1 /* Copyright (c) 2008, 2009 Nicira Networks
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
20 #include <arpa/inet.h>
24 #include <openflow/openflow.h>
29 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include "dynamic-string.h"
41 #include "mac-learning.h"
44 #include "ofp-print.h"
46 #include "ofproto/ofproto.h"
48 #include "poll-loop.h"
49 #include "port-array.h"
50 #include "proc-net-compat.h"
52 #include "socket-util.h"
59 #include "vconn-ssl.h"
60 #include "xenserver.h"
63 #define THIS_MODULE VLM_bridge
71 extern uint64_t mgmt_id;
74 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; /* 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 bridge_unixctl_fdb_show(struct unixctl_conn *, const char *args);
211 static void bond_init(void);
212 static void bond_run(struct bridge *);
213 static void bond_wait(struct bridge *);
214 static void bond_rebalance_port(struct port *);
215 static void bond_send_learning_packets(struct port *);
217 static void port_create(struct bridge *, const char *name);
218 static void port_reconfigure(struct port *);
219 static void port_destroy(struct port *);
220 static struct port *port_lookup(const struct bridge *, const char *name);
221 static struct iface *port_lookup_iface(const struct port *, const char *name);
222 static struct port *port_from_dp_ifidx(const struct bridge *,
224 static void port_update_bond_compat(struct port *);
225 static void port_update_vlan_compat(struct port *);
227 static void mirror_create(struct bridge *, const char *name);
228 static void mirror_destroy(struct mirror *);
229 static void mirror_reconfigure(struct bridge *);
230 static void mirror_reconfigure_one(struct mirror *);
231 static bool vlan_is_mirrored(const struct mirror *, int vlan);
233 static void brstp_reconfigure(struct bridge *);
234 static void brstp_adjust_timers(struct bridge *);
235 static void brstp_run(struct bridge *);
236 static void brstp_wait(struct bridge *);
238 static void iface_create(struct port *, const char *name);
239 static void iface_destroy(struct iface *);
240 static struct iface *iface_lookup(const struct bridge *, const char *name);
241 static struct iface *iface_from_dp_ifidx(const struct bridge *,
244 /* Hooks into ofproto processing. */
245 static struct ofhooks bridge_ofhooks;
247 /* Public functions. */
249 /* Adds the name of each interface used by a bridge, including local and
250 * internal ports, to 'svec'. */
252 bridge_get_ifaces(struct svec *svec)
254 struct bridge *br, *next;
257 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
258 for (i = 0; i < br->n_ports; i++) {
259 struct port *port = br->ports[i];
261 for (j = 0; j < port->n_ifaces; j++) {
262 struct iface *iface = port->ifaces[j];
263 if (iface->dp_ifidx < 0) {
264 VLOG_ERR("%s interface not in datapath %s, ignoring",
265 iface->name, dpif_name(br->dpif));
267 if (iface->dp_ifidx != ODPP_LOCAL) {
268 svec_add(svec, iface->name);
276 /* The caller must already have called cfg_read(). */
280 struct svec dpif_names;
283 unixctl_command_register("fdb/show", bridge_unixctl_fdb_show);
285 dp_enumerate(&dpif_names);
286 for (i = 0; i < dpif_names.n; i++) {
287 const char *dpif_name = dpif_names.names[i];
291 retval = dpif_open(dpif_name, &dpif);
293 struct svec all_names;
296 svec_init(&all_names);
297 dpif_get_all_names(dpif, &all_names);
298 for (j = 0; j < all_names.n; j++) {
299 if (cfg_has("bridge.%s.port", all_names.names[j])) {
305 svec_destroy(&all_names);
311 bridge_reconfigure();
316 config_string_change(const char *key, char **valuep)
318 const char *value = cfg_get_string(0, "%s", key);
319 if (value && (!*valuep || strcmp(value, *valuep))) {
321 *valuep = xstrdup(value);
329 bridge_configure_ssl(void)
331 /* XXX SSL should be configurable on a per-bridge basis.
332 * XXX should be possible to de-configure SSL. */
333 static char *private_key_file;
334 static char *certificate_file;
335 static char *cacert_file;
338 if (config_string_change("ssl.private-key", &private_key_file)) {
339 vconn_ssl_set_private_key_file(private_key_file);
342 if (config_string_change("ssl.certificate", &certificate_file)) {
343 vconn_ssl_set_certificate_file(certificate_file);
346 /* We assume that even if the filename hasn't changed, if the CA cert
347 * file has been removed, that we want to move back into
348 * boot-strapping mode. This opens a small security hole, because
349 * the old certificate will still be trusted until vSwitch is
350 * restarted. We may want to address this in vconn's SSL library. */
351 if (config_string_change("ssl.ca-cert", &cacert_file)
352 || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) {
353 vconn_ssl_set_ca_cert_file(cacert_file,
354 cfg_get_bool(0, "ssl.bootstrap-ca-cert"));
360 bridge_reconfigure(void)
362 struct svec old_br, new_br;
363 struct bridge *br, *next;
366 COVERAGE_INC(bridge_reconfigure);
368 /* Collect old and new bridges. */
371 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
372 svec_add(&old_br, br->name);
374 cfg_get_subsections(&new_br, "bridge");
376 /* Get rid of deleted bridges and add new bridges. */
379 assert(svec_is_unique(&old_br));
380 assert(svec_is_unique(&new_br));
381 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
382 if (!svec_contains(&new_br, br->name)) {
386 for (i = 0; i < new_br.n; i++) {
387 const char *name = new_br.names[i];
388 if (!svec_contains(&old_br, name)) {
392 svec_destroy(&old_br);
393 svec_destroy(&new_br);
397 bridge_configure_ssl();
400 /* Reconfigure all bridges. */
401 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
402 bridge_reconfigure_one(br);
405 /* Add and delete ports on all datapaths.
407 * The kernel will reject any attempt to add a given port to a datapath if
408 * that port already belongs to a different datapath, so we must do all
409 * port deletions before any port additions. */
410 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
411 struct odp_port *dpif_ports;
413 struct svec want_ifaces;
415 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
416 bridge_get_all_ifaces(br, &want_ifaces);
417 for (i = 0; i < n_dpif_ports; i++) {
418 const struct odp_port *p = &dpif_ports[i];
419 if (!svec_contains(&want_ifaces, p->devname)
420 && strcmp(p->devname, br->name)) {
421 int retval = dpif_port_del(br->dpif, p->port);
423 VLOG_ERR("failed to remove %s interface from %s: %s",
424 p->devname, dpif_name(br->dpif),
429 svec_destroy(&want_ifaces);
432 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
433 struct odp_port *dpif_ports;
435 struct svec cur_ifaces, want_ifaces, add_ifaces;
437 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
438 svec_init(&cur_ifaces);
439 for (i = 0; i < n_dpif_ports; i++) {
440 svec_add(&cur_ifaces, dpif_ports[i].devname);
443 svec_sort_unique(&cur_ifaces);
444 bridge_get_all_ifaces(br, &want_ifaces);
445 svec_diff(&want_ifaces, &cur_ifaces, &add_ifaces, NULL, NULL);
447 for (i = 0; i < add_ifaces.n; i++) {
448 const char *if_name = add_ifaces.names[i];
449 int internal = cfg_get_bool(0, "iface.%s.internal", if_name);
450 int flags = internal ? ODP_PORT_INTERNAL : 0;
451 int error = dpif_port_add(br->dpif, if_name, flags, NULL);
452 if (error == EXFULL) {
453 VLOG_ERR("ran out of valid port numbers on %s",
454 dpif_name(br->dpif));
457 VLOG_ERR("failed to add %s interface to %s: %s",
458 if_name, dpif_name(br->dpif), strerror(error));
461 svec_destroy(&cur_ifaces);
462 svec_destroy(&want_ifaces);
463 svec_destroy(&add_ifaces);
465 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
468 struct iface *local_iface = NULL;
470 uint8_t engine_type, engine_id;
471 bool add_id_to_iface = false;
472 struct svec nf_hosts;
474 bridge_fetch_dp_ifaces(br);
475 for (i = 0; i < br->n_ports; ) {
476 struct port *port = br->ports[i];
478 for (j = 0; j < port->n_ifaces; ) {
479 struct iface *iface = port->ifaces[j];
480 if (iface->dp_ifidx < 0) {
481 VLOG_ERR("%s interface not in %s, dropping",
482 iface->name, dpif_name(br->dpif));
483 iface_destroy(iface);
485 if (iface->dp_ifidx == ODPP_LOCAL) {
488 VLOG_DBG("%s has interface %s on port %d",
490 iface->name, iface->dp_ifidx);
494 if (!port->n_ifaces) {
495 VLOG_ERR("%s port has no interfaces, dropping", port->name);
502 /* Pick local port hardware address, datapath ID. */
503 bridge_pick_local_hw_addr(br, ea, &devname);
505 int error = netdev_nodev_set_etheraddr(local_iface->name, ea);
507 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
508 VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge "
509 "Ethernet address: %s",
510 br->name, strerror(error));
514 dpid = bridge_pick_datapath_id(br, ea, devname);
515 ofproto_set_datapath_id(br->ofproto, dpid);
517 /* Set NetFlow configuration on this bridge. */
518 dpif_get_netflow_ids(br->dpif, &engine_type, &engine_id);
519 if (cfg_has("netflow.%s.engine-type", br->name)) {
520 engine_type = cfg_get_int(0, "netflow.%s.engine-type",
523 if (cfg_has("netflow.%s.engine-id", br->name)) {
524 engine_id = cfg_get_int(0, "netflow.%s.engine-id", br->name);
526 if (cfg_has("netflow.%s.add-id-to-iface", br->name)) {
527 add_id_to_iface = cfg_get_bool(0, "netflow.%s.add-id-to-iface",
530 if (add_id_to_iface && engine_id > 0x7f) {
531 VLOG_WARN("bridge %s: netflow port mangling may conflict with "
532 "another vswitch, choose an engine id less than 128",
535 if (add_id_to_iface && br->n_ports > 0x1ff) {
536 VLOG_WARN("bridge %s: netflow port mangling will conflict with "
537 "another port when 512 or more ports are used",
540 svec_init(&nf_hosts);
541 cfg_get_all_keys(&nf_hosts, "netflow.%s.host", br->name);
542 if (ofproto_set_netflow(br->ofproto, &nf_hosts, engine_type,
543 engine_id, add_id_to_iface)) {
544 VLOG_ERR("bridge %s: problem setting netflow collectors",
548 /* Update the controller and related settings. It would be more
549 * straightforward to call this from bridge_reconfigure_one(), but we
550 * can't do it there for two reasons. First, and most importantly, at
551 * that point we don't know the dp_ifidx of any interfaces that have
552 * been added to the bridge (because we haven't actually added them to
553 * the datapath). Second, at that point we haven't set the datapath ID
554 * yet; when a controller is configured, resetting the datapath ID will
555 * immediately disconnect from the controller, so it's better to set
556 * the datapath ID before the controller. */
557 bridge_reconfigure_controller(br);
559 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
560 for (i = 0; i < br->n_ports; i++) {
561 struct port *port = br->ports[i];
562 port_update_vlan_compat(port);
565 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
566 brstp_reconfigure(br);
571 bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN],
572 const char **devname)
574 uint64_t requested_ea;
580 /* Did the user request a particular MAC? */
581 requested_ea = cfg_get_mac(0, "bridge.%s.mac", br->name);
583 eth_addr_from_uint64(requested_ea, ea);
584 if (eth_addr_is_multicast(ea)) {
585 VLOG_ERR("bridge %s: cannot set MAC address to multicast "
586 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
587 } else if (eth_addr_is_zero(ea)) {
588 VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name);
594 /* Otherwise choose the minimum MAC address among all of the interfaces.
595 * (Xen uses FE:FF:FF:FF:FF:FF for virtual interfaces so this will get the
596 * MAC of the physical interface in such an environment.) */
597 memset(ea, 0xff, sizeof ea);
598 for (i = 0; i < br->n_ports; i++) {
599 struct port *port = br->ports[i];
600 if (port->is_mirror_output_port) {
603 for (j = 0; j < port->n_ifaces; j++) {
604 struct iface *iface = port->ifaces[j];
605 uint8_t iface_ea[ETH_ADDR_LEN];
606 if (iface->dp_ifidx == ODPP_LOCAL
607 || cfg_get_bool(0, "iface.%s.internal", iface->name)) {
610 error = netdev_nodev_get_etheraddr(iface->name, iface_ea);
612 if (!eth_addr_is_multicast(iface_ea) &&
613 !eth_addr_is_reserved(iface_ea) &&
614 !eth_addr_is_zero(iface_ea) &&
615 memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) {
616 memcpy(ea, iface_ea, ETH_ADDR_LEN);
617 *devname = iface->name;
620 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
621 VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s",
622 iface->name, strerror(error));
626 if (eth_addr_is_multicast(ea) || eth_addr_is_vif(ea)) {
627 memcpy(ea, br->default_ea, ETH_ADDR_LEN);
629 VLOG_WARN("bridge %s: using default bridge Ethernet "
630 "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea));
632 VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT,
633 br->name, ETH_ADDR_ARGS(ea));
637 /* Choose and returns the datapath ID for bridge 'br' given that the bridge
638 * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of
639 * a network device, then that network device's name must be passed in as
640 * 'devname'; if 'bridge_ea' was derived some other way, then 'devname' must be
641 * passed in as a null pointer. */
643 bridge_pick_datapath_id(struct bridge *br,
644 const uint8_t bridge_ea[ETH_ADDR_LEN],
648 * The procedure for choosing a bridge MAC address will, in the most
649 * ordinary case, also choose a unique MAC that we can use as a datapath
650 * ID. In some special cases, though, multiple bridges will end up with
651 * the same MAC address. This is OK for the bridges, but it will confuse
652 * the OpenFlow controller, because each datapath needs a unique datapath
655 * Datapath IDs must be unique. It is also very desirable that they be
656 * stable from one run to the next, so that policy set on a datapath
661 dpid = cfg_get_dpid(0, "bridge.%s.datapath-id", br->name);
668 if (!netdev_get_vlan_vid(devname, &vlan)) {
670 * A bridge whose MAC address is taken from a VLAN network device
671 * (that is, a network device created with vconfig(8) or similar
672 * tool) will have the same MAC address as a bridge on the VLAN
673 * device's physical network device.
675 * Handle this case by hashing the physical network device MAC
676 * along with the VLAN identifier.
678 uint8_t buf[ETH_ADDR_LEN + 2];
679 memcpy(buf, bridge_ea, ETH_ADDR_LEN);
680 buf[ETH_ADDR_LEN] = vlan >> 8;
681 buf[ETH_ADDR_LEN + 1] = vlan;
682 return dpid_from_hash(buf, sizeof buf);
685 * Assume that this bridge's MAC address is unique, since it
686 * doesn't fit any of the cases we handle specially.
691 * A purely internal bridge, that is, one that has no non-virtual
692 * network devices on it at all, is more difficult because it has no
693 * natural unique identifier at all.
695 * When the host is a XenServer, we handle this case by hashing the
696 * host's UUID with the name of the bridge. Names of bridges are
697 * persistent across XenServer reboots, although they can be reused if
698 * an internal network is destroyed and then a new one is later
699 * created, so this is fairly effective.
701 * When the host is not a XenServer, we punt by using a random MAC
702 * address on each run.
704 const char *host_uuid = xenserver_get_host_uuid();
706 char *combined = xasprintf("%s,%s", host_uuid, br->name);
707 dpid = dpid_from_hash(combined, strlen(combined));
713 return eth_addr_to_uint64(bridge_ea);
717 dpid_from_hash(const void *data, size_t n)
719 uint8_t hash[SHA1_DIGEST_SIZE];
721 BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN);
722 sha1_bytes(data, n, hash);
723 eth_addr_mark_random(hash);
724 return eth_addr_to_uint64(hash);
730 struct bridge *br, *next;
734 LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) {
735 int error = bridge_run_one(br);
737 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
738 VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, "
739 "forcing reconfiguration", br->name);
753 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
754 ofproto_wait(br->ofproto);
755 if (br->controller) {
760 mac_learning_wait(br->ml);
767 /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s
768 * configuration changes. */
770 bridge_flush(struct bridge *br)
772 COVERAGE_INC(bridge_flush);
775 mac_learning_flush(br->ml);
779 /* Bridge unixctl user interface functions. */
781 bridge_unixctl_fdb_show(struct unixctl_conn *conn, const char *args)
783 struct ds ds = DS_EMPTY_INITIALIZER;
784 const struct bridge *br;
786 br = bridge_lookup(args);
788 unixctl_command_reply(conn, 501, "no such bridge");
792 ds_put_cstr(&ds, " port VLAN MAC Age\n");
794 const struct mac_entry *e;
795 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
796 if (e->port < 0 || e->port >= br->n_ports) {
799 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
800 br->ports[e->port]->ifaces[0]->dp_ifidx,
801 e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e));
804 unixctl_command_reply(conn, 200, ds_cstr(&ds));
808 /* Bridge reconfiguration functions. */
810 static struct bridge *
811 bridge_create(const char *name)
816 assert(!bridge_lookup(name));
817 br = xcalloc(1, sizeof *br);
819 error = dpif_create(name, &br->dpif);
820 if (error == EEXIST || error == EBUSY) {
821 error = dpif_open(name, &br->dpif);
823 VLOG_ERR("datapath %s already exists but cannot be opened: %s",
824 name, strerror(error));
828 dpif_flow_flush(br->dpif);
830 VLOG_ERR("failed to create datapath %s: %s", name, strerror(error));
835 error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto);
837 VLOG_ERR("failed to create switch %s: %s", name, strerror(error));
838 dpif_delete(br->dpif);
839 dpif_close(br->dpif);
844 br->name = xstrdup(name);
845 br->ml = mac_learning_create();
846 br->sent_config_request = false;
847 eth_addr_random(br->default_ea);
849 port_array_init(&br->ifaces);
852 br->bond_next_rebalance = time_msec() + 10000;
854 list_push_back(&all_bridges, &br->node);
856 VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif));
862 bridge_destroy(struct bridge *br)
867 while (br->n_ports > 0) {
868 port_destroy(br->ports[br->n_ports - 1]);
870 list_remove(&br->node);
871 error = dpif_delete(br->dpif);
872 if (error && error != ENOENT) {
873 VLOG_ERR("failed to delete %s: %s",
874 dpif_name(br->dpif), strerror(error));
876 dpif_close(br->dpif);
877 ofproto_destroy(br->ofproto);
878 free(br->controller);
879 mac_learning_destroy(br->ml);
880 port_array_destroy(&br->ifaces);
887 static struct bridge *
888 bridge_lookup(const char *name)
892 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
893 if (!strcmp(br->name, name)) {
901 bridge_exists(const char *name)
903 return bridge_lookup(name) ? true : false;
907 bridge_get_datapathid(const char *name)
909 struct bridge *br = bridge_lookup(name);
910 return br ? ofproto_get_datapath_id(br->ofproto) : 0;
914 bridge_run_one(struct bridge *br)
918 error = ofproto_run1(br->ofproto);
924 mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto));
929 error = ofproto_run2(br->ofproto, br->flush);
936 bridge_get_controller(const struct bridge *br)
938 const char *controller;
940 controller = cfg_get_string(0, "bridge.%s.controller", br->name);
942 controller = cfg_get_string(0, "mgmt.controller");
944 return controller && controller[0] ? controller : NULL;
948 bridge_reconfigure_one(struct bridge *br)
950 struct svec old_ports, new_ports, ifaces;
951 struct svec listeners, old_listeners;
952 struct svec snoops, old_snoops;
955 /* Collect old ports. */
956 svec_init(&old_ports);
957 for (i = 0; i < br->n_ports; i++) {
958 svec_add(&old_ports, br->ports[i]->name);
960 svec_sort(&old_ports);
961 assert(svec_is_unique(&old_ports));
963 /* Collect new ports. */
964 svec_init(&new_ports);
965 cfg_get_all_keys(&new_ports, "bridge.%s.port", br->name);
966 svec_sort(&new_ports);
967 if (bridge_get_controller(br)) {
968 char local_name[IF_NAMESIZE];
971 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
972 local_name, sizeof local_name);
973 if (!error && !svec_contains(&new_ports, local_name)) {
974 svec_add(&new_ports, local_name);
975 svec_sort(&new_ports);
978 if (!svec_is_unique(&new_ports)) {
979 VLOG_WARN("bridge %s: %s specified twice as bridge port",
980 br->name, svec_get_duplicate(&new_ports));
981 svec_unique(&new_ports);
984 ofproto_set_mgmt_id(br->ofproto, mgmt_id);
986 /* Get rid of deleted ports and add new ports. */
987 for (i = 0; i < br->n_ports; ) {
988 struct port *port = br->ports[i];
989 if (!svec_contains(&new_ports, port->name)) {
995 for (i = 0; i < new_ports.n; i++) {
996 const char *name = new_ports.names[i];
997 if (!svec_contains(&old_ports, name)) {
998 port_create(br, name);
1001 svec_destroy(&old_ports);
1002 svec_destroy(&new_ports);
1004 /* Reconfigure all ports. */
1005 for (i = 0; i < br->n_ports; i++) {
1006 port_reconfigure(br->ports[i]);
1009 /* Check and delete duplicate interfaces. */
1011 for (i = 0; i < br->n_ports; ) {
1012 struct port *port = br->ports[i];
1013 for (j = 0; j < port->n_ifaces; ) {
1014 struct iface *iface = port->ifaces[j];
1015 if (svec_contains(&ifaces, iface->name)) {
1016 VLOG_ERR("bridge %s: %s interface is on multiple ports, "
1018 br->name, iface->name, port->name);
1019 iface_destroy(iface);
1021 svec_add(&ifaces, iface->name);
1026 if (!port->n_ifaces) {
1027 VLOG_ERR("%s port has no interfaces, dropping", port->name);
1033 svec_destroy(&ifaces);
1035 /* Delete all flows if we're switching from connected to standalone or vice
1036 * versa. (XXX Should we delete all flows if we are switching from one
1037 * controller to another?) */
1039 /* Configure OpenFlow management listeners. */
1040 svec_init(&listeners);
1041 cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name);
1043 svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt",
1044 ovs_rundir, br->name));
1045 } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) {
1046 svec_clear(&listeners);
1048 svec_sort_unique(&listeners);
1050 svec_init(&old_listeners);
1051 ofproto_get_listeners(br->ofproto, &old_listeners);
1052 svec_sort_unique(&old_listeners);
1054 if (!svec_equal(&listeners, &old_listeners)) {
1055 ofproto_set_listeners(br->ofproto, &listeners);
1057 svec_destroy(&listeners);
1058 svec_destroy(&old_listeners);
1060 /* Configure OpenFlow controller connection snooping. */
1062 cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name);
1064 svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop",
1065 ovs_rundir, br->name));
1066 } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) {
1067 svec_clear(&snoops);
1069 svec_sort_unique(&snoops);
1071 svec_init(&old_snoops);
1072 ofproto_get_snoops(br->ofproto, &old_snoops);
1073 svec_sort_unique(&old_snoops);
1075 if (!svec_equal(&snoops, &old_snoops)) {
1076 ofproto_set_snoops(br->ofproto, &snoops);
1078 svec_destroy(&snoops);
1079 svec_destroy(&old_snoops);
1081 mirror_reconfigure(br);
1085 bridge_reconfigure_controller(struct bridge *br)
1087 char *pfx = xasprintf("bridge.%s.controller", br->name);
1088 const char *controller;
1090 controller = bridge_get_controller(br);
1091 if ((br->controller != NULL) != (controller != NULL)) {
1092 ofproto_flush_flows(br->ofproto);
1094 free(br->controller);
1095 br->controller = controller ? xstrdup(controller) : NULL;
1098 const char *fail_mode;
1099 int max_backoff, probe;
1100 int rate_limit, burst_limit;
1102 if (!strcmp(controller, "discover")) {
1103 bool update_resolv_conf = true;
1105 if (cfg_has("%s.update-resolv.conf", pfx)) {
1106 update_resolv_conf = cfg_get_bool(0, "%s.update-resolv.conf",
1109 ofproto_set_discovery(br->ofproto, true,
1110 cfg_get_string(0, "%s.accept-regex", pfx),
1111 update_resolv_conf);
1113 char local_name[IF_NAMESIZE];
1114 struct netdev *netdev;
1118 in_band = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
1120 || cfg_get_bool(0, "%s.in-band", pfx));
1121 ofproto_set_discovery(br->ofproto, false, NULL, NULL);
1122 ofproto_set_in_band(br->ofproto, in_band);
1124 error = dpif_port_get_name(br->dpif, ODPP_LOCAL,
1125 local_name, sizeof local_name);
1127 error = netdev_open(local_name, NETDEV_ETH_TYPE_NONE, &netdev);
1130 if (cfg_is_valid(CFG_IP | CFG_REQUIRED, "%s.ip", pfx)) {
1131 struct in_addr ip, mask, gateway;
1132 ip.s_addr = cfg_get_ip(0, "%s.ip", pfx);
1133 mask.s_addr = cfg_get_ip(0, "%s.netmask", pfx);
1134 gateway.s_addr = cfg_get_ip(0, "%s.gateway", pfx);
1136 netdev_turn_flags_on(netdev, NETDEV_UP, true);
1138 mask.s_addr = guess_netmask(ip.s_addr);
1140 if (!netdev_set_in4(netdev, ip, mask)) {
1141 VLOG_INFO("bridge %s: configured IP address "IP_FMT", "
1143 br->name, IP_ARGS(&ip.s_addr),
1144 IP_ARGS(&mask.s_addr));
1147 if (gateway.s_addr) {
1148 if (!netdev_add_router(gateway)) {
1149 VLOG_INFO("bridge %s: configured gateway "IP_FMT,
1150 br->name, IP_ARGS(&gateway.s_addr));
1154 netdev_close(netdev);
1158 fail_mode = cfg_get_string(0, "%s.fail-mode", pfx);
1160 fail_mode = cfg_get_string(0, "mgmt.fail-mode");
1162 ofproto_set_failure(br->ofproto,
1164 || !strcmp(fail_mode, "standalone")
1165 || !strcmp(fail_mode, "open")));
1167 probe = cfg_get_int(0, "%s.inactivity-probe", pfx);
1169 probe = cfg_get_int(0, "mgmt.inactivity-probe");
1174 ofproto_set_probe_interval(br->ofproto, probe);
1176 max_backoff = cfg_get_int(0, "%s.max-backoff", pfx);
1178 max_backoff = cfg_get_int(0, "mgmt.max-backoff");
1183 ofproto_set_max_backoff(br->ofproto, max_backoff);
1185 rate_limit = cfg_get_int(0, "%s.rate-limit", pfx);
1187 rate_limit = cfg_get_int(0, "mgmt.rate-limit");
1189 burst_limit = cfg_get_int(0, "%s.burst-limit", pfx);
1191 burst_limit = cfg_get_int(0, "mgmt.burst-limit");
1193 ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit);
1195 ofproto_set_stp(br->ofproto, cfg_get_bool(0, "%s.stp", pfx));
1197 if (cfg_has("%s.commands.acl", pfx)) {
1198 struct svec command_acls;
1201 svec_init(&command_acls);
1202 cfg_get_all_strings(&command_acls, "%s.commands.acl", pfx);
1203 command_acl = svec_join(&command_acls, ",", "");
1205 ofproto_set_remote_execution(br->ofproto, command_acl,
1206 cfg_get_string(0, "%s.commands.dir",
1209 svec_destroy(&command_acls);
1212 ofproto_set_remote_execution(br->ofproto, NULL, NULL);
1215 union ofp_action action;
1218 /* Set up a flow that matches every packet and directs them to
1219 * OFPP_NORMAL (which goes to us). */
1220 memset(&action, 0, sizeof action);
1221 action.type = htons(OFPAT_OUTPUT);
1222 action.output.len = htons(sizeof action);
1223 action.output.port = htons(OFPP_NORMAL);
1224 memset(&flow, 0, sizeof flow);
1225 ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0,
1228 ofproto_set_in_band(br->ofproto, false);
1229 ofproto_set_max_backoff(br->ofproto, 1);
1230 ofproto_set_probe_interval(br->ofproto, 5);
1231 ofproto_set_failure(br->ofproto, false);
1232 ofproto_set_stp(br->ofproto, false);
1236 ofproto_set_controller(br->ofproto, br->controller);
1240 bridge_get_all_ifaces(const struct bridge *br, struct svec *ifaces)
1245 for (i = 0; i < br->n_ports; i++) {
1246 struct port *port = br->ports[i];
1247 for (j = 0; j < port->n_ifaces; j++) {
1248 struct iface *iface = port->ifaces[j];
1249 svec_add(ifaces, iface->name);
1253 assert(svec_is_unique(ifaces));
1256 /* For robustness, in case the administrator moves around datapath ports behind
1257 * our back, we re-check all the datapath port numbers here.
1259 * This function will set the 'dp_ifidx' members of interfaces that have
1260 * disappeared to -1, so only call this function from a context where those
1261 * 'struct iface's will be removed from the bridge. Otherwise, the -1
1262 * 'dp_ifidx'es will cause trouble later when we try to send them to the
1263 * datapath, which doesn't support UINT16_MAX+1 ports. */
1265 bridge_fetch_dp_ifaces(struct bridge *br)
1267 struct odp_port *dpif_ports;
1268 size_t n_dpif_ports;
1271 /* Reset all interface numbers. */
1272 for (i = 0; i < br->n_ports; i++) {
1273 struct port *port = br->ports[i];
1274 for (j = 0; j < port->n_ifaces; j++) {
1275 struct iface *iface = port->ifaces[j];
1276 iface->dp_ifidx = -1;
1279 port_array_clear(&br->ifaces);
1281 dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports);
1282 for (i = 0; i < n_dpif_ports; i++) {
1283 struct odp_port *p = &dpif_ports[i];
1284 struct iface *iface = iface_lookup(br, p->devname);
1286 if (iface->dp_ifidx >= 0) {
1287 VLOG_WARN("%s reported interface %s twice",
1288 dpif_name(br->dpif), p->devname);
1289 } else if (iface_from_dp_ifidx(br, p->port)) {
1290 VLOG_WARN("%s reported interface %"PRIu16" twice",
1291 dpif_name(br->dpif), p->port);
1293 port_array_set(&br->ifaces, p->port, iface);
1294 iface->dp_ifidx = p->port;
1301 /* Bridge packet processing functions. */
1304 bond_hash(const uint8_t mac[ETH_ADDR_LEN])
1306 return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK;
1309 static struct bond_entry *
1310 lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN])
1312 return &port->bond_hash[bond_hash(mac)];
1316 bond_choose_iface(const struct port *port)
1319 for (i = 0; i < port->n_ifaces; i++) {
1320 if (port->ifaces[i]->enabled) {
1328 choose_output_iface(const struct port *port, const uint8_t *dl_src,
1329 uint16_t *dp_ifidx, tag_type *tags)
1331 struct iface *iface;
1333 assert(port->n_ifaces);
1334 if (port->n_ifaces == 1) {
1335 iface = port->ifaces[0];
1337 struct bond_entry *e = lookup_bond_entry(port, dl_src);
1338 if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces
1339 || !port->ifaces[e->iface_idx]->enabled) {
1340 /* XXX select interface properly. The current interface selection
1341 * is only good for testing the rebalancing code. */
1342 e->iface_idx = bond_choose_iface(port);
1343 if (e->iface_idx < 0) {
1344 *tags |= port->no_ifaces_tag;
1347 e->iface_tag = tag_create_random();
1349 *tags |= e->iface_tag;
1350 iface = port->ifaces[e->iface_idx];
1352 *dp_ifidx = iface->dp_ifidx;
1353 *tags |= iface->tag; /* Currently only used for bonding. */
1358 bond_link_status_update(struct iface *iface, bool carrier)
1360 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1361 struct port *port = iface->port;
1363 if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) {
1364 /* Nothing to do. */
1367 VLOG_INFO_RL(&rl, "interface %s: carrier %s",
1368 iface->name, carrier ? "detected" : "dropped");
1369 if (carrier == iface->enabled) {
1370 iface->delay_expires = LLONG_MAX;
1371 VLOG_INFO_RL(&rl, "interface %s: will not be %s",
1372 iface->name, carrier ? "disabled" : "enabled");
1373 } else if (carrier && port->updelay && port->active_iface < 0) {
1374 iface->delay_expires = time_msec();
1375 VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no "
1376 "other interface is up", iface->name, port->updelay);
1378 int delay = carrier ? port->updelay : port->downdelay;
1379 iface->delay_expires = time_msec() + delay;
1382 "interface %s: will be %s if it stays %s for %d ms",
1384 carrier ? "enabled" : "disabled",
1385 carrier ? "up" : "down",
1392 bond_choose_active_iface(struct port *port)
1394 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1396 port->active_iface = bond_choose_iface(port);
1397 port->active_iface_tag = tag_create_random();
1398 if (port->active_iface >= 0) {
1399 VLOG_INFO_RL(&rl, "port %s: active interface is now %s",
1400 port->name, port->ifaces[port->active_iface]->name);
1402 VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface",
1408 bond_enable_slave(struct iface *iface, bool enable)
1410 struct port *port = iface->port;
1411 struct bridge *br = port->bridge;
1413 iface->delay_expires = LLONG_MAX;
1414 if (enable == iface->enabled) {
1418 iface->enabled = enable;
1419 if (!iface->enabled) {
1420 VLOG_WARN("interface %s: disabled", iface->name);
1421 ofproto_revalidate(br->ofproto, iface->tag);
1422 if (iface->port_ifidx == port->active_iface) {
1423 ofproto_revalidate(br->ofproto,
1424 port->active_iface_tag);
1425 bond_choose_active_iface(port);
1427 bond_send_learning_packets(port);
1429 VLOG_WARN("interface %s: enabled", iface->name);
1430 if (port->active_iface < 0) {
1431 ofproto_revalidate(br->ofproto, port->no_ifaces_tag);
1432 bond_choose_active_iface(port);
1433 bond_send_learning_packets(port);
1435 iface->tag = tag_create_random();
1440 bond_run(struct bridge *br)
1444 for (i = 0; i < br->n_ports; i++) {
1445 struct port *port = br->ports[i];
1446 if (port->n_ifaces < 2) {
1449 for (j = 0; j < port->n_ifaces; j++) {
1450 struct iface *iface = port->ifaces[j];
1451 if (time_msec() >= iface->delay_expires) {
1452 bond_enable_slave(iface, !iface->enabled);
1459 bond_wait(struct bridge *br)
1463 for (i = 0; i < br->n_ports; i++) {
1464 struct port *port = br->ports[i];
1465 if (port->n_ifaces < 2) {
1468 for (j = 0; j < port->n_ifaces; j++) {
1469 struct iface *iface = port->ifaces[j];
1470 if (iface->delay_expires != LLONG_MAX) {
1471 poll_timer_wait(iface->delay_expires - time_msec());
1478 set_dst(struct dst *p, const flow_t *flow,
1479 const struct port *in_port, const struct port *out_port,
1484 * XXX This uses too many tags: any broadcast flow will get one tag per
1485 * destination port, and thus a broadcast on a switch of any size is likely
1486 * to have all tag bits set. We should figure out a way to be smarter.
1488 * This is OK when STP is disabled, because stp_state_tag is 0 then. */
1489 *tags |= out_port->stp_state_tag;
1490 if (!(out_port->stp_state & (STP_DISABLED | STP_FORWARDING))) {
1494 p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE
1495 : in_port->vlan >= 0 ? in_port->vlan
1496 : ntohs(flow->dl_vlan));
1497 return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags);
1501 swap_dst(struct dst *p, struct dst *q)
1503 struct dst tmp = *p;
1508 /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in
1509 * 'dsts'. (This may help performance by reducing the number of VLAN changes
1510 * that we push to the datapath. We could in fact fully sort the array by
1511 * vlan, but in most cases there are at most two different vlan tags so that's
1512 * possibly overkill.) */
1514 partition_dsts(struct dst *dsts, size_t n_dsts, int vlan)
1516 struct dst *first = dsts;
1517 struct dst *last = dsts + n_dsts;
1519 while (first != last) {
1521 * - All dsts < first have vlan == 'vlan'.
1522 * - All dsts >= last have vlan != 'vlan'.
1523 * - first < last. */
1524 while (first->vlan == vlan) {
1525 if (++first == last) {
1530 /* Same invariants, plus one additional:
1531 * - first->vlan != vlan.
1533 while (last[-1].vlan != vlan) {
1534 if (--last == first) {
1539 /* Same invariants, plus one additional:
1540 * - last[-1].vlan == vlan.*/
1541 swap_dst(first++, --last);
1546 mirror_mask_ffs(mirror_mask_t mask)
1548 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
1553 dst_is_duplicate(const struct dst *dsts, size_t n_dsts,
1554 const struct dst *test)
1557 for (i = 0; i < n_dsts; i++) {
1558 if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) {
1566 port_trunks_vlan(const struct port *port, uint16_t vlan)
1568 return port->vlan < 0 && bitmap_is_set(port->trunks, vlan);
1572 port_includes_vlan(const struct port *port, uint16_t vlan)
1574 return vlan == port->vlan || port_trunks_vlan(port, vlan);
1578 compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan,
1579 const struct port *in_port, const struct port *out_port,
1580 struct dst dsts[], tag_type *tags)
1582 mirror_mask_t mirrors = in_port->src_mirrors;
1583 struct dst *dst = dsts;
1586 *tags |= in_port->stp_state_tag;
1587 if (out_port == FLOOD_PORT) {
1588 /* XXX use ODP_FLOOD if no vlans or bonding. */
1589 /* XXX even better, define each VLAN as a datapath port group */
1590 for (i = 0; i < br->n_ports; i++) {
1591 struct port *port = br->ports[i];
1592 if (port != in_port && port_includes_vlan(port, vlan)
1593 && !port->is_mirror_output_port
1594 && set_dst(dst, flow, in_port, port, tags)) {
1595 mirrors |= port->dst_mirrors;
1599 } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) {
1600 mirrors |= out_port->dst_mirrors;
1605 struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1];
1606 if (!m->n_vlans || vlan_is_mirrored(m, vlan)) {
1608 if (set_dst(dst, flow, in_port, m->out_port, tags)
1609 && !dst_is_duplicate(dsts, dst - dsts, dst)) {
1613 for (i = 0; i < br->n_ports; i++) {
1614 struct port *port = br->ports[i];
1615 if (port_includes_vlan(port, m->out_vlan)
1616 && set_dst(dst, flow, in_port, port, tags)
1617 && !dst_is_duplicate(dsts, dst - dsts, dst))
1619 if (port->vlan < 0) {
1620 dst->vlan = m->out_vlan;
1622 if (dst->dp_ifidx == flow->in_port
1623 && dst->vlan == vlan) {
1624 /* Don't send out input port on same VLAN. */
1632 mirrors &= mirrors - 1;
1635 partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan));
1640 print_dsts(const struct dst *dsts, size_t n)
1642 for (; n--; dsts++) {
1643 printf(">p%"PRIu16, dsts->dp_ifidx);
1644 if (dsts->vlan != OFP_VLAN_NONE) {
1645 printf("v%"PRIu16, dsts->vlan);
1651 compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan,
1652 const struct port *in_port, const struct port *out_port,
1653 tag_type *tags, struct odp_actions *actions)
1655 struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)];
1657 const struct dst *p;
1660 n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags);
1662 cur_vlan = ntohs(flow->dl_vlan);
1663 for (p = dsts; p < &dsts[n_dsts]; p++) {
1664 union odp_action *a;
1665 if (p->vlan != cur_vlan) {
1666 if (p->vlan == OFP_VLAN_NONE) {
1667 odp_actions_add(actions, ODPAT_STRIP_VLAN);
1669 a = odp_actions_add(actions, ODPAT_SET_VLAN_VID);
1670 a->vlan_vid.vlan_vid = htons(p->vlan);
1674 a = odp_actions_add(actions, ODPAT_OUTPUT);
1675 a->output.port = p->dp_ifidx;
1680 is_bcast_arp_reply(const flow_t *flow, const struct ofpbuf *packet)
1682 struct arp_eth_header *arp = (struct arp_eth_header *) packet->data;
1683 return (flow->dl_type == htons(ETH_TYPE_ARP)
1684 && eth_addr_is_broadcast(flow->dl_dst)
1685 && packet->size >= sizeof(struct arp_eth_header)
1686 && arp->ar_op == ARP_OP_REQUEST);
1689 /* If the composed actions may be applied to any packet in the given 'flow',
1690 * returns true. Otherwise, the actions should only be applied to 'packet', or
1691 * not at all, if 'packet' was NULL. */
1693 process_flow(struct bridge *br, const flow_t *flow,
1694 const struct ofpbuf *packet, struct odp_actions *actions,
1697 struct iface *in_iface;
1698 struct port *in_port;
1699 struct port *out_port = NULL; /* By default, drop the packet/flow. */
1702 /* Find the interface and port structure for the received packet. */
1703 in_iface = iface_from_dp_ifidx(br, flow->in_port);
1705 /* No interface? Something fishy... */
1706 if (packet != NULL) {
1707 /* Odd. A few possible reasons here:
1709 * - We deleted an interface but there are still a few packets
1710 * queued up from it.
1712 * - Someone externally added an interface (e.g. with "ovs-dpctl
1713 * add-if") that we don't know about.
1715 * - Packet arrived on the local port but the local port is not
1716 * one of our bridge ports.
1718 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1720 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1721 "interface %"PRIu16, br->name, flow->in_port);
1724 /* Return without adding any actions, to drop packets on this flow. */
1727 in_port = in_iface->port;
1729 /* Figure out what VLAN this packet belongs to.
1731 * Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet
1732 * belongs to VLAN 0, so we should treat both cases identically. (In the
1733 * former case, the packet has an 802.1Q header that specifies VLAN 0,
1734 * presumably to allow a priority to be specified. In the latter case, the
1735 * packet does not have any 802.1Q header.) */
1736 vlan = ntohs(flow->dl_vlan);
1737 if (vlan == OFP_VLAN_NONE) {
1740 if (in_port->vlan >= 0) {
1742 /* XXX support double tagging? */
1743 if (packet != NULL) {
1744 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1745 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
1746 "packet received on port %s configured with "
1747 "implicit VLAN %"PRIu16,
1748 br->name, ntohs(flow->dl_vlan),
1749 in_port->name, in_port->vlan);
1753 vlan = in_port->vlan;
1755 if (!port_includes_vlan(in_port, vlan)) {
1756 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1757 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged "
1758 "packet received on port %s not configured for "
1760 br->name, vlan, in_port->name, vlan);
1765 /* Drop frames for ports that STP wants entirely killed (both for
1766 * forwarding and for learning). Later, after we do learning, we'll drop
1767 * the frames that STP wants to do learning but not forwarding on. */
1768 if (in_port->stp_state & (STP_LISTENING | STP_BLOCKING)) {
1772 /* Drop frames for reserved multicast addresses. */
1773 if (eth_addr_is_reserved(flow->dl_dst)) {
1777 /* Drop frames on ports reserved for mirroring. */
1778 if (in_port->is_mirror_output_port) {
1779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1780 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, "
1781 "which is reserved exclusively for mirroring",
1782 br->name, in_port->name);
1786 /* Multicast (and broadcast) packets on bonds need special attention, to
1787 * avoid receiving duplicates. */
1788 if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
1789 *tags |= in_port->active_iface_tag;
1790 if (in_port->active_iface != in_iface->port_ifidx) {
1791 /* Drop all multicast packets on inactive slaves. */
1794 /* Drop all multicast packets for which we have learned a different
1795 * input port, because we probably sent the packet on one slaves
1796 * and got it back on the active slave. Broadcast ARP replies are
1797 * an exception to this rule: the host has moved to another
1799 int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
1800 if (src_idx != -1 && src_idx != in_port->port_idx) {
1802 if (!is_bcast_arp_reply(flow, packet)) {
1806 /* No way to know whether it's an ARP reply, because the
1807 * flow entry doesn't include enough information and we
1808 * don't have a packet. Punt. */
1816 out_port = FLOOD_PORT;
1820 /* Learn source MAC (but don't try to learn from revalidation). */
1822 tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src,
1823 vlan, in_port->port_idx);
1825 /* The log messages here could actually be useful in debugging,
1826 * so keep the rate limit relatively high. */
1827 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30,
1829 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1830 "on port %s in VLAN %d",
1831 br->name, ETH_ADDR_ARGS(flow->dl_src),
1832 in_port->name, vlan);
1833 ofproto_revalidate(br->ofproto, rev_tag);
1837 /* Determine output port. */
1838 out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan,
1840 if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
1841 out_port = br->ports[out_port_idx];
1845 /* Don't send packets out their input ports. Don't forward frames that STP
1846 * wants us to discard. */
1847 if (in_port == out_port || in_port->stp_state == STP_LEARNING) {
1852 compose_actions(br, flow, vlan, in_port, out_port, tags, actions);
1855 * We send out only a single packet, instead of setting up a flow, if the
1856 * packet is an ARP directed to broadcast that arrived on a bonded
1857 * interface. In such a situation ARP requests and replies must be handled
1858 * differently, but OpenFlow unfortunately can't distinguish them.
1860 return (in_port->n_ifaces < 2
1861 || flow->dl_type != htons(ETH_TYPE_ARP)
1862 || !eth_addr_is_broadcast(flow->dl_dst));
1865 /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port
1868 bridge_port_changed_ofhook_cb(enum ofp_port_reason reason,
1869 const struct ofp_phy_port *opp,
1872 struct bridge *br = br_;
1873 struct iface *iface;
1876 iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no));
1882 if (reason == OFPPR_DELETE) {
1883 VLOG_WARN("bridge %s: interface %s deleted unexpectedly",
1884 br->name, iface->name);
1885 iface_destroy(iface);
1886 if (!port->n_ifaces) {
1887 VLOG_WARN("bridge %s: port %s has no interfaces, dropping",
1888 br->name, port->name);
1894 memcpy(iface->mac, opp->hw_addr, ETH_ADDR_LEN);
1895 if (port->n_ifaces > 1) {
1896 bool up = !(opp->state & OFPPS_LINK_DOWN);
1897 bond_link_status_update(iface, up);
1898 port_update_bond_compat(port);
1904 bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet,
1905 struct odp_actions *actions, tag_type *tags, void *br_)
1907 struct bridge *br = br_;
1910 if (flow->dl_type == htons(OFP_DL_TYPE_NOT_ETH_TYPE)
1911 && eth_addr_equals(flow->dl_dst, stp_eth_addr)) {
1912 brstp_receive(br, flow, payload);
1917 COVERAGE_INC(bridge_process_flow);
1918 return process_flow(br, flow, packet, actions, tags);
1922 bridge_account_flow_ofhook_cb(const flow_t *flow,
1923 const union odp_action *actions,
1924 size_t n_actions, unsigned long long int n_bytes,
1927 struct bridge *br = br_;
1928 const union odp_action *a;
1930 if (!br->has_bonded_ports) {
1934 for (a = actions; a < &actions[n_actions]; a++) {
1935 if (a->type == ODPAT_OUTPUT) {
1936 struct port *port = port_from_dp_ifidx(br, a->output.port);
1937 if (port && port->n_ifaces >= 2) {
1938 struct bond_entry *e = lookup_bond_entry(port, flow->dl_src);
1939 e->tx_bytes += n_bytes;
1946 bridge_account_checkpoint_ofhook_cb(void *br_)
1948 struct bridge *br = br_;
1951 if (!br->has_bonded_ports) {
1955 /* The current ofproto implementation calls this callback at least once a
1956 * second, so this timer implementation is sufficient. */
1957 if (time_msec() < br->bond_next_rebalance) {
1960 br->bond_next_rebalance = time_msec() + 10000;
1962 for (i = 0; i < br->n_ports; i++) {
1963 struct port *port = br->ports[i];
1964 if (port->n_ifaces > 1) {
1965 bond_rebalance_port(port);
1970 static struct ofhooks bridge_ofhooks = {
1971 bridge_port_changed_ofhook_cb,
1972 bridge_normal_ofhook_cb,
1973 bridge_account_flow_ofhook_cb,
1974 bridge_account_checkpoint_ofhook_cb,
1977 /* Bonding functions. */
1979 /* Statistics for a single interface on a bonded port, used for load-based
1980 * bond rebalancing. */
1981 struct slave_balance {
1982 struct iface *iface; /* The interface. */
1983 uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */
1985 /* All the "bond_entry"s that are assigned to this interface, in order of
1986 * increasing tx_bytes. */
1987 struct bond_entry **hashes;
1991 /* Sorts pointers to pointers to bond_entries in ascending order by the
1992 * interface to which they are assigned, and within a single interface in
1993 * ascending order of bytes transmitted. */
1995 compare_bond_entries(const void *a_, const void *b_)
1997 const struct bond_entry *const *ap = a_;
1998 const struct bond_entry *const *bp = b_;
1999 const struct bond_entry *a = *ap;
2000 const struct bond_entry *b = *bp;
2001 if (a->iface_idx != b->iface_idx) {
2002 return a->iface_idx > b->iface_idx ? 1 : -1;
2003 } else if (a->tx_bytes != b->tx_bytes) {
2004 return a->tx_bytes > b->tx_bytes ? 1 : -1;
2010 /* Sorts slave_balances so that enabled ports come first, and otherwise in
2011 * *descending* order by number of bytes transmitted. */
2013 compare_slave_balance(const void *a_, const void *b_)
2015 const struct slave_balance *a = a_;
2016 const struct slave_balance *b = b_;
2017 if (a->iface->enabled != b->iface->enabled) {
2018 return a->iface->enabled ? -1 : 1;
2019 } else if (a->tx_bytes != b->tx_bytes) {
2020 return a->tx_bytes > b->tx_bytes ? -1 : 1;
2027 swap_bals(struct slave_balance *a, struct slave_balance *b)
2029 struct slave_balance tmp = *a;
2034 /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order
2035 * given that 'p' (and only 'p') might be in the wrong location.
2037 * This function invalidates 'p', since it might now be in a different memory
2040 resort_bals(struct slave_balance *p,
2041 struct slave_balance bals[], size_t n_bals)
2044 for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) {
2045 swap_bals(p, p - 1);
2047 for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) {
2048 swap_bals(p, p + 1);
2054 log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port)
2056 if (VLOG_IS_DBG_ENABLED()) {
2057 struct ds ds = DS_EMPTY_INITIALIZER;
2058 const struct slave_balance *b;
2060 for (b = bals; b < bals + n_bals; b++) {
2064 ds_put_char(&ds, ',');
2066 ds_put_format(&ds, " %s %"PRIu64"kB",
2067 b->iface->name, b->tx_bytes / 1024);
2069 if (!b->iface->enabled) {
2070 ds_put_cstr(&ds, " (disabled)");
2072 if (b->n_hashes > 0) {
2073 ds_put_cstr(&ds, " (");
2074 for (i = 0; i < b->n_hashes; i++) {
2075 const struct bond_entry *e = b->hashes[i];
2077 ds_put_cstr(&ds, " + ");
2079 ds_put_format(&ds, "h%td: %"PRIu64"kB",
2080 e - port->bond_hash, e->tx_bytes / 1024);
2082 ds_put_cstr(&ds, ")");
2085 VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds));
2090 /* Shifts 'hash' from 'from' to 'to' within 'port'. */
2092 bond_shift_load(struct slave_balance *from, struct slave_balance *to,
2093 struct bond_entry *hash)
2095 struct port *port = from->iface->port;
2096 uint64_t delta = hash->tx_bytes;
2098 VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) "
2099 "from %s to %s (now carrying %"PRIu64"kB and "
2100 "%"PRIu64"kB load, respectively)",
2101 port->name, delta / 1024, hash - port->bond_hash,
2102 from->iface->name, to->iface->name,
2103 (from->tx_bytes - delta) / 1024,
2104 (to->tx_bytes + delta) / 1024);
2106 /* Delete element from from->hashes.
2108 * We don't bother to add the element to to->hashes because not only would
2109 * it require more work, the only purpose it would be to allow that hash to
2110 * be migrated to another slave in this rebalancing run, and there is no
2111 * point in doing that. */
2112 if (from->hashes[0] == hash) {
2115 int i = hash - from->hashes[0];
2116 memmove(from->hashes + i, from->hashes + i + 1,
2117 (from->n_hashes - (i + 1)) * sizeof *from->hashes);
2121 /* Shift load away from 'from' to 'to'. */
2122 from->tx_bytes -= delta;
2123 to->tx_bytes += delta;
2125 /* Arrange for flows to be revalidated. */
2126 ofproto_revalidate(port->bridge->ofproto, hash->iface_tag);
2127 hash->iface_idx = to->iface->port_ifidx;
2128 hash->iface_tag = tag_create_random();
2132 bond_rebalance_port(struct port *port)
2134 struct slave_balance bals[DP_MAX_PORTS];
2136 struct bond_entry *hashes[BOND_MASK + 1];
2137 struct slave_balance *b, *from, *to;
2138 struct bond_entry *e;
2141 /* Sets up 'bals' to describe each of the port's interfaces, sorted in
2142 * descending order of tx_bytes, so that bals[0] represents the most
2143 * heavily loaded slave and bals[n_bals - 1] represents the least heavily
2146 * The code is a bit tricky: to avoid dynamically allocating a 'hashes'
2147 * array for each slave_balance structure, we sort our local array of
2148 * hashes in order by slave, so that all of the hashes for a given slave
2149 * become contiguous in memory, and then we point each 'hashes' members of
2150 * a slave_balance structure to the start of a contiguous group. */
2151 n_bals = port->n_ifaces;
2152 for (b = bals; b < &bals[n_bals]; b++) {
2153 b->iface = port->ifaces[b - bals];
2158 for (i = 0; i <= BOND_MASK; i++) {
2159 hashes[i] = &port->bond_hash[i];
2161 qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries);
2162 for (i = 0; i <= BOND_MASK; i++) {
2164 if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) {
2165 b = &bals[e->iface_idx];
2166 b->tx_bytes += e->tx_bytes;
2168 b->hashes = &hashes[i];
2173 qsort(bals, n_bals, sizeof *bals, compare_slave_balance);
2174 log_bals(bals, n_bals, port);
2176 /* Discard slaves that aren't enabled (which were sorted to the back of the
2177 * array earlier). */
2178 while (!bals[n_bals - 1].iface->enabled) {
2185 /* Shift load from the most-loaded slaves to the least-loaded slaves. */
2186 to = &bals[n_bals - 1];
2187 for (from = bals; from < to; ) {
2188 uint64_t overload = from->tx_bytes - to->tx_bytes;
2189 if (overload < to->tx_bytes >> 5 || overload < 100000) {
2190 /* The extra load on 'from' (and all less-loaded slaves), compared
2191 * to that of 'to' (the least-loaded slave), is less than ~3%, or
2192 * it is less than ~1Mbps. No point in rebalancing. */
2194 } else if (from->n_hashes == 1) {
2195 /* 'from' only carries a single MAC hash, so we can't shift any
2196 * load away from it, even though we want to. */
2199 /* 'from' is carrying significantly more load than 'to', and that
2200 * load is split across at least two different hashes. Pick a hash
2201 * to migrate to 'to' (the least-loaded slave), given that doing so
2202 * must not cause 'to''s load to exceed 'from''s load.
2204 * The sort order we use means that we prefer to shift away the
2205 * smallest hashes instead of the biggest ones. There is little
2206 * reason behind this decision; we could use the opposite sort
2207 * order to shift away big hashes ahead of small ones. */
2210 for (i = 0; i < from->n_hashes; i++) {
2211 uint64_t delta = from->hashes[i]->tx_bytes;
2212 if (to->tx_bytes + delta < from->tx_bytes - delta) {
2216 if (i < from->n_hashes) {
2217 bond_shift_load(from, to, from->hashes[i]);
2219 /* Re-sort 'bals'. Note that this may make 'from' and 'to'
2220 * point to different slave_balance structures. It is only
2221 * valid to do these two operations in a row at all because we
2222 * know that 'from' will not move past 'to' and vice versa. */
2223 resort_bals(from, bals, n_bals);
2224 resort_bals(to, bals, n_bals);
2231 /* Implement exponentially weighted moving average. A weight of 1/2 causes
2232 * historical data to decay to <1% in 7 rebalancing runs. */
2233 for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) {
2239 bond_send_learning_packets(struct port *port)
2241 struct bridge *br = port->bridge;
2242 struct mac_entry *e;
2243 struct ofpbuf packet;
2244 int error, n_packets, n_errors;
2246 if (!port->n_ifaces || port->active_iface < 0 || !br->ml) {
2250 ofpbuf_init(&packet, 128);
2251 error = n_packets = n_errors = 0;
2252 LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) {
2253 static const char s[] = "Open vSwitch Bond Failover";
2254 union ofp_action actions[2], *a;
2255 struct eth_header *eth;
2256 struct llc_snap_header *llc_snap;
2262 if (e->port == port->port_idx
2263 || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) {
2267 /* Compose packet to send. */
2268 ofpbuf_clear(&packet);
2269 eth = ofpbuf_put_zeros(&packet, ETH_HEADER_LEN);
2270 llc_snap = ofpbuf_put_zeros(&packet, LLC_SNAP_HEADER_LEN);
2271 ofpbuf_put(&packet, s, sizeof s); /* Includes null byte. */
2272 ofpbuf_put(&packet, e->mac, ETH_ADDR_LEN);
2274 memcpy(eth->eth_dst, eth_addr_broadcast, ETH_ADDR_LEN);
2275 memcpy(eth->eth_src, e->mac, ETH_ADDR_LEN);
2276 eth->eth_type = htons(packet.size - ETH_HEADER_LEN);
2278 llc_snap->llc.llc_dsap = LLC_DSAP_SNAP;
2279 llc_snap->llc.llc_ssap = LLC_SSAP_SNAP;
2280 llc_snap->llc.llc_cntl = LLC_CNTL_SNAP;
2281 memcpy(llc_snap->snap.snap_org, "\x00\x23\x20", 3);
2282 llc_snap->snap.snap_type = htons(0xf177); /* Random number. */
2284 /* Compose actions. */
2285 memset(actions, 0, sizeof actions);
2288 a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID);
2289 a->vlan_vid.len = htons(sizeof *a);
2290 a->vlan_vid.vlan_vid = htons(e->vlan);
2293 a->output.type = htons(OFPAT_OUTPUT);
2294 a->output.len = htons(sizeof *a);
2295 a->output.port = htons(odp_port_to_ofp_port(dp_ifidx));
2300 flow_extract(&packet, ODPP_NONE, &flow);
2301 retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions,
2308 ofpbuf_uninit(&packet);
2311 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2312 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2313 "packets, last error was: %s",
2314 port->name, n_errors, n_packets, strerror(error));
2316 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2317 port->name, n_packets);
2321 /* Bonding unixctl user interface functions. */
2324 bond_unixctl_list(struct unixctl_conn *conn, const char *args UNUSED)
2326 struct ds ds = DS_EMPTY_INITIALIZER;
2327 const struct bridge *br;
2329 ds_put_cstr(&ds, "bridge\tbond\tslaves\n");
2331 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2334 for (i = 0; i < br->n_ports; i++) {
2335 const struct port *port = br->ports[i];
2336 if (port->n_ifaces > 1) {
2339 ds_put_format(&ds, "%s\t%s\t", br->name, port->name);
2340 for (j = 0; j < port->n_ifaces; j++) {
2341 const struct iface *iface = port->ifaces[j];
2343 ds_put_cstr(&ds, ", ");
2345 ds_put_cstr(&ds, iface->name);
2347 ds_put_char(&ds, '\n');
2351 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2355 static struct port *
2356 bond_find(const char *name)
2358 const struct bridge *br;
2360 LIST_FOR_EACH (br, struct bridge, node, &all_bridges) {
2363 for (i = 0; i < br->n_ports; i++) {
2364 struct port *port = br->ports[i];
2365 if (!strcmp(port->name, name) && port->n_ifaces > 1) {
2374 bond_unixctl_show(struct unixctl_conn *conn, const char *args)
2376 struct ds ds = DS_EMPTY_INITIALIZER;
2377 const struct port *port;
2380 port = bond_find(args);
2382 unixctl_command_reply(conn, 501, "no such bond");
2386 ds_put_format(&ds, "updelay: %d ms\n", port->updelay);
2387 ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay);
2388 ds_put_format(&ds, "next rebalance: %lld ms\n",
2389 port->bridge->bond_next_rebalance - time_msec());
2390 for (j = 0; j < port->n_ifaces; j++) {
2391 const struct iface *iface = port->ifaces[j];
2392 struct bond_entry *be;
2395 ds_put_format(&ds, "slave %s: %s\n",
2396 iface->name, iface->enabled ? "enabled" : "disabled");
2397 if (j == port->active_iface) {
2398 ds_put_cstr(&ds, "\tactive slave\n");
2400 if (iface->delay_expires != LLONG_MAX) {
2401 ds_put_format(&ds, "\t%s expires in %lld ms\n",
2402 iface->enabled ? "downdelay" : "updelay",
2403 iface->delay_expires - time_msec());
2407 for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) {
2408 int hash = be - port->bond_hash;
2409 struct mac_entry *me;
2411 if (be->iface_idx != j) {
2415 ds_put_format(&ds, "\thash %d: %lld kB load\n",
2416 hash, be->tx_bytes / 1024);
2419 if (!port->bridge->ml) {
2423 LIST_FOR_EACH (me, struct mac_entry, lru_node,
2424 &port->bridge->ml->lrus) {
2427 if (bond_hash(me->mac) == hash
2428 && me->port != port->port_idx
2429 && choose_output_iface(port, me->mac, &dp_ifidx, &tags)
2430 && dp_ifidx == iface->dp_ifidx)
2432 ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n",
2433 ETH_ADDR_ARGS(me->mac));
2438 unixctl_command_reply(conn, 200, ds_cstr(&ds));
2443 bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_)
2445 char *args = (char *) args_;
2446 char *save_ptr = NULL;
2447 char *bond_s, *hash_s, *slave_s;
2448 uint8_t mac[ETH_ADDR_LEN];
2450 struct iface *iface;
2451 struct bond_entry *entry;
2454 bond_s = strtok_r(args, " ", &save_ptr);
2455 hash_s = strtok_r(NULL, " ", &save_ptr);
2456 slave_s = strtok_r(NULL, " ", &save_ptr);
2458 unixctl_command_reply(conn, 501,
2459 "usage: bond/migrate BOND HASH SLAVE");
2463 port = bond_find(bond_s);
2465 unixctl_command_reply(conn, 501, "no such bond");
2469 if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac))
2470 == ETH_ADDR_SCAN_COUNT) {
2471 hash = bond_hash(mac);
2472 } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) {
2473 hash = atoi(hash_s) & BOND_MASK;
2475 unixctl_command_reply(conn, 501, "bad hash");
2479 iface = port_lookup_iface(port, slave_s);
2481 unixctl_command_reply(conn, 501, "no such slave");
2485 if (!iface->enabled) {
2486 unixctl_command_reply(conn, 501, "cannot migrate to disabled slave");
2490 entry = &port->bond_hash[hash];
2491 ofproto_revalidate(port->bridge->ofproto, entry->iface_tag);
2492 entry->iface_idx = iface->port_ifidx;
2493 entry->iface_tag = tag_create_random();
2494 unixctl_command_reply(conn, 200, "migrated");
2498 bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_)
2500 char *args = (char *) args_;
2501 char *save_ptr = NULL;
2502 char *bond_s, *slave_s;
2504 struct iface *iface;
2506 bond_s = strtok_r(args, " ", &save_ptr);
2507 slave_s = strtok_r(NULL, " ", &save_ptr);
2509 unixctl_command_reply(conn, 501,
2510 "usage: bond/set-active-slave BOND SLAVE");
2514 port = bond_find(bond_s);
2516 unixctl_command_reply(conn, 501, "no such bond");
2520 iface = port_lookup_iface(port, slave_s);
2522 unixctl_command_reply(conn, 501, "no such slave");
2526 if (!iface->enabled) {
2527 unixctl_command_reply(conn, 501, "cannot make disabled slave active");
2531 if (port->active_iface != iface->port_ifidx) {
2532 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2533 port->active_iface = iface->port_ifidx;
2534 port->active_iface_tag = tag_create_random();
2535 VLOG_INFO("port %s: active interface is now %s",
2536 port->name, iface->name);
2537 bond_send_learning_packets(port);
2538 unixctl_command_reply(conn, 200, "done");
2540 unixctl_command_reply(conn, 200, "no change");
2545 enable_slave(struct unixctl_conn *conn, const char *args_, bool enable)
2547 char *args = (char *) args_;
2548 char *save_ptr = NULL;
2549 char *bond_s, *slave_s;
2551 struct iface *iface;
2553 bond_s = strtok_r(args, " ", &save_ptr);
2554 slave_s = strtok_r(NULL, " ", &save_ptr);
2556 unixctl_command_reply(conn, 501,
2557 "usage: bond/enable/disable-slave BOND SLAVE");
2561 port = bond_find(bond_s);
2563 unixctl_command_reply(conn, 501, "no such bond");
2567 iface = port_lookup_iface(port, slave_s);
2569 unixctl_command_reply(conn, 501, "no such slave");
2573 bond_enable_slave(iface, enable);
2574 unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled");
2578 bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args)
2580 enable_slave(conn, args, true);
2584 bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args)
2586 enable_slave(conn, args, false);
2592 unixctl_command_register("bond/list", bond_unixctl_list);
2593 unixctl_command_register("bond/show", bond_unixctl_show);
2594 unixctl_command_register("bond/migrate", bond_unixctl_migrate);
2595 unixctl_command_register("bond/set-active-slave",
2596 bond_unixctl_set_active_slave);
2597 unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave);
2598 unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave);
2601 /* Port functions. */
2604 port_create(struct bridge *br, const char *name)
2608 port = xcalloc(1, sizeof *port);
2610 port->port_idx = br->n_ports;
2612 port->trunks = NULL;
2613 port->name = xstrdup(name);
2614 port->active_iface = -1;
2615 port->stp_state = STP_DISABLED;
2616 port->stp_state_tag = 0;
2618 if (br->n_ports >= br->allocated_ports) {
2619 br->ports = x2nrealloc(br->ports, &br->allocated_ports,
2622 br->ports[br->n_ports++] = port;
2624 VLOG_INFO("created port %s on bridge %s", port->name, br->name);
2629 port_reconfigure(struct port *port)
2631 bool bonded = cfg_has_section("bonding.%s", port->name);
2632 struct svec old_ifaces, new_ifaces;
2633 unsigned long *trunks;
2637 /* Collect old and new interfaces. */
2638 svec_init(&old_ifaces);
2639 svec_init(&new_ifaces);
2640 for (i = 0; i < port->n_ifaces; i++) {
2641 svec_add(&old_ifaces, port->ifaces[i]->name);
2643 svec_sort(&old_ifaces);
2645 cfg_get_all_keys(&new_ifaces, "bonding.%s.slave", port->name);
2646 if (!new_ifaces.n) {
2647 VLOG_ERR("port %s: no interfaces specified for bonded port",
2649 } else if (new_ifaces.n == 1) {
2650 VLOG_WARN("port %s: only 1 interface specified for bonded port",
2654 port->updelay = cfg_get_int(0, "bonding.%s.updelay", port->name);
2655 if (port->updelay < 0) {
2658 port->downdelay = cfg_get_int(0, "bonding.%s.downdelay", port->name);
2659 if (port->downdelay < 0) {
2660 port->downdelay = 0;
2663 svec_init(&new_ifaces);
2664 svec_add(&new_ifaces, port->name);
2667 /* Get rid of deleted interfaces and add new interfaces. */
2668 for (i = 0; i < port->n_ifaces; i++) {
2669 struct iface *iface = port->ifaces[i];
2670 if (!svec_contains(&new_ifaces, iface->name)) {
2671 iface_destroy(iface);
2676 for (i = 0; i < new_ifaces.n; i++) {
2677 const char *name = new_ifaces.names[i];
2678 if (!svec_contains(&old_ifaces, name)) {
2679 iface_create(port, name);
2685 if (cfg_has("vlan.%s.tag", port->name)) {
2687 vlan = cfg_get_vlan(0, "vlan.%s.tag", port->name);
2688 if (vlan >= 0 && vlan <= 4095) {
2689 VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan);
2692 /* It's possible that bonded, VLAN-tagged ports make sense. Maybe
2693 * they even work as-is. But they have not been tested. */
2694 VLOG_WARN("port %s: VLAN tags not supported on bonded ports",
2698 if (port->vlan != vlan) {
2700 bridge_flush(port->bridge);
2703 /* Get trunked VLANs. */
2706 size_t n_trunks, n_errors;
2709 trunks = bitmap_allocate(4096);
2710 n_trunks = cfg_count("vlan.%s.trunks", port->name);
2712 for (i = 0; i < n_trunks; i++) {
2713 int trunk = cfg_get_vlan(i, "vlan.%s.trunks", port->name);
2715 bitmap_set1(trunks, trunk);
2721 VLOG_ERR("port %s: invalid values for %zu trunk VLANs",
2722 port->name, n_trunks);
2724 if (n_errors == n_trunks) {
2726 VLOG_ERR("port %s: no valid trunks, trunking all VLANs",
2729 bitmap_set_multiple(trunks, 0, 4096, 1);
2732 if (cfg_has("vlan.%s.trunks", port->name)) {
2733 VLOG_ERR("ignoring vlan.%s.trunks in favor of vlan.%s.vlan",
2734 port->name, port->name);
2738 ? port->trunks != NULL
2739 : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) {
2740 bridge_flush(port->bridge);
2742 bitmap_free(port->trunks);
2743 port->trunks = trunks;
2745 svec_destroy(&old_ifaces);
2746 svec_destroy(&new_ifaces);
2750 port_destroy(struct port *port)
2753 struct bridge *br = port->bridge;
2757 proc_net_compat_update_vlan(port->name, NULL, 0);
2759 for (i = 0; i < MAX_MIRRORS; i++) {
2760 struct mirror *m = br->mirrors[i];
2761 if (m && m->out_port == port) {
2766 while (port->n_ifaces > 0) {
2767 iface_destroy(port->ifaces[port->n_ifaces - 1]);
2770 del = br->ports[port->port_idx] = br->ports[--br->n_ports];
2771 del->port_idx = port->port_idx;
2774 bitmap_free(port->trunks);
2781 static struct port *
2782 port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2784 struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx);
2785 return iface ? iface->port : NULL;
2788 static struct port *
2789 port_lookup(const struct bridge *br, const char *name)
2793 for (i = 0; i < br->n_ports; i++) {
2794 struct port *port = br->ports[i];
2795 if (!strcmp(port->name, name)) {
2802 static struct iface *
2803 port_lookup_iface(const struct port *port, const char *name)
2807 for (j = 0; j < port->n_ifaces; j++) {
2808 struct iface *iface = port->ifaces[j];
2809 if (!strcmp(iface->name, name)) {
2817 port_update_bonding(struct port *port)
2819 if (port->n_ifaces < 2) {
2820 /* Not a bonded port. */
2821 if (port->bond_hash) {
2822 free(port->bond_hash);
2823 port->bond_hash = NULL;
2824 proc_net_compat_update_bond(port->name, NULL);
2827 if (!port->bond_hash) {
2830 port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash);
2831 for (i = 0; i <= BOND_MASK; i++) {
2832 struct bond_entry *e = &port->bond_hash[i];
2836 port->no_ifaces_tag = tag_create_random();
2837 bond_choose_active_iface(port);
2839 port_update_bond_compat(port);
2844 port_update_bond_compat(struct port *port)
2846 struct compat_bond bond;
2849 if (port->n_ifaces < 2) {
2854 bond.updelay = port->updelay;
2855 bond.downdelay = port->downdelay;
2856 bond.n_slaves = port->n_ifaces;
2857 bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves);
2858 for (i = 0; i < port->n_ifaces; i++) {
2859 struct iface *iface = port->ifaces[i];
2860 struct compat_bond_slave *slave = &bond.slaves[i];
2861 slave->name = iface->name;
2862 slave->up = ((iface->enabled && iface->delay_expires == LLONG_MAX) ||
2863 (!iface->enabled && iface->delay_expires != LLONG_MAX));
2867 memcpy(slave->mac, iface->mac, ETH_ADDR_LEN);
2869 proc_net_compat_update_bond(port->name, &bond);
2874 port_update_vlan_compat(struct port *port)
2876 struct bridge *br = port->bridge;
2877 char *vlandev_name = NULL;
2879 if (port->vlan > 0) {
2880 /* Figure out the name that the VLAN device should actually have, if it
2881 * existed. This takes some work because the VLAN device would not
2882 * have port->name in its name; rather, it would have the trunk port's
2883 * name, and 'port' would be attached to a bridge that also had the
2884 * VLAN device one of its ports. So we need to find a trunk port that
2885 * includes port->vlan.
2887 * There might be more than one candidate. This doesn't happen on
2888 * XenServer, so if it happens we just pick the first choice in
2889 * alphabetical order instead of creating multiple VLAN devices. */
2891 for (i = 0; i < br->n_ports; i++) {
2892 struct port *p = br->ports[i];
2893 if (port_trunks_vlan(p, port->vlan)
2895 && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0))
2897 const uint8_t *ea = p->ifaces[0]->mac;
2898 if (!eth_addr_is_multicast(ea) &&
2899 !eth_addr_is_reserved(ea) &&
2900 !eth_addr_is_zero(ea)) {
2901 vlandev_name = p->name;
2906 proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan);
2909 /* Interface functions. */
2912 iface_create(struct port *port, const char *name)
2914 struct iface *iface;
2916 iface = xcalloc(1, sizeof *iface);
2918 iface->port_ifidx = port->n_ifaces;
2919 iface->name = xstrdup(name);
2920 iface->dp_ifidx = -1;
2921 iface->tag = tag_create_random();
2922 iface->delay_expires = LLONG_MAX;
2924 netdev_nodev_get_etheraddr(name, iface->mac);
2925 netdev_nodev_get_carrier(name, &iface->enabled);
2927 if (port->n_ifaces >= port->allocated_ifaces) {
2928 port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces,
2929 sizeof *port->ifaces);
2931 port->ifaces[port->n_ifaces++] = iface;
2932 if (port->n_ifaces > 1) {
2933 port->bridge->has_bonded_ports = true;
2936 VLOG_DBG("attached network device %s to port %s", iface->name, port->name);
2938 port_update_bonding(port);
2939 bridge_flush(port->bridge);
2943 iface_destroy(struct iface *iface)
2946 struct port *port = iface->port;
2947 struct bridge *br = port->bridge;
2948 bool del_active = port->active_iface == iface->port_ifidx;
2951 if (iface->dp_ifidx >= 0) {
2952 port_array_set(&br->ifaces, iface->dp_ifidx, NULL);
2955 del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces];
2956 del->port_ifidx = iface->port_ifidx;
2962 ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag);
2963 bond_choose_active_iface(port);
2964 bond_send_learning_packets(port);
2967 port_update_bonding(port);
2968 bridge_flush(port->bridge);
2972 static struct iface *
2973 iface_lookup(const struct bridge *br, const char *name)
2977 for (i = 0; i < br->n_ports; i++) {
2978 struct port *port = br->ports[i];
2979 for (j = 0; j < port->n_ifaces; j++) {
2980 struct iface *iface = port->ifaces[j];
2981 if (!strcmp(iface->name, name)) {
2989 static struct iface *
2990 iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx)
2992 return port_array_get(&br->ifaces, dp_ifidx);
2995 /* Port mirroring. */
2998 mirror_reconfigure(struct bridge *br)
3000 struct svec old_mirrors, new_mirrors;
3003 /* Collect old and new mirrors. */
3004 svec_init(&old_mirrors);
3005 svec_init(&new_mirrors);
3006 cfg_get_subsections(&new_mirrors, "mirror.%s", br->name);
3007 for (i = 0; i < MAX_MIRRORS; i++) {
3008 if (br->mirrors[i]) {
3009 svec_add(&old_mirrors, br->mirrors[i]->name);
3013 /* Get rid of deleted mirrors and add new mirrors. */
3014 svec_sort(&old_mirrors);
3015 assert(svec_is_unique(&old_mirrors));
3016 svec_sort(&new_mirrors);
3017 assert(svec_is_unique(&new_mirrors));
3018 for (i = 0; i < MAX_MIRRORS; i++) {
3019 struct mirror *m = br->mirrors[i];
3020 if (m && !svec_contains(&new_mirrors, m->name)) {
3024 for (i = 0; i < new_mirrors.n; i++) {
3025 const char *name = new_mirrors.names[i];
3026 if (!svec_contains(&old_mirrors, name)) {
3027 mirror_create(br, name);
3030 svec_destroy(&old_mirrors);
3031 svec_destroy(&new_mirrors);
3033 /* Reconfigure all mirrors. */
3034 for (i = 0; i < MAX_MIRRORS; i++) {
3035 if (br->mirrors[i]) {
3036 mirror_reconfigure_one(br->mirrors[i]);
3040 /* Update port reserved status. */
3041 for (i = 0; i < br->n_ports; i++) {
3042 br->ports[i]->is_mirror_output_port = false;
3044 for (i = 0; i < MAX_MIRRORS; i++) {
3045 struct mirror *m = br->mirrors[i];
3046 if (m && m->out_port) {
3047 m->out_port->is_mirror_output_port = true;
3053 mirror_create(struct bridge *br, const char *name)
3058 for (i = 0; ; i++) {
3059 if (i >= MAX_MIRRORS) {
3060 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
3061 "cannot create %s", br->name, MAX_MIRRORS, name);
3064 if (!br->mirrors[i]) {
3069 VLOG_INFO("created port mirror %s on bridge %s", name, br->name);
3072 br->mirrors[i] = m = xcalloc(1, sizeof *m);
3075 m->name = xstrdup(name);
3076 svec_init(&m->src_ports);
3077 svec_init(&m->dst_ports);
3085 mirror_destroy(struct mirror *m)
3088 struct bridge *br = m->bridge;
3091 for (i = 0; i < br->n_ports; i++) {
3092 br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3093 br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx);
3096 svec_destroy(&m->src_ports);
3097 svec_destroy(&m->dst_ports);
3100 m->bridge->mirrors[m->idx] = NULL;
3108 prune_ports(struct mirror *m, struct svec *ports)
3113 svec_sort_unique(ports);
3116 for (i = 0; i < ports->n; i++) {
3117 const char *name = ports->names[i];
3118 if (port_lookup(m->bridge, name)) {
3119 svec_add(&tmp, name);
3121 VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s",
3122 m->bridge->name, m->name, name);
3125 svec_swap(ports, &tmp);
3130 prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans)
3134 /* This isn't perfect: it won't combine "0" and "00", and the textual sort
3135 * order won't give us numeric sort order. But that's good enough for what
3136 * we need right now. */
3137 svec_sort_unique(vlan_strings);
3139 *vlans = xmalloc(sizeof *vlans * vlan_strings->n);
3141 for (i = 0; i < vlan_strings->n; i++) {
3142 const char *name = vlan_strings->names[i];
3144 if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) {
3145 VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s",
3146 m->bridge->name, m->name, name);
3148 (*vlans)[n_vlans++] = vlan;
3155 vlan_is_mirrored(const struct mirror *m, int vlan)
3159 for (i = 0; i < m->n_vlans; i++) {
3160 if (m->vlans[i] == vlan) {
3168 port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p)
3172 for (i = 0; i < m->n_vlans; i++) {
3173 if (port_trunks_vlan(p, m->vlans[i])) {
3181 mirror_reconfigure_one(struct mirror *m)
3183 char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name);
3184 struct svec src_ports, dst_ports, ports;
3185 struct svec vlan_strings;
3186 mirror_mask_t mirror_bit;
3187 const char *out_port_name;
3188 struct port *out_port;
3193 bool mirror_all_ports;
3195 /* Get output port. */
3196 out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port",
3197 m->bridge->name, m->name);
3198 if (out_port_name) {
3199 out_port = port_lookup(m->bridge, out_port_name);
3201 VLOG_ERR("%s.output.port: bridge %s does not have a port "
3202 "named %s", pfx, m->bridge->name, out_port_name);
3209 if (cfg_has("%s.output.vlan", pfx)) {
3210 VLOG_ERR("%s.output.port and %s.output.vlan both specified; "
3211 "ignoring %s.output.vlan", pfx, pfx, pfx);
3213 } else if (cfg_has("%s.output.vlan", pfx)) {
3215 out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx);
3217 VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, "
3218 "but exactly one is required; disabling port mirror %s",
3219 pfx, pfx, pfx, pfx);
3225 /* Get all the ports, and drop duplicates and ports that don't exist. */
3226 svec_init(&src_ports);
3227 svec_init(&dst_ports);
3229 cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx);
3230 cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx);
3231 cfg_get_all_keys(&ports, "%s.select.port", pfx);
3232 svec_append(&src_ports, &ports);
3233 svec_append(&dst_ports, &ports);
3234 svec_destroy(&ports);
3235 prune_ports(m, &src_ports);
3236 prune_ports(m, &dst_ports);
3238 /* Get all the vlans, and drop duplicate and invalid vlans. */
3239 svec_init(&vlan_strings);
3240 cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx);
3241 n_vlans = prune_vlans(m, &vlan_strings, &vlans);
3242 svec_destroy(&vlan_strings);
3244 /* Update mirror data. */
3245 if (!svec_equal(&m->src_ports, &src_ports)
3246 || !svec_equal(&m->dst_ports, &dst_ports)
3247 || m->n_vlans != n_vlans
3248 || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans)
3249 || m->out_port != out_port
3250 || m->out_vlan != out_vlan) {
3251 bridge_flush(m->bridge);
3253 svec_swap(&m->src_ports, &src_ports);
3254 svec_swap(&m->dst_ports, &dst_ports);
3257 m->n_vlans = n_vlans;
3258 m->out_port = out_port;
3259 m->out_vlan = out_vlan;
3261 /* If no selection criteria have been given, mirror for all ports. */
3262 mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans);
3265 mirror_bit = MIRROR_MASK_C(1) << m->idx;
3266 for (i = 0; i < m->bridge->n_ports; i++) {
3267 struct port *port = m->bridge->ports[i];
3269 if (mirror_all_ports
3270 || svec_contains(&m->src_ports, port->name)
3273 ? port_trunks_any_mirrored_vlan(m, port)
3274 : vlan_is_mirrored(m, port->vlan)))) {
3275 port->src_mirrors |= mirror_bit;
3277 port->src_mirrors &= ~mirror_bit;
3280 if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) {
3281 port->dst_mirrors |= mirror_bit;
3283 port->dst_mirrors &= ~mirror_bit;
3288 svec_destroy(&src_ports);
3289 svec_destroy(&dst_ports);
3293 /* Spanning tree protocol. */
3295 static void brstp_update_port_state(struct port *);
3298 brstp_send_bpdu(struct ofpbuf *pkt, int port_no, void *br_)
3300 struct bridge *br = br_;
3301 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3302 struct iface *iface = iface_from_dp_ifidx(br, port_no);
3304 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
3306 } else if (eth_addr_is_zero(iface->mac)) {
3307 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d with unknown MAC",
3310 union ofp_action action;
3311 struct eth_header *eth = pkt->l2;
3314 memcpy(eth->eth_src, iface->mac, ETH_ADDR_LEN);
3316 memset(&action, 0, sizeof action);
3317 action.type = htons(OFPAT_OUTPUT);
3318 action.output.len = htons(sizeof action);
3319 action.output.port = htons(port_no);
3321 flow_extract(pkt, ODPP_NONE, &flow);
3322 ofproto_send_packet(br->ofproto, &flow, &action, 1, pkt);
3328 brstp_reconfigure(struct bridge *br)
3332 if (!cfg_get_bool(0, "stp.%s.enabled", br->name)) {
3334 stp_destroy(br->stp);
3340 uint64_t bridge_address, bridge_id;
3341 int bridge_priority;
3343 bridge_address = cfg_get_mac(0, "stp.%s.address", br->name);
3344 if (!bridge_address) {
3346 bridge_address = (stp_get_bridge_id(br->stp)
3347 & ((UINT64_C(1) << 48) - 1));
3349 uint8_t mac[ETH_ADDR_LEN];
3350 eth_addr_random(mac);
3351 bridge_address = eth_addr_to_uint64(mac);
3355 if (cfg_is_valid(CFG_INT | CFG_REQUIRED, "stp.%s.priority",
3357 bridge_priority = cfg_get_int(0, "stp.%s.priority", br->name);
3359 bridge_priority = STP_DEFAULT_BRIDGE_PRIORITY;
3362 bridge_id = bridge_address | ((uint64_t) bridge_priority << 48);
3364 br->stp = stp_create(br->name, bridge_id, brstp_send_bpdu, br);
3365 br->stp_last_tick = time_msec();
3368 if (bridge_id != stp_get_bridge_id(br->stp)) {
3369 stp_set_bridge_id(br->stp, bridge_id);
3374 for (i = 0; i < br->n_ports; i++) {
3375 struct port *p = br->ports[i];
3377 struct stp_port *sp;
3378 int path_cost, priority;
3384 dp_ifidx = p->ifaces[0]->dp_ifidx;
3385 if (dp_ifidx < 0 || dp_ifidx >= STP_MAX_PORTS) {
3389 sp = stp_get_port(br->stp, dp_ifidx);
3390 enable = (!cfg_is_valid(CFG_BOOL | CFG_REQUIRED,
3391 "stp.%s.port.%s.enabled",
3393 || cfg_get_bool(0, "stp.%s.port.%s.enabled",
3394 br->name, p->name));
3395 if (p->is_mirror_output_port) {
3398 if (enable != (stp_port_get_state(sp) != STP_DISABLED)) {
3399 bridge_flush(br); /* Might not be necessary. */
3401 stp_port_enable(sp);
3403 stp_port_disable(sp);
3407 path_cost = cfg_get_int(0, "stp.%s.port.%s.path-cost",
3409 stp_port_set_path_cost(sp, path_cost ? path_cost : 19 /* XXX */);
3411 priority = (cfg_is_valid(CFG_INT | CFG_REQUIRED,
3412 "stp.%s.port.%s.priority",
3414 ? cfg_get_int(0, "stp.%s.port.%s.priority",
3416 : STP_DEFAULT_PORT_PRIORITY);
3417 stp_port_set_priority(sp, priority);
3420 brstp_adjust_timers(br);
3422 for (i = 0; i < br->n_ports; i++) {
3423 brstp_update_port_state(br->ports[i]);
3428 brstp_update_port_state(struct port *p)
3430 struct bridge *br = p->bridge;
3431 enum stp_state state;
3433 /* Figure out new state. */
3434 state = STP_DISABLED;
3435 if (br->stp && p->n_ifaces > 0) {
3436 int dp_ifidx = p->ifaces[0]->dp_ifidx;
3437 if (dp_ifidx >= 0 && dp_ifidx < STP_MAX_PORTS) {
3438 state = stp_port_get_state(stp_get_port(br->stp, dp_ifidx));
3443 if (p->stp_state != state) {
3444 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10);
3445 VLOG_INFO_RL(&rl, "port %s: STP state changed from %s to %s",
3446 p->name, stp_state_name(p->stp_state),
3447 stp_state_name(state));
3448 if (p->stp_state == STP_DISABLED) {
3451 ofproto_revalidate(p->bridge->ofproto, p->stp_state_tag);
3453 p->stp_state = state;
3454 p->stp_state_tag = (p->stp_state == STP_DISABLED ? 0
3455 : tag_create_random());
3460 brstp_adjust_timers(struct bridge *br)
3462 int hello_time = cfg_get_int(0, "stp.%s.hello-time", br->name);
3463 int max_age = cfg_get_int(0, "stp.%s.max-age", br->name);
3464 int forward_delay = cfg_get_int(0, "stp.%s.forward-delay", br->name);
3466 stp_set_hello_time(br->stp, hello_time ? hello_time : 2000);
3467 stp_set_max_age(br->stp, max_age ? max_age : 20000);
3468 stp_set_forward_delay(br->stp, forward_delay ? forward_delay : 15000);
3472 brstp_run(struct bridge *br)
3475 long long int now = time_msec();
3476 long long int elapsed = now - br->stp_last_tick;
3477 struct stp_port *sp;
3480 stp_tick(br->stp, MIN(INT_MAX, elapsed));
3481 br->stp_last_tick = now;
3483 while (stp_get_changed_port(br->stp, &sp)) {
3484 struct port *p = port_from_dp_ifidx(br, stp_port_no(sp));
3486 brstp_update_port_state(p);
3493 brstp_wait(struct bridge *br)
3496 poll_timer_wait(1000);