1 /* Copyright (c) 2008, 2009, 2010 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
23 #include <sys/socket.h>
25 #include <openflow/openflow.h>
30 #include <sys/socket.h>
31 #include <sys/types.h>
37 #include "dynamic-string.h"
43 #include "mac-learning.h"
46 #include "ofp-print.h"
48 #include "ofproto/netflow.h"
49 #include "ofproto/ofproto.h"
50 #include "ovsdb-data.h"
52 #include "poll-loop.h"
53 #include "proc-net-compat.h"
57 #include "socket-util.h"
58 #include "stream-ssl.h"
60 #include "system-stats.h"
65 #include "vswitchd/vswitch-idl.h"
66 #include "xenserver.h"
69 #include "sflow_api.h"
71 VLOG_DEFINE_THIS_MODULE(bridge
)
79 /* These members are always valid. */
80 struct port
*port
; /* Containing port. */
81 size_t port_ifidx
; /* Index within containing port. */
82 char *name
; /* Host network device name. */
83 tag_type tag
; /* Tag associated with this interface. */
84 long long delay_expires
; /* Time after which 'enabled' may change. */
86 /* These members are valid only after bridge_reconfigure() causes them to
88 struct hmap_node dp_ifidx_node
; /* In struct bridge's "ifaces" hmap. */
89 int dp_ifidx
; /* Index within kernel datapath. */
90 struct netdev
*netdev
; /* Network device. */
91 bool enabled
; /* May be chosen for flows? */
92 const char *type
; /* Usually same as cfg->type. */
93 const struct ovsrec_interface
*cfg
;
96 #define BOND_MASK 0xff
98 int iface_idx
; /* Index of assigned iface, or -1 if none. */
99 uint64_t tx_bytes
; /* Count of bytes recently transmitted. */
100 tag_type iface_tag
; /* Tag associated with iface_idx. */
103 #define MAX_MIRRORS 32
104 typedef uint32_t mirror_mask_t
;
105 #define MIRROR_MASK_C(X) UINT32_C(X)
106 BUILD_ASSERT_DECL(sizeof(mirror_mask_t
) * CHAR_BIT
>= MAX_MIRRORS
);
108 struct bridge
*bridge
;
111 struct uuid uuid
; /* UUID of this "mirror" record in database. */
113 /* Selection criteria. */
114 struct shash src_ports
; /* Name is port name; data is always NULL. */
115 struct shash dst_ports
; /* Name is port name; data is always NULL. */
120 struct port
*out_port
;
124 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
126 struct bridge
*bridge
;
128 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
129 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
130 * NULL if all VLANs are trunked. */
131 const struct ovsrec_port
*cfg
;
134 /* An ordinary bridge port has 1 interface.
135 * A bridge port for bonding has at least 2 interfaces. */
136 struct iface
**ifaces
;
137 size_t n_ifaces
, allocated_ifaces
;
140 struct bond_entry
*bond_hash
; /* An array of (BOND_MASK + 1) elements. */
141 int active_iface
; /* Ifidx on which bcasts accepted, or -1. */
142 tag_type active_iface_tag
; /* Tag for bcast flows. */
143 tag_type no_ifaces_tag
; /* Tag for flows when all ifaces disabled. */
144 int updelay
, downdelay
; /* Delay before iface goes up/down, in ms. */
145 bool bond_compat_is_stale
; /* Need to call port_update_bond_compat()? */
146 bool bond_fake_iface
; /* Fake a bond interface for legacy compat? */
147 long long int bond_next_fake_iface_update
; /* Time of next update. */
148 int bond_rebalance_interval
; /* Interval between rebalances, in ms. */
149 long long int bond_next_rebalance
; /* Next rebalancing time. */
150 struct netdev_monitor
*monitor
; /* Tracks carrier up/down status. */
152 /* Port mirroring info. */
153 mirror_mask_t src_mirrors
; /* Mirrors triggered when packet received. */
154 mirror_mask_t dst_mirrors
; /* Mirrors triggered when packet sent. */
155 bool is_mirror_output_port
; /* Does port mirroring send frames here? */
158 #define DP_MAX_PORTS 255
160 struct list node
; /* Node in global list of bridges. */
161 char *name
; /* User-specified arbitrary name. */
162 struct mac_learning
*ml
; /* MAC learning table. */
163 uint8_t default_ea
[ETH_ADDR_LEN
]; /* Default MAC. */
164 const struct ovsrec_bridge
*cfg
;
166 /* OpenFlow switch processing. */
167 struct ofproto
*ofproto
; /* OpenFlow switch. */
169 /* Kernel datapath information. */
170 struct dpif
*dpif
; /* Datapath. */
171 struct hmap ifaces
; /* Contains "struct iface"s. */
175 size_t n_ports
, allocated_ports
;
176 struct shash iface_by_name
; /* "struct iface"s indexed by name. */
177 struct shash port_by_name
; /* "struct port"s indexed by name. */
180 bool has_bonded_ports
;
185 /* Port mirroring. */
186 struct mirror
*mirrors
[MAX_MIRRORS
];
189 /* List of all bridges. */
190 static struct list all_bridges
= LIST_INITIALIZER(&all_bridges
);
192 /* OVSDB IDL used to obtain configuration. */
193 static struct ovsdb_idl
*idl
;
195 /* Each time this timer expires, the bridge fetches systems and interface
196 * statistics and pushes them into the database. */
197 #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */
198 static long long int stats_timer
= LLONG_MIN
;
200 static struct bridge
*bridge_create(const struct ovsrec_bridge
*br_cfg
);
201 static void bridge_destroy(struct bridge
*);
202 static struct bridge
*bridge_lookup(const char *name
);
203 static unixctl_cb_func bridge_unixctl_dump_flows
;
204 static unixctl_cb_func bridge_unixctl_reconnect
;
205 static int bridge_run_one(struct bridge
*);
206 static size_t bridge_get_controllers(const struct bridge
*br
,
207 struct ovsrec_controller
***controllersp
);
208 static void bridge_reconfigure_one(struct bridge
*);
209 static void bridge_reconfigure_remotes(struct bridge
*,
210 const struct sockaddr_in
*managers
,
212 static void bridge_get_all_ifaces(const struct bridge
*, struct shash
*ifaces
);
213 static void bridge_fetch_dp_ifaces(struct bridge
*);
214 static void bridge_flush(struct bridge
*);
215 static void bridge_pick_local_hw_addr(struct bridge
*,
216 uint8_t ea
[ETH_ADDR_LEN
],
217 struct iface
**hw_addr_iface
);
218 static uint64_t bridge_pick_datapath_id(struct bridge
*,
219 const uint8_t bridge_ea
[ETH_ADDR_LEN
],
220 struct iface
*hw_addr_iface
);
221 static struct iface
*bridge_get_local_iface(struct bridge
*);
222 static uint64_t dpid_from_hash(const void *, size_t nbytes
);
224 static unixctl_cb_func bridge_unixctl_fdb_show
;
226 static void bond_init(void);
227 static void bond_run(struct bridge
*);
228 static void bond_wait(struct bridge
*);
229 static void bond_rebalance_port(struct port
*);
230 static void bond_send_learning_packets(struct port
*);
231 static void bond_enable_slave(struct iface
*iface
, bool enable
);
233 static struct port
*port_create(struct bridge
*, const char *name
);
234 static void port_reconfigure(struct port
*, const struct ovsrec_port
*);
235 static void port_del_ifaces(struct port
*, const struct ovsrec_port
*);
236 static void port_destroy(struct port
*);
237 static struct port
*port_lookup(const struct bridge
*, const char *name
);
238 static struct iface
*port_lookup_iface(const struct port
*, const char *name
);
239 static struct port
*port_from_dp_ifidx(const struct bridge
*,
241 static void port_update_bond_compat(struct port
*);
242 static void port_update_vlan_compat(struct port
*);
243 static void port_update_bonding(struct port
*);
245 static void mirror_create(struct bridge
*, struct ovsrec_mirror
*);
246 static void mirror_destroy(struct mirror
*);
247 static void mirror_reconfigure(struct bridge
*);
248 static void mirror_reconfigure_one(struct mirror
*, struct ovsrec_mirror
*);
249 static bool vlan_is_mirrored(const struct mirror
*, int vlan
);
251 static struct iface
*iface_create(struct port
*port
,
252 const struct ovsrec_interface
*if_cfg
);
253 static void iface_destroy(struct iface
*);
254 static struct iface
*iface_lookup(const struct bridge
*, const char *name
);
255 static struct iface
*iface_from_dp_ifidx(const struct bridge
*,
257 static void iface_set_mac(struct iface
*);
258 static void iface_set_ofport(const struct ovsrec_interface
*, int64_t ofport
);
259 static void iface_update_qos(struct iface
*, const struct ovsrec_qos
*);
261 static void shash_from_ovs_idl_map(char **keys
, char **values
, size_t n
,
264 /* Hooks into ofproto processing. */
265 static struct ofhooks bridge_ofhooks
;
267 /* Public functions. */
269 /* Initializes the bridge module, configuring it to obtain its configuration
270 * from an OVSDB server accessed over 'remote', which should be a string in a
271 * form acceptable to ovsdb_idl_create(). */
273 bridge_init(const char *remote
)
275 /* Create connection to database. */
276 idl
= ovsdb_idl_create(remote
, &ovsrec_idl_class
);
278 ovsdb_idl_set_write_only(idl
, &ovsrec_open_vswitch_col_cur_cfg
);
279 ovsdb_idl_set_write_only(idl
, &ovsrec_open_vswitch_col_statistics
);
280 ovsdb_idl_omit(idl
, &ovsrec_open_vswitch_col_external_ids
);
282 ovsdb_idl_omit(idl
, &ovsrec_bridge_col_external_ids
);
284 ovsdb_idl_omit(idl
, &ovsrec_port_col_external_ids
);
285 ovsdb_idl_omit(idl
, &ovsrec_port_col_fake_bridge
);
287 ovsdb_idl_set_write_only(idl
, &ovsrec_interface_col_ofport
);
288 ovsdb_idl_set_write_only(idl
, &ovsrec_interface_col_statistics
);
289 ovsdb_idl_omit(idl
, &ovsrec_interface_col_external_ids
);
291 /* Register unixctl commands. */
292 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show
, NULL
);
293 unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows
,
295 unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect
,
300 /* Performs configuration that is only necessary once at ovs-vswitchd startup,
301 * but for which the ovs-vswitchd configuration 'cfg' is required. */
303 bridge_configure_once(const struct ovsrec_open_vswitch
*cfg
)
305 static bool already_configured_once
;
306 struct svec bridge_names
;
307 struct svec dpif_names
, dpif_types
;
310 /* Only do this once per ovs-vswitchd run. */
311 if (already_configured_once
) {
314 already_configured_once
= true;
316 stats_timer
= time_msec() + STATS_INTERVAL
;
318 /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */
319 svec_init(&bridge_names
);
320 for (i
= 0; i
< cfg
->n_bridges
; i
++) {
321 svec_add(&bridge_names
, cfg
->bridges
[i
]->name
);
323 svec_sort(&bridge_names
);
325 /* Iterate over all system dpifs and delete any of them that do not appear
327 svec_init(&dpif_names
);
328 svec_init(&dpif_types
);
329 dp_enumerate_types(&dpif_types
);
330 for (i
= 0; i
< dpif_types
.n
; i
++) {
335 dp_enumerate_names(dpif_types
.names
[i
], &dpif_names
);
337 /* For each dpif... */
338 for (j
= 0; j
< dpif_names
.n
; j
++) {
339 retval
= dpif_open(dpif_names
.names
[j
], dpif_types
.names
[i
], &dpif
);
341 struct svec all_names
;
344 /* ...check whether any of its names is in 'bridge_names'. */
345 svec_init(&all_names
);
346 dpif_get_all_names(dpif
, &all_names
);
347 for (k
= 0; k
< all_names
.n
; k
++) {
348 if (svec_contains(&bridge_names
, all_names
.names
[k
])) {
353 /* No. Delete the dpif. */
357 svec_destroy(&all_names
);
362 svec_destroy(&bridge_names
);
363 svec_destroy(&dpif_names
);
364 svec_destroy(&dpif_types
);
367 /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges
368 * keys from "options" and "other_config", preferring "options" keys over
369 * "other_config" keys.
371 * The value strings in '*options' are taken directly from if_cfg, not copied,
372 * so the caller should not modify or free them. */
374 iface_get_options(const struct ovsrec_interface
*if_cfg
, struct shash
*options
)
378 shash_from_ovs_idl_map(if_cfg
->key_options
, if_cfg
->value_options
,
379 if_cfg
->n_options
, options
);
381 for (i
= 0; i
< if_cfg
->n_other_config
; i
++) {
382 char *key
= if_cfg
->key_other_config
[i
];
383 char *value
= if_cfg
->value_other_config
[i
];
385 if (!shash_find_data(options
, key
)) {
386 shash_add(options
, key
, value
);
388 VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts "
389 "with \"options\" key %s", if_cfg
->name
, key
, key
);
394 /* Returns the type of network device that 'iface' should have. (This is
395 * ordinarily the same type as the interface, but the network devices for
396 * "internal" ports have type "system".) */
398 iface_get_netdev_type(const struct iface
*iface
)
400 return !strcmp(iface
->type
, "internal") ? "system" : iface
->type
;
403 /* Attempt to create the network device for 'iface' through the netdev
406 create_iface_netdev(struct iface
*iface
)
408 struct netdev_options netdev_options
;
409 struct shash options
;
412 memset(&netdev_options
, 0, sizeof netdev_options
);
413 netdev_options
.name
= iface
->cfg
->name
;
414 netdev_options
.type
= iface_get_netdev_type(iface
);
415 netdev_options
.args
= &options
;
416 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
418 iface_get_options(iface
->cfg
, &options
);
420 error
= netdev_open(&netdev_options
, &iface
->netdev
);
423 netdev_get_carrier(iface
->netdev
, &iface
->enabled
);
426 shash_destroy(&options
);
432 reconfigure_iface_netdev(struct iface
*iface
)
434 const char *netdev_type
, *iface_type
;
435 struct shash options
;
438 /* Skip reconfiguration if the device has the wrong type. This shouldn't
440 iface_type
= iface_get_netdev_type(iface
);
441 netdev_type
= netdev_get_type(iface
->netdev
);
442 if (iface_type
&& strcmp(netdev_type
, iface_type
)) {
443 VLOG_WARN("%s: attempting change device type from %s to %s",
444 iface
->cfg
->name
, netdev_type
, iface_type
);
448 /* Reconfigure device. */
449 iface_get_options(iface
->cfg
, &options
);
450 error
= netdev_reconfigure(iface
->netdev
, &options
);
451 shash_destroy(&options
);
456 /* Callback for iterate_and_prune_ifaces(). */
458 check_iface(struct bridge
*br
, struct iface
*iface
, void *aux OVS_UNUSED
)
460 if (!iface
->netdev
) {
461 /* We already reported a related error, don't bother duplicating it. */
465 if (iface
->dp_ifidx
< 0) {
466 VLOG_ERR("%s interface not in %s, dropping",
467 iface
->name
, dpif_name(br
->dpif
));
471 VLOG_DBG("%s has interface %s on port %d", dpif_name(br
->dpif
),
472 iface
->name
, iface
->dp_ifidx
);
476 /* Callback for iterate_and_prune_ifaces(). */
478 set_iface_properties(struct bridge
*br OVS_UNUSED
, struct iface
*iface
,
479 void *aux OVS_UNUSED
)
481 /* Set policing attributes. */
482 netdev_set_policing(iface
->netdev
,
483 iface
->cfg
->ingress_policing_rate
,
484 iface
->cfg
->ingress_policing_burst
);
486 /* Set MAC address of internal interfaces other than the local
488 if (iface
->dp_ifidx
!= ODPP_LOCAL
&& !strcmp(iface
->type
, "internal")) {
489 iface_set_mac(iface
);
495 /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument.
496 * Deletes from 'br' all the interfaces for which 'cb' returns false, and then
497 * deletes from 'br' any ports that no longer have any interfaces. */
499 iterate_and_prune_ifaces(struct bridge
*br
,
500 bool (*cb
)(struct bridge
*, struct iface
*,
506 for (i
= 0; i
< br
->n_ports
; ) {
507 struct port
*port
= br
->ports
[i
];
508 for (j
= 0; j
< port
->n_ifaces
; ) {
509 struct iface
*iface
= port
->ifaces
[j
];
510 if (cb(br
, iface
, aux
)) {
513 iface_set_ofport(iface
->cfg
, -1);
514 iface_destroy(iface
);
518 if (port
->n_ifaces
) {
521 VLOG_ERR("%s port has no interfaces, dropping", port
->name
);
527 /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP
528 * addresses and ports into '*managersp' and '*n_managersp'. The caller is
529 * responsible for freeing '*managersp' (with free()).
531 * You may be asking yourself "why does ovs-vswitchd care?", because
532 * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd
533 * should not be and in fact is not directly involved in that. But
534 * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so
535 * it has to tell in-band control where the managers are to enable that.
538 collect_managers(const struct ovsrec_open_vswitch
*ovs_cfg
,
539 struct sockaddr_in
**managersp
, size_t *n_managersp
)
541 struct sockaddr_in
*managers
= NULL
;
542 size_t n_managers
= 0;
544 if (ovs_cfg
->n_managers
> 0) {
547 managers
= xmalloc(ovs_cfg
->n_managers
* sizeof *managers
);
548 for (i
= 0; i
< ovs_cfg
->n_managers
; i
++) {
549 const char *name
= ovs_cfg
->managers
[i
];
550 struct sockaddr_in
*sin
= &managers
[i
];
552 if ((!strncmp(name
, "tcp:", 4)
553 && inet_parse_active(name
+ 4, JSONRPC_TCP_PORT
, sin
)) ||
554 (!strncmp(name
, "ssl:", 4)
555 && inet_parse_active(name
+ 4, JSONRPC_SSL_PORT
, sin
))) {
561 *managersp
= managers
;
562 *n_managersp
= n_managers
;
566 bridge_reconfigure(const struct ovsrec_open_vswitch
*ovs_cfg
)
568 struct shash old_br
, new_br
;
569 struct shash_node
*node
;
570 struct bridge
*br
, *next
;
571 struct sockaddr_in
*managers
;
574 int sflow_bridge_number
;
576 COVERAGE_INC(bridge_reconfigure
);
578 collect_managers(ovs_cfg
, &managers
, &n_managers
);
580 /* Collect old and new bridges. */
583 LIST_FOR_EACH (br
, node
, &all_bridges
) {
584 shash_add(&old_br
, br
->name
, br
);
586 for (i
= 0; i
< ovs_cfg
->n_bridges
; i
++) {
587 const struct ovsrec_bridge
*br_cfg
= ovs_cfg
->bridges
[i
];
588 if (!shash_add_once(&new_br
, br_cfg
->name
, br_cfg
)) {
589 VLOG_WARN("more than one bridge named %s", br_cfg
->name
);
593 /* Get rid of deleted bridges and add new bridges. */
594 LIST_FOR_EACH_SAFE (br
, next
, node
, &all_bridges
) {
595 struct ovsrec_bridge
*br_cfg
= shash_find_data(&new_br
, br
->name
);
602 SHASH_FOR_EACH (node
, &new_br
) {
603 const char *br_name
= node
->name
;
604 const struct ovsrec_bridge
*br_cfg
= node
->data
;
605 br
= shash_find_data(&old_br
, br_name
);
607 /* If the bridge datapath type has changed, we need to tear it
608 * down and recreate. */
609 if (strcmp(br
->cfg
->datapath_type
, br_cfg
->datapath_type
)) {
611 bridge_create(br_cfg
);
614 bridge_create(br_cfg
);
617 shash_destroy(&old_br
);
618 shash_destroy(&new_br
);
620 /* Reconfigure all bridges. */
621 LIST_FOR_EACH (br
, node
, &all_bridges
) {
622 bridge_reconfigure_one(br
);
625 /* Add and delete ports on all datapaths.
627 * The kernel will reject any attempt to add a given port to a datapath if
628 * that port already belongs to a different datapath, so we must do all
629 * port deletions before any port additions. */
630 LIST_FOR_EACH (br
, node
, &all_bridges
) {
631 struct odp_port
*dpif_ports
;
633 struct shash want_ifaces
;
635 dpif_port_list(br
->dpif
, &dpif_ports
, &n_dpif_ports
);
636 bridge_get_all_ifaces(br
, &want_ifaces
);
637 for (i
= 0; i
< n_dpif_ports
; i
++) {
638 const struct odp_port
*p
= &dpif_ports
[i
];
639 if (!shash_find(&want_ifaces
, p
->devname
)
640 && strcmp(p
->devname
, br
->name
)) {
641 int retval
= dpif_port_del(br
->dpif
, p
->port
);
643 VLOG_ERR("failed to remove %s interface from %s: %s",
644 p
->devname
, dpif_name(br
->dpif
),
649 shash_destroy(&want_ifaces
);
652 LIST_FOR_EACH (br
, node
, &all_bridges
) {
653 struct odp_port
*dpif_ports
;
655 struct shash cur_ifaces
, want_ifaces
;
657 /* Get the set of interfaces currently in this datapath. */
658 dpif_port_list(br
->dpif
, &dpif_ports
, &n_dpif_ports
);
659 shash_init(&cur_ifaces
);
660 for (i
= 0; i
< n_dpif_ports
; i
++) {
661 const char *name
= dpif_ports
[i
].devname
;
662 shash_add_once(&cur_ifaces
, name
, &dpif_ports
[i
]);
665 /* Get the set of interfaces we want on this datapath. */
666 bridge_get_all_ifaces(br
, &want_ifaces
);
668 hmap_clear(&br
->ifaces
);
669 SHASH_FOR_EACH (node
, &want_ifaces
) {
670 const char *if_name
= node
->name
;
671 struct iface
*iface
= node
->data
;
672 bool internal
= !iface
|| !strcmp(iface
->type
, "internal");
673 struct odp_port
*dpif_port
= shash_find_data(&cur_ifaces
, if_name
);
676 /* If we have a port or a netdev already, and it's not the type we
677 * want, then delete the port (if any) and close the netdev (if
680 ? dpif_port
&& !(dpif_port
->flags
& ODP_PORT_INTERNAL
)
682 && strcmp(iface
->type
, netdev_get_type(iface
->netdev
))))
685 error
= ofproto_port_del(br
->ofproto
, dpif_port
->port
);
692 netdev_close(iface
->netdev
);
693 iface
->netdev
= NULL
;
697 /* If it's not an internal port, open (possibly create) the
700 if (!iface
->netdev
) {
701 error
= create_iface_netdev(iface
);
703 VLOG_WARN("could not create iface %s: %s", iface
->name
,
708 reconfigure_iface_netdev(iface
);
712 /* If it's not part of the datapath, add it. */
714 error
= dpif_port_add(br
->dpif
, if_name
,
715 internal
? ODP_PORT_INTERNAL
: 0, NULL
);
716 if (error
== EFBIG
) {
717 VLOG_ERR("ran out of valid port numbers on %s",
718 dpif_name(br
->dpif
));
721 VLOG_ERR("failed to add %s interface to %s: %s",
722 if_name
, dpif_name(br
->dpif
), strerror(error
));
727 /* If it's an internal port, open the netdev. */
729 if (iface
&& !iface
->netdev
) {
730 error
= create_iface_netdev(iface
);
732 VLOG_WARN("could not create iface %s: %s", iface
->name
,
738 assert(iface
->netdev
!= NULL
);
742 shash_destroy(&cur_ifaces
);
743 shash_destroy(&want_ifaces
);
745 sflow_bridge_number
= 0;
746 LIST_FOR_EACH (br
, node
, &all_bridges
) {
749 struct iface
*local_iface
;
750 struct iface
*hw_addr_iface
;
753 bridge_fetch_dp_ifaces(br
);
755 iterate_and_prune_ifaces(br
, check_iface
, NULL
);
757 /* Pick local port hardware address, datapath ID. */
758 bridge_pick_local_hw_addr(br
, ea
, &hw_addr_iface
);
759 local_iface
= bridge_get_local_iface(br
);
761 int error
= netdev_set_etheraddr(local_iface
->netdev
, ea
);
763 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
764 VLOG_ERR_RL(&rl
, "bridge %s: failed to set bridge "
765 "Ethernet address: %s",
766 br
->name
, strerror(error
));
770 dpid
= bridge_pick_datapath_id(br
, ea
, hw_addr_iface
);
771 ofproto_set_datapath_id(br
->ofproto
, dpid
);
773 dpid_string
= xasprintf("%016"PRIx64
, dpid
);
774 ovsrec_bridge_set_datapath_id(br
->cfg
, dpid_string
);
777 /* Set NetFlow configuration on this bridge. */
778 if (br
->cfg
->netflow
) {
779 struct ovsrec_netflow
*nf_cfg
= br
->cfg
->netflow
;
780 struct netflow_options opts
;
782 memset(&opts
, 0, sizeof opts
);
784 dpif_get_netflow_ids(br
->dpif
, &opts
.engine_type
, &opts
.engine_id
);
785 if (nf_cfg
->engine_type
) {
786 opts
.engine_type
= *nf_cfg
->engine_type
;
788 if (nf_cfg
->engine_id
) {
789 opts
.engine_id
= *nf_cfg
->engine_id
;
792 opts
.active_timeout
= nf_cfg
->active_timeout
;
793 if (!opts
.active_timeout
) {
794 opts
.active_timeout
= -1;
795 } else if (opts
.active_timeout
< 0) {
796 VLOG_WARN("bridge %s: active timeout interval set to negative "
797 "value, using default instead (%d seconds)", br
->name
,
798 NF_ACTIVE_TIMEOUT_DEFAULT
);
799 opts
.active_timeout
= -1;
802 opts
.add_id_to_iface
= nf_cfg
->add_id_to_interface
;
803 if (opts
.add_id_to_iface
) {
804 if (opts
.engine_id
> 0x7f) {
805 VLOG_WARN("bridge %s: netflow port mangling may conflict "
806 "with another vswitch, choose an engine id less "
807 "than 128", br
->name
);
809 if (br
->n_ports
> 508) {
810 VLOG_WARN("bridge %s: netflow port mangling will conflict "
811 "with another port when more than 508 ports are "
816 opts
.collectors
.n
= nf_cfg
->n_targets
;
817 opts
.collectors
.names
= nf_cfg
->targets
;
818 if (ofproto_set_netflow(br
->ofproto
, &opts
)) {
819 VLOG_ERR("bridge %s: problem setting netflow collectors",
823 ofproto_set_netflow(br
->ofproto
, NULL
);
826 /* Set sFlow configuration on this bridge. */
827 if (br
->cfg
->sflow
) {
828 const struct ovsrec_sflow
*sflow_cfg
= br
->cfg
->sflow
;
829 struct ovsrec_controller
**controllers
;
830 struct ofproto_sflow_options oso
;
831 size_t n_controllers
;
833 memset(&oso
, 0, sizeof oso
);
835 oso
.targets
.n
= sflow_cfg
->n_targets
;
836 oso
.targets
.names
= sflow_cfg
->targets
;
838 oso
.sampling_rate
= SFL_DEFAULT_SAMPLING_RATE
;
839 if (sflow_cfg
->sampling
) {
840 oso
.sampling_rate
= *sflow_cfg
->sampling
;
843 oso
.polling_interval
= SFL_DEFAULT_POLLING_INTERVAL
;
844 if (sflow_cfg
->polling
) {
845 oso
.polling_interval
= *sflow_cfg
->polling
;
848 oso
.header_len
= SFL_DEFAULT_HEADER_SIZE
;
849 if (sflow_cfg
->header
) {
850 oso
.header_len
= *sflow_cfg
->header
;
853 oso
.sub_id
= sflow_bridge_number
++;
854 oso
.agent_device
= sflow_cfg
->agent
;
856 oso
.control_ip
= NULL
;
857 n_controllers
= bridge_get_controllers(br
, &controllers
);
858 for (i
= 0; i
< n_controllers
; i
++) {
859 if (controllers
[i
]->local_ip
) {
860 oso
.control_ip
= controllers
[i
]->local_ip
;
864 ofproto_set_sflow(br
->ofproto
, &oso
);
866 /* Do not destroy oso.targets because it is owned by sflow_cfg. */
868 ofproto_set_sflow(br
->ofproto
, NULL
);
871 /* Update the controller and related settings. It would be more
872 * straightforward to call this from bridge_reconfigure_one(), but we
873 * can't do it there for two reasons. First, and most importantly, at
874 * that point we don't know the dp_ifidx of any interfaces that have
875 * been added to the bridge (because we haven't actually added them to
876 * the datapath). Second, at that point we haven't set the datapath ID
877 * yet; when a controller is configured, resetting the datapath ID will
878 * immediately disconnect from the controller, so it's better to set
879 * the datapath ID before the controller. */
880 bridge_reconfigure_remotes(br
, managers
, n_managers
);
882 LIST_FOR_EACH (br
, node
, &all_bridges
) {
883 for (i
= 0; i
< br
->n_ports
; i
++) {
884 struct port
*port
= br
->ports
[i
];
887 port_update_vlan_compat(port
);
888 port_update_bonding(port
);
890 for (j
= 0; j
< port
->n_ifaces
; j
++) {
891 iface_update_qos(port
->ifaces
[j
], port
->cfg
->qos
);
895 LIST_FOR_EACH (br
, node
, &all_bridges
) {
896 iterate_and_prune_ifaces(br
, set_iface_properties
, NULL
);
903 get_ovsrec_key_value(const struct ovsdb_idl_row
*row
,
904 const struct ovsdb_idl_column
*column
,
907 const struct ovsdb_datum
*datum
;
908 union ovsdb_atom atom
;
911 datum
= ovsdb_idl_get(row
, column
, OVSDB_TYPE_STRING
, OVSDB_TYPE_STRING
);
912 atom
.string
= (char *) key
;
913 idx
= ovsdb_datum_find_key(datum
, &atom
, OVSDB_TYPE_STRING
);
914 return idx
== UINT_MAX
? NULL
: datum
->values
[idx
].string
;
918 bridge_get_other_config(const struct ovsrec_bridge
*br_cfg
, const char *key
)
920 return get_ovsrec_key_value(&br_cfg
->header_
,
921 &ovsrec_bridge_col_other_config
, key
);
925 bridge_pick_local_hw_addr(struct bridge
*br
, uint8_t ea
[ETH_ADDR_LEN
],
926 struct iface
**hw_addr_iface
)
932 *hw_addr_iface
= NULL
;
934 /* Did the user request a particular MAC? */
935 hwaddr
= bridge_get_other_config(br
->cfg
, "hwaddr");
936 if (hwaddr
&& eth_addr_from_string(hwaddr
, ea
)) {
937 if (eth_addr_is_multicast(ea
)) {
938 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
939 "address "ETH_ADDR_FMT
, br
->name
, ETH_ADDR_ARGS(ea
));
940 } else if (eth_addr_is_zero(ea
)) {
941 VLOG_ERR("bridge %s: cannot set MAC address to zero", br
->name
);
947 /* Otherwise choose the minimum non-local MAC address among all of the
949 memset(ea
, 0xff, sizeof ea
);
950 for (i
= 0; i
< br
->n_ports
; i
++) {
951 struct port
*port
= br
->ports
[i
];
952 uint8_t iface_ea
[ETH_ADDR_LEN
];
955 /* Mirror output ports don't participate. */
956 if (port
->is_mirror_output_port
) {
960 /* Choose the MAC address to represent the port. */
961 if (port
->cfg
->mac
&& eth_addr_from_string(port
->cfg
->mac
, iface_ea
)) {
962 /* Find the interface with this Ethernet address (if any) so that
963 * we can provide the correct devname to the caller. */
965 for (j
= 0; j
< port
->n_ifaces
; j
++) {
966 struct iface
*candidate
= port
->ifaces
[j
];
967 uint8_t candidate_ea
[ETH_ADDR_LEN
];
968 if (!netdev_get_etheraddr(candidate
->netdev
, candidate_ea
)
969 && eth_addr_equals(iface_ea
, candidate_ea
)) {
974 /* Choose the interface whose MAC address will represent the port.
975 * The Linux kernel bonding code always chooses the MAC address of
976 * the first slave added to a bond, and the Fedora networking
977 * scripts always add slaves to a bond in alphabetical order, so
978 * for compatibility we choose the interface with the name that is
979 * first in alphabetical order. */
980 iface
= port
->ifaces
[0];
981 for (j
= 1; j
< port
->n_ifaces
; j
++) {
982 struct iface
*candidate
= port
->ifaces
[j
];
983 if (strcmp(candidate
->name
, iface
->name
) < 0) {
988 /* The local port doesn't count (since we're trying to choose its
989 * MAC address anyway). */
990 if (iface
->dp_ifidx
== ODPP_LOCAL
) {
995 error
= netdev_get_etheraddr(iface
->netdev
, iface_ea
);
997 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
998 VLOG_ERR_RL(&rl
, "failed to obtain Ethernet address of %s: %s",
999 iface
->name
, strerror(error
));
1004 /* Compare against our current choice. */
1005 if (!eth_addr_is_multicast(iface_ea
) &&
1006 !eth_addr_is_local(iface_ea
) &&
1007 !eth_addr_is_reserved(iface_ea
) &&
1008 !eth_addr_is_zero(iface_ea
) &&
1009 memcmp(iface_ea
, ea
, ETH_ADDR_LEN
) < 0)
1011 memcpy(ea
, iface_ea
, ETH_ADDR_LEN
);
1012 *hw_addr_iface
= iface
;
1015 if (eth_addr_is_multicast(ea
)) {
1016 memcpy(ea
, br
->default_ea
, ETH_ADDR_LEN
);
1017 *hw_addr_iface
= NULL
;
1018 VLOG_WARN("bridge %s: using default bridge Ethernet "
1019 "address "ETH_ADDR_FMT
, br
->name
, ETH_ADDR_ARGS(ea
));
1021 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT
,
1022 br
->name
, ETH_ADDR_ARGS(ea
));
1026 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
1027 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
1028 * an interface on 'br', then that interface must be passed in as
1029 * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then
1030 * 'hw_addr_iface' must be passed in as a null pointer. */
1032 bridge_pick_datapath_id(struct bridge
*br
,
1033 const uint8_t bridge_ea
[ETH_ADDR_LEN
],
1034 struct iface
*hw_addr_iface
)
1037 * The procedure for choosing a bridge MAC address will, in the most
1038 * ordinary case, also choose a unique MAC that we can use as a datapath
1039 * ID. In some special cases, though, multiple bridges will end up with
1040 * the same MAC address. This is OK for the bridges, but it will confuse
1041 * the OpenFlow controller, because each datapath needs a unique datapath
1044 * Datapath IDs must be unique. It is also very desirable that they be
1045 * stable from one run to the next, so that policy set on a datapath
1048 const char *datapath_id
;
1051 datapath_id
= bridge_get_other_config(br
->cfg
, "datapath-id");
1052 if (datapath_id
&& dpid_from_string(datapath_id
, &dpid
)) {
1056 if (hw_addr_iface
) {
1058 if (!netdev_get_vlan_vid(hw_addr_iface
->netdev
, &vlan
)) {
1060 * A bridge whose MAC address is taken from a VLAN network device
1061 * (that is, a network device created with vconfig(8) or similar
1062 * tool) will have the same MAC address as a bridge on the VLAN
1063 * device's physical network device.
1065 * Handle this case by hashing the physical network device MAC
1066 * along with the VLAN identifier.
1068 uint8_t buf
[ETH_ADDR_LEN
+ 2];
1069 memcpy(buf
, bridge_ea
, ETH_ADDR_LEN
);
1070 buf
[ETH_ADDR_LEN
] = vlan
>> 8;
1071 buf
[ETH_ADDR_LEN
+ 1] = vlan
;
1072 return dpid_from_hash(buf
, sizeof buf
);
1075 * Assume that this bridge's MAC address is unique, since it
1076 * doesn't fit any of the cases we handle specially.
1081 * A purely internal bridge, that is, one that has no non-virtual
1082 * network devices on it at all, is more difficult because it has no
1083 * natural unique identifier at all.
1085 * When the host is a XenServer, we handle this case by hashing the
1086 * host's UUID with the name of the bridge. Names of bridges are
1087 * persistent across XenServer reboots, although they can be reused if
1088 * an internal network is destroyed and then a new one is later
1089 * created, so this is fairly effective.
1091 * When the host is not a XenServer, we punt by using a random MAC
1092 * address on each run.
1094 const char *host_uuid
= xenserver_get_host_uuid();
1096 char *combined
= xasprintf("%s,%s", host_uuid
, br
->name
);
1097 dpid
= dpid_from_hash(combined
, strlen(combined
));
1103 return eth_addr_to_uint64(bridge_ea
);
1107 dpid_from_hash(const void *data
, size_t n
)
1109 uint8_t hash
[SHA1_DIGEST_SIZE
];
1111 BUILD_ASSERT_DECL(sizeof hash
>= ETH_ADDR_LEN
);
1112 sha1_bytes(data
, n
, hash
);
1113 eth_addr_mark_random(hash
);
1114 return eth_addr_to_uint64(hash
);
1118 iface_refresh_stats(struct iface
*iface
)
1124 static const struct iface_stat iface_stats
[] = {
1125 { "rx_packets", offsetof(struct netdev_stats
, rx_packets
) },
1126 { "tx_packets", offsetof(struct netdev_stats
, tx_packets
) },
1127 { "rx_bytes", offsetof(struct netdev_stats
, rx_bytes
) },
1128 { "tx_bytes", offsetof(struct netdev_stats
, tx_bytes
) },
1129 { "rx_dropped", offsetof(struct netdev_stats
, rx_dropped
) },
1130 { "tx_dropped", offsetof(struct netdev_stats
, tx_dropped
) },
1131 { "rx_errors", offsetof(struct netdev_stats
, rx_errors
) },
1132 { "tx_errors", offsetof(struct netdev_stats
, tx_errors
) },
1133 { "rx_frame_err", offsetof(struct netdev_stats
, rx_frame_errors
) },
1134 { "rx_over_err", offsetof(struct netdev_stats
, rx_over_errors
) },
1135 { "rx_crc_err", offsetof(struct netdev_stats
, rx_crc_errors
) },
1136 { "collisions", offsetof(struct netdev_stats
, collisions
) },
1138 enum { N_STATS
= ARRAY_SIZE(iface_stats
) };
1139 const struct iface_stat
*s
;
1141 char *keys
[N_STATS
];
1142 int64_t values
[N_STATS
];
1145 struct netdev_stats stats
;
1147 /* Intentionally ignore return value, since errors will set 'stats' to
1148 * all-1s, and we will deal with that correctly below. */
1149 netdev_get_stats(iface
->netdev
, &stats
);
1152 for (s
= iface_stats
; s
< &iface_stats
[N_STATS
]; s
++) {
1153 uint64_t value
= *(uint64_t *) (((char *) &stats
) + s
->offset
);
1154 if (value
!= UINT64_MAX
) {
1161 ovsrec_interface_set_statistics(iface
->cfg
, keys
, values
, n
);
1165 refresh_system_stats(const struct ovsrec_open_vswitch
*cfg
)
1167 struct ovsdb_datum datum
;
1171 get_system_stats(&stats
);
1173 ovsdb_datum_from_shash(&datum
, &stats
);
1174 ovsdb_idl_txn_write(&cfg
->header_
, &ovsrec_open_vswitch_col_statistics
,
1181 const struct ovsrec_open_vswitch
*cfg
;
1183 bool datapath_destroyed
;
1184 bool database_changed
;
1187 /* Let each bridge do the work that it needs to do. */
1188 datapath_destroyed
= false;
1189 LIST_FOR_EACH (br
, node
, &all_bridges
) {
1190 int error
= bridge_run_one(br
);
1192 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1193 VLOG_ERR_RL(&rl
, "bridge %s: datapath was destroyed externally, "
1194 "forcing reconfiguration", br
->name
);
1195 datapath_destroyed
= true;
1199 /* (Re)configure if necessary. */
1200 database_changed
= ovsdb_idl_run(idl
);
1201 cfg
= ovsrec_open_vswitch_first(idl
);
1202 if (database_changed
|| datapath_destroyed
) {
1204 struct ovsdb_idl_txn
*txn
= ovsdb_idl_txn_create(idl
);
1206 bridge_configure_once(cfg
);
1207 bridge_reconfigure(cfg
);
1209 ovsrec_open_vswitch_set_cur_cfg(cfg
, cfg
->next_cfg
);
1210 ovsdb_idl_txn_commit(txn
);
1211 ovsdb_idl_txn_destroy(txn
); /* XXX */
1213 /* We still need to reconfigure to avoid dangling pointers to
1214 * now-destroyed ovsrec structures inside bridge data. */
1215 static const struct ovsrec_open_vswitch null_cfg
;
1217 bridge_reconfigure(&null_cfg
);
1222 /* Re-configure SSL. We do this on every trip through the main loop,
1223 * instead of just when the database changes, because the contents of the
1224 * key and certificate files can change without the database changing. */
1225 if (cfg
&& cfg
->ssl
) {
1226 const struct ovsrec_ssl
*ssl
= cfg
->ssl
;
1228 stream_ssl_set_key_and_cert(ssl
->private_key
, ssl
->certificate
);
1229 stream_ssl_set_ca_cert_file(ssl
->ca_cert
, ssl
->bootstrap_ca_cert
);
1233 /* Refresh system and interface stats if necessary. */
1234 if (time_msec() >= stats_timer
) {
1236 struct ovsdb_idl_txn
*txn
;
1238 txn
= ovsdb_idl_txn_create(idl
);
1239 LIST_FOR_EACH (br
, node
, &all_bridges
) {
1242 for (i
= 0; i
< br
->n_ports
; i
++) {
1243 struct port
*port
= br
->ports
[i
];
1246 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1247 struct iface
*iface
= port
->ifaces
[j
];
1248 iface_refresh_stats(iface
);
1252 refresh_system_stats(cfg
);
1253 ovsdb_idl_txn_commit(txn
);
1254 ovsdb_idl_txn_destroy(txn
); /* XXX */
1257 stats_timer
= time_msec() + STATS_INTERVAL
;
1266 LIST_FOR_EACH (br
, node
, &all_bridges
) {
1267 ofproto_wait(br
->ofproto
);
1268 if (ofproto_has_primary_controller(br
->ofproto
)) {
1272 mac_learning_wait(br
->ml
);
1275 ovsdb_idl_wait(idl
);
1276 poll_timer_wait_until(stats_timer
);
1279 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
1280 * configuration changes. */
1282 bridge_flush(struct bridge
*br
)
1284 COVERAGE_INC(bridge_flush
);
1286 mac_learning_flush(br
->ml
);
1289 /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no
1290 * such interface. */
1291 static struct iface
*
1292 bridge_get_local_iface(struct bridge
*br
)
1296 for (i
= 0; i
< br
->n_ports
; i
++) {
1297 struct port
*port
= br
->ports
[i
];
1298 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1299 struct iface
*iface
= port
->ifaces
[j
];
1300 if (iface
->dp_ifidx
== ODPP_LOCAL
) {
1309 /* Bridge unixctl user interface functions. */
1311 bridge_unixctl_fdb_show(struct unixctl_conn
*conn
,
1312 const char *args
, void *aux OVS_UNUSED
)
1314 struct ds ds
= DS_EMPTY_INITIALIZER
;
1315 const struct bridge
*br
;
1316 const struct mac_entry
*e
;
1318 br
= bridge_lookup(args
);
1320 unixctl_command_reply(conn
, 501, "no such bridge");
1324 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
1325 LIST_FOR_EACH (e
, lru_node
, &br
->ml
->lrus
) {
1326 if (e
->port
< 0 || e
->port
>= br
->n_ports
) {
1329 ds_put_format(&ds
, "%5d %4d "ETH_ADDR_FMT
" %3d\n",
1330 br
->ports
[e
->port
]->ifaces
[0]->dp_ifidx
,
1331 e
->vlan
, ETH_ADDR_ARGS(e
->mac
), mac_entry_age(e
));
1333 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
1337 /* Bridge reconfiguration functions. */
1338 static struct bridge
*
1339 bridge_create(const struct ovsrec_bridge
*br_cfg
)
1344 assert(!bridge_lookup(br_cfg
->name
));
1345 br
= xzalloc(sizeof *br
);
1347 error
= dpif_create_and_open(br_cfg
->name
, br_cfg
->datapath_type
,
1353 dpif_flow_flush(br
->dpif
);
1355 error
= ofproto_create(br_cfg
->name
, br_cfg
->datapath_type
, &bridge_ofhooks
,
1358 VLOG_ERR("failed to create switch %s: %s", br_cfg
->name
,
1360 dpif_delete(br
->dpif
);
1361 dpif_close(br
->dpif
);
1366 br
->name
= xstrdup(br_cfg
->name
);
1368 br
->ml
= mac_learning_create();
1369 eth_addr_nicira_random(br
->default_ea
);
1371 hmap_init(&br
->ifaces
);
1373 shash_init(&br
->port_by_name
);
1374 shash_init(&br
->iface_by_name
);
1378 list_push_back(&all_bridges
, &br
->node
);
1380 VLOG_INFO("created bridge %s on %s", br
->name
, dpif_name(br
->dpif
));
1386 bridge_destroy(struct bridge
*br
)
1391 while (br
->n_ports
> 0) {
1392 port_destroy(br
->ports
[br
->n_ports
- 1]);
1394 list_remove(&br
->node
);
1395 error
= dpif_delete(br
->dpif
);
1396 if (error
&& error
!= ENOENT
) {
1397 VLOG_ERR("failed to delete %s: %s",
1398 dpif_name(br
->dpif
), strerror(error
));
1400 dpif_close(br
->dpif
);
1401 ofproto_destroy(br
->ofproto
);
1402 mac_learning_destroy(br
->ml
);
1403 hmap_destroy(&br
->ifaces
);
1404 shash_destroy(&br
->port_by_name
);
1405 shash_destroy(&br
->iface_by_name
);
1412 static struct bridge
*
1413 bridge_lookup(const char *name
)
1417 LIST_FOR_EACH (br
, node
, &all_bridges
) {
1418 if (!strcmp(br
->name
, name
)) {
1425 /* Handle requests for a listing of all flows known by the OpenFlow
1426 * stack, including those normally hidden. */
1428 bridge_unixctl_dump_flows(struct unixctl_conn
*conn
,
1429 const char *args
, void *aux OVS_UNUSED
)
1434 br
= bridge_lookup(args
);
1436 unixctl_command_reply(conn
, 501, "Unknown bridge");
1441 ofproto_get_all_flows(br
->ofproto
, &results
);
1443 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
1444 ds_destroy(&results
);
1447 /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller
1448 * connections and reconnect. If BRIDGE is not specified, then all bridges
1449 * drop their controller connections and reconnect. */
1451 bridge_unixctl_reconnect(struct unixctl_conn
*conn
,
1452 const char *args
, void *aux OVS_UNUSED
)
1455 if (args
[0] != '\0') {
1456 br
= bridge_lookup(args
);
1458 unixctl_command_reply(conn
, 501, "Unknown bridge");
1461 ofproto_reconnect_controllers(br
->ofproto
);
1463 LIST_FOR_EACH (br
, node
, &all_bridges
) {
1464 ofproto_reconnect_controllers(br
->ofproto
);
1467 unixctl_command_reply(conn
, 200, NULL
);
1471 bridge_run_one(struct bridge
*br
)
1475 error
= ofproto_run1(br
->ofproto
);
1480 mac_learning_run(br
->ml
, ofproto_get_revalidate_set(br
->ofproto
));
1483 error
= ofproto_run2(br
->ofproto
, br
->flush
);
1490 bridge_get_controllers(const struct bridge
*br
,
1491 struct ovsrec_controller
***controllersp
)
1493 struct ovsrec_controller
**controllers
;
1494 size_t n_controllers
;
1496 controllers
= br
->cfg
->controller
;
1497 n_controllers
= br
->cfg
->n_controller
;
1499 if (n_controllers
== 1 && !strcmp(controllers
[0]->target
, "none")) {
1505 *controllersp
= controllers
;
1507 return n_controllers
;
1511 bridge_reconfigure_one(struct bridge
*br
)
1513 struct shash old_ports
, new_ports
;
1514 struct svec snoops
, old_snoops
;
1515 struct shash_node
*node
;
1516 enum ofproto_fail_mode fail_mode
;
1519 /* Collect old ports. */
1520 shash_init(&old_ports
);
1521 for (i
= 0; i
< br
->n_ports
; i
++) {
1522 shash_add(&old_ports
, br
->ports
[i
]->name
, br
->ports
[i
]);
1525 /* Collect new ports. */
1526 shash_init(&new_ports
);
1527 for (i
= 0; i
< br
->cfg
->n_ports
; i
++) {
1528 const char *name
= br
->cfg
->ports
[i
]->name
;
1529 if (!shash_add_once(&new_ports
, name
, br
->cfg
->ports
[i
])) {
1530 VLOG_WARN("bridge %s: %s specified twice as bridge port",
1535 /* If we have a controller, then we need a local port. Complain if the
1536 * user didn't specify one.
1538 * XXX perhaps we should synthesize a port ourselves in this case. */
1539 if (bridge_get_controllers(br
, NULL
)) {
1540 char local_name
[IF_NAMESIZE
];
1543 error
= dpif_port_get_name(br
->dpif
, ODPP_LOCAL
,
1544 local_name
, sizeof local_name
);
1545 if (!error
&& !shash_find(&new_ports
, local_name
)) {
1546 VLOG_WARN("bridge %s: controller specified but no local port "
1547 "(port named %s) defined",
1548 br
->name
, local_name
);
1552 /* Get rid of deleted ports.
1553 * Get rid of deleted interfaces on ports that still exist. */
1554 SHASH_FOR_EACH (node
, &old_ports
) {
1555 struct port
*port
= node
->data
;
1556 const struct ovsrec_port
*port_cfg
;
1558 port_cfg
= shash_find_data(&new_ports
, node
->name
);
1562 port_del_ifaces(port
, port_cfg
);
1566 /* Create new ports.
1567 * Add new interfaces to existing ports.
1568 * Reconfigure existing ports. */
1569 SHASH_FOR_EACH (node
, &new_ports
) {
1570 struct port
*port
= shash_find_data(&old_ports
, node
->name
);
1572 port
= port_create(br
, node
->name
);
1575 port_reconfigure(port
, node
->data
);
1576 if (!port
->n_ifaces
) {
1577 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1578 br
->name
, port
->name
);
1582 shash_destroy(&old_ports
);
1583 shash_destroy(&new_ports
);
1585 /* Set the fail-mode */
1586 fail_mode
= !br
->cfg
->fail_mode
1587 || !strcmp(br
->cfg
->fail_mode
, "standalone")
1588 ? OFPROTO_FAIL_STANDALONE
1589 : OFPROTO_FAIL_SECURE
;
1590 if (ofproto_get_fail_mode(br
->ofproto
) != fail_mode
1591 && !ofproto_has_primary_controller(br
->ofproto
)) {
1592 ofproto_flush_flows(br
->ofproto
);
1594 ofproto_set_fail_mode(br
->ofproto
, fail_mode
);
1596 /* Delete all flows if we're switching from connected to standalone or vice
1597 * versa. (XXX Should we delete all flows if we are switching from one
1598 * controller to another?) */
1600 /* Configure OpenFlow controller connection snooping. */
1602 svec_add_nocopy(&snoops
, xasprintf("punix:%s/%s.snoop",
1603 ovs_rundir
, br
->name
));
1604 svec_init(&old_snoops
);
1605 ofproto_get_snoops(br
->ofproto
, &old_snoops
);
1606 if (!svec_equal(&snoops
, &old_snoops
)) {
1607 ofproto_set_snoops(br
->ofproto
, &snoops
);
1609 svec_destroy(&snoops
);
1610 svec_destroy(&old_snoops
);
1612 mirror_reconfigure(br
);
1615 /* Initializes 'oc' appropriately as a management service controller for
1618 * The caller must free oc->target when it is no longer needed. */
1620 bridge_ofproto_controller_for_mgmt(const struct bridge
*br
,
1621 struct ofproto_controller
*oc
)
1623 oc
->target
= xasprintf("punix:%s/%s.mgmt", ovs_rundir
, br
->name
);
1624 oc
->max_backoff
= 0;
1625 oc
->probe_interval
= 60;
1626 oc
->band
= OFPROTO_OUT_OF_BAND
;
1627 oc
->accept_re
= NULL
;
1628 oc
->update_resolv_conf
= false;
1630 oc
->burst_limit
= 0;
1633 /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */
1635 bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller
*c
,
1636 struct ofproto_controller
*oc
)
1638 oc
->target
= c
->target
;
1639 oc
->max_backoff
= c
->max_backoff
? *c
->max_backoff
/ 1000 : 8;
1640 oc
->probe_interval
= c
->inactivity_probe
? *c
->inactivity_probe
/ 1000 : 5;
1641 oc
->band
= (!c
->connection_mode
|| !strcmp(c
->connection_mode
, "in-band")
1642 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
1643 oc
->accept_re
= c
->discover_accept_regex
;
1644 oc
->update_resolv_conf
= c
->discover_update_resolv_conf
;
1645 oc
->rate_limit
= c
->controller_rate_limit
? *c
->controller_rate_limit
: 0;
1646 oc
->burst_limit
= (c
->controller_burst_limit
1647 ? *c
->controller_burst_limit
: 0);
1650 /* Configures the IP stack for 'br''s local interface properly according to the
1651 * configuration in 'c'. */
1653 bridge_configure_local_iface_netdev(struct bridge
*br
,
1654 struct ovsrec_controller
*c
)
1656 struct netdev
*netdev
;
1657 struct in_addr mask
, gateway
;
1659 struct iface
*local_iface
;
1662 /* Controller discovery does its own TCP/IP configuration later. */
1663 if (strcmp(c
->target
, "discover")) {
1667 /* If there's no local interface or no IP address, give up. */
1668 local_iface
= bridge_get_local_iface(br
);
1669 if (!local_iface
|| !c
->local_ip
|| !inet_aton(c
->local_ip
, &ip
)) {
1673 /* Bring up the local interface. */
1674 netdev
= local_iface
->netdev
;
1675 netdev_turn_flags_on(netdev
, NETDEV_UP
, true);
1677 /* Configure the IP address and netmask. */
1678 if (!c
->local_netmask
1679 || !inet_aton(c
->local_netmask
, &mask
)
1681 mask
.s_addr
= guess_netmask(ip
.s_addr
);
1683 if (!netdev_set_in4(netdev
, ip
, mask
)) {
1684 VLOG_INFO("bridge %s: configured IP address "IP_FMT
", netmask "IP_FMT
,
1685 br
->name
, IP_ARGS(&ip
.s_addr
), IP_ARGS(&mask
.s_addr
));
1688 /* Configure the default gateway. */
1689 if (c
->local_gateway
1690 && inet_aton(c
->local_gateway
, &gateway
)
1691 && gateway
.s_addr
) {
1692 if (!netdev_add_router(netdev
, gateway
)) {
1693 VLOG_INFO("bridge %s: configured gateway "IP_FMT
,
1694 br
->name
, IP_ARGS(&gateway
.s_addr
));
1700 bridge_reconfigure_remotes(struct bridge
*br
,
1701 const struct sockaddr_in
*managers
,
1704 struct ovsrec_controller
**controllers
;
1705 size_t n_controllers
;
1707 const char *disable_ib_str
;
1708 bool disable_in_band
= false;
1710 struct ofproto_controller
*ocs
;
1715 /* Check if we should disable in-band control on this bridge. */
1716 disable_ib_str
= bridge_get_other_config(br
->cfg
, "disable-in-band");
1717 if (disable_ib_str
&& !strcmp(disable_ib_str
, "true")) {
1718 disable_in_band
= true;
1721 if (disable_in_band
) {
1722 ofproto_set_extra_in_band_remotes(br
->ofproto
, NULL
, 0);
1724 ofproto_set_extra_in_band_remotes(br
->ofproto
, managers
, n_managers
);
1726 had_primary
= ofproto_has_primary_controller(br
->ofproto
);
1728 n_controllers
= bridge_get_controllers(br
, &controllers
);
1730 ocs
= xmalloc((n_controllers
+ 1) * sizeof *ocs
);
1733 bridge_ofproto_controller_for_mgmt(br
, &ocs
[n_ocs
++]);
1734 for (i
= 0; i
< n_controllers
; i
++) {
1735 struct ovsrec_controller
*c
= controllers
[i
];
1737 if (!strncmp(c
->target
, "punix:", 6)
1738 || !strncmp(c
->target
, "unix:", 5)) {
1739 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1741 /* Prevent remote ovsdb-server users from accessing arbitrary Unix
1742 * domain sockets and overwriting arbitrary local files. */
1743 VLOG_ERR_RL(&rl
, "%s: not adding Unix domain socket controller "
1744 "\"%s\" due to possibility for remote exploit",
1745 dpif_name(br
->dpif
), c
->target
);
1749 bridge_configure_local_iface_netdev(br
, c
);
1750 bridge_ofproto_controller_from_ovsrec(c
, &ocs
[n_ocs
]);
1751 if (disable_in_band
) {
1752 ocs
[n_ocs
].band
= OFPROTO_OUT_OF_BAND
;
1757 ofproto_set_controllers(br
->ofproto
, ocs
, n_ocs
);
1758 free(ocs
[0].target
); /* From bridge_ofproto_controller_for_mgmt(). */
1761 if (had_primary
!= ofproto_has_primary_controller(br
->ofproto
)) {
1762 ofproto_flush_flows(br
->ofproto
);
1765 /* If there are no controllers and the bridge is in standalone
1766 * mode, set up a flow that matches every packet and directs
1767 * them to OFPP_NORMAL (which goes to us). Otherwise, the
1768 * switch is in secure mode and we won't pass any traffic until
1769 * a controller has been defined and it tells us to do so. */
1771 && ofproto_get_fail_mode(br
->ofproto
) == OFPROTO_FAIL_STANDALONE
) {
1772 union ofp_action action
;
1775 memset(&action
, 0, sizeof action
);
1776 action
.type
= htons(OFPAT_OUTPUT
);
1777 action
.output
.len
= htons(sizeof action
);
1778 action
.output
.port
= htons(OFPP_NORMAL
);
1779 memset(&flow
, 0, sizeof flow
);
1780 ofproto_add_flow(br
->ofproto
, &flow
, OVSFW_ALL
, 0, &action
, 1, 0);
1785 bridge_get_all_ifaces(const struct bridge
*br
, struct shash
*ifaces
)
1790 for (i
= 0; i
< br
->n_ports
; i
++) {
1791 struct port
*port
= br
->ports
[i
];
1792 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1793 struct iface
*iface
= port
->ifaces
[j
];
1794 shash_add_once(ifaces
, iface
->name
, iface
);
1796 if (port
->n_ifaces
> 1 && port
->cfg
->bond_fake_iface
) {
1797 shash_add_once(ifaces
, port
->name
, NULL
);
1802 /* For robustness, in case the administrator moves around datapath ports behind
1803 * our back, we re-check all the datapath port numbers here.
1805 * This function will set the 'dp_ifidx' members of interfaces that have
1806 * disappeared to -1, so only call this function from a context where those
1807 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1808 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1809 * datapath, which doesn't support UINT16_MAX+1 ports. */
1811 bridge_fetch_dp_ifaces(struct bridge
*br
)
1813 struct odp_port
*dpif_ports
;
1814 size_t n_dpif_ports
;
1817 /* Reset all interface numbers. */
1818 for (i
= 0; i
< br
->n_ports
; i
++) {
1819 struct port
*port
= br
->ports
[i
];
1820 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1821 struct iface
*iface
= port
->ifaces
[j
];
1822 iface
->dp_ifidx
= -1;
1825 hmap_clear(&br
->ifaces
);
1827 dpif_port_list(br
->dpif
, &dpif_ports
, &n_dpif_ports
);
1828 for (i
= 0; i
< n_dpif_ports
; i
++) {
1829 struct odp_port
*p
= &dpif_ports
[i
];
1830 struct iface
*iface
= iface_lookup(br
, p
->devname
);
1832 if (iface
->dp_ifidx
>= 0) {
1833 VLOG_WARN("%s reported interface %s twice",
1834 dpif_name(br
->dpif
), p
->devname
);
1835 } else if (iface_from_dp_ifidx(br
, p
->port
)) {
1836 VLOG_WARN("%s reported interface %"PRIu16
" twice",
1837 dpif_name(br
->dpif
), p
->port
);
1839 iface
->dp_ifidx
= p
->port
;
1840 hmap_insert(&br
->ifaces
, &iface
->dp_ifidx_node
,
1841 hash_int(iface
->dp_ifidx
, 0));
1844 iface_set_ofport(iface
->cfg
,
1845 (iface
->dp_ifidx
>= 0
1846 ? odp_port_to_ofp_port(iface
->dp_ifidx
)
1853 /* Bridge packet processing functions. */
1856 bond_hash(const uint8_t mac
[ETH_ADDR_LEN
])
1858 return hash_bytes(mac
, ETH_ADDR_LEN
, 0) & BOND_MASK
;
1861 static struct bond_entry
*
1862 lookup_bond_entry(const struct port
*port
, const uint8_t mac
[ETH_ADDR_LEN
])
1864 return &port
->bond_hash
[bond_hash(mac
)];
1868 bond_choose_iface(const struct port
*port
)
1870 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1871 size_t i
, best_down_slave
= -1;
1872 long long next_delay_expiration
= LLONG_MAX
;
1874 for (i
= 0; i
< port
->n_ifaces
; i
++) {
1875 struct iface
*iface
= port
->ifaces
[i
];
1877 if (iface
->enabled
) {
1879 } else if (iface
->delay_expires
< next_delay_expiration
) {
1880 best_down_slave
= i
;
1881 next_delay_expiration
= iface
->delay_expires
;
1885 if (best_down_slave
!= -1) {
1886 struct iface
*iface
= port
->ifaces
[best_down_slave
];
1888 VLOG_INFO_RL(&rl
, "interface %s: skipping remaining %lli ms updelay "
1889 "since no other interface is up", iface
->name
,
1890 iface
->delay_expires
- time_msec());
1891 bond_enable_slave(iface
, true);
1894 return best_down_slave
;
1898 choose_output_iface(const struct port
*port
, const uint8_t *dl_src
,
1899 uint16_t *dp_ifidx
, tag_type
*tags
)
1901 struct iface
*iface
;
1903 assert(port
->n_ifaces
);
1904 if (port
->n_ifaces
== 1) {
1905 iface
= port
->ifaces
[0];
1907 struct bond_entry
*e
= lookup_bond_entry(port
, dl_src
);
1908 if (e
->iface_idx
< 0 || e
->iface_idx
>= port
->n_ifaces
1909 || !port
->ifaces
[e
->iface_idx
]->enabled
) {
1910 /* XXX select interface properly. The current interface selection
1911 * is only good for testing the rebalancing code. */
1912 e
->iface_idx
= bond_choose_iface(port
);
1913 if (e
->iface_idx
< 0) {
1914 *tags
|= port
->no_ifaces_tag
;
1917 e
->iface_tag
= tag_create_random();
1918 ((struct port
*) port
)->bond_compat_is_stale
= true;
1920 *tags
|= e
->iface_tag
;
1921 iface
= port
->ifaces
[e
->iface_idx
];
1923 *dp_ifidx
= iface
->dp_ifidx
;
1924 *tags
|= iface
->tag
; /* Currently only used for bonding. */
1929 bond_link_status_update(struct iface
*iface
, bool carrier
)
1931 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1932 struct port
*port
= iface
->port
;
1934 if ((carrier
== iface
->enabled
) == (iface
->delay_expires
== LLONG_MAX
)) {
1935 /* Nothing to do. */
1938 VLOG_INFO_RL(&rl
, "interface %s: carrier %s",
1939 iface
->name
, carrier
? "detected" : "dropped");
1940 if (carrier
== iface
->enabled
) {
1941 iface
->delay_expires
= LLONG_MAX
;
1942 VLOG_INFO_RL(&rl
, "interface %s: will not be %s",
1943 iface
->name
, carrier
? "disabled" : "enabled");
1944 } else if (carrier
&& port
->active_iface
< 0) {
1945 bond_enable_slave(iface
, true);
1946 if (port
->updelay
) {
1947 VLOG_INFO_RL(&rl
, "interface %s: skipping %d ms updelay since no "
1948 "other interface is up", iface
->name
, port
->updelay
);
1951 int delay
= carrier
? port
->updelay
: port
->downdelay
;
1952 iface
->delay_expires
= time_msec() + delay
;
1955 "interface %s: will be %s if it stays %s for %d ms",
1957 carrier
? "enabled" : "disabled",
1958 carrier
? "up" : "down",
1965 bond_choose_active_iface(struct port
*port
)
1967 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1969 port
->active_iface
= bond_choose_iface(port
);
1970 port
->active_iface_tag
= tag_create_random();
1971 if (port
->active_iface
>= 0) {
1972 VLOG_INFO_RL(&rl
, "port %s: active interface is now %s",
1973 port
->name
, port
->ifaces
[port
->active_iface
]->name
);
1975 VLOG_WARN_RL(&rl
, "port %s: all ports disabled, no active interface",
1981 bond_enable_slave(struct iface
*iface
, bool enable
)
1983 struct port
*port
= iface
->port
;
1984 struct bridge
*br
= port
->bridge
;
1986 /* This acts as a recursion check. If the act of disabling a slave
1987 * causes a different slave to be enabled, the flag will allow us to
1988 * skip redundant work when we reenter this function. It must be
1989 * cleared on exit to keep things safe with multiple bonds. */
1990 static bool moving_active_iface
= false;
1992 iface
->delay_expires
= LLONG_MAX
;
1993 if (enable
== iface
->enabled
) {
1997 iface
->enabled
= enable
;
1998 if (!iface
->enabled
) {
1999 VLOG_WARN("interface %s: disabled", iface
->name
);
2000 ofproto_revalidate(br
->ofproto
, iface
->tag
);
2001 if (iface
->port_ifidx
== port
->active_iface
) {
2002 ofproto_revalidate(br
->ofproto
,
2003 port
->active_iface_tag
);
2005 /* Disabling a slave can lead to another slave being immediately
2006 * enabled if there will be no active slaves but one is waiting
2007 * on an updelay. In this case we do not need to run most of the
2008 * code for the newly enabled slave since there was no period
2009 * without an active slave and it is redundant with the disabling
2011 moving_active_iface
= true;
2012 bond_choose_active_iface(port
);
2014 bond_send_learning_packets(port
);
2016 VLOG_WARN("interface %s: enabled", iface
->name
);
2017 if (port
->active_iface
< 0 && !moving_active_iface
) {
2018 ofproto_revalidate(br
->ofproto
, port
->no_ifaces_tag
);
2019 bond_choose_active_iface(port
);
2020 bond_send_learning_packets(port
);
2022 iface
->tag
= tag_create_random();
2025 moving_active_iface
= false;
2026 port
->bond_compat_is_stale
= true;
2029 /* Attempts to make the sum of the bond slaves' statistics appear on the fake
2030 * bond interface. */
2032 bond_update_fake_iface_stats(struct port
*port
)
2034 struct netdev_stats bond_stats
;
2035 struct netdev
*bond_dev
;
2038 memset(&bond_stats
, 0, sizeof bond_stats
);
2040 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2041 struct netdev_stats slave_stats
;
2043 if (!netdev_get_stats(port
->ifaces
[i
]->netdev
, &slave_stats
)) {
2044 /* XXX: We swap the stats here because they are swapped back when
2045 * reported by the internal device. The reason for this is
2046 * internal devices normally represent packets going into the system
2047 * but when used as fake bond device they represent packets leaving
2048 * the system. We really should do this in the internal device
2049 * itself because changing it here reverses the counts from the
2050 * perspective of the switch. However, the internal device doesn't
2051 * know what type of device it represents so we have to do it here
2053 bond_stats
.tx_packets
+= slave_stats
.rx_packets
;
2054 bond_stats
.tx_bytes
+= slave_stats
.rx_bytes
;
2055 bond_stats
.rx_packets
+= slave_stats
.tx_packets
;
2056 bond_stats
.rx_bytes
+= slave_stats
.tx_bytes
;
2060 if (!netdev_open_default(port
->name
, &bond_dev
)) {
2061 netdev_set_stats(bond_dev
, &bond_stats
);
2062 netdev_close(bond_dev
);
2067 bond_run(struct bridge
*br
)
2071 for (i
= 0; i
< br
->n_ports
; i
++) {
2072 struct port
*port
= br
->ports
[i
];
2074 if (port
->n_ifaces
>= 2) {
2077 /* Track carrier going up and down on interfaces. */
2078 while (!netdev_monitor_poll(port
->monitor
, &devname
)) {
2079 struct iface
*iface
;
2082 iface
= port_lookup_iface(port
, devname
);
2083 if (iface
&& !netdev_get_carrier(iface
->netdev
, &carrier
)) {
2084 bond_link_status_update(iface
, carrier
);
2085 port_update_bond_compat(port
);
2090 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2091 struct iface
*iface
= port
->ifaces
[j
];
2092 if (time_msec() >= iface
->delay_expires
) {
2093 bond_enable_slave(iface
, !iface
->enabled
);
2097 if (port
->bond_fake_iface
2098 && time_msec() >= port
->bond_next_fake_iface_update
) {
2099 bond_update_fake_iface_stats(port
);
2100 port
->bond_next_fake_iface_update
= time_msec() + 1000;
2104 if (port
->bond_compat_is_stale
) {
2105 port
->bond_compat_is_stale
= false;
2106 port_update_bond_compat(port
);
2112 bond_wait(struct bridge
*br
)
2116 for (i
= 0; i
< br
->n_ports
; i
++) {
2117 struct port
*port
= br
->ports
[i
];
2118 if (port
->n_ifaces
< 2) {
2121 netdev_monitor_poll_wait(port
->monitor
);
2122 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2123 struct iface
*iface
= port
->ifaces
[j
];
2124 if (iface
->delay_expires
!= LLONG_MAX
) {
2125 poll_timer_wait_until(iface
->delay_expires
);
2128 if (port
->bond_fake_iface
) {
2129 poll_timer_wait_until(port
->bond_next_fake_iface_update
);
2135 set_dst(struct dst
*p
, const struct flow
*flow
,
2136 const struct port
*in_port
, const struct port
*out_port
,
2139 p
->vlan
= (out_port
->vlan
>= 0 ? OFP_VLAN_NONE
2140 : in_port
->vlan
>= 0 ? in_port
->vlan
2141 : ntohs(flow
->dl_vlan
));
2142 return choose_output_iface(out_port
, flow
->dl_src
, &p
->dp_ifidx
, tags
);
2146 swap_dst(struct dst
*p
, struct dst
*q
)
2148 struct dst tmp
= *p
;
2153 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
2154 * 'dsts'. (This may help performance by reducing the number of VLAN changes
2155 * that we push to the datapath. We could in fact fully sort the array by
2156 * vlan, but in most cases there are at most two different vlan tags so that's
2157 * possibly overkill.) */
2159 partition_dsts(struct dst
*dsts
, size_t n_dsts
, int vlan
)
2161 struct dst
*first
= dsts
;
2162 struct dst
*last
= dsts
+ n_dsts
;
2164 while (first
!= last
) {
2166 * - All dsts < first have vlan == 'vlan'.
2167 * - All dsts >= last have vlan != 'vlan'.
2168 * - first < last. */
2169 while (first
->vlan
== vlan
) {
2170 if (++first
== last
) {
2175 /* Same invariants, plus one additional:
2176 * - first->vlan != vlan.
2178 while (last
[-1].vlan
!= vlan
) {
2179 if (--last
== first
) {
2184 /* Same invariants, plus one additional:
2185 * - last[-1].vlan == vlan.*/
2186 swap_dst(first
++, --last
);
2191 mirror_mask_ffs(mirror_mask_t mask
)
2193 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask
));
2198 dst_is_duplicate(const struct dst
*dsts
, size_t n_dsts
,
2199 const struct dst
*test
)
2202 for (i
= 0; i
< n_dsts
; i
++) {
2203 if (dsts
[i
].vlan
== test
->vlan
&& dsts
[i
].dp_ifidx
== test
->dp_ifidx
) {
2211 port_trunks_vlan(const struct port
*port
, uint16_t vlan
)
2213 return (port
->vlan
< 0
2214 && (!port
->trunks
|| bitmap_is_set(port
->trunks
, vlan
)));
2218 port_includes_vlan(const struct port
*port
, uint16_t vlan
)
2220 return vlan
== port
->vlan
|| port_trunks_vlan(port
, vlan
);
2224 port_is_floodable(const struct port
*port
)
2228 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2229 if (!ofproto_port_is_floodable(port
->bridge
->ofproto
,
2230 port
->ifaces
[i
]->dp_ifidx
)) {
2238 compose_dsts(const struct bridge
*br
, const struct flow
*flow
, uint16_t vlan
,
2239 const struct port
*in_port
, const struct port
*out_port
,
2240 struct dst dsts
[], tag_type
*tags
, uint16_t *nf_output_iface
)
2242 mirror_mask_t mirrors
= in_port
->src_mirrors
;
2243 struct dst
*dst
= dsts
;
2246 if (out_port
== FLOOD_PORT
) {
2247 /* XXX use ODP_FLOOD if no vlans or bonding. */
2248 /* XXX even better, define each VLAN as a datapath port group */
2249 for (i
= 0; i
< br
->n_ports
; i
++) {
2250 struct port
*port
= br
->ports
[i
];
2252 && port_is_floodable(port
)
2253 && port_includes_vlan(port
, vlan
)
2254 && !port
->is_mirror_output_port
2255 && set_dst(dst
, flow
, in_port
, port
, tags
)) {
2256 mirrors
|= port
->dst_mirrors
;
2260 *nf_output_iface
= NF_OUT_FLOOD
;
2261 } else if (out_port
&& set_dst(dst
, flow
, in_port
, out_port
, tags
)) {
2262 *nf_output_iface
= dst
->dp_ifidx
;
2263 mirrors
|= out_port
->dst_mirrors
;
2268 struct mirror
*m
= br
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
2269 if (!m
->n_vlans
|| vlan_is_mirrored(m
, vlan
)) {
2271 if (set_dst(dst
, flow
, in_port
, m
->out_port
, tags
)
2272 && !dst_is_duplicate(dsts
, dst
- dsts
, dst
)) {
2276 for (i
= 0; i
< br
->n_ports
; i
++) {
2277 struct port
*port
= br
->ports
[i
];
2278 if (port_includes_vlan(port
, m
->out_vlan
)
2279 && set_dst(dst
, flow
, in_port
, port
, tags
))
2283 if (port
->vlan
< 0) {
2284 dst
->vlan
= m
->out_vlan
;
2286 if (dst_is_duplicate(dsts
, dst
- dsts
, dst
)) {
2290 /* Use the vlan tag on the original flow instead of
2291 * the one passed in the vlan parameter. This ensures
2292 * that we compare the vlan from before any implicit
2293 * tagging tags place. This is necessary because
2294 * dst->vlan is the final vlan, after removing implicit
2296 flow_vlan
= ntohs(flow
->dl_vlan
);
2297 if (flow_vlan
== 0) {
2298 flow_vlan
= OFP_VLAN_NONE
;
2300 if (port
== in_port
&& dst
->vlan
== flow_vlan
) {
2301 /* Don't send out input port on same VLAN. */
2309 mirrors
&= mirrors
- 1;
2312 partition_dsts(dsts
, dst
- dsts
, ntohs(flow
->dl_vlan
));
2316 static void OVS_UNUSED
2317 print_dsts(const struct dst
*dsts
, size_t n
)
2319 for (; n
--; dsts
++) {
2320 printf(">p%"PRIu16
, dsts
->dp_ifidx
);
2321 if (dsts
->vlan
!= OFP_VLAN_NONE
) {
2322 printf("v%"PRIu16
, dsts
->vlan
);
2328 compose_actions(struct bridge
*br
, const struct flow
*flow
, uint16_t vlan
,
2329 const struct port
*in_port
, const struct port
*out_port
,
2330 tag_type
*tags
, struct odp_actions
*actions
,
2331 uint16_t *nf_output_iface
)
2333 struct dst dsts
[DP_MAX_PORTS
* (MAX_MIRRORS
+ 1)];
2335 const struct dst
*p
;
2338 n_dsts
= compose_dsts(br
, flow
, vlan
, in_port
, out_port
, dsts
, tags
,
2341 cur_vlan
= ntohs(flow
->dl_vlan
);
2342 for (p
= dsts
; p
< &dsts
[n_dsts
]; p
++) {
2343 union odp_action
*a
;
2344 if (p
->vlan
!= cur_vlan
) {
2345 if (p
->vlan
== OFP_VLAN_NONE
) {
2346 odp_actions_add(actions
, ODPAT_STRIP_VLAN
);
2348 a
= odp_actions_add(actions
, ODPAT_SET_DL_TCI
);
2349 a
->dl_tci
.tci
= htons(p
->vlan
& VLAN_VID_MASK
);
2350 a
->dl_tci
.tci
|= htons(flow
->dl_vlan_pcp
<< VLAN_PCP_SHIFT
);
2354 a
= odp_actions_add(actions
, ODPAT_OUTPUT
);
2355 a
->output
.port
= p
->dp_ifidx
;
2359 /* Returns the effective vlan of a packet, taking into account both the
2360 * 802.1Q header and implicitly tagged ports. A value of 0 indicates that
2361 * the packet is untagged and -1 indicates it has an invalid header and
2362 * should be dropped. */
2363 static int flow_get_vlan(struct bridge
*br
, const struct flow
*flow
,
2364 struct port
*in_port
, bool have_packet
)
2366 /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
2367 * belongs to VLAN 0, so we should treat both cases identically. (In the
2368 * former case, the packet has an 802.1Q header that specifies VLAN 0,
2369 * presumably to allow a priority to be specified. In the latter case, the
2370 * packet does not have any 802.1Q header.) */
2371 int vlan
= ntohs(flow
->dl_vlan
);
2372 if (vlan
== OFP_VLAN_NONE
) {
2375 if (in_port
->vlan
>= 0) {
2377 /* XXX support double tagging? */
2379 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2380 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" tagged "
2381 "packet received on port %s configured with "
2382 "implicit VLAN %"PRIu16
,
2383 br
->name
, ntohs(flow
->dl_vlan
),
2384 in_port
->name
, in_port
->vlan
);
2388 vlan
= in_port
->vlan
;
2390 if (!port_includes_vlan(in_port
, vlan
)) {
2392 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2393 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %d tagged "
2394 "packet received on port %s not configured for "
2396 br
->name
, vlan
, in_port
->name
, vlan
);
2405 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2406 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2407 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2409 is_gratuitous_arp(const struct flow
*flow
)
2411 return (flow
->dl_type
== htons(ETH_TYPE_ARP
)
2412 && eth_addr_is_broadcast(flow
->dl_dst
)
2413 && (flow
->nw_proto
== ARP_OP_REPLY
2414 || (flow
->nw_proto
== ARP_OP_REQUEST
2415 && flow
->nw_src
== flow
->nw_dst
)));
2419 update_learning_table(struct bridge
*br
, const struct flow
*flow
, int vlan
,
2420 struct port
*in_port
)
2422 enum grat_arp_lock_type lock_type
;
2425 /* We don't want to learn from gratuitous ARP packets that are reflected
2426 * back over bond slaves so we lock the learning table. */
2427 lock_type
= !is_gratuitous_arp(flow
) ? GRAT_ARP_LOCK_NONE
:
2428 (in_port
->n_ifaces
== 1) ? GRAT_ARP_LOCK_SET
:
2429 GRAT_ARP_LOCK_CHECK
;
2431 rev_tag
= mac_learning_learn(br
->ml
, flow
->dl_src
, vlan
, in_port
->port_idx
,
2434 /* The log messages here could actually be useful in debugging,
2435 * so keep the rate limit relatively high. */
2436 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30,
2438 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
2439 "on port %s in VLAN %d",
2440 br
->name
, ETH_ADDR_ARGS(flow
->dl_src
),
2441 in_port
->name
, vlan
);
2442 ofproto_revalidate(br
->ofproto
, rev_tag
);
2446 /* Determines whether packets in 'flow' within 'br' should be forwarded or
2447 * dropped. Returns true if they may be forwarded, false if they should be
2450 * If 'have_packet' is true, it indicates that the caller is processing a
2451 * received packet. If 'have_packet' is false, then the caller is just
2452 * revalidating an existing flow because configuration has changed. Either
2453 * way, 'have_packet' only affects logging (there is no point in logging errors
2454 * during revalidation).
2456 * Sets '*in_portp' to the input port. This will be a null pointer if
2457 * flow->in_port does not designate a known input port (in which case
2458 * is_admissible() returns false).
2460 * When returning true, sets '*vlanp' to the effective VLAN of the input
2461 * packet, as returned by flow_get_vlan().
2463 * May also add tags to '*tags', although the current implementation only does
2464 * so in one special case.
2467 is_admissible(struct bridge
*br
, const struct flow
*flow
, bool have_packet
,
2468 tag_type
*tags
, int *vlanp
, struct port
**in_portp
)
2470 struct iface
*in_iface
;
2471 struct port
*in_port
;
2474 /* Find the interface and port structure for the received packet. */
2475 in_iface
= iface_from_dp_ifidx(br
, flow
->in_port
);
2477 /* No interface? Something fishy... */
2479 /* Odd. A few possible reasons here:
2481 * - We deleted an interface but there are still a few packets
2482 * queued up from it.
2484 * - Someone externally added an interface (e.g. with "ovs-dpctl
2485 * add-if") that we don't know about.
2487 * - Packet arrived on the local port but the local port is not
2488 * one of our bridge ports.
2490 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2492 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
2493 "interface %"PRIu16
, br
->name
, flow
->in_port
);
2499 *in_portp
= in_port
= in_iface
->port
;
2500 *vlanp
= vlan
= flow_get_vlan(br
, flow
, in_port
, have_packet
);
2505 /* Drop frames for reserved multicast addresses. */
2506 if (eth_addr_is_reserved(flow
->dl_dst
)) {
2510 /* Drop frames on ports reserved for mirroring. */
2511 if (in_port
->is_mirror_output_port
) {
2513 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2514 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
2515 "%s, which is reserved exclusively for mirroring",
2516 br
->name
, in_port
->name
);
2521 /* Packets received on bonds need special attention to avoid duplicates. */
2522 if (in_port
->n_ifaces
> 1) {
2524 bool is_grat_arp_locked
;
2526 if (eth_addr_is_multicast(flow
->dl_dst
)) {
2527 *tags
|= in_port
->active_iface_tag
;
2528 if (in_port
->active_iface
!= in_iface
->port_ifidx
) {
2529 /* Drop all multicast packets on inactive slaves. */
2534 /* Drop all packets for which we have learned a different input
2535 * port, because we probably sent the packet on one slave and got
2536 * it back on the other. Gratuitous ARP packets are an exception
2537 * to this rule: the host has moved to another switch. The exception
2538 * to the exception is if we locked the learning table to avoid
2539 * reflections on bond slaves. If this is the case, just drop the
2541 src_idx
= mac_learning_lookup(br
->ml
, flow
->dl_src
, vlan
,
2542 &is_grat_arp_locked
);
2543 if (src_idx
!= -1 && src_idx
!= in_port
->port_idx
&&
2544 (!is_gratuitous_arp(flow
) || is_grat_arp_locked
)) {
2552 /* If the composed actions may be applied to any packet in the given 'flow',
2553 * returns true. Otherwise, the actions should only be applied to 'packet', or
2554 * not at all, if 'packet' was NULL. */
2556 process_flow(struct bridge
*br
, const struct flow
*flow
,
2557 const struct ofpbuf
*packet
, struct odp_actions
*actions
,
2558 tag_type
*tags
, uint16_t *nf_output_iface
)
2560 struct port
*in_port
;
2561 struct port
*out_port
;
2565 /* Check whether we should drop packets in this flow. */
2566 if (!is_admissible(br
, flow
, packet
!= NULL
, tags
, &vlan
, &in_port
)) {
2571 /* Learn source MAC (but don't try to learn from revalidation). */
2573 update_learning_table(br
, flow
, vlan
, in_port
);
2576 /* Determine output port. */
2577 out_port_idx
= mac_learning_lookup_tag(br
->ml
, flow
->dl_dst
, vlan
, tags
,
2579 if (out_port_idx
>= 0 && out_port_idx
< br
->n_ports
) {
2580 out_port
= br
->ports
[out_port_idx
];
2581 } else if (!packet
&& !eth_addr_is_multicast(flow
->dl_dst
)) {
2582 /* If we are revalidating but don't have a learning entry then
2583 * eject the flow. Installing a flow that floods packets opens
2584 * up a window of time where we could learn from a packet reflected
2585 * on a bond and blackhole packets before the learning table is
2586 * updated to reflect the correct port. */
2589 out_port
= FLOOD_PORT
;
2592 /* Don't send packets out their input ports. */
2593 if (in_port
== out_port
) {
2599 compose_actions(br
, flow
, vlan
, in_port
, out_port
, tags
, actions
,
2607 bridge_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
2608 struct odp_actions
*actions
, tag_type
*tags
,
2609 uint16_t *nf_output_iface
, void *br_
)
2611 struct bridge
*br
= br_
;
2613 COVERAGE_INC(bridge_process_flow
);
2615 return process_flow(br
, flow
, packet
, actions
, tags
, nf_output_iface
);
2619 bridge_account_flow_ofhook_cb(const struct flow
*flow
, tag_type tags
,
2620 const union odp_action
*actions
,
2621 size_t n_actions
, unsigned long long int n_bytes
,
2624 struct bridge
*br
= br_
;
2625 const union odp_action
*a
;
2626 struct port
*in_port
;
2630 /* Feed information from the active flows back into the learning table to
2631 * ensure that table is always in sync with what is actually flowing
2632 * through the datapath.
2634 * We test that 'tags' is nonzero to ensure that only flows that include an
2635 * OFPP_NORMAL action are used for learning. This works because
2636 * bridge_normal_ofhook_cb() always sets a nonzero tag value. */
2637 if (tags
&& is_admissible(br
, flow
, false, &dummy
, &vlan
, &in_port
)) {
2638 update_learning_table(br
, flow
, vlan
, in_port
);
2641 /* Account for bond slave utilization. */
2642 if (!br
->has_bonded_ports
) {
2645 for (a
= actions
; a
< &actions
[n_actions
]; a
++) {
2646 if (a
->type
== ODPAT_OUTPUT
) {
2647 struct port
*out_port
= port_from_dp_ifidx(br
, a
->output
.port
);
2648 if (out_port
&& out_port
->n_ifaces
>= 2) {
2649 struct bond_entry
*e
= lookup_bond_entry(out_port
,
2651 e
->tx_bytes
+= n_bytes
;
2658 bridge_account_checkpoint_ofhook_cb(void *br_
)
2660 struct bridge
*br
= br_
;
2664 if (!br
->has_bonded_ports
) {
2669 for (i
= 0; i
< br
->n_ports
; i
++) {
2670 struct port
*port
= br
->ports
[i
];
2671 if (port
->n_ifaces
> 1 && now
>= port
->bond_next_rebalance
) {
2672 port
->bond_next_rebalance
= now
+ port
->bond_rebalance_interval
;
2673 bond_rebalance_port(port
);
2678 static struct ofhooks bridge_ofhooks
= {
2679 bridge_normal_ofhook_cb
,
2680 bridge_account_flow_ofhook_cb
,
2681 bridge_account_checkpoint_ofhook_cb
,
2684 /* Bonding functions. */
2686 /* Statistics for a single interface on a bonded port, used for load-based
2687 * bond rebalancing. */
2688 struct slave_balance
{
2689 struct iface
*iface
; /* The interface. */
2690 uint64_t tx_bytes
; /* Sum of hashes[*]->tx_bytes. */
2692 /* All the "bond_entry"s that are assigned to this interface, in order of
2693 * increasing tx_bytes. */
2694 struct bond_entry
**hashes
;
2698 /* Sorts pointers to pointers to bond_entries in ascending order by the
2699 * interface to which they are assigned, and within a single interface in
2700 * ascending order of bytes transmitted. */
2702 compare_bond_entries(const void *a_
, const void *b_
)
2704 const struct bond_entry
*const *ap
= a_
;
2705 const struct bond_entry
*const *bp
= b_
;
2706 const struct bond_entry
*a
= *ap
;
2707 const struct bond_entry
*b
= *bp
;
2708 if (a
->iface_idx
!= b
->iface_idx
) {
2709 return a
->iface_idx
> b
->iface_idx
? 1 : -1;
2710 } else if (a
->tx_bytes
!= b
->tx_bytes
) {
2711 return a
->tx_bytes
> b
->tx_bytes
? 1 : -1;
2717 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2718 * *descending* order by number of bytes transmitted. */
2720 compare_slave_balance(const void *a_
, const void *b_
)
2722 const struct slave_balance
*a
= a_
;
2723 const struct slave_balance
*b
= b_
;
2724 if (a
->iface
->enabled
!= b
->iface
->enabled
) {
2725 return a
->iface
->enabled
? -1 : 1;
2726 } else if (a
->tx_bytes
!= b
->tx_bytes
) {
2727 return a
->tx_bytes
> b
->tx_bytes
? -1 : 1;
2734 swap_bals(struct slave_balance
*a
, struct slave_balance
*b
)
2736 struct slave_balance tmp
= *a
;
2741 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2742 * given that 'p' (and only 'p') might be in the wrong location.
2744 * This function invalidates 'p', since it might now be in a different memory
2747 resort_bals(struct slave_balance
*p
,
2748 struct slave_balance bals
[], size_t n_bals
)
2751 for (; p
> bals
&& p
->tx_bytes
> p
[-1].tx_bytes
; p
--) {
2752 swap_bals(p
, p
- 1);
2754 for (; p
< &bals
[n_bals
- 1] && p
->tx_bytes
< p
[1].tx_bytes
; p
++) {
2755 swap_bals(p
, p
+ 1);
2761 log_bals(const struct slave_balance
*bals
, size_t n_bals
, struct port
*port
)
2763 if (VLOG_IS_DBG_ENABLED()) {
2764 struct ds ds
= DS_EMPTY_INITIALIZER
;
2765 const struct slave_balance
*b
;
2767 for (b
= bals
; b
< bals
+ n_bals
; b
++) {
2771 ds_put_char(&ds
, ',');
2773 ds_put_format(&ds
, " %s %"PRIu64
"kB",
2774 b
->iface
->name
, b
->tx_bytes
/ 1024);
2776 if (!b
->iface
->enabled
) {
2777 ds_put_cstr(&ds
, " (disabled)");
2779 if (b
->n_hashes
> 0) {
2780 ds_put_cstr(&ds
, " (");
2781 for (i
= 0; i
< b
->n_hashes
; i
++) {
2782 const struct bond_entry
*e
= b
->hashes
[i
];
2784 ds_put_cstr(&ds
, " + ");
2786 ds_put_format(&ds
, "h%td: %"PRIu64
"kB",
2787 e
- port
->bond_hash
, e
->tx_bytes
/ 1024);
2789 ds_put_cstr(&ds
, ")");
2792 VLOG_DBG("bond %s:%s", port
->name
, ds_cstr(&ds
));
2797 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2799 bond_shift_load(struct slave_balance
*from
, struct slave_balance
*to
,
2802 struct bond_entry
*hash
= from
->hashes
[hash_idx
];
2803 struct port
*port
= from
->iface
->port
;
2804 uint64_t delta
= hash
->tx_bytes
;
2806 VLOG_INFO("bond %s: shift %"PRIu64
"kB of load (with hash %td) "
2807 "from %s to %s (now carrying %"PRIu64
"kB and "
2808 "%"PRIu64
"kB load, respectively)",
2809 port
->name
, delta
/ 1024, hash
- port
->bond_hash
,
2810 from
->iface
->name
, to
->iface
->name
,
2811 (from
->tx_bytes
- delta
) / 1024,
2812 (to
->tx_bytes
+ delta
) / 1024);
2814 /* Delete element from from->hashes.
2816 * We don't bother to add the element to to->hashes because not only would
2817 * it require more work, the only purpose it would be to allow that hash to
2818 * be migrated to another slave in this rebalancing run, and there is no
2819 * point in doing that. */
2820 if (hash_idx
== 0) {
2823 memmove(from
->hashes
+ hash_idx
, from
->hashes
+ hash_idx
+ 1,
2824 (from
->n_hashes
- (hash_idx
+ 1)) * sizeof *from
->hashes
);
2828 /* Shift load away from 'from' to 'to'. */
2829 from
->tx_bytes
-= delta
;
2830 to
->tx_bytes
+= delta
;
2832 /* Arrange for flows to be revalidated. */
2833 ofproto_revalidate(port
->bridge
->ofproto
, hash
->iface_tag
);
2834 hash
->iface_idx
= to
->iface
->port_ifidx
;
2835 hash
->iface_tag
= tag_create_random();
2839 bond_rebalance_port(struct port
*port
)
2841 struct slave_balance bals
[DP_MAX_PORTS
];
2843 struct bond_entry
*hashes
[BOND_MASK
+ 1];
2844 struct slave_balance
*b
, *from
, *to
;
2845 struct bond_entry
*e
;
2848 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2849 * descending order of tx_bytes, so that bals[0] represents the most
2850 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2853 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2854 * array for each slave_balance structure, we sort our local array of
2855 * hashes in order by slave, so that all of the hashes for a given slave
2856 * become contiguous in memory, and then we point each 'hashes' members of
2857 * a slave_balance structure to the start of a contiguous group. */
2858 n_bals
= port
->n_ifaces
;
2859 for (b
= bals
; b
< &bals
[n_bals
]; b
++) {
2860 b
->iface
= port
->ifaces
[b
- bals
];
2865 for (i
= 0; i
<= BOND_MASK
; i
++) {
2866 hashes
[i
] = &port
->bond_hash
[i
];
2868 qsort(hashes
, BOND_MASK
+ 1, sizeof *hashes
, compare_bond_entries
);
2869 for (i
= 0; i
<= BOND_MASK
; i
++) {
2871 if (e
->iface_idx
>= 0 && e
->iface_idx
< port
->n_ifaces
) {
2872 b
= &bals
[e
->iface_idx
];
2873 b
->tx_bytes
+= e
->tx_bytes
;
2875 b
->hashes
= &hashes
[i
];
2880 qsort(bals
, n_bals
, sizeof *bals
, compare_slave_balance
);
2881 log_bals(bals
, n_bals
, port
);
2883 /* Discard slaves that aren't enabled (which were sorted to the back of the
2884 * array earlier). */
2885 while (!bals
[n_bals
- 1].iface
->enabled
) {
2892 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2893 to
= &bals
[n_bals
- 1];
2894 for (from
= bals
; from
< to
; ) {
2895 uint64_t overload
= from
->tx_bytes
- to
->tx_bytes
;
2896 if (overload
< to
->tx_bytes
>> 5 || overload
< 100000) {
2897 /* The extra load on 'from' (and all less-loaded slaves), compared
2898 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2899 * it is less than ~1Mbps. No point in rebalancing. */
2901 } else if (from
->n_hashes
== 1) {
2902 /* 'from' only carries a single MAC hash, so we can't shift any
2903 * load away from it, even though we want to. */
2906 /* 'from' is carrying significantly more load than 'to', and that
2907 * load is split across at least two different hashes. Pick a hash
2908 * to migrate to 'to' (the least-loaded slave), given that doing so
2909 * must decrease the ratio of the load on the two slaves by at
2912 * The sort order we use means that we prefer to shift away the
2913 * smallest hashes instead of the biggest ones. There is little
2914 * reason behind this decision; we could use the opposite sort
2915 * order to shift away big hashes ahead of small ones. */
2918 for (i
= 0; i
< from
->n_hashes
; i
++) {
2919 double old_ratio
, new_ratio
;
2920 uint64_t delta
= from
->hashes
[i
]->tx_bytes
;
2922 if (delta
== 0 || from
->tx_bytes
- delta
== 0) {
2923 /* Pointless move. */
2927 order_swapped
= from
->tx_bytes
- delta
< to
->tx_bytes
+ delta
;
2929 if (to
->tx_bytes
== 0) {
2930 /* Nothing on the new slave, move it. */
2934 old_ratio
= (double)from
->tx_bytes
/ to
->tx_bytes
;
2935 new_ratio
= (double)(from
->tx_bytes
- delta
) /
2936 (to
->tx_bytes
+ delta
);
2938 if (new_ratio
== 0) {
2939 /* Should already be covered but check to prevent division
2944 if (new_ratio
< 1) {
2945 new_ratio
= 1 / new_ratio
;
2948 if (old_ratio
- new_ratio
> 0.1) {
2949 /* Would decrease the ratio, move it. */
2953 if (i
< from
->n_hashes
) {
2954 bond_shift_load(from
, to
, i
);
2955 port
->bond_compat_is_stale
= true;
2957 /* If the result of the migration changed the relative order of
2958 * 'from' and 'to' swap them back to maintain invariants. */
2959 if (order_swapped
) {
2960 swap_bals(from
, to
);
2963 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2964 * point to different slave_balance structures. It is only
2965 * valid to do these two operations in a row at all because we
2966 * know that 'from' will not move past 'to' and vice versa. */
2967 resort_bals(from
, bals
, n_bals
);
2968 resort_bals(to
, bals
, n_bals
);
2975 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2976 * historical data to decay to <1% in 7 rebalancing runs. */
2977 for (e
= &port
->bond_hash
[0]; e
<= &port
->bond_hash
[BOND_MASK
]; e
++) {
2983 bond_send_learning_packets(struct port
*port
)
2985 struct bridge
*br
= port
->bridge
;
2986 struct mac_entry
*e
;
2987 struct ofpbuf packet
;
2988 int error
, n_packets
, n_errors
;
2990 if (!port
->n_ifaces
|| port
->active_iface
< 0) {
2994 ofpbuf_init(&packet
, 128);
2995 error
= n_packets
= n_errors
= 0;
2996 LIST_FOR_EACH (e
, lru_node
, &br
->ml
->lrus
) {
2997 union ofp_action actions
[2], *a
;
3003 if (e
->port
== port
->port_idx
3004 || !choose_output_iface(port
, e
->mac
, &dp_ifidx
, &tags
)) {
3008 /* Compose actions. */
3009 memset(actions
, 0, sizeof actions
);
3012 a
->vlan_vid
.type
= htons(OFPAT_SET_VLAN_VID
);
3013 a
->vlan_vid
.len
= htons(sizeof *a
);
3014 a
->vlan_vid
.vlan_vid
= htons(e
->vlan
);
3017 a
->output
.type
= htons(OFPAT_OUTPUT
);
3018 a
->output
.len
= htons(sizeof *a
);
3019 a
->output
.port
= htons(odp_port_to_ofp_port(dp_ifidx
));
3024 compose_benign_packet(&packet
, "Open vSwitch Bond Failover", 0xf177,
3026 flow_extract(&packet
, 0, ODPP_NONE
, &flow
);
3027 retval
= ofproto_send_packet(br
->ofproto
, &flow
, actions
, a
- actions
,
3034 ofpbuf_uninit(&packet
);
3037 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3038 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
3039 "packets, last error was: %s",
3040 port
->name
, n_errors
, n_packets
, strerror(error
));
3042 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
3043 port
->name
, n_packets
);
3047 /* Bonding unixctl user interface functions. */
3050 bond_unixctl_list(struct unixctl_conn
*conn
,
3051 const char *args OVS_UNUSED
, void *aux OVS_UNUSED
)
3053 struct ds ds
= DS_EMPTY_INITIALIZER
;
3054 const struct bridge
*br
;
3056 ds_put_cstr(&ds
, "bridge\tbond\tslaves\n");
3058 LIST_FOR_EACH (br
, node
, &all_bridges
) {
3061 for (i
= 0; i
< br
->n_ports
; i
++) {
3062 const struct port
*port
= br
->ports
[i
];
3063 if (port
->n_ifaces
> 1) {
3066 ds_put_format(&ds
, "%s\t%s\t", br
->name
, port
->name
);
3067 for (j
= 0; j
< port
->n_ifaces
; j
++) {
3068 const struct iface
*iface
= port
->ifaces
[j
];
3070 ds_put_cstr(&ds
, ", ");
3072 ds_put_cstr(&ds
, iface
->name
);
3074 ds_put_char(&ds
, '\n');
3078 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
3082 static struct port
*
3083 bond_find(const char *name
)
3085 const struct bridge
*br
;
3087 LIST_FOR_EACH (br
, node
, &all_bridges
) {
3090 for (i
= 0; i
< br
->n_ports
; i
++) {
3091 struct port
*port
= br
->ports
[i
];
3092 if (!strcmp(port
->name
, name
) && port
->n_ifaces
> 1) {
3101 bond_unixctl_show(struct unixctl_conn
*conn
,
3102 const char *args
, void *aux OVS_UNUSED
)
3104 struct ds ds
= DS_EMPTY_INITIALIZER
;
3105 const struct port
*port
;
3108 port
= bond_find(args
);
3110 unixctl_command_reply(conn
, 501, "no such bond");
3114 ds_put_format(&ds
, "updelay: %d ms\n", port
->updelay
);
3115 ds_put_format(&ds
, "downdelay: %d ms\n", port
->downdelay
);
3116 ds_put_format(&ds
, "next rebalance: %lld ms\n",
3117 port
->bond_next_rebalance
- time_msec());
3118 for (j
= 0; j
< port
->n_ifaces
; j
++) {
3119 const struct iface
*iface
= port
->ifaces
[j
];
3120 struct bond_entry
*be
;
3123 ds_put_format(&ds
, "slave %s: %s\n",
3124 iface
->name
, iface
->enabled
? "enabled" : "disabled");
3125 if (j
== port
->active_iface
) {
3126 ds_put_cstr(&ds
, "\tactive slave\n");
3128 if (iface
->delay_expires
!= LLONG_MAX
) {
3129 ds_put_format(&ds
, "\t%s expires in %lld ms\n",
3130 iface
->enabled
? "downdelay" : "updelay",
3131 iface
->delay_expires
- time_msec());
3135 for (be
= port
->bond_hash
; be
<= &port
->bond_hash
[BOND_MASK
]; be
++) {
3136 int hash
= be
- port
->bond_hash
;
3137 struct mac_entry
*me
;
3139 if (be
->iface_idx
!= j
) {
3143 ds_put_format(&ds
, "\thash %d: %"PRIu64
" kB load\n",
3144 hash
, be
->tx_bytes
/ 1024);
3147 LIST_FOR_EACH (me
, lru_node
, &port
->bridge
->ml
->lrus
) {
3150 if (bond_hash(me
->mac
) == hash
3151 && me
->port
!= port
->port_idx
3152 && choose_output_iface(port
, me
->mac
, &dp_ifidx
, &tags
)
3153 && dp_ifidx
== iface
->dp_ifidx
)
3155 ds_put_format(&ds
, "\t\t"ETH_ADDR_FMT
"\n",
3156 ETH_ADDR_ARGS(me
->mac
));
3161 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
3166 bond_unixctl_migrate(struct unixctl_conn
*conn
, const char *args_
,
3167 void *aux OVS_UNUSED
)
3169 char *args
= (char *) args_
;
3170 char *save_ptr
= NULL
;
3171 char *bond_s
, *hash_s
, *slave_s
;
3172 uint8_t mac
[ETH_ADDR_LEN
];
3174 struct iface
*iface
;
3175 struct bond_entry
*entry
;
3178 bond_s
= strtok_r(args
, " ", &save_ptr
);
3179 hash_s
= strtok_r(NULL
, " ", &save_ptr
);
3180 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
3182 unixctl_command_reply(conn
, 501,
3183 "usage: bond/migrate BOND HASH SLAVE");
3187 port
= bond_find(bond_s
);
3189 unixctl_command_reply(conn
, 501, "no such bond");
3193 if (sscanf(hash_s
, ETH_ADDR_SCAN_FMT
, ETH_ADDR_SCAN_ARGS(mac
))
3194 == ETH_ADDR_SCAN_COUNT
) {
3195 hash
= bond_hash(mac
);
3196 } else if (strspn(hash_s
, "0123456789") == strlen(hash_s
)) {
3197 hash
= atoi(hash_s
) & BOND_MASK
;
3199 unixctl_command_reply(conn
, 501, "bad hash");
3203 iface
= port_lookup_iface(port
, slave_s
);
3205 unixctl_command_reply(conn
, 501, "no such slave");
3209 if (!iface
->enabled
) {
3210 unixctl_command_reply(conn
, 501, "cannot migrate to disabled slave");
3214 entry
= &port
->bond_hash
[hash
];
3215 ofproto_revalidate(port
->bridge
->ofproto
, entry
->iface_tag
);
3216 entry
->iface_idx
= iface
->port_ifidx
;
3217 entry
->iface_tag
= tag_create_random();
3218 port
->bond_compat_is_stale
= true;
3219 unixctl_command_reply(conn
, 200, "migrated");
3223 bond_unixctl_set_active_slave(struct unixctl_conn
*conn
, const char *args_
,
3224 void *aux OVS_UNUSED
)
3226 char *args
= (char *) args_
;
3227 char *save_ptr
= NULL
;
3228 char *bond_s
, *slave_s
;
3230 struct iface
*iface
;
3232 bond_s
= strtok_r(args
, " ", &save_ptr
);
3233 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
3235 unixctl_command_reply(conn
, 501,
3236 "usage: bond/set-active-slave BOND SLAVE");
3240 port
= bond_find(bond_s
);
3242 unixctl_command_reply(conn
, 501, "no such bond");
3246 iface
= port_lookup_iface(port
, slave_s
);
3248 unixctl_command_reply(conn
, 501, "no such slave");
3252 if (!iface
->enabled
) {
3253 unixctl_command_reply(conn
, 501, "cannot make disabled slave active");
3257 if (port
->active_iface
!= iface
->port_ifidx
) {
3258 ofproto_revalidate(port
->bridge
->ofproto
, port
->active_iface_tag
);
3259 port
->active_iface
= iface
->port_ifidx
;
3260 port
->active_iface_tag
= tag_create_random();
3261 VLOG_INFO("port %s: active interface is now %s",
3262 port
->name
, iface
->name
);
3263 bond_send_learning_packets(port
);
3264 unixctl_command_reply(conn
, 200, "done");
3266 unixctl_command_reply(conn
, 200, "no change");
3271 enable_slave(struct unixctl_conn
*conn
, const char *args_
, bool enable
)
3273 char *args
= (char *) args_
;
3274 char *save_ptr
= NULL
;
3275 char *bond_s
, *slave_s
;
3277 struct iface
*iface
;
3279 bond_s
= strtok_r(args
, " ", &save_ptr
);
3280 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
3282 unixctl_command_reply(conn
, 501,
3283 "usage: bond/enable/disable-slave BOND SLAVE");
3287 port
= bond_find(bond_s
);
3289 unixctl_command_reply(conn
, 501, "no such bond");
3293 iface
= port_lookup_iface(port
, slave_s
);
3295 unixctl_command_reply(conn
, 501, "no such slave");
3299 bond_enable_slave(iface
, enable
);
3300 unixctl_command_reply(conn
, 501, enable
? "enabled" : "disabled");
3304 bond_unixctl_enable_slave(struct unixctl_conn
*conn
, const char *args
,
3305 void *aux OVS_UNUSED
)
3307 enable_slave(conn
, args
, true);
3311 bond_unixctl_disable_slave(struct unixctl_conn
*conn
, const char *args
,
3312 void *aux OVS_UNUSED
)
3314 enable_slave(conn
, args
, false);
3318 bond_unixctl_hash(struct unixctl_conn
*conn
, const char *args
,
3319 void *aux OVS_UNUSED
)
3321 uint8_t mac
[ETH_ADDR_LEN
];
3325 if (sscanf(args
, ETH_ADDR_SCAN_FMT
, ETH_ADDR_SCAN_ARGS(mac
))
3326 == ETH_ADDR_SCAN_COUNT
) {
3327 hash
= bond_hash(mac
);
3329 hash_cstr
= xasprintf("%u", hash
);
3330 unixctl_command_reply(conn
, 200, hash_cstr
);
3333 unixctl_command_reply(conn
, 501, "invalid mac");
3340 unixctl_command_register("bond/list", bond_unixctl_list
, NULL
);
3341 unixctl_command_register("bond/show", bond_unixctl_show
, NULL
);
3342 unixctl_command_register("bond/migrate", bond_unixctl_migrate
, NULL
);
3343 unixctl_command_register("bond/set-active-slave",
3344 bond_unixctl_set_active_slave
, NULL
);
3345 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave
,
3347 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave
,
3349 unixctl_command_register("bond/hash", bond_unixctl_hash
, NULL
);
3352 /* Port functions. */
3354 static struct port
*
3355 port_create(struct bridge
*br
, const char *name
)
3359 port
= xzalloc(sizeof *port
);
3361 port
->port_idx
= br
->n_ports
;
3363 port
->trunks
= NULL
;
3364 port
->name
= xstrdup(name
);
3365 port
->active_iface
= -1;
3367 if (br
->n_ports
>= br
->allocated_ports
) {
3368 br
->ports
= x2nrealloc(br
->ports
, &br
->allocated_ports
,
3371 br
->ports
[br
->n_ports
++] = port
;
3372 shash_add_assert(&br
->port_by_name
, port
->name
, port
);
3374 VLOG_INFO("created port %s on bridge %s", port
->name
, br
->name
);
3381 get_port_other_config(const struct ovsrec_port
*port
, const char *key
,
3382 const char *default_value
)
3386 value
= get_ovsrec_key_value(&port
->header_
, &ovsrec_port_col_other_config
,
3388 return value
? value
: default_value
;
3392 port_del_ifaces(struct port
*port
, const struct ovsrec_port
*cfg
)
3394 struct shash new_ifaces
;
3397 /* Collect list of new interfaces. */
3398 shash_init(&new_ifaces
);
3399 for (i
= 0; i
< cfg
->n_interfaces
; i
++) {
3400 const char *name
= cfg
->interfaces
[i
]->name
;
3401 shash_add_once(&new_ifaces
, name
, NULL
);
3404 /* Get rid of deleted interfaces. */
3405 for (i
= 0; i
< port
->n_ifaces
; ) {
3406 if (!shash_find(&new_ifaces
, cfg
->interfaces
[i
]->name
)) {
3407 iface_destroy(port
->ifaces
[i
]);
3413 shash_destroy(&new_ifaces
);
3417 port_reconfigure(struct port
*port
, const struct ovsrec_port
*cfg
)
3419 struct shash new_ifaces
;
3420 long long int next_rebalance
;
3421 unsigned long *trunks
;
3427 /* Update settings. */
3428 port
->updelay
= cfg
->bond_updelay
;
3429 if (port
->updelay
< 0) {
3432 port
->downdelay
= cfg
->bond_downdelay
;
3433 if (port
->downdelay
< 0) {
3434 port
->downdelay
= 0;
3436 port
->bond_rebalance_interval
= atoi(
3437 get_port_other_config(cfg
, "bond-rebalance-interval", "10000"));
3438 if (port
->bond_rebalance_interval
< 1000) {
3439 port
->bond_rebalance_interval
= 1000;
3441 next_rebalance
= time_msec() + port
->bond_rebalance_interval
;
3442 if (port
->bond_next_rebalance
> next_rebalance
) {
3443 port
->bond_next_rebalance
= next_rebalance
;
3446 /* Add new interfaces and update 'cfg' member of existing ones. */
3447 shash_init(&new_ifaces
);
3448 for (i
= 0; i
< cfg
->n_interfaces
; i
++) {
3449 const struct ovsrec_interface
*if_cfg
= cfg
->interfaces
[i
];
3450 struct iface
*iface
;
3452 if (!shash_add_once(&new_ifaces
, if_cfg
->name
, NULL
)) {
3453 VLOG_WARN("port %s: %s specified twice as port interface",
3454 port
->name
, if_cfg
->name
);
3455 iface_set_ofport(if_cfg
, -1);
3459 iface
= iface_lookup(port
->bridge
, if_cfg
->name
);
3461 if (iface
->port
!= port
) {
3462 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
3464 port
->bridge
->name
, if_cfg
->name
, iface
->port
->name
);
3467 iface
->cfg
= if_cfg
;
3469 iface
= iface_create(port
, if_cfg
);
3472 /* Determine interface type. The local port always has type
3473 * "internal". Other ports take their type from the database and
3474 * default to "system" if none is specified. */
3475 iface
->type
= (!strcmp(if_cfg
->name
, port
->bridge
->name
) ? "internal"
3476 : if_cfg
->type
[0] ? if_cfg
->type
3479 shash_destroy(&new_ifaces
);
3484 if (port
->n_ifaces
< 2) {
3486 if (vlan
>= 0 && vlan
<= 4095) {
3487 VLOG_DBG("port %s: assigning VLAN tag %d", port
->name
, vlan
);
3492 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
3493 * they even work as-is. But they have not been tested. */
3494 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
3498 if (port
->vlan
!= vlan
) {
3500 bridge_flush(port
->bridge
);
3503 /* Get trunked VLANs. */
3505 if (vlan
< 0 && cfg
->n_trunks
) {
3508 trunks
= bitmap_allocate(4096);
3510 for (i
= 0; i
< cfg
->n_trunks
; i
++) {
3511 int trunk
= cfg
->trunks
[i
];
3513 bitmap_set1(trunks
, trunk
);
3519 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
3520 port
->name
, cfg
->n_trunks
);
3522 if (n_errors
== cfg
->n_trunks
) {
3523 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
3525 bitmap_free(trunks
);
3528 } else if (vlan
>= 0 && cfg
->n_trunks
) {
3529 VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan",
3533 ? port
->trunks
!= NULL
3534 : port
->trunks
== NULL
|| !bitmap_equal(trunks
, port
->trunks
, 4096)) {
3535 bridge_flush(port
->bridge
);
3537 bitmap_free(port
->trunks
);
3538 port
->trunks
= trunks
;
3542 port_destroy(struct port
*port
)
3545 struct bridge
*br
= port
->bridge
;
3549 proc_net_compat_update_vlan(port
->name
, NULL
, 0);
3550 proc_net_compat_update_bond(port
->name
, NULL
);
3552 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3553 struct mirror
*m
= br
->mirrors
[i
];
3554 if (m
&& m
->out_port
== port
) {
3559 while (port
->n_ifaces
> 0) {
3560 iface_destroy(port
->ifaces
[port
->n_ifaces
- 1]);
3563 shash_find_and_delete_assert(&br
->port_by_name
, port
->name
);
3565 del
= br
->ports
[port
->port_idx
] = br
->ports
[--br
->n_ports
];
3566 del
->port_idx
= port
->port_idx
;
3568 VLOG_INFO("destroyed port %s on bridge %s", port
->name
, br
->name
);
3570 netdev_monitor_destroy(port
->monitor
);
3572 bitmap_free(port
->trunks
);
3579 static struct port
*
3580 port_from_dp_ifidx(const struct bridge
*br
, uint16_t dp_ifidx
)
3582 struct iface
*iface
= iface_from_dp_ifidx(br
, dp_ifidx
);
3583 return iface
? iface
->port
: NULL
;
3586 static struct port
*
3587 port_lookup(const struct bridge
*br
, const char *name
)
3589 return shash_find_data(&br
->port_by_name
, name
);
3592 static struct iface
*
3593 port_lookup_iface(const struct port
*port
, const char *name
)
3595 struct iface
*iface
= iface_lookup(port
->bridge
, name
);
3596 return iface
&& iface
->port
== port
? iface
: NULL
;
3600 port_update_bonding(struct port
*port
)
3602 if (port
->monitor
) {
3603 netdev_monitor_destroy(port
->monitor
);
3604 port
->monitor
= NULL
;
3606 if (port
->n_ifaces
< 2) {
3607 /* Not a bonded port. */
3608 if (port
->bond_hash
) {
3609 free(port
->bond_hash
);
3610 port
->bond_hash
= NULL
;
3611 port
->bond_compat_is_stale
= true;
3612 port
->bond_fake_iface
= false;
3617 if (!port
->bond_hash
) {
3618 port
->bond_hash
= xcalloc(BOND_MASK
+ 1, sizeof *port
->bond_hash
);
3619 for (i
= 0; i
<= BOND_MASK
; i
++) {
3620 struct bond_entry
*e
= &port
->bond_hash
[i
];
3624 port
->no_ifaces_tag
= tag_create_random();
3625 bond_choose_active_iface(port
);
3626 port
->bond_next_rebalance
3627 = time_msec() + port
->bond_rebalance_interval
;
3629 if (port
->cfg
->bond_fake_iface
) {
3630 port
->bond_next_fake_iface_update
= time_msec();
3633 port
->bond_compat_is_stale
= true;
3634 port
->bond_fake_iface
= port
->cfg
->bond_fake_iface
;
3636 port
->monitor
= netdev_monitor_create();
3637 for (i
= 0; i
< port
->n_ifaces
; i
++) {
3638 netdev_monitor_add(port
->monitor
, port
->ifaces
[i
]->netdev
);
3644 port_update_bond_compat(struct port
*port
)
3646 struct compat_bond_hash compat_hashes
[BOND_MASK
+ 1];
3647 struct compat_bond bond
;
3650 if (port
->n_ifaces
< 2) {
3651 proc_net_compat_update_bond(port
->name
, NULL
);
3656 bond
.updelay
= port
->updelay
;
3657 bond
.downdelay
= port
->downdelay
;
3660 bond
.hashes
= compat_hashes
;
3661 if (port
->bond_hash
) {
3662 const struct bond_entry
*e
;
3663 for (e
= port
->bond_hash
; e
<= &port
->bond_hash
[BOND_MASK
]; e
++) {
3664 if (e
->iface_idx
>= 0 && e
->iface_idx
< port
->n_ifaces
) {
3665 struct compat_bond_hash
*cbh
= &bond
.hashes
[bond
.n_hashes
++];
3666 cbh
->hash
= e
- port
->bond_hash
;
3667 cbh
->netdev_name
= port
->ifaces
[e
->iface_idx
]->name
;
3672 bond
.n_slaves
= port
->n_ifaces
;
3673 bond
.slaves
= xmalloc(port
->n_ifaces
* sizeof *bond
.slaves
);
3674 for (i
= 0; i
< port
->n_ifaces
; i
++) {
3675 struct iface
*iface
= port
->ifaces
[i
];
3676 struct compat_bond_slave
*slave
= &bond
.slaves
[i
];
3677 slave
->name
= iface
->name
;
3679 /* We need to make the same determination as the Linux bonding
3680 * code to determine whether a slave should be consider "up".
3681 * The Linux function bond_miimon_inspect() supports four
3682 * BOND_LINK_* states:
3684 * - BOND_LINK_UP: carrier detected, updelay has passed.
3685 * - BOND_LINK_FAIL: carrier lost, downdelay in progress.
3686 * - BOND_LINK_DOWN: carrier lost, downdelay has passed.
3687 * - BOND_LINK_BACK: carrier detected, updelay in progress.
3689 * The function bond_info_show_slave() only considers BOND_LINK_UP
3690 * to be "up" and anything else to be "down".
3692 slave
->up
= iface
->enabled
&& iface
->delay_expires
== LLONG_MAX
;
3696 netdev_get_etheraddr(iface
->netdev
, slave
->mac
);
3699 if (port
->bond_fake_iface
) {
3700 struct netdev
*bond_netdev
;
3702 if (!netdev_open_default(port
->name
, &bond_netdev
)) {
3704 netdev_turn_flags_on(bond_netdev
, NETDEV_UP
, true);
3706 netdev_turn_flags_off(bond_netdev
, NETDEV_UP
, true);
3708 netdev_close(bond_netdev
);
3712 proc_net_compat_update_bond(port
->name
, &bond
);
3717 port_update_vlan_compat(struct port
*port
)
3719 struct bridge
*br
= port
->bridge
;
3720 char *vlandev_name
= NULL
;
3722 if (port
->vlan
> 0) {
3723 /* Figure out the name that the VLAN device should actually have, if it
3724 * existed. This takes some work because the VLAN device would not
3725 * have port->name in its name; rather, it would have the trunk port's
3726 * name, and 'port' would be attached to a bridge that also had the
3727 * VLAN device one of its ports. So we need to find a trunk port that
3728 * includes port->vlan.
3730 * There might be more than one candidate. This doesn't happen on
3731 * XenServer, so if it happens we just pick the first choice in
3732 * alphabetical order instead of creating multiple VLAN devices. */
3734 for (i
= 0; i
< br
->n_ports
; i
++) {
3735 struct port
*p
= br
->ports
[i
];
3736 if (port_trunks_vlan(p
, port
->vlan
)
3738 && (!vlandev_name
|| strcmp(p
->name
, vlandev_name
) <= 0))
3740 uint8_t ea
[ETH_ADDR_LEN
];
3741 netdev_get_etheraddr(p
->ifaces
[0]->netdev
, ea
);
3742 if (!eth_addr_is_multicast(ea
) &&
3743 !eth_addr_is_reserved(ea
) &&
3744 !eth_addr_is_zero(ea
)) {
3745 vlandev_name
= p
->name
;
3750 proc_net_compat_update_vlan(port
->name
, vlandev_name
, port
->vlan
);
3753 /* Interface functions. */
3755 static struct iface
*
3756 iface_create(struct port
*port
, const struct ovsrec_interface
*if_cfg
)
3758 struct bridge
*br
= port
->bridge
;
3759 struct iface
*iface
;
3760 char *name
= if_cfg
->name
;
3762 iface
= xzalloc(sizeof *iface
);
3764 iface
->port_ifidx
= port
->n_ifaces
;
3765 iface
->name
= xstrdup(name
);
3766 iface
->dp_ifidx
= -1;
3767 iface
->tag
= tag_create_random();
3768 iface
->delay_expires
= LLONG_MAX
;
3769 iface
->netdev
= NULL
;
3770 iface
->cfg
= if_cfg
;
3772 shash_add_assert(&br
->iface_by_name
, iface
->name
, iface
);
3774 if (port
->n_ifaces
>= port
->allocated_ifaces
) {
3775 port
->ifaces
= x2nrealloc(port
->ifaces
, &port
->allocated_ifaces
,
3776 sizeof *port
->ifaces
);
3778 port
->ifaces
[port
->n_ifaces
++] = iface
;
3779 if (port
->n_ifaces
> 1) {
3780 br
->has_bonded_ports
= true;
3783 VLOG_DBG("attached network device %s to port %s", iface
->name
, port
->name
);
3791 iface_destroy(struct iface
*iface
)
3794 struct port
*port
= iface
->port
;
3795 struct bridge
*br
= port
->bridge
;
3796 bool del_active
= port
->active_iface
== iface
->port_ifidx
;
3799 shash_find_and_delete_assert(&br
->iface_by_name
, iface
->name
);
3801 if (iface
->dp_ifidx
>= 0) {
3802 hmap_remove(&br
->ifaces
, &iface
->dp_ifidx_node
);
3805 del
= port
->ifaces
[iface
->port_ifidx
] = port
->ifaces
[--port
->n_ifaces
];
3806 del
->port_ifidx
= iface
->port_ifidx
;
3808 netdev_close(iface
->netdev
);
3811 ofproto_revalidate(port
->bridge
->ofproto
, port
->active_iface_tag
);
3812 bond_choose_active_iface(port
);
3813 bond_send_learning_packets(port
);
3819 bridge_flush(port
->bridge
);
3823 static struct iface
*
3824 iface_lookup(const struct bridge
*br
, const char *name
)
3826 return shash_find_data(&br
->iface_by_name
, name
);
3829 static struct iface
*
3830 iface_from_dp_ifidx(const struct bridge
*br
, uint16_t dp_ifidx
)
3832 struct iface
*iface
;
3834 HMAP_FOR_EACH_IN_BUCKET (iface
, dp_ifidx_node
,
3835 hash_int(dp_ifidx
, 0), &br
->ifaces
) {
3836 if (iface
->dp_ifidx
== dp_ifidx
) {
3843 /* Set Ethernet address of 'iface', if one is specified in the configuration
3846 iface_set_mac(struct iface
*iface
)
3848 uint8_t ea
[ETH_ADDR_LEN
];
3850 if (iface
->cfg
->mac
&& eth_addr_from_string(iface
->cfg
->mac
, ea
)) {
3851 if (eth_addr_is_multicast(ea
)) {
3852 VLOG_ERR("interface %s: cannot set MAC to multicast address",
3854 } else if (iface
->dp_ifidx
== ODPP_LOCAL
) {
3855 VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead",
3856 iface
->name
, iface
->name
);
3858 int error
= netdev_set_etheraddr(iface
->netdev
, ea
);
3860 VLOG_ERR("interface %s: setting MAC failed (%s)",
3861 iface
->name
, strerror(error
));
3867 /* Sets the ofport column of 'if_cfg' to 'ofport'. */
3869 iface_set_ofport(const struct ovsrec_interface
*if_cfg
, int64_t ofport
)
3872 ovsrec_interface_set_ofport(if_cfg
, &ofport
, 1);
3876 /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'.
3878 * The value strings in '*shash' are taken directly from values[], not copied,
3879 * so the caller should not modify or free them. */
3881 shash_from_ovs_idl_map(char **keys
, char **values
, size_t n
,
3882 struct shash
*shash
)
3887 for (i
= 0; i
< n
; i
++) {
3888 shash_add(shash
, keys
[i
], values
[i
]);
3892 struct iface_delete_queues_cbdata
{
3893 struct netdev
*netdev
;
3894 const struct ovsdb_datum
*queues
;
3898 queue_ids_include(const struct ovsdb_datum
*queues
, int64_t target
)
3900 union ovsdb_atom atom
;
3902 atom
.integer
= target
;
3903 return ovsdb_datum_find_key(queues
, &atom
, OVSDB_TYPE_INTEGER
) != UINT_MAX
;
3907 iface_delete_queues(unsigned int queue_id
,
3908 const struct shash
*details OVS_UNUSED
, void *cbdata_
)
3910 struct iface_delete_queues_cbdata
*cbdata
= cbdata_
;
3912 if (!queue_ids_include(cbdata
->queues
, queue_id
)) {
3913 netdev_delete_queue(cbdata
->netdev
, queue_id
);
3918 iface_update_qos(struct iface
*iface
, const struct ovsrec_qos
*qos
)
3920 if (!qos
|| qos
->type
[0] == '\0') {
3921 netdev_set_qos(iface
->netdev
, NULL
, NULL
);
3923 struct iface_delete_queues_cbdata cbdata
;
3924 struct shash details
;
3927 /* Configure top-level Qos for 'iface'. */
3928 shash_from_ovs_idl_map(qos
->key_other_config
, qos
->value_other_config
,
3929 qos
->n_other_config
, &details
);
3930 netdev_set_qos(iface
->netdev
, qos
->type
, &details
);
3931 shash_destroy(&details
);
3933 /* Deconfigure queues that were deleted. */
3934 cbdata
.netdev
= iface
->netdev
;
3935 cbdata
.queues
= ovsrec_qos_get_queues(qos
, OVSDB_TYPE_INTEGER
,
3937 netdev_dump_queues(iface
->netdev
, iface_delete_queues
, &cbdata
);
3939 /* Configure queues for 'iface'. */
3940 for (i
= 0; i
< qos
->n_queues
; i
++) {
3941 const struct ovsrec_queue
*queue
= qos
->value_queues
[i
];
3942 unsigned int queue_id
= qos
->key_queues
[i
];
3944 shash_from_ovs_idl_map(queue
->key_other_config
,
3945 queue
->value_other_config
,
3946 queue
->n_other_config
, &details
);
3947 netdev_set_queue(iface
->netdev
, queue_id
, &details
);
3948 shash_destroy(&details
);
3953 /* Port mirroring. */
3955 static struct mirror
*
3956 mirror_find_by_uuid(struct bridge
*br
, const struct uuid
*uuid
)
3960 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3961 struct mirror
*m
= br
->mirrors
[i
];
3962 if (m
&& uuid_equals(uuid
, &m
->uuid
)) {
3970 mirror_reconfigure(struct bridge
*br
)
3972 unsigned long *rspan_vlans
;
3975 /* Get rid of deleted mirrors. */
3976 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3977 struct mirror
*m
= br
->mirrors
[i
];
3979 const struct ovsdb_datum
*mc
;
3980 union ovsdb_atom atom
;
3982 mc
= ovsrec_bridge_get_mirrors(br
->cfg
, OVSDB_TYPE_UUID
);
3983 atom
.uuid
= br
->mirrors
[i
]->uuid
;
3984 if (ovsdb_datum_find_key(mc
, &atom
, OVSDB_TYPE_UUID
) == UINT_MAX
) {
3990 /* Add new mirrors and reconfigure existing ones. */
3991 for (i
= 0; i
< br
->cfg
->n_mirrors
; i
++) {
3992 struct ovsrec_mirror
*cfg
= br
->cfg
->mirrors
[i
];
3993 struct mirror
*m
= mirror_find_by_uuid(br
, &cfg
->header_
.uuid
);
3995 mirror_reconfigure_one(m
, cfg
);
3997 mirror_create(br
, cfg
);
4001 /* Update port reserved status. */
4002 for (i
= 0; i
< br
->n_ports
; i
++) {
4003 br
->ports
[i
]->is_mirror_output_port
= false;
4005 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
4006 struct mirror
*m
= br
->mirrors
[i
];
4007 if (m
&& m
->out_port
) {
4008 m
->out_port
->is_mirror_output_port
= true;
4012 /* Update flooded vlans (for RSPAN). */
4014 if (br
->cfg
->n_flood_vlans
) {
4015 rspan_vlans
= bitmap_allocate(4096);
4017 for (i
= 0; i
< br
->cfg
->n_flood_vlans
; i
++) {
4018 int64_t vlan
= br
->cfg
->flood_vlans
[i
];
4019 if (vlan
>= 0 && vlan
< 4096) {
4020 bitmap_set1(rspan_vlans
, vlan
);
4021 VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64
,
4024 VLOG_ERR("bridge %s: invalid value %"PRId64
"for flood VLAN",
4029 if (mac_learning_set_flood_vlans(br
->ml
, rspan_vlans
)) {
4035 mirror_create(struct bridge
*br
, struct ovsrec_mirror
*cfg
)
4040 for (i
= 0; ; i
++) {
4041 if (i
>= MAX_MIRRORS
) {
4042 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
4043 "cannot create %s", br
->name
, MAX_MIRRORS
, cfg
->name
);
4046 if (!br
->mirrors
[i
]) {
4051 VLOG_INFO("created port mirror %s on bridge %s", cfg
->name
, br
->name
);
4054 br
->mirrors
[i
] = m
= xzalloc(sizeof *m
);
4057 m
->name
= xstrdup(cfg
->name
);
4058 shash_init(&m
->src_ports
);
4059 shash_init(&m
->dst_ports
);
4065 mirror_reconfigure_one(m
, cfg
);
4069 mirror_destroy(struct mirror
*m
)
4072 struct bridge
*br
= m
->bridge
;
4075 for (i
= 0; i
< br
->n_ports
; i
++) {
4076 br
->ports
[i
]->src_mirrors
&= ~(MIRROR_MASK_C(1) << m
->idx
);
4077 br
->ports
[i
]->dst_mirrors
&= ~(MIRROR_MASK_C(1) << m
->idx
);
4080 shash_destroy(&m
->src_ports
);
4081 shash_destroy(&m
->dst_ports
);
4084 m
->bridge
->mirrors
[m
->idx
] = NULL
;
4093 mirror_collect_ports(struct mirror
*m
, struct ovsrec_port
**ports
, int n_ports
,
4094 struct shash
*names
)
4098 for (i
= 0; i
< n_ports
; i
++) {
4099 const char *name
= ports
[i
]->name
;
4100 if (port_lookup(m
->bridge
, name
)) {
4101 shash_add_once(names
, name
, NULL
);
4103 VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent "
4104 "port %s", m
->bridge
->name
, m
->name
, name
);
4110 mirror_collect_vlans(struct mirror
*m
, const struct ovsrec_mirror
*cfg
,
4116 *vlans
= xmalloc(sizeof **vlans
* cfg
->n_select_vlan
);
4118 for (i
= 0; i
< cfg
->n_select_vlan
; i
++) {
4119 int64_t vlan
= cfg
->select_vlan
[i
];
4120 if (vlan
< 0 || vlan
> 4095) {
4121 VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64
,
4122 m
->bridge
->name
, m
->name
, vlan
);
4124 (*vlans
)[n_vlans
++] = vlan
;
4131 vlan_is_mirrored(const struct mirror
*m
, int vlan
)
4135 for (i
= 0; i
< m
->n_vlans
; i
++) {
4136 if (m
->vlans
[i
] == vlan
) {
4144 port_trunks_any_mirrored_vlan(const struct mirror
*m
, const struct port
*p
)
4148 for (i
= 0; i
< m
->n_vlans
; i
++) {
4149 if (port_trunks_vlan(p
, m
->vlans
[i
])) {
4157 mirror_reconfigure_one(struct mirror
*m
, struct ovsrec_mirror
*cfg
)
4159 struct shash src_ports
, dst_ports
;
4160 mirror_mask_t mirror_bit
;
4161 struct port
*out_port
;
4168 if (strcmp(cfg
->name
, m
->name
)) {
4170 m
->name
= xstrdup(cfg
->name
);
4173 /* Get output port. */
4174 if (cfg
->output_port
) {
4175 out_port
= port_lookup(m
->bridge
, cfg
->output_port
->name
);
4177 VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge",
4178 m
->bridge
->name
, m
->name
);
4184 if (cfg
->output_vlan
) {
4185 VLOG_ERR("bridge %s: mirror %s specifies both output port and "
4186 "output vlan; ignoring output vlan",
4187 m
->bridge
->name
, m
->name
);
4189 } else if (cfg
->output_vlan
) {
4191 out_vlan
= *cfg
->output_vlan
;
4193 VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring",
4194 m
->bridge
->name
, m
->name
);
4199 shash_init(&src_ports
);
4200 shash_init(&dst_ports
);
4201 if (cfg
->select_all
) {
4202 for (i
= 0; i
< m
->bridge
->n_ports
; i
++) {
4203 const char *name
= m
->bridge
->ports
[i
]->name
;
4204 shash_add_once(&src_ports
, name
, NULL
);
4205 shash_add_once(&dst_ports
, name
, NULL
);
4210 /* Get ports, and drop duplicates and ports that don't exist. */
4211 mirror_collect_ports(m
, cfg
->select_src_port
, cfg
->n_select_src_port
,
4213 mirror_collect_ports(m
, cfg
->select_dst_port
, cfg
->n_select_dst_port
,
4216 /* Get all the vlans, and drop duplicate and invalid vlans. */
4217 n_vlans
= mirror_collect_vlans(m
, cfg
, &vlans
);
4220 /* Update mirror data. */
4221 if (!shash_equal_keys(&m
->src_ports
, &src_ports
)
4222 || !shash_equal_keys(&m
->dst_ports
, &dst_ports
)
4223 || m
->n_vlans
!= n_vlans
4224 || memcmp(m
->vlans
, vlans
, sizeof *vlans
* n_vlans
)
4225 || m
->out_port
!= out_port
4226 || m
->out_vlan
!= out_vlan
) {
4227 bridge_flush(m
->bridge
);
4229 shash_swap(&m
->src_ports
, &src_ports
);
4230 shash_swap(&m
->dst_ports
, &dst_ports
);
4233 m
->n_vlans
= n_vlans
;
4234 m
->out_port
= out_port
;
4235 m
->out_vlan
= out_vlan
;
4238 mirror_bit
= MIRROR_MASK_C(1) << m
->idx
;
4239 for (i
= 0; i
< m
->bridge
->n_ports
; i
++) {
4240 struct port
*port
= m
->bridge
->ports
[i
];
4242 if (shash_find(&m
->src_ports
, port
->name
)
4245 ? port_trunks_any_mirrored_vlan(m
, port
)
4246 : vlan_is_mirrored(m
, port
->vlan
)))) {
4247 port
->src_mirrors
|= mirror_bit
;
4249 port
->src_mirrors
&= ~mirror_bit
;
4252 if (shash_find(&m
->dst_ports
, port
->name
)) {
4253 port
->dst_mirrors
|= mirror_bit
;
4255 port
->dst_mirrors
&= ~mirror_bit
;
4260 shash_destroy(&src_ports
);
4261 shash_destroy(&dst_ports
);