FOR_EACH (me
, struct mac_entry
, lru_node
,
2584 &port
->bridge
->ml
->lrus
) {
2587 if (bond_hash(me
->mac
) == hash
2588 && me
->port
!= port
->port_idx
2589 && choose_output_iface(port
, me
->mac
, &dp_ifidx
, &tags
)
2590 && dp_ifidx
== iface
->dp_ifidx
)
2592 ds_put_format(&ds
, "\t\t"ETH_ADDR_FMT
"\n",
2593 ETH_ADDR_ARGS(me
->mac
));
2598 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
2603 bond_unixctl_migrate(struct unixctl_conn
*conn
, const char *args_
)
2605 char *args
= (char *) args_
;
2606 char *save_ptr
= NULL
;
2607 char *bond_s
, *hash_s
, *slave_s
;
2608 uint8_t mac
[ETH_ADDR_LEN
];
2610 struct iface
*iface
;
2611 struct bond_entry
*entry
;
2614 bond_s
= strtok_r(args
, " ", &save_ptr
);
2615 hash_s
= strtok_r(NULL
, " ", &save_ptr
);
2616 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
2618 unixctl_command_reply(conn
, 501,
2619 "usage: bond/migrate BOND HASH SLAVE");
2623 port
= bond_find(bond_s
);
2625 unixctl_command_reply(conn
, 501, "no such bond");
2629 if (sscanf(hash_s
, ETH_ADDR_SCAN_FMT
, ETH_ADDR_SCAN_ARGS(mac
))
2630 == ETH_ADDR_SCAN_COUNT
) {
2631 hash
= bond_hash(mac
);
2632 } else if (strspn(hash_s
, "0123456789") == strlen(hash_s
)) {
2633 hash
= atoi(hash_s
) & BOND_MASK
;
2635 unixctl_command_reply(conn
, 501, "bad hash");
2639 iface
= port_lookup_iface(port
, slave_s
);
2641 unixctl_command_reply(conn
, 501, "no such slave");
2645 if (!iface
->enabled
) {
2646 unixctl_command_reply(conn
, 501, "cannot migrate to disabled slave");
2650 entry
= &port
->bond_hash
[hash
];
2651 ofproto_revalidate(port
->bridge
->ofproto
, entry
->iface_tag
);
2652 entry
->iface_idx
= iface
->port_ifidx
;
2653 entry
->iface_tag
= tag_create_random();
2654 port
->bond_compat_is_stale
= true;
2655 unixctl_command_reply(conn
, 200, "migrated");
2659 bond_unixctl_set_active_slave(struct unixctl_conn
*conn
, const char *args_
)
2661 char *args
= (char *) args_
;
2662 char *save_ptr
= NULL
;
2663 char *bond_s
, *slave_s
;
2665 struct iface
*iface
;
2667 bond_s
= strtok_r(args
, " ", &save_ptr
);
2668 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
2670 unixctl_command_reply(conn
, 501,
2671 "usage: bond/set-active-slave BOND SLAVE");
2675 port
= bond_find(bond_s
);
2677 unixctl_command_reply(conn
, 501, "no such bond");
2681 iface
= port_lookup_iface(port
, slave_s
);
2683 unixctl_command_reply(conn
, 501, "no such slave");
2687 if (!iface
->enabled
) {
2688 unixctl_command_reply(conn
, 501, "cannot make disabled slave active");
2692 if (port
->active_iface
!= iface
->port_ifidx
) {
2693 ofproto_revalidate(port
->bridge
->ofproto
, port
->active_iface_tag
);
2694 port
->active_iface
= iface
->port_ifidx
;
2695 port
->active_iface_tag
= tag_create_random();
2696 VLOG_INFO("port %s: active interface is now %s",
2697 port
->name
, iface
->name
);
2698 bond_send_learning_packets(port
);
2699 unixctl_command_reply(conn
, 200, "done");
2701 unixctl_command_reply(conn
, 200, "no change");
2706 enable_slave(struct unixctl_conn
*conn
, const char *args_
, bool enable
)
2708 char *args
= (char *) args_
;
2709 char *save_ptr
= NULL
;
2710 char *bond_s
, *slave_s
;
2712 struct iface
*iface
;
2714 bond_s
= strtok_r(args
, " ", &save_ptr
);
2715 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
2717 unixctl_command_reply(conn
, 501,
2718 "usage: bond/enable/disable-slave BOND SLAVE");
2722 port
= bond_find(bond_s
);
2724 unixctl_command_reply(conn
, 501, "no such bond");
2728 iface
= port_lookup_iface(port
, slave_s
);
2730 unixctl_command_reply(conn
, 501, "no such slave");
2734 bond_enable_slave(iface
, enable
);
2735 unixctl_command_reply(conn
, 501, enable
? "enabled" : "disabled");
2739 bond_unixctl_enable_slave(struct unixctl_conn
*conn
, const char *args
)
2741 enable_slave(conn
, args
, true);
2745 bond_unixctl_disable_slave(struct unixctl_conn
*conn
, const char *args
)
2747 enable_slave(conn
, args
, false);
2753 unixctl_command_register("bond/list", bond_unixctl_list
);
2754 unixctl_command_register("bond/show", bond_unixctl_show
);
2755 unixctl_command_register("bond/migrate", bond_unixctl_migrate
);
2756 unixctl_command_register("bond/set-active-slave",
2757 bond_unixctl_set_active_slave
);
2758 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave
);
2759 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave
);
2762 /* Port functions. */
2765 port_create(struct bridge
*br
, const char *name
)
2769 port
= xcalloc(1, sizeof *port
);
2771 port
->port_idx
= br
->n_ports
;
2773 port
->trunks
= NULL
;
2774 port
->name
= xstrdup(name
);
2775 port
->active_iface
= -1;
2776 port
->stp_state
= STP_DISABLED
;
2777 port
->stp_state_tag
= 0;
2779 if (br
->n_ports
>= br
->allocated_ports
) {
2780 br
->ports
= x2nrealloc(br
->ports
, &br
->allocated_ports
,
2783 br
->ports
[br
->n_ports
++] = port
;
2785 VLOG_INFO("created port %s on bridge %s", port
->name
, br
->name
);
2790 port_reconfigure(struct port
*port
)
2792 bool bonded
= cfg_has_section("bonding.%s", port
->name
);
2793 struct svec old_ifaces
, new_ifaces
;
2794 unsigned long *trunks
;
2798 /* Collect old and new interfaces. */
2799 svec_init(&old_ifaces
);
2800 svec_init(&new_ifaces
);
2801 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2802 svec_add(&old_ifaces
, port
->ifaces
[i
]->name
);
2804 svec_sort(&old_ifaces
);
2806 cfg_get_all_keys(&new_ifaces
, "bonding.%s.slave", port
->name
);
2807 if (!new_ifaces
.n
) {
2808 VLOG_ERR("port %s: no interfaces specified for bonded port",
2810 } else if (new_ifaces
.n
== 1) {
2811 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2815 port
->updelay
= cfg_get_int(0, "bonding.%s.updelay", port
->name
);
2816 if (port
->updelay
< 0) {
2819 port
->downdelay
= cfg_get_int(0, "bonding.%s.downdelay", port
->name
);
2820 if (port
->downdelay
< 0) {
2821 port
->downdelay
= 0;
2824 svec_init(&new_ifaces
);
2825 svec_add(&new_ifaces
, port
->name
);
2828 /* Get rid of deleted interfaces and add new interfaces. */
2829 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2830 struct iface
*iface
= port
->ifaces
[i
];
2831 if (!svec_contains(&new_ifaces
, iface
->name
)) {
2832 iface_destroy(iface
);
2837 for (i
= 0; i
< new_ifaces
.n
; i
++) {
2838 const char *name
= new_ifaces
.names
[i
];
2839 if (!svec_contains(&old_ifaces
, name
)) {
2840 iface_create(port
, name
);
2846 if (cfg_has("vlan.%s.tag", port
->name
)) {
2848 vlan
= cfg_get_vlan(0, "vlan.%s.tag", port
->name
);
2849 if (vlan
>= 0 && vlan
<= 4095) {
2850 VLOG_DBG("port %s: assigning VLAN tag %d", port
->name
, vlan
);
2853 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2854 * they even work as-is. But they have not been tested. */
2855 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2859 if (port
->vlan
!= vlan
) {
2861 bridge_flush(port
->bridge
);
2864 /* Get trunked VLANs. */
2867 size_t n_trunks
, n_errors
;
2870 trunks
= bitmap_allocate(4096);
2871 n_trunks
= cfg_count("vlan.%s.trunks", port
->name
);
2873 for (i
= 0; i
< n_trunks
; i
++) {
2874 int trunk
= cfg_get_vlan(i
, "vlan.%s.trunks", port
->name
);
2876 bitmap_set1(trunks
, trunk
);
2882 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2883 port
->name
, n_trunks
);
2885 if (n_errors
== n_trunks
) {
2887 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2890 bitmap_set_multiple(trunks
, 0, 4096, 1);
2893 if (cfg_has("vlan.%s.trunks", port
->name
)) {
2894 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2895 port
->name
, port
->name
);
2899 ? port
->trunks
!= NULL
2900 : port
->trunks
== NULL
|| !bitmap_equal(trunks
, port
->trunks
, 4096)) {
2901 bridge_flush(port
->bridge
);
2903 bitmap_free(port
->trunks
);
2904 port
->trunks
= trunks
;
2906 svec_destroy(&old_ifaces
);
2907 svec_destroy(&new_ifaces
);
2911 port_destroy(struct port
*port
)
2914 struct bridge
*br
= port
->bridge
;
2918 proc_net_compat_update_vlan(port
->name
, NULL
, 0);
2919 proc_net_compat_update_bond(port
->name
, NULL
);
2921 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2922 struct mirror
*m
= br
->mirrors
[i
];
2923 if (m
&& m
->out_port
== port
) {
2928 while (port
->n_ifaces
> 0) {
2929 iface_destroy(port
->ifaces
[port
->n_ifaces
- 1]);
2932 del
= br
->ports
[port
->port_idx
] = br
->ports
[--br
->n_ports
];
2933 del
->port_idx
= port
->port_idx
;
2936 bitmap_free(port
->trunks
);
2943 static struct port
*
2944 port_from_dp_ifidx(const struct bridge
*br
, uint16_t dp_ifidx
)
2946 struct iface
*iface
= iface_from_dp_ifidx(br
, dp_ifidx
);
2947 return iface
? iface
->port
: NULL
;
2950 static struct port
*
2951 port_lookup(const struct bridge
*br
, const char *name
)
2955 for (i
= 0; i
< br
->n_ports
; i
++) {
2956 struct port
*port
= br
->ports
[i
];
2957 if (!strcmp(port
->name
, name
)) {
2964 static struct iface
*
2965 port_lookup_iface(const struct port
*port
, const char *name
)
2969 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2970 struct iface
*iface
= port
->ifaces
[j
];
2971 if (!strcmp(iface
->name
, name
)) {
2979 port_update_bonding(struct port
*port
)
2981 if (port
->n_ifaces
< 2) {
2982 /* Not a bonded port. */
2983 if (port
->bond_hash
) {
2984 free(port
->bond_hash
);
2985 port
->bond_hash
= NULL
;
2986 port
->bond_compat_is_stale
= true;
2989 if (!port
->bond_hash
) {
2992 port
->bond_hash
= xcalloc(BOND_MASK
+ 1, sizeof *port
->bond_hash
);
2993 for (i
= 0; i
<= BOND_MASK
; i
++) {
2994 struct bond_entry
*e
= &port
->bond_hash
[i
];
2998 port
->no_ifaces_tag
= tag_create_random();
2999 bond_choose_active_iface(port
);
3001 port
->bond_compat_is_stale
= true;
3006 port_update_bond_compat(struct port
*port
)
3008 struct compat_bond_hash compat_hashes
[BOND_MASK
+ 1];
3009 struct compat_bond bond
;
3012 if (port
->n_ifaces
< 2) {
3013 proc_net_compat_update_bond(port
->name
, NULL
);
3018 bond
.updelay
= port
->updelay
;
3019 bond
.downdelay
= port
->downdelay
;
3022 bond
.hashes
= compat_hashes
;
3023 if (port
->bond_hash
) {
3024 const struct bond_entry
*e
;
3025 for (e
= port
->bond_hash
; e
<= &port
->bond_hash
[BOND_MASK
]; e
++) {
3026 if (e
->iface_idx
>= 0 && e
->iface_idx
< port
->n_ifaces
) {
3027 struct compat_bond_hash
*cbh
= &bond
.hashes
[bond
.n_hashes
++];
3028 cbh
->hash
= e
- port
->bond_hash
;
3029 cbh
->netdev_name
= port
->ifaces
[e
->iface_idx
]->name
;
3034 bond
.n_slaves
= port
->n_ifaces
;
3035 bond
.slaves
= xmalloc(port
->n_ifaces
* sizeof *bond
.slaves
);
3036 for (i
= 0; i
< port
->n_ifaces
; i
++) {
3037 struct iface
*iface
= port
->ifaces
[i
];
3038 struct compat_bond_slave
*slave
= &bond
.slaves
[i
];
3039 slave
->name
= iface
->name
;
3040 slave
->up
= ((iface
->enabled
&& iface
->delay_expires
== LLONG_MAX
) ||
3041 (!iface
->enabled
&& iface
->delay_expires
!= LLONG_MAX
));
3045 netdev_get_etheraddr(iface
->netdev
, slave
->mac
);
3048 proc_net_compat_update_bond(port
->name
, &bond
);
3053 port_update_vlan_compat(struct port
*port
)
3055 struct bridge
*br
= port
->bridge
;
3056 char *vlandev_name
= NULL
;
3058 if (port
->vlan
> 0) {
3059 /* Figure out the name that the VLAN device should actually have, if it
3060 * existed. This takes some work because the VLAN device would not
3061 * have port->name in its name; rather, it would have the trunk port's
3062 * name, and 'port' would be attached to a bridge that also had the
3063 * VLAN device one of its ports. So we need to find a trunk port that
3064 * includes port->vlan.
3066 * There might be more than one candidate. This doesn't happen on
3067 * XenServer, so if it happens we just pick the first choice in
3068 * alphabetical order instead of creating multiple VLAN devices. */
3070 for (i
= 0; i
< br
->n_ports
; i
++) {
3071 struct port
*p
= br
->ports
[i
];
3072 if (port_trunks_vlan(p
, port
->vlan
)
3074 && (!vlandev_name
|| strcmp(p
->name
, vlandev_name
) <= 0))
3076 uint8_t ea
[ETH_ADDR_LEN
];
3077 netdev_get_etheraddr(p
->ifaces
[0]->netdev
, ea
);
3078 if (!eth_addr_is_multicast(ea
) &&
3079 !eth_addr_is_reserved(ea
) &&
3080 !eth_addr_is_zero(ea
)) {
3081 vlandev_name
= p
->name
;
3086 proc_net_compat_update_vlan(port
->name
, vlandev_name
, port
->vlan
);
3089 /* Interface functions. */
3092 iface_create(struct port
*port
, const char *name
)
3094 struct iface
*iface
;
3096 iface
= xcalloc(1, sizeof *iface
);
3098 iface
->port_ifidx
= port
->n_ifaces
;
3099 iface
->name
= xstrdup(name
);
3100 iface
->dp_ifidx
= -1;
3101 iface
->tag
= tag_create_random();
3102 iface
->delay_expires
= LLONG_MAX
;
3103 iface
->netdev
= NULL
;
3105 if (port
->n_ifaces
>= port
->allocated_ifaces
) {
3106 port
->ifaces
= x2nrealloc(port
->ifaces
, &port
->allocated_ifaces
,
3107 sizeof *port
->ifaces
);
3109 port
->ifaces
[port
->n_ifaces
++] = iface
;
3110 if (port
->n_ifaces
> 1) {
3111 port
->bridge
->has_bonded_ports
= true;
3114 VLOG_DBG("attached network device %s to port %s", iface
->name
, port
->name
);
3116 bridge_flush(port
->bridge
);
3120 iface_destroy(struct iface
*iface
)
3123 struct port
*port
= iface
->port
;
3124 struct bridge
*br
= port
->bridge
;
3125 bool del_active
= port
->active_iface
== iface
->port_ifidx
;
3128 if (iface
->dp_ifidx
>= 0) {
3129 port_array_set(&br
->ifaces
, iface
->dp_ifidx
, NULL
);
3132 del
= port
->ifaces
[iface
->port_ifidx
] = port
->ifaces
[--port
->n_ifaces
];
3133 del
->port_ifidx
= iface
->port_ifidx
;
3135 netdev_close(iface
->netdev
);
3140 ofproto_revalidate(port
->bridge
->ofproto
, port
->active_iface_tag
);
3141 bond_choose_active_iface(port
);
3142 bond_send_learning_packets(port
);
3145 bridge_flush(port
->bridge
);
3149 static struct iface
*
3150 iface_lookup(const struct bridge
*br
, const char *name
)
3154 for (i
= 0; i
< br
->n_ports
; i
++) {
3155 struct port
*port
= br
->ports
[i
];
3156 for (j
= 0; j
< port
->n_ifaces
; j
++) {
3157 struct iface
*iface
= port
->ifaces
[j
];
3158 if (!strcmp(iface
->name
, name
)) {
3166 static struct iface
*
3167 iface_from_dp_ifidx(const struct bridge
*br
, uint16_t dp_ifidx
)
3169 return port_array_get(&br
->ifaces
, dp_ifidx
);
3172 /* Port mirroring. */
3175 mirror_reconfigure(struct bridge
*br
)
3177 struct svec old_mirrors
, new_mirrors
;
3180 /* Collect old and new mirrors. */
3181 svec_init(&old_mirrors
);
3182 svec_init(&new_mirrors
);
3183 cfg_get_subsections(&new_mirrors
, "mirror.%s", br
->name
);
3184 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3185 if (br
->mirrors
[i
]) {
3186 svec_add(&old_mirrors
, br
->mirrors
[i
]->name
);
3190 /* Get rid of deleted mirrors and add new mirrors. */
3191 svec_sort(&old_mirrors
);
3192 assert(svec_is_unique(&old_mirrors
));
3193 svec_sort(&new_mirrors
);
3194 assert(svec_is_unique(&new_mirrors
));
3195 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3196 struct mirror
*m
= br
->mirrors
[i
];
3197 if (m
&& !svec_contains(&new_mirrors
, m
->name
)) {
3201 for (i
= 0; i
< new_mirrors
.n
; i
++) {
3202 const char *name
= new_mirrors
.names
[i
];
3203 if (!svec_contains(&old_mirrors
, name
)) {
3204 mirror_create(br
, name
);
3207 svec_destroy(&old_mirrors
);
3208 svec_destroy(&new_mirrors
);
3210 /* Reconfigure all mirrors. */
3211 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3212 if (br
->mirrors
[i
]) {
3213 mirror_reconfigure_one(br
->mirrors
[i
]);
3217 /* Update port reserved status. */
3218 for (i
= 0; i
< br
->n_ports
; i
++) {
3219 br
->ports
[i
]->is_mirror_output_port
= false;
3221 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3222 struct mirror
*m
= br
->mirrors
[i
];
3223 if (m
&& m
->out_port
) {
3224 m
->out_port
->is_mirror_output_port
= true;
3230 mirror_create(struct bridge
*br
, const char *name
)
3235 for (i
= 0; ; i
++) {
3236 if (i
>= MAX_MIRRORS
) {
3237 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3238 "cannot create %s", br
->name
, MAX_MIRRORS
, name
);
3241 if (!br
->mirrors
[i
]) {
3246 VLOG_INFO("created port mirror %s on bridge %s", name
, br
->name
);
3249 br
->mirrors
[i
] = m
= xcalloc(1, sizeof *m
);
3252 m
->name
= xstrdup(name
);
3253 svec_init(&m
->src_ports
);
3254 svec_init(&m
->dst_ports
);
3262 mirror_destroy(struct mirror
*m
)
3265 struct bridge
*br
= m
->bridge
;
3268 for (i
= 0; i
< br
->n_ports
; i
++) {
3269 br
->ports
[i
]->src_mirrors
&= ~(MIRROR_MASK_C(1) << m
->idx
);
3270 br
->ports
[i
]->dst_mirrors
&= ~(MIRROR_MASK_C(1) << m
->idx
);
3273 svec_destroy(&m
->src_ports
);
3274 svec_destroy(&m
->dst_ports
);
3277 m
->bridge
->mirrors
[m
->idx
] = NULL
;
3285 prune_ports(struct mirror
*m
, struct svec
*ports
)
3290 svec_sort_unique(ports
);
3293 for (i
= 0; i
< ports
->n
; i
++) {
3294 const char *name
= ports
->names
[i
];
3295 if (port_lookup(m
->bridge
, name
)) {
3296 svec_add(&tmp
, name
);
3298 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3299 m
->bridge
->name
, m
->name
, name
);
3302 svec_swap(ports
, &tmp
);
3307 prune_vlans(struct mirror
*m
, struct svec
*vlan_strings
, int **vlans
)
3311 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3312 * order won't give us numeric sort order. But that's good enough for what
3313 * we need right now. */
3314 svec_sort_unique(vlan_strings
);
3316 *vlans
= xmalloc(sizeof *vlans
* vlan_strings
->n
);
3318 for (i
= 0; i
< vlan_strings
->n
; i
++) {
3319 const char *name
= vlan_strings
->names
[i
];
3321 if (!str_to_int(name
, 10, &vlan
) || vlan
< 0 || vlan
> 4095) {
3322 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3323 m
->bridge
->name
, m
->name
, name
);
3325 (*vlans
)[n_vlans
++] = vlan
;
3332 vlan_is_mirrored(const struct mirror
*m
, int vlan
)
3336 for (i
= 0; i
< m
->n_vlans
; i
++) {
3337 if (m
->vlans
[i
] == vlan
) {
3345 port_trunks_any_mirrored_vlan(const struct mirror
*m
, const struct port
*p
)
3349 for (i
= 0; i
< m
->n_vlans
; i
++) {
3350 if (port_trunks_vlan(p
, m
->vlans
[i
])) {
3358 mirror_reconfigure_one(struct mirror
*m
)
3360 char *pfx
= xasprintf("mirror.%s.%s", m
->bridge
->name
, m
->name
);
3361 struct svec src_ports
, dst_ports
, ports
;
3362 struct svec vlan_strings
;
3363 mirror_mask_t mirror_bit
;
3364 const char *out_port_name
;
3365 struct port
*out_port
;
3370 bool mirror_all_ports
;
3372 /* Get output port. */
3373 out_port_name
= cfg_get_key(0, "mirror.%s.%s.output.port",
3374 m
->bridge
->name
, m
->name
);
3375 if (out_port_name
) {
3376 out_port
= port_lookup(m
->bridge
, out_port_name
);
3378 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3379 "named %s", pfx
, m
->bridge
->name
, out_port_name
);
3386 if (cfg_has("%s.output.vlan", pfx
)) {
3387 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3388 "ignoring %s.output.vlan", pfx
, pfx
, pfx
);
3390 } else if (cfg_has("%s.output.vlan", pfx
)) {
3392 out_vlan
= cfg_get_vlan(0, "%s.output.vlan", pfx
);
3394 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3395 "but exactly one is required; disabling port mirror %s",
3396 pfx
, pfx
, pfx
, pfx
);
3402 /* Get all the ports, and drop duplicates and ports that don't exist. */
3403 svec_init(&src_ports
);
3404 svec_init(&dst_ports
);
3406 cfg_get_all_keys(&src_ports
, "%s.select.src-port", pfx
);
3407 cfg_get_all_keys(&dst_ports
, "%s.select.dst-port", pfx
);
3408 cfg_get_all_keys(&ports
, "%s.select.port", pfx
);
3409 svec_append(&src_ports
, &ports
);
3410 svec_append(&dst_ports
, &ports
);
3411 svec_destroy(&ports
);
3412 prune_ports(m
, &src_ports
);
3413 prune_ports(m
, &dst_ports
);
3415 /* Get all the vlans, and drop duplicate and invalid vlans. */
3416 svec_init(&vlan_strings
);
3417 cfg_get_all_keys(&vlan_strings
, "%s.select.vlan", pfx
);
3418 n_vlans
= prune_vlans(m
, &vlan_strings
, &vlans
);
3419 svec_destroy(&vlan_strings
);
3421 /* Update mirror data. */
3422 if (!svec_equal(&m
->src_ports
, &src_ports
)
3423 || !svec_equal(&m
->dst_ports
, &dst_ports
)
3424 || m
->n_vlans
!= n_vlans
3425 || memcmp(m
->vlans
, vlans
, sizeof *vlans
* n_vlans
)
3426 || m
->out_port
!= out_port
3427 || m
->out_vlan
!= out_vlan
) {
3428 bridge_flush(m
->bridge
);
3430 svec_swap(&m
->src_ports
, &src_ports
);
3431 svec_swap(&m
->dst_ports
, &dst_ports
);
3434 m
->n_vlans
= n_vlans
;
3435 m
->out_port
= out_port
;
3436 m
->out_vlan
= out_vlan
;
3438 /* If no selection criteria have been given, mirror for all ports. */
3439 mirror_all_ports
= (!m
->src_ports
.n
) && (!m
->dst_ports
.n
) && (!m
->n_vlans
);
3442 mirror_bit
= MIRROR_MASK_C(1) << m
->idx
;
3443 for (i
= 0; i
< m
->bridge
->n_ports
; i
++) {
3444 struct port
*port
= m
->bridge
->ports
[i
];
3446 if (mirror_all_ports
3447 || svec_contains(&m
->src_ports
, port
->name
)
3450 ? port_trunks_any_mirrored_vlan(m
, port
)
3451 : vlan_is_mirrored(m
, port
->vlan
)))) {
3452 port
->src_mirrors
|= mirror_bit
;
3454 port
->src_mirrors
&= ~mirror_bit
;
3457 if (mirror_all_ports
|| svec_contains(&m
->dst_ports
, port
->name
)) {
3458 port
->dst_mirrors
|= mirror_bit
;
3460 port
->dst_mirrors
&= ~mirror_bit
;
3465 svec_destroy(&src_ports
);
3466 svec_destroy(&dst_ports
);
3470 /* Spanning tree protocol. */
3472 static void brstp_update_port_state(struct port
*);
3475 brstp_send_bpdu(struct ofpbuf
*pkt
, int port_no
, void *br_
)
3477 struct bridge
*br
= br_
;
3478 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3479 struct iface
*iface
= iface_from_dp_ifidx(br
, port_no
);
3481 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
3484 struct eth_header
*eth
= pkt
->l2
;
3486 netdev_get_etheraddr(iface
->netdev
, eth
->eth_src
);
3487 if (eth_addr_is_zero(eth
->eth_src
)) {
3488 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
3489 "with unknown MAC", br
->name
, port_no
);
3491 union ofp_action action
;
3494 memset(&action
, 0, sizeof action
);
3495 action
.type
= htons(OFPAT_OUTPUT
);
3496 action
.output
.len
= htons(sizeof action
);
3497 action
.output
.port
= htons(port_no
);
3499 flow_extract(pkt
, ODPP_NONE
, &flow
);
3500 ofproto_send_packet(br
->ofproto
, &flow
, &action
, 1, pkt
);
3507 brstp_reconfigure(struct bridge
*br
)
3511 if (!cfg_get_bool(0, "stp.%s.enabled", br
->name
)) {
3513 stp_destroy(br
->stp
);
3519 uint64_t bridge_address
, bridge_id
;
3520 int bridge_priority
;
3522 bridge_address
= cfg_get_mac(0, "stp.%s.address", br
->name
);
3523 if (!bridge_address
) {
3525 bridge_address
= (stp_get_bridge_id(br
->stp
)
3526 & ((UINT64_C(1) << 48) - 1));
3528 uint8_t mac
[ETH_ADDR_LEN
];
3529 eth_addr_random(mac
);
3530 bridge_address
= eth_addr_to_uint64(mac
);
3534 if (cfg_is_valid(CFG_INT
| CFG_REQUIRED
, "stp.%s.priority",
3536 bridge_priority
= cfg_get_int(0, "stp.%s.priority", br
->name
);
3538 bridge_priority
= STP_DEFAULT_BRIDGE_PRIORITY
;
3541 bridge_id
= bridge_address
| ((uint64_t) bridge_priority
<< 48);
3543 br
->stp
= stp_create(br
->name
, bridge_id
, brstp_send_bpdu
, br
);
3544 br
->stp_last_tick
= time_msec();
3547 if (bridge_id
!= stp_get_bridge_id(br
->stp
)) {
3548 stp_set_bridge_id(br
->stp
, bridge_id
);
3553 for (i
= 0; i
< br
->n_ports
; i
++) {
3554 struct port
*p
= br
->ports
[i
];
3556 struct stp_port
*sp
;
3557 int path_cost
, priority
;
3563 dp_ifidx
= p
->ifaces
[0]->dp_ifidx
;
3564 if (dp_ifidx
< 0 || dp_ifidx
>= STP_MAX_PORTS
) {
3568 sp
= stp_get_port(br
->stp
, dp_ifidx
);
3569 enable
= (!cfg_is_valid(CFG_BOOL
| CFG_REQUIRED
,
3570 "stp.%s.port.%s.enabled",
3572 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3573 br
->name
, p
->name
));
3574 if (p
->is_mirror_output_port
) {
3577 if (enable
!= (stp_port_get_state(sp
) != STP_DISABLED
)) {
3578 bridge_flush(br
); /* Might not be necessary. */
3580 stp_port_enable(sp
);
3582 stp_port_disable(sp
);
3586 path_cost
= cfg_get_int(0, "stp.%s.port.%s.path-cost",
3588 stp_port_set_path_cost(sp
, path_cost
? path_cost
: 19 /* XXX */);
3590 priority
= (cfg_is_valid(CFG_INT
| CFG_REQUIRED
,
3591 "stp.%s.port.%s.priority",
3593 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3595 : STP_DEFAULT_PORT_PRIORITY
);
3596 stp_port_set_priority(sp
, priority
);
3599 brstp_adjust_timers(br
);
3601 for (i
= 0; i
< br
->n_ports
; i
++) {
3602 brstp_update_port_state(br
->ports
[i
]);
3607 brstp_update_port_state(struct port
*p
)
3609 struct bridge
*br
= p
->bridge
;
3610 enum stp_state state
;
3612 /* Figure out new state. */
3613 state
= STP_DISABLED
;
3614 if (br
->stp
&& p
->n_ifaces
> 0) {
3615 int dp_ifidx
= p
->ifaces
[0]->dp_ifidx
;
3616 if (dp_ifidx
>= 0 && dp_ifidx
< STP_MAX_PORTS
) {
3617 state
= stp_port_get_state(stp_get_port(br
->stp
, dp_ifidx
));
3622 if (p
->stp_state
!= state
) {
3623 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3624 VLOG_INFO_RL(&rl
, "port %s: STP state changed from %s to %s",
3625 p
->name
, stp_state_name(p
->stp_state
),
3626 stp_state_name(state
));
3627 if (p
->stp_state
== STP_DISABLED
) {
3630 ofproto_revalidate(p
->bridge
->ofproto
, p
->stp_state_tag
);
3632 p
->stp_state
= state
;
3633 p
->stp_state_tag
= (p
->stp_state
== STP_DISABLED
? 0
3634 : tag_create_random());
3639 brstp_adjust_timers(struct bridge
*br
)
3641 int hello_time
= cfg_get_int(0, "stp.%s.hello-time", br
->name
);
3642 int max_age
= cfg_get_int(0, "stp.%s.max-age", br
->name
);
3643 int forward_delay
= cfg_get_int(0, "stp.%s.forward-delay", br
->name
);
3645 stp_set_hello_time(br
->stp
, hello_time
? hello_time
: 2000);
3646 stp_set_max_age(br
->stp
, max_age
? max_age
: 20000);
3647 stp_set_forward_delay(br
->stp
, forward_delay
? forward_delay
: 15000);
3651 brstp_run(struct bridge
*br
)
3654 long long int now
= time_msec();
3655 long long int elapsed
= now
- br
->stp_last_tick
;
3656 struct stp_port
*sp
;
3659 stp_tick(br
->stp
, MIN(INT_MAX
, elapsed
));
3660 br
->stp_last_tick
= now
;
3662 while (stp_get_changed_port(br
->stp
, &sp
)) {
3663 struct port
*p
= port_from_dp_ifidx(br
, stp_port_no(sp
));
3665 brstp_update_port_state(p
);
3672 brstp_wait(struct bridge
*br
)
3675 poll_timer_wait(1000);