1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
47 #include "poll-loop.h"
48 #include "port-array.h"
49 #include "proc-net-compat.h"
51 #include "secchan/ofproto.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id
;
74 struct port
*port
; /* Containing port. */
75 size_t port_ifidx
; /* Index within containing port. */
77 char *name
; /* Host network device name. */
78 int dp_ifidx
; /* Index within kernel datapath. */
80 uint8_t mac
[ETH_ADDR_LEN
]; /* Ethernet address (all zeros if unknowns). */
82 tag_type tag
; /* Tag associated with this interface. */
83 bool enabled
; /* May be chosen for flows? */
84 long long delay_expires
; /* Time after which 'enabled' may change. */
87 #define BOND_MASK 0xff
89 int iface_idx
; /* Index of assigned iface, or -1 if none. */
90 uint64_t tx_bytes
; /* Count of bytes recently transmitted. */
91 tag_type iface_tag
; /* Tag associated with iface_idx. */
94 #define MAX_MIRRORS 32
95 typedef uint32_t mirror_mask_t
;
96 #define MIRROR_MASK_C(X) UINT32_C(X)
97 BUILD_ASSERT_DECL(sizeof(mirror_mask_t
) * CHAR_BIT
>= MAX_MIRRORS
);
99 struct bridge
*bridge
;
103 /* Selection criteria. */
104 struct svec src_ports
;
105 struct svec dst_ports
;
110 struct port
*out_port
;
114 #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */
116 struct bridge
*bridge
;
118 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1. */
122 /* An ordinary bridge port has 1 interface.
123 * A bridge port for bonding has at least 2 interfaces. */
124 struct iface
**ifaces
;
125 size_t n_ifaces
, allocated_ifaces
;
128 struct bond_entry
*bond_hash
; /* An array of (BOND_MASK + 1) elements. */
129 int active_iface
; /* Ifidx on which bcasts accepted, or -1. */
130 tag_type active_iface_tag
; /* Tag for bcast flows. */
131 tag_type no_ifaces_tag
; /* Tag for flows when all ifaces disabled. */
132 int updelay
, downdelay
; /* Delay before iface goes up/down, in ms. */
134 /* Port mirroring info. */
135 mirror_mask_t src_mirrors
; /* Mirrors triggered when packet received. */
136 mirror_mask_t dst_mirrors
; /* Mirrors triggered when packet sent. */
137 bool is_mirror_output_port
; /* Does port mirroring send frames here? */
139 /* Spanning tree info. */
140 enum stp_state stp_state
; /* Always STP_FORWARDING if STP not in use. */
141 tag_type stp_state_tag
; /* Tag for STP state change. */
144 #define DP_MAX_PORTS 255
146 struct list node
; /* Node in global list of bridges. */
147 char *name
; /* User-specified arbitrary name. */
148 struct mac_learning
*ml
; /* MAC learning table, or null not to learn. */
149 bool sent_config_request
; /* Successfully sent config request? */
150 uint8_t default_ea
[ETH_ADDR_LEN
]; /* Default MAC. */
152 /* Support for remote controllers. */
153 char *controller
; /* NULL if there is no remote controller;
154 * "discover" to do controller discovery;
155 * otherwise a vconn name. */
157 /* OpenFlow switch processing. */
158 struct ofproto
*ofproto
; /* OpenFlow switch. */
160 /* Kernel datapath information. */
161 struct dpif dpif
; /* Kernel datapath. */
162 struct port_array ifaces
; /* Indexed by kernel datapath port number. */
166 size_t n_ports
, allocated_ports
;
169 bool has_bonded_ports
;
170 long long int bond_next_rebalance
;
175 /* Flow statistics gathering. */
176 time_t next_stats_request
;
178 /* Port mirroring. */
179 struct mirror
*mirrors
[MAX_MIRRORS
];
183 long long int stp_last_tick
;
186 /* List of all bridges. */
187 static struct list all_bridges
= LIST_INITIALIZER(&all_bridges
);
189 /* Maximum number of datapaths. */
190 enum { DP_MAX
= 256 };
192 static struct bridge
*bridge_create(const char *name
);
193 static void bridge_destroy(struct bridge
*);
194 static struct bridge
*bridge_lookup(const char *name
);
195 static int bridge_run_one(struct bridge
*);
196 static void bridge_reconfigure_one(struct bridge
*);
197 static void bridge_reconfigure_controller(struct bridge
*);
198 static void bridge_get_all_ifaces(const struct bridge
*, struct svec
*ifaces
);
199 static void bridge_fetch_dp_ifaces(struct bridge
*);
200 static void bridge_flush(struct bridge
*);
201 static void bridge_pick_local_hw_addr(struct bridge
*,
202 uint8_t ea
[ETH_ADDR_LEN
],
203 const char **devname
);
204 static uint64_t bridge_pick_datapath_id(struct bridge
*,
205 const uint8_t bridge_ea
[ETH_ADDR_LEN
],
206 const char *devname
);
207 static uint64_t dpid_from_hash(const void *, size_t nbytes
);
209 static void bond_init(void);
210 static void bond_run(struct bridge
*);
211 static void bond_wait(struct bridge
*);
212 static void bond_rebalance_port(struct port
*);
213 static void bond_send_learning_packets(struct port
*);
215 static void port_create(struct bridge
*, const char *name
);
216 static void port_reconfigure(struct port
*);
217 static void port_destroy(struct port
*);
218 static struct port
*port_lookup(const struct bridge
*, const char *name
);
219 static struct iface
*port_lookup_iface(const struct port
*, const char *name
);
220 static struct port
*port_from_dp_ifidx(const struct bridge
*,
222 static void port_update_bond_compat(struct port
*);
223 static void port_update_vlan_compat(struct port
*);
225 static void mirror_create(struct bridge
*, const char *name
);
226 static void mirror_destroy(struct mirror
*);
227 static void mirror_reconfigure(struct bridge
*);
228 static void mirror_reconfigure_one(struct mirror
*);
229 static bool vlan_is_mirrored(const struct mirror
*, int vlan
);
231 static void brstp_reconfigure(struct bridge
*);
232 static void brstp_adjust_timers(struct bridge
*);
233 static void brstp_run(struct bridge
*);
234 static void brstp_wait(struct bridge
*);
236 static void iface_create(struct port
*, const char *name
);
237 static void iface_destroy(struct iface
*);
238 static struct iface
*iface_lookup(const struct bridge
*, const char *name
);
239 static struct iface
*iface_from_dp_ifidx(const struct bridge
*,
242 /* Hooks into ofproto processing. */
243 static struct ofhooks bridge_ofhooks
;
245 /* Public functions. */
247 /* Adds the name of each interface used by a bridge, including local and
248 * internal ports, to 'svec'. */
250 bridge_get_ifaces(struct svec
*svec
)
252 struct bridge
*br
, *next
;
255 LIST_FOR_EACH_SAFE (br
, next
, struct bridge
, node
, &all_bridges
) {
256 for (i
= 0; i
< br
->n_ports
; i
++) {
257 struct port
*port
= br
->ports
[i
];
259 for (j
= 0; j
< port
->n_ifaces
; j
++) {
260 struct iface
*iface
= port
->ifaces
[j
];
261 if (iface
->dp_ifidx
< 0) {
262 VLOG_ERR("%s interface not in dp%u, ignoring",
263 iface
->name
, dpif_id(&br
->dpif
));
265 if (iface
->dp_ifidx
!= ODPP_LOCAL
) {
266 svec_add(svec
, iface
->name
);
274 /* The caller must already have called cfg_read(). */
283 for (i
= 0; i
< DP_MAX
; i
++) {
287 sprintf(devname
, "dp%d", i
);
288 retval
= dpif_open(devname
, &dpif
);
290 char dpif_name
[IF_NAMESIZE
];
291 if (dpif_get_name(&dpif
, dpif_name
, sizeof dpif_name
)
292 || !cfg_has("bridge.%s.port", dpif_name
)) {
296 } else if (retval
!= ENODEV
) {
297 VLOG_ERR("failed to delete datapath dp%d: %s",
298 i
, strerror(retval
));
302 bridge_reconfigure();
307 config_string_change(const char *key
, char **valuep
)
309 const char *value
= cfg_get_string(0, "%s", key
);
310 if (value
&& (!*valuep
|| strcmp(value
, *valuep
))) {
312 *valuep
= xstrdup(value
);
320 bridge_configure_ssl(void)
322 /* XXX SSL should be configurable on a per-bridge basis.
323 * XXX should be possible to de-configure SSL. */
324 static char *private_key_file
;
325 static char *certificate_file
;
326 static char *cacert_file
;
329 if (config_string_change("ssl.private-key", &private_key_file
)) {
330 vconn_ssl_set_private_key_file(private_key_file
);
333 if (config_string_change("ssl.certificate", &certificate_file
)) {
334 vconn_ssl_set_certificate_file(certificate_file
);
337 /* We assume that even if the filename hasn't changed, if the CA cert
338 * file has been removed, that we want to move back into
339 * boot-strapping mode. This opens a small security hole, because
340 * the old certificate will still be trusted until vSwitch is
341 * restarted. We may want to address this in vconn's SSL library. */
342 if (config_string_change("ssl.ca-cert", &cacert_file
)
343 || (stat(cacert_file
, &s
) && errno
== ENOENT
)) {
344 vconn_ssl_set_ca_cert_file(cacert_file
,
345 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
351 bridge_reconfigure(void)
353 struct svec old_br
, new_br
, raw_new_br
;
354 struct bridge
*br
, *next
;
357 COVERAGE_INC(bridge_reconfigure
);
359 /* Collect old bridges. */
361 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
362 svec_add(&old_br
, br
->name
);
365 /* Collect new bridges. */
366 svec_init(&raw_new_br
);
367 cfg_get_subsections(&raw_new_br
, "bridge");
369 for (i
= 0; i
< raw_new_br
.n
; i
++) {
370 const char *name
= raw_new_br
.names
[i
];
371 if ((!strncmp(name
, "dp", 2) && isdigit(name
[2])) ||
372 (!strncmp(name
, "nl:", 3) && isdigit(name
[3]))) {
373 VLOG_ERR("%s is not a valid bridge name (bridges may not be "
374 "named \"dp\" or \"nl:\" followed by a digit)", name
);
376 svec_add(&new_br
, name
);
379 svec_destroy(&raw_new_br
);
381 /* Get rid of deleted bridges and add new bridges. */
384 assert(svec_is_unique(&old_br
));
385 assert(svec_is_unique(&new_br
));
386 LIST_FOR_EACH_SAFE (br
, next
, struct bridge
, node
, &all_bridges
) {
387 if (!svec_contains(&new_br
, br
->name
)) {
391 for (i
= 0; i
< new_br
.n
; i
++) {
392 const char *name
= new_br
.names
[i
];
393 if (!svec_contains(&old_br
, name
)) {
397 svec_destroy(&old_br
);
398 svec_destroy(&new_br
);
402 bridge_configure_ssl();
405 /* Reconfigure all bridges. */
406 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
407 bridge_reconfigure_one(br
);
410 /* Add and delete ports on all datapaths.
412 * The kernel will reject any attempt to add a given port to a datapath if
413 * that port already belongs to a different datapath, so we must do all
414 * port deletions before any port additions. */
415 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
416 struct odp_port
*dpif_ports
;
418 struct svec want_ifaces
;
420 dpif_port_list(&br
->dpif
, &dpif_ports
, &n_dpif_ports
);
421 bridge_get_all_ifaces(br
, &want_ifaces
);
422 for (i
= 0; i
< n_dpif_ports
; i
++) {
423 const struct odp_port
*p
= &dpif_ports
[i
];
424 if (!svec_contains(&want_ifaces
, p
->devname
)
425 && strcmp(p
->devname
, br
->name
)) {
426 int retval
= dpif_port_del(&br
->dpif
, p
->port
);
428 VLOG_ERR("failed to remove %s interface from dp%u: %s",
429 p
->devname
, dpif_id(&br
->dpif
), strerror(retval
));
433 svec_destroy(&want_ifaces
);
436 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
437 struct odp_port
*dpif_ports
;
439 struct svec cur_ifaces
, want_ifaces
, add_ifaces
;
442 dpif_port_list(&br
->dpif
, &dpif_ports
, &n_dpif_ports
);
443 svec_init(&cur_ifaces
);
444 for (i
= 0; i
< n_dpif_ports
; i
++) {
445 svec_add(&cur_ifaces
, dpif_ports
[i
].devname
);
448 svec_sort_unique(&cur_ifaces
);
449 bridge_get_all_ifaces(br
, &want_ifaces
);
450 svec_diff(&want_ifaces
, &cur_ifaces
, &add_ifaces
, NULL
, NULL
);
453 for (i
= 0; i
< add_ifaces
.n
; i
++) {
454 const char *if_name
= add_ifaces
.names
[i
];
456 int internal
= cfg_get_bool(0, "iface.%s.internal", if_name
);
457 int error
= dpif_port_add(&br
->dpif
, if_name
, next_port_no
++,
458 internal
? ODP_PORT_INTERNAL
: 0);
459 if (error
!= EEXIST
) {
460 if (next_port_no
>= 256) {
461 VLOG_ERR("ran out of valid port numbers on dp%u",
466 VLOG_ERR("failed to add %s interface to dp%u: %s",
467 if_name
, dpif_id(&br
->dpif
), strerror(error
));
474 svec_destroy(&cur_ifaces
);
475 svec_destroy(&want_ifaces
);
476 svec_destroy(&add_ifaces
);
478 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
481 struct iface
*local_iface
= NULL
;
483 uint8_t engine_type
= br
->dpif
.minor
;
484 uint8_t engine_id
= br
->dpif
.minor
;
485 bool add_id_to_iface
= false;
486 struct svec nf_hosts
;
488 bridge_fetch_dp_ifaces(br
);
489 for (i
= 0; i
< br
->n_ports
; ) {
490 struct port
*port
= br
->ports
[i
];
492 for (j
= 0; j
< port
->n_ifaces
; ) {
493 struct iface
*iface
= port
->ifaces
[j
];
494 if (iface
->dp_ifidx
< 0) {
495 VLOG_ERR("%s interface not in dp%u, dropping",
496 iface
->name
, dpif_id(&br
->dpif
));
497 iface_destroy(iface
);
499 if (iface
->dp_ifidx
== ODPP_LOCAL
) {
502 VLOG_DBG("dp%u has interface %s on port %d",
503 dpif_id(&br
->dpif
), iface
->name
, iface
->dp_ifidx
);
507 if (!port
->n_ifaces
) {
508 VLOG_ERR("%s port has no interfaces, dropping", port
->name
);
515 /* Pick local port hardware address, datapath ID. */
516 bridge_pick_local_hw_addr(br
, ea
, &devname
);
518 int error
= netdev_nodev_set_etheraddr(local_iface
->name
, ea
);
520 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
521 VLOG_ERR_RL(&rl
, "bridge %s: failed to set bridge "
522 "Ethernet address: %s",
523 br
->name
, strerror(error
));
527 dpid
= bridge_pick_datapath_id(br
, ea
, devname
);
528 ofproto_set_datapath_id(br
->ofproto
, dpid
);
530 /* Set NetFlow configuration on this bridge. */
531 if (cfg_has("netflow.%s.engine-type", br
->name
)) {
532 engine_type
= cfg_get_int(0, "netflow.%s.engine-type",
535 if (cfg_has("netflow.%s.engine-id", br
->name
)) {
536 engine_id
= cfg_get_int(0, "netflow.%s.engine-id", br
->name
);
538 if (cfg_has("netflow.%s.add-id-to-iface", br
->name
)) {
539 add_id_to_iface
= cfg_get_bool(0, "netflow.%s.add-id-to-iface",
542 if (add_id_to_iface
&& engine_id
> 0x7f) {
543 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
544 "another vswitch, choose an engine id less than 128",
547 if (add_id_to_iface
&& br
->n_ports
> 0x1ff) {
548 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
549 "another port when 512 or more ports are used",
552 svec_init(&nf_hosts
);
553 cfg_get_all_keys(&nf_hosts
, "netflow.%s.host", br
->name
);
554 if (ofproto_set_netflow(br
->ofproto
, &nf_hosts
, engine_type
,
555 engine_id
, add_id_to_iface
)) {
556 VLOG_ERR("bridge %s: problem setting netflow collectors",
560 /* Update the controller and related settings. It would be more
561 * straightforward to call this from bridge_reconfigure_one(), but we
562 * can't do it there for two reasons. First, and most importantly, at
563 * that point we don't know the dp_ifidx of any interfaces that have
564 * been added to the bridge (because we haven't actually added them to
565 * the datapath). Second, at that point we haven't set the datapath ID
566 * yet; when a controller is configured, resetting the datapath ID will
567 * immediately disconnect from the controller, so it's better to set
568 * the datapath ID before the controller. */
569 bridge_reconfigure_controller(br
);
571 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
572 for (i
= 0; i
< br
->n_ports
; i
++) {
573 struct port
*port
= br
->ports
[i
];
574 port_update_vlan_compat(port
);
577 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
578 brstp_reconfigure(br
);
583 bridge_pick_local_hw_addr(struct bridge
*br
, uint8_t ea
[ETH_ADDR_LEN
],
584 const char **devname
)
586 uint64_t requested_ea
;
592 /* Did the user request a particular MAC? */
593 requested_ea
= cfg_get_mac(0, "bridge.%s.mac", br
->name
);
595 eth_addr_from_uint64(requested_ea
, ea
);
596 if (eth_addr_is_multicast(ea
)) {
597 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
598 "address "ETH_ADDR_FMT
, br
->name
, ETH_ADDR_ARGS(ea
));
599 } else if (eth_addr_is_zero(ea
)) {
600 VLOG_ERR("bridge %s: cannot set MAC address to zero", br
->name
);
606 /* Otherwise choose the minimum MAC address among all of the interfaces.
607 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
608 * MAC of the physical interface in such an environment.) */
609 memset(ea
, 0xff, sizeof ea
);
610 for (i
= 0; i
< br
->n_ports
; i
++) {
611 struct port
*port
= br
->ports
[i
];
612 if (port
->is_mirror_output_port
) {
615 for (j
= 0; j
< port
->n_ifaces
; j
++) {
616 struct iface
*iface
= port
->ifaces
[j
];
617 uint8_t iface_ea
[ETH_ADDR_LEN
];
618 if (iface
->dp_ifidx
== ODPP_LOCAL
619 || cfg_get_bool(0, "iface.%s.internal", iface
->name
)) {
622 error
= netdev_nodev_get_etheraddr(iface
->name
, iface_ea
);
624 if (!eth_addr_is_multicast(iface_ea
) &&
625 !eth_addr_is_reserved(iface_ea
) &&
626 !eth_addr_is_zero(iface_ea
) &&
627 memcmp(iface_ea
, ea
, ETH_ADDR_LEN
) < 0) {
628 memcpy(ea
, iface_ea
, ETH_ADDR_LEN
);
629 *devname
= iface
->name
;
632 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
633 VLOG_ERR_RL(&rl
, "failed to obtain Ethernet address of %s: %s",
634 iface
->name
, strerror(error
));
638 if (eth_addr_is_multicast(ea
) || eth_addr_is_vif(ea
)) {
639 memcpy(ea
, br
->default_ea
, ETH_ADDR_LEN
);
641 VLOG_WARN("bridge %s: using default bridge Ethernet "
642 "address "ETH_ADDR_FMT
, br
->name
, ETH_ADDR_ARGS(ea
));
644 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT
,
645 br
->name
, ETH_ADDR_ARGS(ea
));
649 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
650 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
651 * a network device, then that network device's name must be passed in as
652 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
653 * passed in as a null pointer. */
655 bridge_pick_datapath_id(struct bridge
*br
,
656 const uint8_t bridge_ea
[ETH_ADDR_LEN
],
660 * The procedure for choosing a bridge MAC address will, in the most
661 * ordinary case, also choose a unique MAC that we can use as a datapath
662 * ID. In some special cases, though, multiple bridges will end up with
663 * the same MAC address. This is OK for the bridges, but it will confuse
664 * the OpenFlow controller, because each datapath needs a unique datapath
667 * Datapath IDs must be unique. It is also very desirable that they be
668 * stable from one run to the next, so that policy set on a datapath
673 dpid
= cfg_get_dpid(0, "bridge.%s.datapath-id", br
->name
);
680 if (!netdev_get_vlan_vid(devname
, &vlan
)) {
682 * A bridge whose MAC address is taken from a VLAN network device
683 * (that is, a network device created with vconfig(8) or similar
684 * tool) will have the same MAC address as a bridge on the VLAN
685 * device's physical network device.
687 * Handle this case by hashing the physical network device MAC
688 * along with the VLAN identifier.
690 uint8_t buf
[ETH_ADDR_LEN
+ 2];
691 memcpy(buf
, bridge_ea
, ETH_ADDR_LEN
);
692 buf
[ETH_ADDR_LEN
] = vlan
>> 8;
693 buf
[ETH_ADDR_LEN
+ 1] = vlan
;
694 return dpid_from_hash(buf
, sizeof buf
);
697 * Assume that this bridge's MAC address is unique, since it
698 * doesn't fit any of the cases we handle specially.
703 * A purely internal bridge, that is, one that has no non-virtual
704 * network devices on it at all, is more difficult because it has no
705 * natural unique identifier at all.
707 * When the host is a XenServer, we handle this case by hashing the
708 * host's UUID with the name of the bridge. Names of bridges are
709 * persistent across XenServer reboots, although they can be reused if
710 * an internal network is destroyed and then a new one is later
711 * created, so this is fairly effective.
713 * When the host is not a XenServer, we punt by using a random MAC
714 * address on each run.
716 const char *host_uuid
= xenserver_get_host_uuid();
718 char *combined
= xasprintf("%s,%s", host_uuid
, br
->name
);
719 dpid
= dpid_from_hash(combined
, strlen(combined
));
725 return eth_addr_to_uint64(bridge_ea
);
729 dpid_from_hash(const void *data
, size_t n
)
731 uint8_t hash
[SHA1_DIGEST_SIZE
];
733 BUILD_ASSERT_DECL(sizeof hash
>= ETH_ADDR_LEN
);
734 sha1_bytes(data
, n
, hash
);
735 eth_addr_mark_random(hash
);
736 return eth_addr_to_uint64(hash
);
742 struct bridge
*br
, *next
;
746 LIST_FOR_EACH_SAFE (br
, next
, struct bridge
, node
, &all_bridges
) {
747 int error
= bridge_run_one(br
);
749 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
750 VLOG_ERR_RL(&rl
, "bridge %s: datapath was destroyed externally, "
751 "forcing reconfiguration", br
->name
);
765 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
766 ofproto_wait(br
->ofproto
);
767 if (br
->controller
) {
772 mac_learning_wait(br
->ml
);
779 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
780 * configuration changes. */
782 bridge_flush(struct bridge
*br
)
784 COVERAGE_INC(bridge_flush
);
787 mac_learning_flush(br
->ml
);
791 /* Bridge reconfiguration functions. */
793 static struct bridge
*
794 bridge_create(const char *name
)
799 assert(!bridge_lookup(name
));
800 br
= xcalloc(1, sizeof *br
);
802 error
= dpif_create(name
, &br
->dpif
);
803 if (error
== EEXIST
) {
804 error
= dpif_open(name
, &br
->dpif
);
806 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
807 name
, strerror(error
));
811 dpif_flow_flush(&br
->dpif
);
813 VLOG_ERR("failed to create datapath %s: %s", name
, strerror(error
));
818 error
= ofproto_create(name
, &bridge_ofhooks
, br
, &br
->ofproto
);
820 VLOG_ERR("failed to create switch %s: %s", name
, strerror(error
));
821 dpif_delete(&br
->dpif
);
822 dpif_close(&br
->dpif
);
827 br
->name
= xstrdup(name
);
828 br
->ml
= mac_learning_create();
829 br
->sent_config_request
= false;
830 eth_addr_random(br
->default_ea
);
832 port_array_init(&br
->ifaces
);
835 br
->bond_next_rebalance
= time_msec() + 10000;
837 list_push_back(&all_bridges
, &br
->node
);
839 VLOG_INFO("created bridge %s on dp%u", br
->name
, dpif_id(&br
->dpif
));
845 bridge_destroy(struct bridge
*br
)
850 while (br
->n_ports
> 0) {
851 port_destroy(br
->ports
[br
->n_ports
- 1]);
853 list_remove(&br
->node
);
854 error
= dpif_delete(&br
->dpif
);
855 if (error
&& error
!= ENOENT
) {
856 VLOG_ERR("failed to delete dp%u: %s",
857 dpif_id(&br
->dpif
), strerror(error
));
859 dpif_close(&br
->dpif
);
860 ofproto_destroy(br
->ofproto
);
861 free(br
->controller
);
862 mac_learning_destroy(br
->ml
);
863 port_array_destroy(&br
->ifaces
);
870 static struct bridge
*
871 bridge_lookup(const char *name
)
875 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
876 if (!strcmp(br
->name
, name
)) {
884 bridge_exists(const char *name
)
886 return bridge_lookup(name
) ? true : false;
890 bridge_get_datapathid(const char *name
)
892 struct bridge
*br
= bridge_lookup(name
);
893 return br
? ofproto_get_datapath_id(br
->ofproto
) : 0;
897 bridge_run_one(struct bridge
*br
)
901 error
= ofproto_run1(br
->ofproto
);
907 mac_learning_run(br
->ml
, ofproto_get_revalidate_set(br
->ofproto
));
912 error
= ofproto_run2(br
->ofproto
, br
->flush
);
919 bridge_get_controller(const struct bridge
*br
)
921 const char *controller
;
923 controller
= cfg_get_string(0, "bridge.%s.controller", br
->name
);
925 controller
= cfg_get_string(0, "mgmt.controller");
927 return controller
&& controller
[0] ? controller
: NULL
;
931 bridge_reconfigure_one(struct bridge
*br
)
933 struct svec old_ports
, new_ports
, ifaces
;
934 struct svec listeners
, old_listeners
;
935 struct svec snoops
, old_snoops
;
938 /* Collect old ports. */
939 svec_init(&old_ports
);
940 for (i
= 0; i
< br
->n_ports
; i
++) {
941 svec_add(&old_ports
, br
->ports
[i
]->name
);
943 svec_sort(&old_ports
);
944 assert(svec_is_unique(&old_ports
));
946 /* Collect new ports. */
947 svec_init(&new_ports
);
948 cfg_get_all_keys(&new_ports
, "bridge.%s.port", br
->name
);
949 svec_sort(&new_ports
);
950 if (bridge_get_controller(br
) && !svec_contains(&new_ports
, br
->name
)) {
951 svec_add(&new_ports
, br
->name
);
952 svec_sort(&new_ports
);
954 if (!svec_is_unique(&new_ports
)) {
955 VLOG_WARN("bridge %s: %s specified twice as bridge port",
956 br
->name
, svec_get_duplicate(&new_ports
));
957 svec_unique(&new_ports
);
960 ofproto_set_mgmt_id(br
->ofproto
, mgmt_id
);
962 /* Get rid of deleted ports and add new ports. */
963 for (i
= 0; i
< br
->n_ports
; ) {
964 struct port
*port
= br
->ports
[i
];
965 if (!svec_contains(&new_ports
, port
->name
)) {
971 for (i
= 0; i
< new_ports
.n
; i
++) {
972 const char *name
= new_ports
.names
[i
];
973 if (!svec_contains(&old_ports
, name
)) {
974 port_create(br
, name
);
977 svec_destroy(&old_ports
);
978 svec_destroy(&new_ports
);
980 /* Reconfigure all ports. */
981 for (i
= 0; i
< br
->n_ports
; i
++) {
982 port_reconfigure(br
->ports
[i
]);
985 /* Check and delete duplicate interfaces. */
987 for (i
= 0; i
< br
->n_ports
; ) {
988 struct port
*port
= br
->ports
[i
];
989 for (j
= 0; j
< port
->n_ifaces
; ) {
990 struct iface
*iface
= port
->ifaces
[j
];
991 if (svec_contains(&ifaces
, iface
->name
)) {
992 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
994 br
->name
, iface
->name
, port
->name
);
995 iface_destroy(iface
);
997 svec_add(&ifaces
, iface
->name
);
1002 if (!port
->n_ifaces
) {
1003 VLOG_ERR("%s port has no interfaces, dropping", port
->name
);
1009 svec_destroy(&ifaces
);
1011 /* Delete all flows if we're switching from connected to standalone or vice
1012 * versa. (XXX Should we delete all flows if we are switching from one
1013 * controller to another?) */
1015 /* Configure OpenFlow management listeners. */
1016 svec_init(&listeners
);
1017 cfg_get_all_strings(&listeners
, "bridge.%s.openflow.listeners", br
->name
);
1019 svec_add_nocopy(&listeners
, xasprintf("punix:%s/%s.mgmt",
1020 ovs_rundir
, br
->name
));
1021 } else if (listeners
.n
== 1 && !strcmp(listeners
.names
[0], "none")) {
1022 svec_clear(&listeners
);
1024 svec_sort_unique(&listeners
);
1026 svec_init(&old_listeners
);
1027 ofproto_get_listeners(br
->ofproto
, &old_listeners
);
1028 svec_sort_unique(&old_listeners
);
1030 if (!svec_equal(&listeners
, &old_listeners
)) {
1031 ofproto_set_listeners(br
->ofproto
, &listeners
);
1033 svec_destroy(&listeners
);
1034 svec_destroy(&old_listeners
);
1036 /* Configure OpenFlow controller connection snooping. */
1038 cfg_get_all_strings(&snoops
, "bridge.%s.openflow.snoops", br
->name
);
1040 svec_add_nocopy(&snoops
, xasprintf("punix:%s/%s.snoop",
1041 ovs_rundir
, br
->name
));
1042 } else if (snoops
.n
== 1 && !strcmp(snoops
.names
[0], "none")) {
1043 svec_clear(&snoops
);
1045 svec_sort_unique(&snoops
);
1047 svec_init(&old_snoops
);
1048 ofproto_get_snoops(br
->ofproto
, &old_snoops
);
1049 svec_sort_unique(&old_snoops
);
1051 if (!svec_equal(&snoops
, &old_snoops
)) {
1052 ofproto_set_snoops(br
->ofproto
, &snoops
);
1054 svec_destroy(&snoops
);
1055 svec_destroy(&old_snoops
);
1057 mirror_reconfigure(br
);
1061 bridge_reconfigure_controller(struct bridge
*br
)
1063 char *pfx
= xasprintf("bridge.%s.controller", br
->name
);
1064 const char *controller
;
1066 controller
= bridge_get_controller(br
);
1067 if ((br
->controller
!= NULL
) != (controller
!= NULL
)) {
1068 ofproto_flush_flows(br
->ofproto
);
1070 free(br
->controller
);
1071 br
->controller
= controller
? xstrdup(controller
) : NULL
;
1074 const char *fail_mode
;
1075 int max_backoff
, probe
;
1076 int rate_limit
, burst_limit
;
1078 if (!strcmp(controller
, "discover")) {
1079 bool update_resolv_conf
= true;
1081 if (cfg_has("%s.update-resolv.conf", pfx
)) {
1082 update_resolv_conf
= cfg_get_bool(0, "%s.update-resolv.conf",
1085 ofproto_set_discovery(br
->ofproto
, true,
1086 cfg_get_string(0, "%s.accept-regex", pfx
),
1087 update_resolv_conf
);
1089 struct netdev
*netdev
;
1093 in_band
= (!cfg_is_valid(CFG_BOOL
| CFG_REQUIRED
,
1095 || cfg_get_bool(0, "%s.in-band", pfx
));
1096 ofproto_set_discovery(br
->ofproto
, false, NULL
, NULL
);
1097 ofproto_set_in_band(br
->ofproto
, in_band
);
1099 error
= netdev_open(br
->name
, NETDEV_ETH_TYPE_NONE
, &netdev
);
1101 if (cfg_is_valid(CFG_IP
| CFG_REQUIRED
, "%s.ip", pfx
)) {
1102 struct in_addr ip
, mask
, gateway
;
1103 ip
.s_addr
= cfg_get_ip(0, "%s.ip", pfx
);
1104 mask
.s_addr
= cfg_get_ip(0, "%s.netmask", pfx
);
1105 gateway
.s_addr
= cfg_get_ip(0, "%s.gateway", pfx
);
1107 netdev_turn_flags_on(netdev
, NETDEV_UP
, true);
1109 mask
.s_addr
= guess_netmask(ip
.s_addr
);
1111 if (!netdev_set_in4(netdev
, ip
, mask
)) {
1112 VLOG_INFO("bridge %s: configured IP address "IP_FMT
", "
1114 br
->name
, IP_ARGS(&ip
.s_addr
),
1115 IP_ARGS(&mask
.s_addr
));
1118 if (gateway
.s_addr
) {
1119 if (!netdev_add_router(gateway
)) {
1120 VLOG_INFO("bridge %s: configured gateway "IP_FMT
,
1121 br
->name
, IP_ARGS(&gateway
.s_addr
));
1125 netdev_close(netdev
);
1129 fail_mode
= cfg_get_string(0, "%s.fail-mode", pfx
);
1131 fail_mode
= cfg_get_string(0, "mgmt.fail-mode");
1133 ofproto_set_failure(br
->ofproto
,
1135 || !strcmp(fail_mode
, "standalone")
1136 || !strcmp(fail_mode
, "open")));
1138 probe
= cfg_get_int(0, "%s.inactivity-probe", pfx
);
1140 probe
= cfg_get_int(0, "mgmt.inactivity-probe");
1145 ofproto_set_probe_interval(br
->ofproto
, probe
);
1147 max_backoff
= cfg_get_int(0, "%s.max-backoff", pfx
);
1149 max_backoff
= cfg_get_int(0, "mgmt.max-backoff");
1154 ofproto_set_max_backoff(br
->ofproto
, max_backoff
);
1156 rate_limit
= cfg_get_int(0, "%s.rate-limit", pfx
);
1158 rate_limit
= cfg_get_int(0, "mgmt.rate-limit");
1160 burst_limit
= cfg_get_int(0, "%s.burst-limit", pfx
);
1162 burst_limit
= cfg_get_int(0, "mgmt.burst-limit");
1164 ofproto_set_rate_limit(br
->ofproto
, rate_limit
, burst_limit
);
1166 ofproto_set_stp(br
->ofproto
, cfg_get_bool(0, "%s.stp", pfx
));
1168 if (cfg_has("%s.commands.acl", pfx
)) {
1169 struct svec command_acls
;
1172 svec_init(&command_acls
);
1173 cfg_get_all_strings(&command_acls
, "%s.commands.acl", pfx
);
1174 command_acl
= svec_join(&command_acls
, ",", "");
1176 ofproto_set_remote_execution(br
->ofproto
, command_acl
,
1177 cfg_get_string(0, "%s.commands.dir",
1180 svec_destroy(&command_acls
);
1183 ofproto_set_remote_execution(br
->ofproto
, NULL
, NULL
);
1186 union ofp_action action
;
1189 /* Set up a flow that matches every packet and directs them to
1190 * OFPP_NORMAL (which goes to us). */
1191 memset(&action
, 0, sizeof action
);
1192 action
.type
= htons(OFPAT_OUTPUT
);
1193 action
.output
.len
= htons(sizeof action
);
1194 action
.output
.port
= htons(OFPP_NORMAL
);
1195 memset(&flow
, 0, sizeof flow
);
1196 ofproto_add_flow(br
->ofproto
, &flow
, OFPFW_ALL
, 0,
1199 ofproto_set_in_band(br
->ofproto
, false);
1200 ofproto_set_max_backoff(br
->ofproto
, 1);
1201 ofproto_set_probe_interval(br
->ofproto
, 5);
1202 ofproto_set_failure(br
->ofproto
, false);
1203 ofproto_set_stp(br
->ofproto
, false);
1207 ofproto_set_controller(br
->ofproto
, br
->controller
);
1211 bridge_get_all_ifaces(const struct bridge
*br
, struct svec
*ifaces
)
1216 for (i
= 0; i
< br
->n_ports
; i
++) {
1217 struct port
*port
= br
->ports
[i
];
1218 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1219 struct iface
*iface
= port
->ifaces
[j
];
1220 svec_add(ifaces
, iface
->name
);
1224 assert(svec_is_unique(ifaces
));
1227 /* For robustness, in case the administrator moves around datapath ports behind
1228 * our back, we re-check all the datapath port numbers here.
1230 * This function will set the 'dp_ifidx' members of interfaces that have
1231 * disappeared to -1, so only call this function from a context where those
1232 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1233 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1234 * datapath, which doesn't support UINT16_MAX+1 ports. */
1236 bridge_fetch_dp_ifaces(struct bridge
*br
)
1238 struct odp_port
*dpif_ports
;
1239 size_t n_dpif_ports
;
1242 /* Reset all interface numbers. */
1243 for (i
= 0; i
< br
->n_ports
; i
++) {
1244 struct port
*port
= br
->ports
[i
];
1245 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1246 struct iface
*iface
= port
->ifaces
[j
];
1247 iface
->dp_ifidx
= -1;
1250 port_array_clear(&br
->ifaces
);
1252 dpif_port_list(&br
->dpif
, &dpif_ports
, &n_dpif_ports
);
1253 for (i
= 0; i
< n_dpif_ports
; i
++) {
1254 struct odp_port
*p
= &dpif_ports
[i
];
1255 struct iface
*iface
= iface_lookup(br
, p
->devname
);
1257 if (iface
->dp_ifidx
>= 0) {
1258 VLOG_WARN("dp%u reported interface %s twice",
1259 dpif_id(&br
->dpif
), p
->devname
);
1260 } else if (iface_from_dp_ifidx(br
, p
->port
)) {
1261 VLOG_WARN("dp%u reported interface %"PRIu16
" twice",
1262 dpif_id(&br
->dpif
), p
->port
);
1264 port_array_set(&br
->ifaces
, p
->port
, iface
);
1265 iface
->dp_ifidx
= p
->port
;
1272 /* Bridge packet processing functions. */
1275 bond_hash(const uint8_t mac
[ETH_ADDR_LEN
])
1277 return hash_bytes(mac
, ETH_ADDR_LEN
, 0) & BOND_MASK
;
1280 static struct bond_entry
*
1281 lookup_bond_entry(const struct port
*port
, const uint8_t mac
[ETH_ADDR_LEN
])
1283 return &port
->bond_hash
[bond_hash(mac
)];
1287 bond_choose_iface(const struct port
*port
)
1290 for (i
= 0; i
< port
->n_ifaces
; i
++) {
1291 if (port
->ifaces
[i
]->enabled
) {
1299 choose_output_iface(const struct port
*port
, const uint8_t *dl_src
,
1300 uint16_t *dp_ifidx
, tag_type
*tags
)
1302 struct iface
*iface
;
1304 assert(port
->n_ifaces
);
1305 if (port
->n_ifaces
== 1) {
1306 iface
= port
->ifaces
[0];
1308 struct bond_entry
*e
= lookup_bond_entry(port
, dl_src
);
1309 if (e
->iface_idx
< 0 || e
->iface_idx
>= port
->n_ifaces
1310 || !port
->ifaces
[e
->iface_idx
]->enabled
) {
1311 /* XXX select interface properly. The current interface selection
1312 * is only good for testing the rebalancing code. */
1313 e
->iface_idx
= bond_choose_iface(port
);
1314 if (e
->iface_idx
< 0) {
1315 *tags
|= port
->no_ifaces_tag
;
1318 e
->iface_tag
= tag_create_random();
1320 *tags
|= e
->iface_tag
;
1321 iface
= port
->ifaces
[e
->iface_idx
];
1323 *dp_ifidx
= iface
->dp_ifidx
;
1324 *tags
|= iface
->tag
; /* Currently only used for bonding. */
1329 bond_link_status_update(struct iface
*iface
, bool carrier
)
1331 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1332 struct port
*port
= iface
->port
;
1334 if ((carrier
== iface
->enabled
) == (iface
->delay_expires
== LLONG_MAX
)) {
1335 /* Nothing to do. */
1338 VLOG_INFO_RL(&rl
, "interface %s: carrier %s",
1339 iface
->name
, carrier
? "detected" : "dropped");
1340 if (carrier
== iface
->enabled
) {
1341 iface
->delay_expires
= LLONG_MAX
;
1342 VLOG_INFO_RL(&rl
, "interface %s: will not be %s",
1343 iface
->name
, carrier
? "disabled" : "enabled");
1345 int delay
= carrier
? port
->updelay
: port
->downdelay
;
1346 iface
->delay_expires
= time_msec() + delay
;
1349 "interface %s: will be %s if it stays %s for %d ms",
1351 carrier
? "enabled" : "disabled",
1352 carrier
? "up" : "down",
1359 bond_choose_active_iface(struct port
*port
)
1361 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(5, 20);
1363 port
->active_iface
= bond_choose_iface(port
);
1364 port
->active_iface_tag
= tag_create_random();
1365 if (port
->active_iface
>= 0) {
1366 VLOG_INFO_RL(&rl
, "port %s: active interface is now %s",
1367 port
->name
, port
->ifaces
[port
->active_iface
]->name
);
1369 VLOG_WARN_RL(&rl
, "port %s: all ports disabled, no active interface",
1375 bond_enable_slave(struct iface
*iface
, bool enable
)
1377 struct port
*port
= iface
->port
;
1378 struct bridge
*br
= port
->bridge
;
1380 iface
->delay_expires
= LLONG_MAX
;
1381 if (enable
== iface
->enabled
) {
1385 iface
->enabled
= enable
;
1386 if (!iface
->enabled
) {
1387 VLOG_WARN("interface %s: disabled", iface
->name
);
1388 ofproto_revalidate(br
->ofproto
, iface
->tag
);
1389 if (iface
->port_ifidx
== port
->active_iface
) {
1390 ofproto_revalidate(br
->ofproto
,
1391 port
->active_iface_tag
);
1392 bond_choose_active_iface(port
);
1394 bond_send_learning_packets(port
);
1396 VLOG_WARN("interface %s: enabled", iface
->name
);
1397 if (port
->active_iface
< 0) {
1398 ofproto_revalidate(br
->ofproto
, port
->no_ifaces_tag
);
1399 bond_choose_active_iface(port
);
1400 bond_send_learning_packets(port
);
1402 iface
->tag
= tag_create_random();
1407 bond_run(struct bridge
*br
)
1411 for (i
= 0; i
< br
->n_ports
; i
++) {
1412 struct port
*port
= br
->ports
[i
];
1413 if (port
->n_ifaces
< 2) {
1416 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1417 struct iface
*iface
= port
->ifaces
[j
];
1418 if (time_msec() >= iface
->delay_expires
) {
1419 bond_enable_slave(iface
, !iface
->enabled
);
1426 bond_wait(struct bridge
*br
)
1430 for (i
= 0; i
< br
->n_ports
; i
++) {
1431 struct port
*port
= br
->ports
[i
];
1432 if (port
->n_ifaces
< 2) {
1435 for (j
= 0; j
< port
->n_ifaces
; j
++) {
1436 struct iface
*iface
= port
->ifaces
[j
];
1437 if (iface
->delay_expires
!= LLONG_MAX
) {
1438 poll_timer_wait(iface
->delay_expires
- time_msec());
1445 set_dst(struct dst
*p
, const flow_t
*flow
,
1446 const struct port
*in_port
, const struct port
*out_port
,
1451 * XXX This uses too many tags: any broadcast flow will get one tag per
1452 * destination port, and thus a broadcast on a switch of any size is likely
1453 * to have all tag bits set. We should figure out a way to be smarter.
1455 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1456 *tags
|= out_port
->stp_state_tag
;
1457 if (!(out_port
->stp_state
& (STP_DISABLED
| STP_FORWARDING
))) {
1461 p
->vlan
= (out_port
->vlan
>= 0 ? OFP_VLAN_NONE
1462 : in_port
->vlan
>= 0 ? in_port
->vlan
1463 : ntohs(flow
->dl_vlan
));
1464 return choose_output_iface(out_port
, flow
->dl_src
, &p
->dp_ifidx
, tags
);
1468 swap_dst(struct dst
*p
, struct dst
*q
)
1470 struct dst tmp
= *p
;
1475 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1476 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1477 * that we push to the datapath. We could in fact fully sort the array by
1478 * vlan, but in most cases there are at most two different vlan tags so that's
1479 * possibly overkill.) */
1481 partition_dsts(struct dst
*dsts
, size_t n_dsts
, int vlan
)
1483 struct dst
*first
= dsts
;
1484 struct dst
*last
= dsts
+ n_dsts
;
1486 while (first
!= last
) {
1488 * - All dsts < first have vlan == 'vlan'.
1489 * - All dsts >= last have vlan != 'vlan'.
1490 * - first < last. */
1491 while (first
->vlan
== vlan
) {
1492 if (++first
== last
) {
1497 /* Same invariants, plus one additional:
1498 * - first->vlan != vlan.
1500 while (last
[-1].vlan
!= vlan
) {
1501 if (--last
== first
) {
1506 /* Same invariants, plus one additional:
1507 * - last[-1].vlan == vlan.*/
1508 swap_dst(first
++, --last
);
1513 mirror_mask_ffs(mirror_mask_t mask
)
1515 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask
));
1520 dst_is_duplicate(const struct dst
*dsts
, size_t n_dsts
,
1521 const struct dst
*test
)
1524 for (i
= 0; i
< n_dsts
; i
++) {
1525 if (dsts
[i
].vlan
== test
->vlan
&& dsts
[i
].dp_ifidx
== test
->dp_ifidx
) {
1533 port_trunks_vlan(const struct port
*port
, uint16_t vlan
)
1535 return port
->vlan
< 0 && bitmap_is_set(port
->trunks
, vlan
);
1539 port_includes_vlan(const struct port
*port
, uint16_t vlan
)
1541 return vlan
== port
->vlan
|| port_trunks_vlan(port
, vlan
);
1545 compose_dsts(const struct bridge
*br
, const flow_t
*flow
, uint16_t vlan
,
1546 const struct port
*in_port
, const struct port
*out_port
,
1547 struct dst dsts
[], tag_type
*tags
)
1549 mirror_mask_t mirrors
= in_port
->src_mirrors
;
1550 struct dst
*dst
= dsts
;
1553 *tags
|= in_port
->stp_state_tag
;
1554 if (out_port
== FLOOD_PORT
) {
1555 /* XXX use ODP_FLOOD if no vlans or bonding. */
1556 /* XXX even better, define each VLAN as a datapath port group */
1557 for (i
= 0; i
< br
->n_ports
; i
++) {
1558 struct port
*port
= br
->ports
[i
];
1559 if (port
!= in_port
&& port_includes_vlan(port
, vlan
)
1560 && !port
->is_mirror_output_port
1561 && set_dst(dst
, flow
, in_port
, port
, tags
)) {
1562 mirrors
|= port
->dst_mirrors
;
1566 } else if (out_port
&& set_dst(dst
, flow
, in_port
, out_port
, tags
)) {
1567 mirrors
|= out_port
->dst_mirrors
;
1572 struct mirror
*m
= br
->mirrors
[mirror_mask_ffs(mirrors
) - 1];
1573 if (!m
->n_vlans
|| vlan_is_mirrored(m
, vlan
)) {
1575 if (set_dst(dst
, flow
, in_port
, m
->out_port
, tags
)
1576 && !dst_is_duplicate(dsts
, dst
- dsts
, dst
)) {
1580 for (i
= 0; i
< br
->n_ports
; i
++) {
1581 struct port
*port
= br
->ports
[i
];
1582 if (port_includes_vlan(port
, m
->out_vlan
)
1583 && set_dst(dst
, flow
, in_port
, port
, tags
)
1584 && !dst_is_duplicate(dsts
, dst
- dsts
, dst
))
1586 if (port
->vlan
< 0) {
1587 dst
->vlan
= m
->out_vlan
;
1589 if (dst
->dp_ifidx
== flow
->in_port
1590 && dst
->vlan
== vlan
) {
1591 /* Don't send out input port on same VLAN. */
1599 mirrors
&= mirrors
- 1;
1602 partition_dsts(dsts
, dst
- dsts
, ntohs(flow
->dl_vlan
));
1607 print_dsts(const struct dst
*dsts
, size_t n
)
1609 for (; n
--; dsts
++) {
1610 printf(">p%"PRIu16
, dsts
->dp_ifidx
);
1611 if (dsts
->vlan
!= OFP_VLAN_NONE
) {
1612 printf("v%"PRIu16
, dsts
->vlan
);
1618 compose_actions(struct bridge
*br
, const flow_t
*flow
, uint16_t vlan
,
1619 const struct port
*in_port
, const struct port
*out_port
,
1620 tag_type
*tags
, struct odp_actions
*actions
)
1622 struct dst dsts
[DP_MAX_PORTS
* (MAX_MIRRORS
+ 1)];
1624 const struct dst
*p
;
1627 n_dsts
= compose_dsts(br
, flow
, vlan
, in_port
, out_port
, dsts
, tags
);
1629 cur_vlan
= ntohs(flow
->dl_vlan
);
1630 for (p
= dsts
; p
< &dsts
[n_dsts
]; p
++) {
1631 union odp_action
*a
;
1632 if (p
->vlan
!= cur_vlan
) {
1633 if (p
->vlan
== OFP_VLAN_NONE
) {
1634 odp_actions_add(actions
, ODPAT_STRIP_VLAN
);
1636 a
= odp_actions_add(actions
, ODPAT_SET_VLAN_VID
);
1637 a
->vlan_vid
.vlan_vid
= htons(p
->vlan
);
1641 a
= odp_actions_add(actions
, ODPAT_OUTPUT
);
1642 a
->output
.port
= p
->dp_ifidx
;
1647 is_bcast_arp_reply(const flow_t
*flow
, const struct ofpbuf
*packet
)
1649 struct arp_eth_header
*arp
= (struct arp_eth_header
*) packet
->data
;
1650 return (flow
->dl_type
== htons(ETH_TYPE_ARP
)
1651 && eth_addr_is_broadcast(flow
->dl_dst
)
1652 && packet
->size
>= sizeof(struct arp_eth_header
)
1653 && arp
->ar_op
== ARP_OP_REQUEST
);
1656 /* If the composed actions may be applied to any packet in the given 'flow',
1657 * returns true. Otherwise, the actions should only be applied to 'packet', or
1658 * not at all, if 'packet' was NULL. */
1660 process_flow(struct bridge
*br
, const flow_t
*flow
,
1661 const struct ofpbuf
*packet
, struct odp_actions
*actions
,
1664 struct iface
*in_iface
;
1665 struct port
*in_port
;
1666 struct port
*out_port
= NULL
; /* By default, drop the packet/flow. */
1669 /* Find the interface and port structure for the received packet. */
1670 in_iface
= iface_from_dp_ifidx(br
, flow
->in_port
);
1672 /* No interface? Something fishy... */
1673 if (packet
!= NULL
) {
1674 /* Odd. A few possible reasons here:
1676 * - We deleted an interface but there are still a few packets
1677 * queued up from it.
1679 * - Someone externally added an interface (e.g. with "ovs-dpctl
1680 * add-if") that we don't know about.
1682 * - Packet arrived on the local port but the local port is not
1683 * one of our bridge ports.
1685 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1687 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
1688 "interface %"PRIu16
, br
->name
, flow
->in_port
);
1691 /* Return without adding any actions, to drop packets on this flow. */
1694 in_port
= in_iface
->port
;
1696 /* Figure out what VLAN this packet belongs to.
1698 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1699 * belongs to VLAN 0, so we should treat both cases identically. (In the
1700 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1701 * presumably to allow a priority to be specified. In the latter case, the
1702 * packet does not have any 802.1Q header.) */
1703 vlan
= ntohs(flow
->dl_vlan
);
1704 if (vlan
== OFP_VLAN_NONE
) {
1707 if (in_port
->vlan
>= 0) {
1709 /* XXX support double tagging? */
1710 if (packet
!= NULL
) {
1711 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1712 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %"PRIu16
" tagged "
1713 "packet received on port %s configured with "
1714 "implicit VLAN %"PRIu16
,
1715 br
->name
, ntohs(flow
->dl_vlan
),
1716 in_port
->name
, in_port
->vlan
);
1720 vlan
= in_port
->vlan
;
1722 if (!port_includes_vlan(in_port
, vlan
)) {
1723 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1724 VLOG_WARN_RL(&rl
, "bridge %s: dropping VLAN %d tagged "
1725 "packet received on port %s not configured for "
1727 br
->name
, vlan
, in_port
->name
, vlan
);
1732 /* Drop frames for ports that STP wants entirely killed (both for
1733 * forwarding and for learning). Later, after we do learning, we'll drop
1734 * the frames that STP wants to do learning but not forwarding on. */
1735 if (in_port
->stp_state
& (STP_LISTENING
| STP_BLOCKING
)) {
1739 /* Drop frames for reserved multicast addresses. */
1740 if (eth_addr_is_reserved(flow
->dl_dst
)) {
1744 /* Drop frames on ports reserved for mirroring. */
1745 if (in_port
->is_mirror_output_port
) {
1746 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1747 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port %s, "
1748 "which is reserved exclusively for mirroring",
1749 br
->name
, in_port
->name
);
1753 /* Multicast (and broadcast) packets on bonds need special attention, to
1754 * avoid receiving duplicates. */
1755 if (in_port
->n_ifaces
> 1 && eth_addr_is_multicast(flow
->dl_dst
)) {
1756 *tags
|= in_port
->active_iface_tag
;
1757 if (in_port
->active_iface
!= in_iface
->port_ifidx
) {
1758 /* Drop all multicast packets on inactive slaves. */
1761 /* Drop all multicast packets for which we have learned a different
1762 * input port, because we probably sent the packet on one slaves
1763 * and got it back on the active slave. Broadcast ARP replies are
1764 * an exception to this rule: the host has moved to another
1766 int src_idx
= mac_learning_lookup(br
->ml
, flow
->dl_src
, vlan
);
1767 if (src_idx
!= -1 && src_idx
!= in_port
->port_idx
) {
1769 if (!is_bcast_arp_reply(flow
, packet
)) {
1773 /* No way to know whether it's an ARP reply, because the
1774 * flow entry doesn't include enough information and we
1775 * don't have a packet. Punt. */
1783 out_port
= FLOOD_PORT
;
1787 /* Learn source MAC (but don't try to learn from revalidation). */
1789 tag_type rev_tag
= mac_learning_learn(br
->ml
, flow
->dl_src
,
1790 vlan
, in_port
->port_idx
);
1792 /* The log messages here could actually be useful in debugging,
1793 * so keep the rate limit relatively high. */
1794 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30,
1796 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
1797 "on port %s in VLAN %d",
1798 br
->name
, ETH_ADDR_ARGS(flow
->dl_src
),
1799 in_port
->name
, vlan
);
1800 ofproto_revalidate(br
->ofproto
, rev_tag
);
1804 /* Determine output port. */
1805 out_port_idx
= mac_learning_lookup_tag(br
->ml
, flow
->dl_dst
, vlan
,
1807 if (out_port_idx
>= 0 && out_port_idx
< br
->n_ports
) {
1808 out_port
= br
->ports
[out_port_idx
];
1812 /* Don't send packets out their input ports. Don't forward frames that STP
1813 * wants us to discard. */
1814 if (in_port
== out_port
|| in_port
->stp_state
== STP_LEARNING
) {
1819 compose_actions(br
, flow
, vlan
, in_port
, out_port
, tags
, actions
);
1822 * We send out only a single packet, instead of setting up a flow, if the
1823 * packet is an ARP directed to broadcast that arrived on a bonded
1824 * interface. In such a situation ARP requests and replies must be handled
1825 * differently, but OpenFlow unfortunately can't distinguish them.
1827 return (in_port
->n_ifaces
< 2
1828 || flow
->dl_type
!= htons(ETH_TYPE_ARP
)
1829 || !eth_addr_is_broadcast(flow
->dl_dst
));
1832 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1835 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason
,
1836 const struct ofp_phy_port
*opp
,
1839 struct bridge
*br
= br_
;
1840 struct iface
*iface
;
1843 iface
= iface_from_dp_ifidx(br
, ofp_port_to_odp_port(opp
->port_no
));
1849 if (reason
== OFPPR_DELETE
) {
1850 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1851 br
->name
, iface
->name
);
1852 iface_destroy(iface
);
1853 if (!port
->n_ifaces
) {
1854 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1855 br
->name
, port
->name
);
1861 memcpy(iface
->mac
, opp
->hw_addr
, ETH_ADDR_LEN
);
1862 if (port
->n_ifaces
> 1) {
1863 bool up
= !(opp
->state
& OFPPS_LINK_DOWN
);
1864 bond_link_status_update(iface
, up
);
1865 port_update_bond_compat(port
);
1871 bridge_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
1872 struct odp_actions
*actions
, tag_type
*tags
, void *br_
)
1874 struct bridge
*br
= br_
;
1877 if (flow
->dl_type
== htons(OFP_DL_TYPE_NOT_ETH_TYPE
)
1878 && eth_addr_equals(flow
->dl_dst
, stp_eth_addr
)) {
1879 brstp_receive(br
, flow
, payload
);
1884 COVERAGE_INC(bridge_process_flow
);
1885 return process_flow(br
, flow
, packet
, actions
, tags
);
1889 bridge_account_flow_ofhook_cb(const flow_t
*flow
,
1890 const union odp_action
*actions
,
1891 size_t n_actions
, unsigned long long int n_bytes
,
1894 struct bridge
*br
= br_
;
1895 const union odp_action
*a
;
1897 if (!br
->has_bonded_ports
) {
1901 for (a
= actions
; a
< &actions
[n_actions
]; a
++) {
1902 if (a
->type
== ODPAT_OUTPUT
) {
1903 struct port
*port
= port_from_dp_ifidx(br
, a
->output
.port
);
1904 if (port
&& port
->n_ifaces
>= 2) {
1905 struct bond_entry
*e
= lookup_bond_entry(port
, flow
->dl_src
);
1906 e
->tx_bytes
+= n_bytes
;
1913 bridge_account_checkpoint_ofhook_cb(void *br_
)
1915 struct bridge
*br
= br_
;
1918 if (!br
->has_bonded_ports
) {
1922 /* The current ofproto implementation calls this callback at least once a
1923 * second, so this timer implementation is sufficient. */
1924 if (time_msec() < br
->bond_next_rebalance
) {
1927 br
->bond_next_rebalance
= time_msec() + 10000;
1929 for (i
= 0; i
< br
->n_ports
; i
++) {
1930 struct port
*port
= br
->ports
[i
];
1931 if (port
->n_ifaces
> 1) {
1932 bond_rebalance_port(port
);
1937 static struct ofhooks bridge_ofhooks
= {
1938 bridge_port_changed_ofhook_cb
,
1939 bridge_normal_ofhook_cb
,
1940 bridge_account_flow_ofhook_cb
,
1941 bridge_account_checkpoint_ofhook_cb
,
1944 /* Bonding functions. */
1946 /* Statistics for a single interface on a bonded port, used for load-based
1947 * bond rebalancing. */
1948 struct slave_balance
{
1949 struct iface
*iface
; /* The interface. */
1950 uint64_t tx_bytes
; /* Sum of hashes[*]->tx_bytes. */
1952 /* All the "bond_entry"s that are assigned to this interface, in order of
1953 * increasing tx_bytes. */
1954 struct bond_entry
**hashes
;
1958 /* Sorts pointers to pointers to bond_entries in ascending order by the
1959 * interface to which they are assigned, and within a single interface in
1960 * ascending order of bytes transmitted. */
1962 compare_bond_entries(const void *a_
, const void *b_
)
1964 const struct bond_entry
*const *ap
= a_
;
1965 const struct bond_entry
*const *bp
= b_
;
1966 const struct bond_entry
*a
= *ap
;
1967 const struct bond_entry
*b
= *bp
;
1968 if (a
->iface_idx
!= b
->iface_idx
) {
1969 return a
->iface_idx
> b
->iface_idx
? 1 : -1;
1970 } else if (a
->tx_bytes
!= b
->tx_bytes
) {
1971 return a
->tx_bytes
> b
->tx_bytes
? 1 : -1;
1977 /* Sorts slave_balances so that enabled ports come first, and otherwise in
1978 * *descending* order by number of bytes transmitted. */
1980 compare_slave_balance(const void *a_
, const void *b_
)
1982 const struct slave_balance
*a
= a_
;
1983 const struct slave_balance
*b
= b_
;
1984 if (a
->iface
->enabled
!= b
->iface
->enabled
) {
1985 return a
->iface
->enabled
? -1 : 1;
1986 } else if (a
->tx_bytes
!= b
->tx_bytes
) {
1987 return a
->tx_bytes
> b
->tx_bytes
? -1 : 1;
1994 swap_bals(struct slave_balance
*a
, struct slave_balance
*b
)
1996 struct slave_balance tmp
= *a
;
2001 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2002 * given that 'p' (and only 'p') might be in the wrong location.
2004 * This function invalidates 'p', since it might now be in a different memory
2007 resort_bals(struct slave_balance
*p
,
2008 struct slave_balance bals
[], size_t n_bals
)
2011 for (; p
> bals
&& p
->tx_bytes
> p
[-1].tx_bytes
; p
--) {
2012 swap_bals(p
, p
- 1);
2014 for (; p
< &bals
[n_bals
- 1] && p
->tx_bytes
< p
[1].tx_bytes
; p
++) {
2015 swap_bals(p
, p
+ 1);
2021 log_bals(const struct slave_balance
*bals
, size_t n_bals
, struct port
*port
)
2023 if (VLOG_IS_DBG_ENABLED()) {
2024 struct ds ds
= DS_EMPTY_INITIALIZER
;
2025 const struct slave_balance
*b
;
2027 for (b
= bals
; b
< bals
+ n_bals
; b
++) {
2031 ds_put_char(&ds
, ',');
2033 ds_put_format(&ds
, " %s %"PRIu64
"kB",
2034 b
->iface
->name
, b
->tx_bytes
/ 1024);
2036 if (!b
->iface
->enabled
) {
2037 ds_put_cstr(&ds
, " (disabled)");
2039 if (b
->n_hashes
> 0) {
2040 ds_put_cstr(&ds
, " (");
2041 for (i
= 0; i
< b
->n_hashes
; i
++) {
2042 const struct bond_entry
*e
= b
->hashes
[i
];
2044 ds_put_cstr(&ds
, " + ");
2046 ds_put_format(&ds
, "h%td: %"PRIu64
"kB",
2047 e
- port
->bond_hash
, e
->tx_bytes
/ 1024);
2049 ds_put_cstr(&ds
, ")");
2052 VLOG_DBG("bond %s:%s", port
->name
, ds_cstr(&ds
));
2057 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2059 bond_shift_load(struct slave_balance
*from
, struct slave_balance
*to
,
2060 struct bond_entry
*hash
)
2062 struct port
*port
= from
->iface
->port
;
2063 uint64_t delta
= hash
->tx_bytes
;
2065 VLOG_INFO("bond %s: shift %"PRIu64
"kB of load (with hash %td) "
2066 "from %s to %s (now carrying %"PRIu64
"kB and "
2067 "%"PRIu64
"kB load, respectively)",
2068 port
->name
, delta
/ 1024, hash
- port
->bond_hash
,
2069 from
->iface
->name
, to
->iface
->name
,
2070 (from
->tx_bytes
- delta
) / 1024,
2071 (to
->tx_bytes
+ delta
) / 1024);
2073 /* Delete element from from->hashes.
2075 * We don't bother to add the element to to->hashes because not only would
2076 * it require more work, the only purpose it would be to allow that hash to
2077 * be migrated to another slave in this rebalancing run, and there is no
2078 * point in doing that. */
2079 if (from
->hashes
[0] == hash
) {
2082 int i
= hash
- from
->hashes
[0];
2083 memmove(from
->hashes
+ i
, from
->hashes
+ i
+ 1,
2084 (from
->n_hashes
- (i
+ 1)) * sizeof *from
->hashes
);
2088 /* Shift load away from 'from' to 'to'. */
2089 from
->tx_bytes
-= delta
;
2090 to
->tx_bytes
+= delta
;
2092 /* Arrange for flows to be revalidated. */
2093 ofproto_revalidate(port
->bridge
->ofproto
, hash
->iface_tag
);
2094 hash
->iface_idx
= to
->iface
->port_ifidx
;
2095 hash
->iface_tag
= tag_create_random();
2099 bond_rebalance_port(struct port
*port
)
2101 struct slave_balance bals
[DP_MAX_PORTS
];
2103 struct bond_entry
*hashes
[BOND_MASK
+ 1];
2104 struct slave_balance
*b
, *from
, *to
;
2105 struct bond_entry
*e
;
2108 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2109 * descending order of tx_bytes, so that bals[0] represents the most
2110 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2113 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2114 * array for each slave_balance structure, we sort our local array of
2115 * hashes in order by slave, so that all of the hashes for a given slave
2116 * become contiguous in memory, and then we point each 'hashes' members of
2117 * a slave_balance structure to the start of a contiguous group. */
2118 n_bals
= port
->n_ifaces
;
2119 for (b
= bals
; b
< &bals
[n_bals
]; b
++) {
2120 b
->iface
= port
->ifaces
[b
- bals
];
2125 for (i
= 0; i
<= BOND_MASK
; i
++) {
2126 hashes
[i
] = &port
->bond_hash
[i
];
2128 qsort(hashes
, BOND_MASK
+ 1, sizeof *hashes
, compare_bond_entries
);
2129 for (i
= 0; i
<= BOND_MASK
; i
++) {
2131 if (e
->iface_idx
>= 0 && e
->iface_idx
< port
->n_ifaces
) {
2132 b
= &bals
[e
->iface_idx
];
2133 b
->tx_bytes
+= e
->tx_bytes
;
2135 b
->hashes
= &hashes
[i
];
2140 qsort(bals
, n_bals
, sizeof *bals
, compare_slave_balance
);
2141 log_bals(bals
, n_bals
, port
);
2143 /* Discard slaves that aren't enabled (which were sorted to the back of the
2144 * array earlier). */
2145 while (!bals
[n_bals
- 1].iface
->enabled
) {
2152 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2153 to
= &bals
[n_bals
- 1];
2154 for (from
= bals
; from
< to
; ) {
2155 uint64_t overload
= from
->tx_bytes
- to
->tx_bytes
;
2156 if (overload
< to
->tx_bytes
>> 5 || overload
< 100000) {
2157 /* The extra load on 'from' (and all less-loaded slaves), compared
2158 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2159 * it is less than ~1Mbps. No point in rebalancing. */
2161 } else if (from
->n_hashes
== 1) {
2162 /* 'from' only carries a single MAC hash, so we can't shift any
2163 * load away from it, even though we want to. */
2166 /* 'from' is carrying significantly more load than 'to', and that
2167 * load is split across at least two different hashes. Pick a hash
2168 * to migrate to 'to' (the least-loaded slave), given that doing so
2169 * must not cause 'to''s load to exceed 'from''s load.
2171 * The sort order we use means that we prefer to shift away the
2172 * smallest hashes instead of the biggest ones. There is little
2173 * reason behind this decision; we could use the opposite sort
2174 * order to shift away big hashes ahead of small ones. */
2177 for (i
= 0; i
< from
->n_hashes
; i
++) {
2178 uint64_t delta
= from
->hashes
[i
]->tx_bytes
;
2179 if (to
->tx_bytes
+ delta
< from
->tx_bytes
- delta
) {
2183 if (i
< from
->n_hashes
) {
2184 bond_shift_load(from
, to
, from
->hashes
[i
]);
2186 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2187 * point to different slave_balance structures. It is only
2188 * valid to do these two operations in a row at all because we
2189 * know that 'from' will not move past 'to' and vice versa. */
2190 resort_bals(from
, bals
, n_bals
);
2191 resort_bals(to
, bals
, n_bals
);
2198 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2199 * historical data to decay to <1% in 7 rebalancing runs. */
2200 for (e
= &port
->bond_hash
[0]; e
<= &port
->bond_hash
[BOND_MASK
]; e
++) {
2206 bond_send_learning_packets(struct port
*port
)
2208 struct bridge
*br
= port
->bridge
;
2209 struct mac_entry
*e
;
2210 struct ofpbuf packet
;
2211 int error
, n_packets
, n_errors
;
2213 if (!port
->n_ifaces
|| port
->active_iface
< 0 || !br
->ml
) {
2217 ofpbuf_init(&packet
, 128);
2218 error
= n_packets
= n_errors
= 0;
2219 LIST_FOR_EACH (e
, struct mac_entry
, lru_node
, &br
->ml
->lrus
) {
2220 static const char s
[] = "Open vSwitch Bond Failover";
2221 union ofp_action actions
[2], *a
;
2222 struct eth_header
*eth
;
2223 struct llc_snap_header
*llc_snap
;
2229 if (e
->port
== port
->port_idx
2230 || !choose_output_iface(port
, e
->mac
, &dp_ifidx
, &tags
)) {
2234 /* Compose packet to send. */
2235 ofpbuf_clear(&packet
);
2236 eth
= ofpbuf_put_zeros(&packet
, ETH_HEADER_LEN
);
2237 llc_snap
= ofpbuf_put_zeros(&packet
, LLC_SNAP_HEADER_LEN
);
2238 ofpbuf_put(&packet
, s
, sizeof s
); /* Includes null byte. */
2239 ofpbuf_put(&packet
, e
->mac
, ETH_ADDR_LEN
);
2241 memcpy(eth
->eth_dst
, eth_addr_broadcast
, ETH_ADDR_LEN
);
2242 memcpy(eth
->eth_src
, e
->mac
, ETH_ADDR_LEN
);
2243 eth
->eth_type
= htons(packet
.size
- ETH_HEADER_LEN
);
2245 llc_snap
->llc
.llc_dsap
= LLC_DSAP_SNAP
;
2246 llc_snap
->llc
.llc_ssap
= LLC_SSAP_SNAP
;
2247 llc_snap
->llc
.llc_cntl
= LLC_CNTL_SNAP
;
2248 memcpy(llc_snap
->snap
.snap_org
, "\x00\x23\x20", 3);
2249 llc_snap
->snap
.snap_type
= htons(0xf177); /* Random number. */
2251 /* Compose actions. */
2252 memset(actions
, 0, sizeof actions
);
2255 a
->vlan_vid
.type
= htons(OFPAT_SET_VLAN_VID
);
2256 a
->vlan_vid
.len
= htons(sizeof *a
);
2257 a
->vlan_vid
.vlan_vid
= htons(e
->vlan
);
2260 a
->output
.type
= htons(OFPAT_OUTPUT
);
2261 a
->output
.len
= htons(sizeof *a
);
2262 a
->output
.port
= htons(odp_port_to_ofp_port(dp_ifidx
));
2267 flow_extract(&packet
, ODPP_NONE
, &flow
);
2268 retval
= ofproto_send_packet(br
->ofproto
, &flow
, actions
, a
- actions
,
2275 ofpbuf_uninit(&packet
);
2278 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2279 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2280 "packets, last error was: %s",
2281 port
->name
, n_errors
, n_packets
, strerror(error
));
2283 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2284 port
->name
, n_packets
);
2288 /* Bonding unixctl user interface functions. */
2291 bond_unixctl_list(struct unixctl_conn
*conn
, const char *args UNUSED
)
2293 struct ds ds
= DS_EMPTY_INITIALIZER
;
2294 const struct bridge
*br
;
2296 ds_put_cstr(&ds
, "bridge\tbond\tslaves\n");
2298 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
2301 for (i
= 0; i
< br
->n_ports
; i
++) {
2302 const struct port
*port
= br
->ports
[i
];
2303 if (port
->n_ifaces
> 1) {
2306 ds_put_format(&ds
, "%s\t%s\t", br
->name
, port
->name
);
2307 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2308 const struct iface
*iface
= port
->ifaces
[j
];
2310 ds_put_cstr(&ds
, ", ");
2312 ds_put_cstr(&ds
, iface
->name
);
2314 ds_put_char(&ds
, '\n');
2318 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
2322 static struct port
*
2323 bond_find(const char *name
)
2325 const struct bridge
*br
;
2327 LIST_FOR_EACH (br
, struct bridge
, node
, &all_bridges
) {
2330 for (i
= 0; i
< br
->n_ports
; i
++) {
2331 struct port
*port
= br
->ports
[i
];
2332 if (!strcmp(port
->name
, name
) && port
->n_ifaces
> 1) {
2341 bond_unixctl_show(struct unixctl_conn
*conn
, const char *args
)
2343 struct ds ds
= DS_EMPTY_INITIALIZER
;
2344 const struct port
*port
;
2347 port
= bond_find(args
);
2349 unixctl_command_reply(conn
, 501, "no such bond");
2353 ds_put_format(&ds
, "updelay: %d ms\n", port
->updelay
);
2354 ds_put_format(&ds
, "downdelay: %d ms\n", port
->downdelay
);
2355 ds_put_format(&ds
, "next rebalance: %lld ms\n",
2356 port
->bridge
->bond_next_rebalance
- time_msec());
2357 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2358 const struct iface
*iface
= port
->ifaces
[j
];
2359 struct bond_entry
*be
;
2362 ds_put_format(&ds
, "slave %s: %s\n",
2363 iface
->name
, iface
->enabled
? "enabled" : "disabled");
2364 if (j
== port
->active_iface
) {
2365 ds_put_cstr(&ds
, "\tactive slave\n");
2367 if (iface
->delay_expires
!= LLONG_MAX
) {
2368 ds_put_format(&ds
, "\t%s expires in %lld ms\n",
2369 iface
->enabled
? "downdelay" : "updelay",
2370 iface
->delay_expires
- time_msec());
2374 for (be
= port
->bond_hash
; be
<= &port
->bond_hash
[BOND_MASK
]; be
++) {
2375 int hash
= be
- port
->bond_hash
;
2376 struct mac_entry
*me
;
2378 if (be
->iface_idx
!= j
) {
2382 ds_put_format(&ds
, "\thash %d: %lld kB load\n",
2383 hash
, be
->tx_bytes
/ 1024);
2386 if (!port
->bridge
->ml
) {
2390 LIST_FOR_EACH (me
, struct mac_entry
, lru_node
,
2391 &port
->bridge
->ml
->lrus
) {
2394 if (bond_hash(me
->mac
) == hash
2395 && me
->port
!= port
->port_idx
2396 && choose_output_iface(port
, me
->mac
, &dp_ifidx
, &tags
)
2397 && dp_ifidx
== iface
->dp_ifidx
)
2399 ds_put_format(&ds
, "\t\t"ETH_ADDR_FMT
"\n",
2400 ETH_ADDR_ARGS(me
->mac
));
2405 unixctl_command_reply(conn
, 200, ds_cstr(&ds
));
2410 bond_unixctl_migrate(struct unixctl_conn
*conn
, const char *args_
)
2412 char *args
= (char *) args_
;
2413 char *save_ptr
= NULL
;
2414 char *bond_s
, *hash_s
, *slave_s
;
2415 uint8_t mac
[ETH_ADDR_LEN
];
2417 struct iface
*iface
;
2418 struct bond_entry
*entry
;
2421 bond_s
= strtok_r(args
, " ", &save_ptr
);
2422 hash_s
= strtok_r(NULL
, " ", &save_ptr
);
2423 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
2425 unixctl_command_reply(conn
, 501,
2426 "usage: bond/migrate BOND HASH SLAVE");
2430 port
= bond_find(bond_s
);
2432 unixctl_command_reply(conn
, 501, "no such bond");
2436 if (sscanf(hash_s
, "%"SCNx8
":%"SCNx8
":%"SCNx8
":%"SCNx8
":%"SCNx8
":%"SCNx8
,
2437 &mac
[0], &mac
[1], &mac
[2], &mac
[3], &mac
[4], &mac
[5]) == 6) {
2438 hash
= bond_hash(mac
);
2439 } else if (strspn(hash_s
, "0123456789") == strlen(hash_s
)) {
2440 hash
= atoi(hash_s
) & BOND_MASK
;
2442 unixctl_command_reply(conn
, 501, "bad hash");
2446 iface
= port_lookup_iface(port
, slave_s
);
2448 unixctl_command_reply(conn
, 501, "no such slave");
2452 if (!iface
->enabled
) {
2453 unixctl_command_reply(conn
, 501, "cannot migrate to disabled slave");
2457 entry
= &port
->bond_hash
[hash
];
2458 ofproto_revalidate(port
->bridge
->ofproto
, entry
->iface_tag
);
2459 entry
->iface_idx
= iface
->port_ifidx
;
2460 entry
->iface_tag
= tag_create_random();
2461 unixctl_command_reply(conn
, 200, "migrated");
2465 bond_unixctl_set_active_slave(struct unixctl_conn
*conn
, const char *args_
)
2467 char *args
= (char *) args_
;
2468 char *save_ptr
= NULL
;
2469 char *bond_s
, *slave_s
;
2471 struct iface
*iface
;
2473 bond_s
= strtok_r(args
, " ", &save_ptr
);
2474 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
2476 unixctl_command_reply(conn
, 501,
2477 "usage: bond/set-active-slave BOND SLAVE");
2481 port
= bond_find(bond_s
);
2483 unixctl_command_reply(conn
, 501, "no such bond");
2487 iface
= port_lookup_iface(port
, slave_s
);
2489 unixctl_command_reply(conn
, 501, "no such slave");
2493 if (!iface
->enabled
) {
2494 unixctl_command_reply(conn
, 501, "cannot make disabled slave active");
2498 if (port
->active_iface
!= iface
->port_ifidx
) {
2499 ofproto_revalidate(port
->bridge
->ofproto
, port
->active_iface_tag
);
2500 port
->active_iface
= iface
->port_ifidx
;
2501 port
->active_iface_tag
= tag_create_random();
2502 VLOG_INFO("port %s: active interface is now %s",
2503 port
->name
, iface
->name
);
2504 bond_send_learning_packets(port
);
2505 unixctl_command_reply(conn
, 200, "done");
2507 unixctl_command_reply(conn
, 200, "no change");
2512 enable_slave(struct unixctl_conn
*conn
, const char *args_
, bool enable
)
2514 char *args
= (char *) args_
;
2515 char *save_ptr
= NULL
;
2516 char *bond_s
, *slave_s
;
2518 struct iface
*iface
;
2520 bond_s
= strtok_r(args
, " ", &save_ptr
);
2521 slave_s
= strtok_r(NULL
, " ", &save_ptr
);
2523 unixctl_command_reply(conn
, 501,
2524 "usage: bond/enable/disable-slave BOND SLAVE");
2528 port
= bond_find(bond_s
);
2530 unixctl_command_reply(conn
, 501, "no such bond");
2534 iface
= port_lookup_iface(port
, slave_s
);
2536 unixctl_command_reply(conn
, 501, "no such slave");
2540 bond_enable_slave(iface
, enable
);
2541 unixctl_command_reply(conn
, 501, enable
? "enabled" : "disabled");
2545 bond_unixctl_enable_slave(struct unixctl_conn
*conn
, const char *args
)
2547 enable_slave(conn
, args
, true);
2551 bond_unixctl_disable_slave(struct unixctl_conn
*conn
, const char *args
)
2553 enable_slave(conn
, args
, false);
2559 unixctl_command_register("bond/list", bond_unixctl_list
);
2560 unixctl_command_register("bond/show", bond_unixctl_show
);
2561 unixctl_command_register("bond/migrate", bond_unixctl_migrate
);
2562 unixctl_command_register("bond/set-active-slave",
2563 bond_unixctl_set_active_slave
);
2564 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave
);
2565 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave
);
2568 /* Port functions. */
2571 port_create(struct bridge
*br
, const char *name
)
2575 port
= xcalloc(1, sizeof *port
);
2577 port
->port_idx
= br
->n_ports
;
2579 port
->trunks
= NULL
;
2580 port
->name
= xstrdup(name
);
2581 port
->active_iface
= -1;
2582 port
->stp_state
= STP_DISABLED
;
2583 port
->stp_state_tag
= 0;
2585 if (br
->n_ports
>= br
->allocated_ports
) {
2586 br
->ports
= x2nrealloc(br
->ports
, &br
->allocated_ports
,
2589 br
->ports
[br
->n_ports
++] = port
;
2591 VLOG_INFO("created port %s on bridge %s", port
->name
, br
->name
);
2596 port_reconfigure(struct port
*port
)
2598 bool bonded
= cfg_has_section("bonding.%s", port
->name
);
2599 struct svec old_ifaces
, new_ifaces
;
2600 unsigned long *trunks
;
2604 /* Collect old and new interfaces. */
2605 svec_init(&old_ifaces
);
2606 svec_init(&new_ifaces
);
2607 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2608 svec_add(&old_ifaces
, port
->ifaces
[i
]->name
);
2610 svec_sort(&old_ifaces
);
2612 cfg_get_all_keys(&new_ifaces
, "bonding.%s.slave", port
->name
);
2613 if (!new_ifaces
.n
) {
2614 VLOG_ERR("port %s: no interfaces specified for bonded port",
2616 } else if (new_ifaces
.n
== 1) {
2617 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2621 port
->updelay
= cfg_get_int(0, "bonding.%s.updelay", port
->name
);
2622 if (port
->updelay
< 0) {
2625 port
->downdelay
= cfg_get_int(0, "bonding.%s.downdelay", port
->name
);
2626 if (port
->downdelay
< 0) {
2627 port
->downdelay
= 0;
2630 svec_init(&new_ifaces
);
2631 svec_add(&new_ifaces
, port
->name
);
2634 /* Get rid of deleted interfaces and add new interfaces. */
2635 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2636 struct iface
*iface
= port
->ifaces
[i
];
2637 if (!svec_contains(&new_ifaces
, iface
->name
)) {
2638 iface_destroy(iface
);
2643 for (i
= 0; i
< new_ifaces
.n
; i
++) {
2644 const char *name
= new_ifaces
.names
[i
];
2645 if (!svec_contains(&old_ifaces
, name
)) {
2646 iface_create(port
, name
);
2652 if (cfg_has("vlan.%s.tag", port
->name
)) {
2654 vlan
= cfg_get_vlan(0, "vlan.%s.tag", port
->name
);
2655 if (vlan
>= 0 && vlan
<= 4095) {
2656 VLOG_DBG("port %s: assigning VLAN tag %d", port
->name
, vlan
);
2659 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2660 * they even work as-is. But they have not been tested. */
2661 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2665 if (port
->vlan
!= vlan
) {
2667 bridge_flush(port
->bridge
);
2670 /* Get trunked VLANs. */
2673 size_t n_trunks
, n_errors
;
2676 trunks
= bitmap_allocate(4096);
2677 n_trunks
= cfg_count("vlan.%s.trunks", port
->name
);
2679 for (i
= 0; i
< n_trunks
; i
++) {
2680 int trunk
= cfg_get_vlan(i
, "vlan.%s.trunks", port
->name
);
2682 bitmap_set1(trunks
, trunk
);
2688 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2689 port
->name
, n_trunks
);
2691 if (n_errors
== n_trunks
) {
2693 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2696 bitmap_set_multiple(trunks
, 0, 4096, 1);
2699 if (cfg_has("vlan.%s.trunks", port
->name
)) {
2700 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2701 port
->name
, port
->name
);
2705 ? port
->trunks
!= NULL
2706 : port
->trunks
== NULL
|| !bitmap_equal(trunks
, port
->trunks
, 4096)) {
2707 bridge_flush(port
->bridge
);
2709 bitmap_free(port
->trunks
);
2710 port
->trunks
= trunks
;
2712 svec_destroy(&old_ifaces
);
2713 svec_destroy(&new_ifaces
);
2717 port_destroy(struct port
*port
)
2720 struct bridge
*br
= port
->bridge
;
2724 proc_net_compat_update_vlan(port
->name
, NULL
, 0);
2726 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2727 struct mirror
*m
= br
->mirrors
[i
];
2728 if (m
&& m
->out_port
== port
) {
2733 while (port
->n_ifaces
> 0) {
2734 iface_destroy(port
->ifaces
[port
->n_ifaces
- 1]);
2737 del
= br
->ports
[port
->port_idx
] = br
->ports
[--br
->n_ports
];
2738 del
->port_idx
= port
->port_idx
;
2741 bitmap_free(port
->trunks
);
2748 static struct port
*
2749 port_from_dp_ifidx(const struct bridge
*br
, uint16_t dp_ifidx
)
2751 struct iface
*iface
= iface_from_dp_ifidx(br
, dp_ifidx
);
2752 return iface
? iface
->port
: NULL
;
2755 static struct port
*
2756 port_lookup(const struct bridge
*br
, const char *name
)
2760 for (i
= 0; i
< br
->n_ports
; i
++) {
2761 struct port
*port
= br
->ports
[i
];
2762 if (!strcmp(port
->name
, name
)) {
2769 static struct iface
*
2770 port_lookup_iface(const struct port
*port
, const char *name
)
2774 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2775 struct iface
*iface
= port
->ifaces
[j
];
2776 if (!strcmp(iface
->name
, name
)) {
2784 port_update_bonding(struct port
*port
)
2786 if (port
->n_ifaces
< 2) {
2787 /* Not a bonded port. */
2788 if (port
->bond_hash
) {
2789 free(port
->bond_hash
);
2790 port
->bond_hash
= NULL
;
2791 proc_net_compat_update_bond(port
->name
, NULL
);
2794 if (!port
->bond_hash
) {
2797 port
->bond_hash
= xcalloc(BOND_MASK
+ 1, sizeof *port
->bond_hash
);
2798 for (i
= 0; i
<= BOND_MASK
; i
++) {
2799 struct bond_entry
*e
= &port
->bond_hash
[i
];
2803 port
->no_ifaces_tag
= tag_create_random();
2804 bond_choose_active_iface(port
);
2806 port_update_bond_compat(port
);
2811 port_update_bond_compat(struct port
*port
)
2813 struct compat_bond bond
;
2816 if (port
->n_ifaces
< 2) {
2821 bond
.updelay
= port
->updelay
;
2822 bond
.downdelay
= port
->downdelay
;
2823 bond
.n_slaves
= port
->n_ifaces
;
2824 bond
.slaves
= xmalloc(port
->n_ifaces
* sizeof *bond
.slaves
);
2825 for (i
= 0; i
< port
->n_ifaces
; i
++) {
2826 struct iface
*iface
= port
->ifaces
[i
];
2827 struct compat_bond_slave
*slave
= &bond
.slaves
[i
];
2828 slave
->name
= iface
->name
;
2829 slave
->up
= ((iface
->enabled
&& iface
->delay_expires
== LLONG_MAX
) ||
2830 (!iface
->enabled
&& iface
->delay_expires
!= LLONG_MAX
));
2834 memcpy(slave
->mac
, iface
->mac
, ETH_ADDR_LEN
);
2836 proc_net_compat_update_bond(port
->name
, &bond
);
2841 port_update_vlan_compat(struct port
*port
)
2843 struct bridge
*br
= port
->bridge
;
2844 char *vlandev_name
= NULL
;
2846 if (port
->vlan
> 0) {
2847 /* Figure out the name that the VLAN device should actually have, if it
2848 * existed. This takes some work because the VLAN device would not
2849 * have port->name in its name; rather, it would have the trunk port's
2850 * name, and 'port' would be attached to a bridge that also had the
2851 * VLAN device one of its ports. So we need to find a trunk port that
2852 * includes port->vlan.
2854 * There might be more than one candidate. This doesn't happen on
2855 * XenServer, so if it happens we just pick the first choice in
2856 * alphabetical order instead of creating multiple VLAN devices. */
2858 for (i
= 0; i
< br
->n_ports
; i
++) {
2859 struct port
*p
= br
->ports
[i
];
2860 if (port_trunks_vlan(p
, port
->vlan
)
2862 && (!vlandev_name
|| strcmp(p
->name
, vlandev_name
) <= 0))
2864 const uint8_t *ea
= p
->ifaces
[0]->mac
;
2865 if (!eth_addr_is_multicast(ea
) &&
2866 !eth_addr_is_reserved(ea
) &&
2867 !eth_addr_is_zero(ea
)) {
2868 vlandev_name
= p
->name
;
2873 proc_net_compat_update_vlan(port
->name
, vlandev_name
, port
->vlan
);
2876 /* Interface functions. */
2879 iface_create(struct port
*port
, const char *name
)
2881 struct iface
*iface
;
2883 iface
= xcalloc(1, sizeof *iface
);
2885 iface
->port_ifidx
= port
->n_ifaces
;
2886 iface
->name
= xstrdup(name
);
2887 iface
->dp_ifidx
= -1;
2888 iface
->tag
= tag_create_random();
2889 iface
->delay_expires
= LLONG_MAX
;
2891 netdev_nodev_get_etheraddr(name
, iface
->mac
);
2892 netdev_nodev_get_carrier(name
, &iface
->enabled
);
2894 if (port
->n_ifaces
>= port
->allocated_ifaces
) {
2895 port
->ifaces
= x2nrealloc(port
->ifaces
, &port
->allocated_ifaces
,
2896 sizeof *port
->ifaces
);
2898 port
->ifaces
[port
->n_ifaces
++] = iface
;
2899 if (port
->n_ifaces
> 1) {
2900 port
->bridge
->has_bonded_ports
= true;
2903 VLOG_DBG("attached network device %s to port %s", iface
->name
, port
->name
);
2905 port_update_bonding(port
);
2906 bridge_flush(port
->bridge
);
2910 iface_destroy(struct iface
*iface
)
2913 struct port
*port
= iface
->port
;
2914 struct bridge
*br
= port
->bridge
;
2915 bool del_active
= port
->active_iface
== iface
->port_ifidx
;
2918 if (iface
->dp_ifidx
>= 0) {
2919 port_array_set(&br
->ifaces
, iface
->dp_ifidx
, NULL
);
2922 del
= port
->ifaces
[iface
->port_ifidx
] = port
->ifaces
[--port
->n_ifaces
];
2923 del
->port_ifidx
= iface
->port_ifidx
;
2929 ofproto_revalidate(port
->bridge
->ofproto
, port
->active_iface_tag
);
2930 bond_choose_active_iface(port
);
2931 bond_send_learning_packets(port
);
2934 port_update_bonding(port
);
2935 bridge_flush(port
->bridge
);
2939 static struct iface
*
2940 iface_lookup(const struct bridge
*br
, const char *name
)
2944 for (i
= 0; i
< br
->n_ports
; i
++) {
2945 struct port
*port
= br
->ports
[i
];
2946 for (j
= 0; j
< port
->n_ifaces
; j
++) {
2947 struct iface
*iface
= port
->ifaces
[j
];
2948 if (!strcmp(iface
->name
, name
)) {
2956 static struct iface
*
2957 iface_from_dp_ifidx(const struct bridge
*br
, uint16_t dp_ifidx
)
2959 return port_array_get(&br
->ifaces
, dp_ifidx
);
2962 /* Port mirroring. */
2965 mirror_reconfigure(struct bridge
*br
)
2967 struct svec old_mirrors
, new_mirrors
;
2970 /* Collect old and new mirrors. */
2971 svec_init(&old_mirrors
);
2972 svec_init(&new_mirrors
);
2973 cfg_get_subsections(&new_mirrors
, "mirror.%s", br
->name
);
2974 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2975 if (br
->mirrors
[i
]) {
2976 svec_add(&old_mirrors
, br
->mirrors
[i
]->name
);
2980 /* Get rid of deleted mirrors and add new mirrors. */
2981 svec_sort(&old_mirrors
);
2982 assert(svec_is_unique(&old_mirrors
));
2983 svec_sort(&new_mirrors
);
2984 assert(svec_is_unique(&new_mirrors
));
2985 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
2986 struct mirror
*m
= br
->mirrors
[i
];
2987 if (m
&& !svec_contains(&new_mirrors
, m
->name
)) {
2991 for (i
= 0; i
< new_mirrors
.n
; i
++) {
2992 const char *name
= new_mirrors
.names
[i
];
2993 if (!svec_contains(&old_mirrors
, name
)) {
2994 mirror_create(br
, name
);
2997 svec_destroy(&old_mirrors
);
2998 svec_destroy(&new_mirrors
);
3000 /* Reconfigure all mirrors. */
3001 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3002 if (br
->mirrors
[i
]) {
3003 mirror_reconfigure_one(br
->mirrors
[i
]);
3007 /* Update port reserved status. */
3008 for (i
= 0; i
< br
->n_ports
; i
++) {
3009 br
->ports
[i
]->is_mirror_output_port
= false;
3011 for (i
= 0; i
< MAX_MIRRORS
; i
++) {
3012 struct mirror
*m
= br
->mirrors
[i
];
3013 if (m
&& m
->out_port
) {
3014 m
->out_port
->is_mirror_output_port
= true;
3020 mirror_create(struct bridge
*br
, const char *name
)
3025 for (i
= 0; ; i
++) {
3026 if (i
>= MAX_MIRRORS
) {
3027 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3028 "cannot create %s", br
->name
, MAX_MIRRORS
, name
);
3031 if (!br
->mirrors
[i
]) {
3036 VLOG_INFO("created port mirror %s on bridge %s", name
, br
->name
);
3039 br
->mirrors
[i
] = m
= xcalloc(1, sizeof *m
);
3042 m
->name
= xstrdup(name
);
3043 svec_init(&m
->src_ports
);
3044 svec_init(&m
->dst_ports
);
3052 mirror_destroy(struct mirror
*m
)
3055 struct bridge
*br
= m
->bridge
;
3058 for (i
= 0; i
< br
->n_ports
; i
++) {
3059 br
->ports
[i
]->src_mirrors
&= ~(MIRROR_MASK_C(1) << m
->idx
);
3060 br
->ports
[i
]->dst_mirrors
&= ~(MIRROR_MASK_C(1) << m
->idx
);
3063 svec_destroy(&m
->src_ports
);
3064 svec_destroy(&m
->dst_ports
);
3067 m
->bridge
->mirrors
[m
->idx
] = NULL
;
3075 prune_ports(struct mirror
*m
, struct svec
*ports
)
3080 svec_sort_unique(ports
);
3083 for (i
= 0; i
< ports
->n
; i
++) {
3084 const char *name
= ports
->names
[i
];
3085 if (port_lookup(m
->bridge
, name
)) {
3086 svec_add(&tmp
, name
);
3088 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3089 m
->bridge
->name
, m
->name
, name
);
3092 svec_swap(ports
, &tmp
);
3097 prune_vlans(struct mirror
*m
, struct svec
*vlan_strings
, int **vlans
)
3101 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3102 * order won't give us numeric sort order. But that's good enough for what
3103 * we need right now. */
3104 svec_sort_unique(vlan_strings
);
3106 *vlans
= xmalloc(sizeof *vlans
* vlan_strings
->n
);
3108 for (i
= 0; i
< vlan_strings
->n
; i
++) {
3109 const char *name
= vlan_strings
->names
[i
];
3111 if (!str_to_int(name
, 10, &vlan
) || vlan
< 0 || vlan
> 4095) {
3112 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3113 m
->bridge
->name
, m
->name
, name
);
3115 (*vlans
)[n_vlans
++] = vlan
;
3122 vlan_is_mirrored(const struct mirror
*m
, int vlan
)
3126 for (i
= 0; i
< m
->n_vlans
; i
++) {
3127 if (m
->vlans
[i
] == vlan
) {
3135 port_trunks_any_mirrored_vlan(const struct mirror
*m
, const struct port
*p
)
3139 for (i
= 0; i
< m
->n_vlans
; i
++) {
3140 if (port_trunks_vlan(p
, m
->vlans
[i
])) {
3148 mirror_reconfigure_one(struct mirror
*m
)
3150 char *pfx
= xasprintf("mirror.%s.%s", m
->bridge
->name
, m
->name
);
3151 struct svec src_ports
, dst_ports
, ports
;
3152 struct svec vlan_strings
;
3153 mirror_mask_t mirror_bit
;
3154 const char *out_port_name
;
3155 struct port
*out_port
;
3160 bool mirror_all_ports
;
3162 /* Get output port. */
3163 out_port_name
= cfg_get_key(0, "mirror.%s.%s.output.port",
3164 m
->bridge
->name
, m
->name
);
3165 if (out_port_name
) {
3166 out_port
= port_lookup(m
->bridge
, out_port_name
);
3168 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3169 "named %s", pfx
, m
->bridge
->name
, out_port_name
);
3176 if (cfg_has("%s.output.vlan", pfx
)) {
3177 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3178 "ignoring %s.output.vlan", pfx
, pfx
, pfx
);
3180 } else if (cfg_has("%s.output.vlan", pfx
)) {
3182 out_vlan
= cfg_get_vlan(0, "%s.output.vlan", pfx
);
3184 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3185 "but exactly one is required; disabling port mirror %s",
3186 pfx
, pfx
, pfx
, pfx
);
3192 /* Get all the ports, and drop duplicates and ports that don't exist. */
3193 svec_init(&src_ports
);
3194 svec_init(&dst_ports
);
3196 cfg_get_all_keys(&src_ports
, "%s.select.src-port", pfx
);
3197 cfg_get_all_keys(&dst_ports
, "%s.select.dst-port", pfx
);
3198 cfg_get_all_keys(&ports
, "%s.select.port", pfx
);
3199 svec_append(&src_ports
, &ports
);
3200 svec_append(&dst_ports
, &ports
);
3201 svec_destroy(&ports
);
3202 prune_ports(m
, &src_ports
);
3203 prune_ports(m
, &dst_ports
);
3205 /* Get all the vlans, and drop duplicate and invalid vlans. */
3206 svec_init(&vlan_strings
);
3207 cfg_get_all_keys(&vlan_strings
, "%s.select.vlan", pfx
);
3208 n_vlans
= prune_vlans(m
, &vlan_strings
, &vlans
);
3209 svec_destroy(&vlan_strings
);
3211 /* Update mirror data. */
3212 if (!svec_equal(&m
->src_ports
, &src_ports
)
3213 || !svec_equal(&m
->dst_ports
, &dst_ports
)
3214 || m
->n_vlans
!= n_vlans
3215 || memcmp(m
->vlans
, vlans
, sizeof *vlans
* n_vlans
)
3216 || m
->out_port
!= out_port
3217 || m
->out_vlan
!= out_vlan
) {
3218 bridge_flush(m
->bridge
);
3220 svec_swap(&m
->src_ports
, &src_ports
);
3221 svec_swap(&m
->dst_ports
, &dst_ports
);
3224 m
->n_vlans
= n_vlans
;
3225 m
->out_port
= out_port
;
3226 m
->out_vlan
= out_vlan
;
3228 /* If no selection criteria have been given, mirror for all ports. */
3229 mirror_all_ports
= (!m
->src_ports
.n
) && (!m
->dst_ports
.n
) && (!m
->n_vlans
);
3232 mirror_bit
= MIRROR_MASK_C(1) << m
->idx
;
3233 for (i
= 0; i
< m
->bridge
->n_ports
; i
++) {
3234 struct port
*port
= m
->bridge
->ports
[i
];
3236 if (mirror_all_ports
3237 || svec_contains(&m
->src_ports
, port
->name
)
3240 ? port_trunks_any_mirrored_vlan(m
, port
)
3241 : vlan_is_mirrored(m
, port
->vlan
)))) {
3242 port
->src_mirrors
|= mirror_bit
;
3244 port
->src_mirrors
&= ~mirror_bit
;
3247 if (mirror_all_ports
|| svec_contains(&m
->dst_ports
, port
->name
)) {
3248 port
->dst_mirrors
|= mirror_bit
;
3250 port
->dst_mirrors
&= ~mirror_bit
;
3255 svec_destroy(&src_ports
);
3256 svec_destroy(&dst_ports
);
3260 /* Spanning tree protocol. */
3262 static void brstp_update_port_state(struct port
*);
3265 brstp_send_bpdu(struct ofpbuf
*pkt
, int port_no
, void *br_
)
3267 struct bridge
*br
= br_
;
3268 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3269 struct iface
*iface
= iface_from_dp_ifidx(br
, port_no
);
3271 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
3273 } else if (eth_addr_is_zero(iface
->mac
)) {
3274 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d with unknown MAC",
3277 union ofp_action action
;
3278 struct eth_header
*eth
= pkt
->l2
;
3281 memcpy(eth
->eth_src
, iface
->mac
, ETH_ADDR_LEN
);
3283 memset(&action
, 0, sizeof action
);
3284 action
.type
= htons(OFPAT_OUTPUT
);
3285 action
.output
.len
= htons(sizeof action
);
3286 action
.output
.port
= htons(port_no
);
3288 flow_extract(pkt
, ODPP_NONE
, &flow
);
3289 ofproto_send_packet(br
->ofproto
, &flow
, &action
, 1, pkt
);
3295 brstp_reconfigure(struct bridge
*br
)
3299 if (!cfg_get_bool(0, "stp.%s.enabled", br
->name
)) {
3301 stp_destroy(br
->stp
);
3307 uint64_t bridge_address
, bridge_id
;
3308 int bridge_priority
;
3310 bridge_address
= cfg_get_mac(0, "stp.%s.address", br
->name
);
3311 if (!bridge_address
) {
3313 bridge_address
= (stp_get_bridge_id(br
->stp
)
3314 & ((UINT64_C(1) << 48) - 1));
3316 uint8_t mac
[ETH_ADDR_LEN
];
3317 eth_addr_random(mac
);
3318 bridge_address
= eth_addr_to_uint64(mac
);
3322 if (cfg_is_valid(CFG_INT
| CFG_REQUIRED
, "stp.%s.priority",
3324 bridge_priority
= cfg_get_int(0, "stp.%s.priority", br
->name
);
3326 bridge_priority
= STP_DEFAULT_BRIDGE_PRIORITY
;
3329 bridge_id
= bridge_address
| ((uint64_t) bridge_priority
<< 48);
3331 br
->stp
= stp_create(br
->name
, bridge_id
, brstp_send_bpdu
, br
);
3332 br
->stp_last_tick
= time_msec();
3335 if (bridge_id
!= stp_get_bridge_id(br
->stp
)) {
3336 stp_set_bridge_id(br
->stp
, bridge_id
);
3341 for (i
= 0; i
< br
->n_ports
; i
++) {
3342 struct port
*p
= br
->ports
[i
];
3344 struct stp_port
*sp
;
3345 int path_cost
, priority
;
3351 dp_ifidx
= p
->ifaces
[0]->dp_ifidx
;
3352 if (dp_ifidx
< 0 || dp_ifidx
>= STP_MAX_PORTS
) {
3356 sp
= stp_get_port(br
->stp
, dp_ifidx
);
3357 enable
= (!cfg_is_valid(CFG_BOOL
| CFG_REQUIRED
,
3358 "stp.%s.port.%s.enabled",
3360 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3361 br
->name
, p
->name
));
3362 if (p
->is_mirror_output_port
) {
3365 if (enable
!= (stp_port_get_state(sp
) != STP_DISABLED
)) {
3366 bridge_flush(br
); /* Might not be necessary. */
3368 stp_port_enable(sp
);
3370 stp_port_disable(sp
);
3374 path_cost
= cfg_get_int(0, "stp.%s.port.%s.path-cost",
3376 stp_port_set_path_cost(sp
, path_cost
? path_cost
: 19 /* XXX */);
3378 priority
= (cfg_is_valid(CFG_INT
| CFG_REQUIRED
,
3379 "stp.%s.port.%s.priority",
3381 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3383 : STP_DEFAULT_PORT_PRIORITY
);
3384 stp_port_set_priority(sp
, priority
);
3387 brstp_adjust_timers(br
);
3389 for (i
= 0; i
< br
->n_ports
; i
++) {
3390 brstp_update_port_state(br
->ports
[i
]);
3395 brstp_update_port_state(struct port
*p
)
3397 struct bridge
*br
= p
->bridge
;
3398 enum stp_state state
;
3400 /* Figure out new state. */
3401 state
= STP_DISABLED
;
3402 if (br
->stp
&& p
->n_ifaces
> 0) {
3403 int dp_ifidx
= p
->ifaces
[0]->dp_ifidx
;
3404 if (dp_ifidx
>= 0 && dp_ifidx
< STP_MAX_PORTS
) {
3405 state
= stp_port_get_state(stp_get_port(br
->stp
, dp_ifidx
));
3410 if (p
->stp_state
!= state
) {
3411 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(10, 10);
3412 VLOG_INFO_RL(&rl
, "port %s: STP state changed from %s to %s",
3413 p
->name
, stp_state_name(p
->stp_state
),
3414 stp_state_name(state
));
3415 if (p
->stp_state
== STP_DISABLED
) {
3418 ofproto_revalidate(p
->bridge
->ofproto
, p
->stp_state_tag
);
3420 p
->stp_state
= state
;
3421 p
->stp_state_tag
= (p
->stp_state
== STP_DISABLED
? 0
3422 : tag_create_random());
3427 brstp_adjust_timers(struct bridge
*br
)
3429 int hello_time
= cfg_get_int(0, "stp.%s.hello-time", br
->name
);
3430 int max_age
= cfg_get_int(0, "stp.%s.max-age", br
->name
);
3431 int forward_delay
= cfg_get_int(0, "stp.%s.forward-delay", br
->name
);
3433 stp_set_hello_time(br
->stp
, hello_time
? hello_time
: 2000);
3434 stp_set_max_age(br
->stp
, max_age
? max_age
: 20000);
3435 stp_set_forward_delay(br
->stp
, forward_delay
? forward_delay
: 15000);
3439 brstp_run(struct bridge
*br
)
3442 long long int now
= time_msec();
3443 long long int elapsed
= now
- br
->stp_last_tick
;
3444 struct stp_port
*sp
;
3447 stp_tick(br
->stp
, MIN(INT_MAX
, elapsed
));
3448 br
->stp_last_tick
= now
;
3450 while (stp_get_changed_port(br
->stp
, &sp
)) {
3451 struct port
*p
= port_from_dp_ifidx(br
, stp_port_no(sp
));
3453 brstp_update_port_state(p
);
3460 brstp_wait(struct bridge
*br
)
3463 poll_timer_wait(1000);